15811-9781455221950.pdf - imf elibrary

©International Monetary Fund. Not for Redistribution

STAFF PAPERS

PETER HOLE

Clwit; £di10ria/ Co111mitt

0AVID M. CHENEY

Editor and Depwy Chai

MARINA PRIMORAC Assistant Editor

F. Charles Adams William E. Alexandet Tamim Bayoumi Sharmini Coorey Peter lsard

Editorial Committee

Paul R. Masson Donald J. Mathieson Guy M. Meredith Ratna Sahay Howell H. Zee

Among the re spo nsib ilitie:. of the International Mo netar y Fund. a' \Cl fo rth in its Articles of Ag ree ment. is the ob ligat io n t o "act as a centre for the collection and

exchange of information on mo ne tary �nd financial pro blems." Slt!O. Papers make� available to a wider audience papers prepare d by the member., of the Fund .-tafT. The

view� pr ese nt ed in the papers a re those of the author� and arc not to be i nte rpreted a� nece ssar ily i ndic at ing the po sition of the Exec utive Board or of the Fund.

To facili tat e electronic storage and retrieval of bibliographic data, S1ajf Papers has adopted the !o.ubject clal-sification �cheme developed by the Jouma/ of Economic Lilerature.

Subscription: US$54.00 a volume or the approximate e quivale nt in the currcm:ie' of most countries. Four numbers constitute a volume. Single copie:. may be purch a-.ed at $18.00. Individual academic rate to full-time professor:. and student" of universitic' and co lleges: $27 a volume. Subscriptions and or der s should be J-cnt to:

International Monetary Fund Pub lica tio n Service:. 700 19th Street. N.W. Washington. D.C. 20431. U.S.A.

Telephone: (202) 623-7430 Telefax: (202) 623-720 I E-mail: [email protected] Internet: hup://www.imf.org

http://www.imf.org


INTERNATIONAL MONETARY FUND

STAFF

PAPERS

Vol. 45 No. 4 DECEMBER 1998


The Editor invites from contributors outside the Fund brief comments (not more than I ,000 words) on published articles in Staff Papers. These comments should be addressed to the Editor, who will forward them to the author of the original article for reply. Both the comments and the reply will be considered for publication.

The term "country;' as used in this publication, may not refer to a territorial entity that is a state as understood by international law and practice; the term may also cover some territorial entities that are not states but for which statistical data are maintained and provided internationally on a separate and independent basis.

© 1998 by the Intemational Monetary Fund Intemational Standard Serial Number: ISSN 0020-8027

The U.S. Librcu:1· of Congre.1·s has ca/CIIoged this serial puf7/icarion t1.1 .fol/oH·s:

International Monetary Fund Staff papers- International Monetary Fund. v. 1- Feb. I 950-

1 Washington] International Monetary Fund.

v. �able>. diagr�. 23 �rn.

TI1rec no. a year, 1950-1977: four· no. ;1 ye;or. 1978-Indexes:

Yob. 1-27, 1950-80. I v. ISSN 0020-8027 = Swff papers- lntcmalronal Monetary Fund.

HG3810.15

I. Foreign cxch<mgc-Pcriodic;ols. 2. Commerce-Periodical>.

3. Currency question-Periodical>.

332.082 53-35483


CONTENTS

Vol. 45 No. 4 December 1998

Corruption Around the World: Causes, Consequences, Scope, and Cures

VITO TANZI • 559

Count ries' Repayment Performance Vis-a-Vis the IMF: An Empirical Analysis

LYNN AYLWARD and RUPERT THORNE • 595

Virtual Deficits and the Patinkin Effect

ELlANA CARDOSO • 619

Anticipation and Surprises in Central Bank Interest Rate Policy: The Case of the Bundesbank

DANIEL HARDY • 647

Waming: Inflation May Be Harmful to Your Growth

ATISH GHOSH and STEYEN PHILLIPS • 672

IMF Working Papers • 711

Papers on Policy Analysis and Assessment • 714

Volume 45 Index· 715

Ill

This page intentionally left blank



IMF Staff Papers Vol. 45. No. 4 (December 1998)

© 1998 International Monc1ary Fund

Corruption Around the World

Causes, Consequences, Scope, and Cures

VITO TANZI*

Corruption is attracting a lot of aflention around the world. This paper surveys and discusses issues related to the causes, consequences, and scope of corruption, and possible corrective actions. It emphasizes the costs of corruption in terms of economic growth. lt also emphasizes thca the _fight against corruption may not be cheap and cannot be independent from the reform of the state. If certain reforms are not made, corruption is likely to continue to be a problem regardless of actions directly aimed at curtailing it. [JEL £62, H l . H3, K4, N4]

I. The Growth of Corruption

IN RECENT YEARS, and especially in the 1 990s, a phenomenon broadly referred to as corruption has attracted a great deal of attention. In coun

tries developed and developing. large or small. market-oriented or otherwise, governments have fallen because of accusations of corruption, prominent politicians ( including presidents of countries and prime ministers) have lost their official positions, and, in some cases. whole political classes have been replaced. For examples, see Johnston ( 1997).

Corruption is not a new phenomenon. Two thousand years ago, Kautilya, the prime minister of an Indian kingdom, had already written a book, Arthashastra. discussing it. Seven centuries ago. Dante placed bribers in the deepest parts of Hell, reflecting the medieval distaste for

* Vito Tanzi is the Director of the IMF's Fiscal Affairs Department. An earlier draft of this paper was wriuen while he was spending a short sabbatical at Collegium Budapest. Institute for Advanced Study ( Budapest). The author thanks Ham id Davoodi and Shing-Jin Wei for their comments.

559


560 VITO TANZI

corrupt behavior. Shakespeare gave corruption a prominent role in some of his plays; and the American Constitution made bribery and treason the two explicitly mentioned crimes that could justify the impeachment of a U .S. president. ' However. the degree of attention paid to corruption in recent years is unprecedented. For example. in its end-of-year editorial on December 3 1 , 1 995, The Financial Times characterized 1995 as the year of corruption. The fol lowing three years could have earned the same title. The writing of books on corruption has become a growth industry in various countries.

The degree of attention now paid to corruption leads naturally to the question of why. Why so much attention now? Is it because there is more corruption than in the past? Or is it because more attention is being paid to a phenomenon that had always existed but had been largely. though nor completely, ignored? The answer is not obvious, and there are no reliable statistics that would make possible a definitive answer.

Several arguments can be advanced that suggest that corruption is simply attracting more attention now than in the past.

First, the end of the Cold War has stopped the political hypocrisy that had made the decision makers in some industrial countries ignore the political corruption that existed in particular countries, such as Za"ire (now the Democratic Republic of the Congo). As long as the latter were in the right political camp. there was a tendency to overlook obvious cases of high-level corruption.

Second. perhaps because of lack of information, or reluctance to talk about it by those familiar with these countries, there was also a tendency not to focus on corruption in the centrally planned economies.2 It is now widely known that centrally planned economies, such as the Soviet Union, or those imitating them through highly regimented economic activities. such as Nicaragua and Tanzania, experienced a great deal of corrupt practices. However. these practices were either ignored or not widely reported at the time. Donor countries also tended to play down this problem in countries that they assisted financially, even in the face of misuse or misappropriation of foreign aid.

Third. the increase in recent years in the number of countries with democratic governments and free and active media has created an environment in which discussion of corruption is no longer a taboo. In some countries, such as Russia, the media has responded with a vengeance to

1 See Noonan, Jr .• ( 1984) for a very interesting historical overview of corruption i n different societies.

z However, much information was available on the corruption in centrally planned economies. See. for example, Si m is ( 1982), Galasi and Gertesi ( 1987). Gross man ( 1982). and Rem nick ( I 994 ) .


CORRUPTION AROUND THE WORLD 561

this newly acquired freedom.3 In some other countries. political changes have increased the reporting of cases of corruption. See Davigo ( 1998).

Fourth, in all its ramifications, globalization has brought individuals from countries with little corruption into frequent contact with those from countries where corruption is endemic. These contacts have increased the international attention paid to corruption. especially when some companies believed that they were cut out of some contracts because the winning company had paid a bribe.

Fifth, a growing role has been played by nongovernmental organizations, such as Transparency International, in publicizing the problems of corruption and in trying to create anticorruption movements in many countries. Recently the international financial institutions, such as the IMF and the World Bank, and other international organizations have been playing a growing role in the anticorruption movement. In addition, empirical studies of corruption have contributed to a greater awareness of the economic costs of this problem.

Sixth, the greater reliance on the market for economic decisions and the increased need to be competitive have created an environment in which the pursuit of efficiency has acquired greater importance and distortions attributed to corruption attract more attention.

Finally. the role played by the United States, especially through its influence in some international institutions, has been important. American policymakers have argued that American exporters have lost out in foreign deals because they have not been allowed by law to pay bribes to foreign officials. For American companies, the payment of bribes to foreign officials is a criminal act, and, of course, the bribes paid cannot be deducted as costs for tax purposes.� This has not been the case in other OECD countries, although recently, under the sponsorship of the OECD, the situation has started to change. In several other countries. bribing a foreign official was not illegal and bribes paid could be considered a deductible business cost.

A case can also be made that the increased attention now paid to conllption reflects the growth of that phenomenon in recent decades, a growth that culminated in a peak in corruption activities in the 1990s. Let me briefly consider a few arguments that support this alternative hypothesis .

.1 An attempt by the author of this paper to create a corruption index on the basis of newspaper stories reported by the Internet found that for some countries these Internet entries amount to tens of thousands.

4 See, for example, the remarks by (lhen) Secretary of Commerce, Michael Kantor, to the Detroit Economic Club (July 25, 1996) in which he stated thal since 1994 American companies had lost international contracts worth $45 bi llion because of bribes paid by foreign contractors to the officials of foreign counLries. See also Hines. Jr. (1995).


562 vrro TANZI

Recent studies have shown the extent to which the role of the government in the economy has grown in recent decades.5 The environment that prevailed in these years brought about (I) a large increase in the level o f taxation in many countries; (2) a large increase in the level of public spending; and (3) probably, though not statistically ascertainable, a large increase in regulations and controls on economic activities on the part of governments. In recent decades, in a significant number o f countries, many economic operations or activities have required various kinds of permits or authorizations on the part of, often, several public offices. This gave the bureaucrats charged with giving the authorizations the opportunity to request bribes or to accept offered bribes.

l would hypothesize that the impact that high taxes, a high level of spending, and new regulations have o n acts of corruption is not immediate but. rather, a function of time, given the established norms of behavior.6 In a country with traditionally well-functioning and honest bureaucracy, the short-term impact of a larger government role on public officials will be limited. For some time, public officials will not be asked to perform corrupt acts and will reject bribery attempts, and they will not initiate such acts. In countries without such tradition, the more invasive role of government, played through higher taxes, higher public spending. and, especially, more widespread regulations, would have a more immediate impact on the behavior of civil servants and on corruption. This will be particularly true if fiscal policy suffers from lack of transparency in policymaking, fiscal reporting, and the assignment of responsibilities to public institutions. See Kopits and Craig ( I 998) and Tanzi ( 1 998).

However, with the passing of time, and with increasing frequency, some government officials would be approached by bribers and asked to bend rules or even to break Jaws to obtain a government benefit or to avoid a government-imposed cost. Some will respond and will get compensation from the bribers for their actions. Others will start emulating them. The process is likely to be cumulative over time and resemble the spreading of a contagious disease. Acts of corruption that might have appeared shocking earlier will begin to look less shocking, and may even begin to be tolerated. The government may respond to this situation not by punishing the officials who bend or break the rules, but by reducing wages on the assumption that officials are getting extra compensation.7 It is easy to see where this process could lead to if not checked.

5 See, for example, Tanzi and Schuknccht ( 1 997). 6These norms of behavior may be different between countries and are likely to

change only slowly over time. 7 This has actually happened in some countries vis-a-vis public employees work

ing in particular areas such as customs administration.



Two other factors may have had an impact on corruption in recent years: the growth of international trade and business and the economic changes that have taken place in many countries and especially in the economies in transition.

The growth of international trade and business has created many situations in which the payment of bribes (often euphemistically called "commissions") may be highly beneficial to the companies that pay them by giving them access to profitable contracts over competitors. Large bribes have been reported to have been paid to get foreign contracts or to get privileged access to markets or to particular benefits such as tax incentives. Le Monde of March 17, 1995, reported that the bribes paid abroad by French companies in 1 994 had been estimated at FF I O billion in a confidential government report. World Business of March 4, 1 996, reported that the bribes paid abroad by German companies had been estimated to exceed US$3 billion a year.s These were not the only countries in which companies had paid bribes to foreign officials. Some experts have estimated that as much as 1 5 percent o f the total money spent for weapons acquisition may be ··commissions" that fill somebody's pockets. Here, again, contagion is important. When the economic operators of some countries begin to pay bribes, they put pressure on those from other countries to do the same. The cost of not doing so is lost contracts. as Kantor argued.

Among the economic changes that have taken place in recent years. privatization has been most closely linked with corruption. There is no question that public or state enterprises have been a major source of corruption and especially of political corruption because they have occasionally been used to finance the activities of political parties and to provide jobs to the clienteles of particular political groups. This was clearly the case in Italy, before tangentopoli,9 and in many Latin American countries. Privatization of nonnatural monopolies is necessary to reduce this form of corruption because it eliminates an instrument often used especially in political corruption. Unfortunately. the process of privatizing public or state enterprises has itself created situations whereby some individuals (ministers, high political officials) have the discretion to make the basic decisions while others (managers and other insiders) have information not available to outsiders so that they can use privatization to benefit themselves. These problems have been observed and reported in all regions of the world, but the abuses appear to have been particularly significant in the transition economies.'0 [n the

8 Reported in Galtung ( 1997). 9 See Nordio ( 1 997). Carlo Nordio was one of the leading judges in the Italian

fight against political corruption. "Tangentopoli" or '·bribe city" is the term that was given to the Italian corruption scandal that shook Italy.

10 See Kaufmann and Siegelbaum ( 1996) and Goldman ( 1997). For a review of the Latin American experience, see Manzetti and Blake ( 1997).


564 VITOTANZI

latter, terms such as asset stripping and nomenklatura privatization have been used to describe the abuses associated with the transfer of state enterprises to private ownership. In these countries some individuals have become enormously rich because of these abuses. Two examples from the Russian experience come to mind. In the privatization of large monopolies, such as Gazprom, many close to the corridors of power received highly valued shares at very low prices. And the "loans-for-share" scheme made some banks shareholders of enterprises by extending loans to the firms. These developments have made many Russian citizens highly skeptical about the virtues of a market economy.

Thus, several arguments lead to the conclusion that the current interest in corruption probably reflects an increase in the scope of the phenomenon over the years and not just a greater awareness of an age-old problem.

II. The Definition of Corruption

Corruption has been defined in many different ways, each lacking in some aspect. A few years ago. the question of definition absorbed a large proportion of the time spent on discussions of corruption at conferences and meetings. However, like an elephant, while it may be difficult to describe, corruption is generally not difficult to recognize when observed. In most cases, different observers would agree on whether a particular behavior connotes corruption. Unfortunately, the behavior is often difficult to observe because acts of corruption do not typically take place in broad daylight.

The most popular and simplest definition of corruption is that it is the abuse of public power for private benefit. This is the definition used by the World Bank.1 1 From this definition it should not be concluded that cornlption cannot exist within private sector activities. Especially in large private enterprises, corruption clearly does ex.ist, as for example in procurement or even in hiring. It also exists in private activities regulated by the government.12 Sometimes. the abuse of public power is not necessarily for one's private benefit but for the benefit of one's party. class, tribe. friends, family, and so on. In fact. in many countries some proceeds of corruption go to finance the activities of the political parties.

1 1 A more neutral definition is that corruption is the intentional noncompliance with arm's leng1h relationship aimed at deriving some advantage from this behavior for oneself or for related individuals. See Tanzi ( 1995a). For other definitions. see Theobald ( 1990). 12 For example. when a taxi driver charges the passenger more than the regula1ed price or when a doctor in a hospital charges for services not rendered.



Not all acts of corruption result in the payment of bribes. For example, a public employee who claims to be sick but goes on vacation is abusing his public position for personal use. Thus. he is engaging in an act of conuption even though no bribe is paid. The president of a country who has an airport built in his small hometown is also engaging in an act of corruption that does not involve the payment of a bribe.13

It is important to distinguish bribes from gifts. In many instances, bribes can be disguised as gifts. A bribe implies reciprocity while a gift should not. 1� However, even though the distinction is fundamemal, it is at times difficult to make. 15 At what point does a gift become a bribe? Does the distinction depend on the size of the gift? What about cultural differences that can explain different sizes of gifts? What if a large gift is given not to the person who provides the favor but to a relative of that person? Does the distinction depend on whether the gift is given in broad daylight, for everyone to see, or privately? Clearly, the identification of a bribe is not always simple.

Acts of corruption can be classified in different categories. Corruption can be

• bureaucratic (or "petty") or political (or "grand"): for example, cor-ruption by the bureaucracy or by the political leadership;

• cost-reducing (to the briber) or benefit enhancing; • briber-initiated or bribee-initiated; • coercive or collusive: • centralized or decentralized: • predictable or arbitrary: and • involving cash payments or not.

Undoubtedly, other classifications could be added to this list.

IJI. Factors Contributing Directly to Corruption

Corruption is generally connected with the activities of the state and especially with the monopoly and discretionary power of the state. Therefore. as Gary Seeker, Nobel Laureate in economics. pointed out in

1� It becomes difficult to draw a distinction between some forms of rent seeking and corruption . See Krueger ( 1974).

14 ln practice, those who give gifts may expect some form of payment for themfor example, we expect love or good behavior from our children when we give them gifts-but the recipienLs of the gifts do not have an obligation to reciprocate.

15 For an elaboration of some of these points. see Tanzi ( 1 995a).


566 VITOTANZI

one of his Business Week columns, if we abolish the state, we abolish corruption. But, of course. quite apart from the fact that corruption can exist in the private sector, a civilized society cannot function without a state, and in modern, advanced societies, the state must have many functions. The Beck er argument seems to collide with the reality that some of the least corrupt countries in the world, such as Canada, Denmark, Finland, the Netherlands, and Sweden, have some of the largest public sectors, measured as shares of tax revenue or public spending in gross domestic product. Thus. the solution to the problem of corruption may not be as simple as just reducing the level of taxation or public spending. Rather, the way the state operates and carries out its functions is far more important than the size of public sector activity measured in the traditional way.16 Particular aspects of governmental activities create a fertile ground for corruption. Let us look at this issue in more detail.

Regulations and Authorizations

In many countries, and especially in developing countries, the role of the state is often carried out through the use of numerous rules or regulations. In these countries, licenses, permits, and authorizations of various sorts are required to engage in many activities. Opening a shop and keeping it open, borrowing money, investing, driving a car, owning a car, building a house, engaging in foreign trade, obtaining foreign exchange, getting a passport, going abroad, and so on require specific documents or authorizations. Often several government offices must be contacted to authorize the activity.

The existence of these regulations and authorizations gives a kind of monopoly power to the officials who must authorize or inspect the activities. These officials may refuse the authorizations or may simply sit on a decision for months or even years. Thus, they can use their public power to extract bribes from those who need the authorizations or permits. In India, for example, the expression "licence raj" referred to the individual who sold the permits needed to engage in many forms of economic activities. In some countries, individuals become middlemen or facilitators for obtaining these permits. The fact that in some cases the regulations are nontransparent or are not even publicly available and that an authorization can be obtained only from a specific office or individual-that is, there is no competition in the granting of these authorizations-gives the

16The state can exercise its role through various instruments. Some of these lend themselves more easily to acts of corruption. See Tanzi ( 1995a). For an empirical analysis that links market strucLUre and rents to the level of corruption, see Ades and Di Tella (forthcoming).


CORRUPTION AROUND 11-!E WORLD 567

bureaucrats a great amount of power and a good opportunity to extract bribes. 17

The existence of these regulations generates the need for frequent contacts between citizens and bureaucrats. It also requires an enormous amount of time to be spent by the citizens in acquiring these permits and in dealing with public officials. Surveys from different countries and especially from developing and transition countries indicate that much of the time of the managers of enterprises, and especially of small enterprises, is spent dealing with public bureaucracies. This time that is taken away from managing the enterprises can be reduced through the payment of bribes.

Taxation

Taxes based on clear laws and not requiring contacts between taxpayers and tax inspectors are much less likely to lead to acts of corruption. However, when the following situations arise, corruption is likely to be a major problem in tax and customs administrations (see Tanzi. 1998):

• the laws are difficult to understand and can be interpreted differently so that taxpayers need assistance in complying with them;

• the payment of taxes requires frequent contacts between taxpayers and tax administrators;

• the wages of the tax administrators are low; • acts of corruption on the part of the tax administrators are ignored, not

easily discovered, or when discovered penalized only mildly; • the admjnistrative procedures (e.g .. the criteria for the selection of tax

payers for audits) lack transparency and are not closely monitored within the tax or customs administrations:

• tax administrators have discretion over important decisions, such as those related to the provision of tax incentives, determination of tax liabilities, selection of audits, litigations, and so on; and

• more broadly, the controls from the state (the principal) on the agents charged with carrying out its functions are weak. 18

17 Some economists have argued that this kind of corruption can be eliminated by selling up several offices all authorized to provide the authorizations or permits. This would remove the monopoly power from the bureaucrats. See ShJeifer and Vishny ( 1993). Unfortunately, the setting up of several offices may be costly. In some cases, particular activities (say, yearly inspections of cars) can be privatized.

'8 In cases of political corruption, those who represent the state (president, prime minister. ministers) or their close relatives may use the tax and customs administrations to pursue rent seeking and corrupt practices.


568 VJTO TANZI

In some countries. at one time, corruption became so endemic in the tax administration (e.g., Peru and Uganda) that the government decided to close down the existing administrations and to replace them by new ones. In several countries. customs administrations have been very corrupt, leading to the jailing of the director of customs and in several cases resulting in the replacement of the domestic customs organizations with the services of foreign companies engaged in preshipment inspections.

Reports from several countries indicate that the number of applicants for poorly paid jobs in administering taxes or in customs has been unusually large, pointing to the possibility that applicants know these jobs create opportunities for extra incomes.19

Spending Decisions

Corruption can also affect public expenditure. Corruption related to the provision by the government of goods at below-market prices is discussed below, but I will now discuss other aspects of public expenditure.

Investment projects have lent themselves to frequent acts of high-level corruption. Because of the discretion that some high-level public officials have over decisions regarding public investment projects, this type of public spending can become much distorted, both in size and in composition, by corruption.20 Public projects have, at times, been carried out specifically to provide opportunities to some individuals or political groups to get "commissions'' from those who are chosen to execute the projects. This has reduced the productivity of such expenditure and has resulted in projects that would not have been justified on objective criteria of investment selection such as cost-benefit analysis.

Procurement spending, that is, the purchase of goods and services on the part of the government, is another area affected by corruption. To reduce corruption possibilities, some countries have developed complex and costly procedures that may have reduced corruption at the cost of sharply increasing the prices at which some goods are purchased.21

Extrabudgetary accounts are common in many countries. Some of them have legitimacy and are set up for specific purposes (pension funds, road funds, etc.). Others are set up to reduce the political and administrative controls that are more likely to accompany spending that goes through the budget. In some countries, the money received from foreign

19There have even been reports that in some coumries these jobs can be bought. 10 See Tanzi and Davoodi ( 1997). 21 The notorious $600 hammers bought by the Pentagon could be explained by such procedures.


CORRUPTION AROUND TilE WORLD 569

aid or from the sale of natural resources such as oil and tin is channeled toward special accounts that tend to be less transparent and less controlled than the money channeled through the budget. Some of this money may go into illegitimate uses or pockets.22

In all these areas. Jack of transparency and of effective institutional controls are the main factors leading to corruption.

Provision for Goods and Services at Below-Market Prices

In most countries, the government engages in the provision of goods, services, and resources at below-market prices-for example, foreign exchange, credit, electricity, water. public housing, some rationed goods. access to educational and health facilities, access to public land, and so on. Even access to some forms of pensions, such as those for disability, fall into this category because the individuals who get them have paid less in contributions to the pension funds over time than the pension they get once their disability status is approved. In some countries, disability pensions have been a fertile ground for corruption. In others, some individuals benefited enormously when they were able to get access to large amounts of credit or foreign exchange at below-market prices.

Sometimes, because of limited supply, rationing or queuing becomes unavoidable. Excess demand is created and decisions have to be made to apportion the limited supply. These decisions are often made by public employees. Those who want these goods (the users) would be willing to pay a bribe to get access (or a higher access) to what the government is providing. It is thus not surprising that in all the areas mentioned above, cases of corruption have been reported.

Other Discretionary Decisions

Besides the areas mentioned above, in many countries public officials can find themselves in positions where they have discretion over important decisions; in these situations. corruption, including high-level or political corruption, can play a major role. The most important of these discretionary decisions are as follows:

• provision of tax incentives against income taxes, value-added taxes, and foreign trade taxes, which may be worth millions of dollars in terms of

22 Because of the variation of the price of commodities even within a day. it may be difficult to ascertain at which price a transaction takes place. Some of the difference between the actual price and the declared price may be channeled into foreign accounts.


570 VITO TANZI

the present value of reduced future liabilities to those who benefit from them;23

• decisions as to the particular use of private land (zoning Jaws) that determine whether a piece of land can be used only for agriculture, and thus have low market value, or for high-rise buildings, and thus be very expensive;

• decisions as to the use of government-owned land, for example, for logging purposes, which may also be worth a lot to the recipients. Major cases of corruption related to permissions to cut trees from publicly owned forests, or to exploit public lands for their mineral wealth, have been reported in several countries;

• decisions that authorize major foreign investments, often undertaken in connection with domestic interests, which often provide the privileged investors with monopoly power;

• decisions related to the sale of public sector assets, including the right to extract natural resources;

• decisions on the privatization of state-owned enterprises and on the conditions attached to that process, such as the degree of regulation of the industry; and

• decisions providing monopoly power to particular export, import. or domestic activities.

Decisions such as those described above are often worth a lot to individuals or enterprises. It is natural that attempts will be made by some of them to get favorable decisions, in some cases by paying bribes and in other cases by simply exploiting close personal relations with public officials. The bribes may be paid to public officials whose salaries may be very low and whose "temptation price" may be far less than the value of the potential benefit from a favorable decision to the bribers.

Financing of Parties

Some time before the tangentopoli scandal exploded in Italy, Minister Martell i , an important member of the socialist party, candidly admitted in a speech that the Italian political parties had on their payrolls a small army of employees. The salaries for these employees had to be paid. He implied that the needed money had to come from somewhere. Minister Martelli had put his finger on a major problem for democracies-the need to finance the activities, including the electoral campaigns, of the

23ln some countries these incentives have been provided outside the normal legal process, by high-level public officials, to favored individuals.



political parties.24 When public money is not available for the activities of the political parties. enormous pressures will build up to generate funds. The recent controversy concerning political donations in the United Stares is an example of this problem. As Susan Rose-Ackerman ( 1997, p. 40) has put it: "Democracy gives citizens a role in choosing their political leaders. Thus corrupt elected officials can be voted out of office. But democracy is not necessarily a cure for corruption."

IV. Indirect Causes of Corruption

Besides rhe factors that promote conuption directly, discussed in the previous section, other factors can contribute to corruption indirectly. Some of these are discussed briefly in Lhis section.

Quality of the B u reaucracy

The quality of the bureaucracy varies greatly among countries. In some, public sector jobs give a lot of prestige and status; in others, much less so. Many factors contribute to that quality. Many years ago Max Weber ( 1947), the outstanding German sociologist, described what should be the characteristics of an ideal bureaucracy. He was aware that most bureaucracies are not ideal. Tradition and the effect that it has on the pride that individuals have in working for the government may explain why, all things being equal, some bureaucracies are much more efficient and much less vulnerable to corruption than others.25 Rauch and Evans ( 1997) have gathered information on the degree to which civil servants' recruitment and promotions are merit based for 35 developing countries. Their results indicate that the less recruitment and promotion are based on merit, the higher is the extent of corruption.

Absence of politically motivated hiring, patronage, and nepotism. and clear rules on promotions and hiring, i n addition to some of the factors discussed separately below, all contribute to the quality of a bureaucracy.

24 One of the leading judges of mani pulite (clean hands), the investigation in the Italian corruption scandal, has recently described the arrangements among the parties to share the proceeds from corruption. See Nordio ( 1 997). On the issue of political corruption, see also Cazzola ( 1988); .lohnstOn ( 1997); and Ferrero and Brosio ( 1997).

25 See von Klimo ( 1 997). V on Klimo compares the public conception of an inefficient and corrupt public admjnistration in 1 9th century Italy with the ·'myth of absolute efficiency and incorruptibility" enjoyed by the administration of the Prussian slate.


572 VITO TANZI

The incentive structure plus tradition go a long way toward explaining why some bureaucracies are much less corTupt than others.26

Level of Public Sector Wages

Over the years many observers have speculated that the wages paid to civil servants are important in determining the degree of corruption. For example, Assar Lindbeck ( 1 998) attributes the low corruption in Sweden in this century partly to the fact that at the turn of the century, high-level administrators earned 12-15 times the salary of an average industrial worker. One can speculate that there may be corruption due to greed and corruption due to need. In Figure I. CC' represents the trade-off between the level of corruption and the level of wages. The higher the wage level, the lower is corruption. Assume that OR represents a level of wage consistent with the minimum required by the family of a public employee for a decent living. It can be assumed that OA is corruption due to greed, while corruption beyond OA would be due to need. Figure I also implies that, regardless of the wage level, some public officials will be corrupt perhaps because of their own psychological or moral makeup, or because some of the bribes offered may be too large for some officials to resist. Thus, i t implies, realistically, that not all officials respond in the same way to the same incentives. ln theoretical jargon. agents are heterogeneous.

The relationship between wage level and corruption index has been tested empirically by Van Rijckeghem and Weder ( 1997). See also Hague and Sahay ( 1996). With the use of cross-sectional data, they have been able to support the common intuition by finding a statistically significant relationship between corruption and wage levels, similar to that shown by the CC' curve in Figure I. They have speculated that while an increase in the wage level is likely to reduce corruption, a very large increase would be necessary to reduce i t to minimal levels. In other words, the fight against corruption, pursued exclusively on the basis of wage increases, can be very costly to the budget of a country and can achieve only part of the objective. Furthermore, as argued above. even at high wages some individuals may continue to engage in corrupt practices.

In recent years, several countries (Argentina, Peru. etc.) have attempted to reduce corruption in particularly sensitive areas, such as customs and tax administrations, by increasing the level of salaries for the public

261n some countries, public sector hiring has had the reduction of unemployment as its main objective. rather than improving the quality of the public administration.



Figure I . Wage LHel Corruptio11 Trade-Off

Wage level

c

R

· - -- - - - - - - - - c·

0 A Corruption

employees in these areasY These countries have also increased salary differentials to be able to retain and attract more able, productive. and honest individuals. Over the years, Singapore has pursued a wage policy aimed at reducing the temptation for public officials to engage in acts of corruption. Reportedly, the salaries of ministers and other high-level officials in Singapore are among the highest in the world.�8

There has been some speculation in the theoretical economic literature that high wages may reduce the number of corrupt acts, while they may lead to demands for higher bribes on the part of those who continue to be corTupt. The reason is that high wages raise the opportunity cost of losing one's job, while they do not eliminate the greed on the part of some officials. Thus. while the number of corrupt acts is reduced, the total amount of corruption money paid may not necessarily fall.

171n Peru, the wage structure in the tax administration became similar to that of the central bank and thus somewhat higher than the wage structure of the rest of the civil service. Also. an incentive system was introduced that assigned to the tax administration (SUN AT) a share of the tax revenue. The average age of the employees of the tax administration fell dramatically.

1s A common belief is that in situations of low wages but high possibilities of corruption. less honest individuals will be attracted to the civil service.


574 VITO TANZl

Penalty Systems

Following Gary Becker's ( 1 968) classic analysis of crime prevention, given the probability that the perpetrator of a crime would be caught, the penalty imposed plays an important role in determining the probability that criminal or illegal acts would take place.29 In theory, all things being equal, corruption could be reduced by increasing the penalties on those who get caught. This analysis implies d1at the penalty structure existing in a country is an impo11ant factor in determining the extent of corruption in that country. But once again, at least theoretically, higher penalties may reduce the number of acts of corruption, but they may lead to demands for higher bribes on the corrupt acts that still take place.

In the real world. relatively few people are punished for acts of corruption. in spite of the extent of the phenomenon. Furthermore, with the exception of a few countries, there seems to be a wide gap between the penalties specified in the laws and regulations and the penalties that are effectively imposed.30 Generally, effective penalties tend to be more lenient than the statutory ones. The administrative procedures followed before a public employee is punished for acts of corruption are slow and cumbersome. Often legal, political, or administrative impediment prevent the full or quick application of the penalties. Due process and the need to provide incontrovertible evidence are major hurdles. The potential accusers are often reluctant to come forward and to spend the time and effort to go through the full process required to punish someone. Also, when corruption is widespread, the costs to the accusers in terms of social capital, such as lost friends, can be high.31 Furthermore, the judges who will impose the penalties may themselves be accessible to corruption or may have political biases, so that they may be bought by the accused or may put obstacles in the way of the proceedings. All these factors limit the role that penalties actually play in many countries, especially when corruption is partly politically motivated. This attitude brings a toleration for small acts of corruption that can in time encourage bigger acts.32

29 For an econometric application of Seeker's theory to the Netherlands, see van Tulden and van der Torre ( 1997).

3°China has recently gone as far as applying the death penalty to some individuals accused of corruption. However, many acts of corruption still go unpunished, so that uncertainty prevails on the treatment of individuals accused of corruption. This may lead to the perception that penalties are applied selectively or arbitrarily.

11 Even in coumries with relatively little corruption. so-called '"whistle blowers·· do not seem to have an easy time.

·12 Reluctance to apply harsh penalties may also be due to concerns that the penalties might be applied selectively, to political opponents.



Institutional Controls

The other important ingredient in Gary Becker"s analysi� is the probability that those who commit crimes would be caught. Thb leads to the role of institutional controls. The existence of these controb reflects to a large extent the attitude of the political body toward this problem. Generally, the most effective controls should be those that exist inside inslitlllions. Thi� i� really the first line of defense. Honest and effective supervisors. good auditing offices, and clear rules on ethical behavior should be able to discourage or discover con·upt activities. Good and transparent procedures should make i t easier for these offices to exercise their controls. Supervisors should be able to monitor the activities of their subordinates and they should themselves be held accountable for acts of corruption in their offices that go unpunished. These characteristics vary from country to country. In some. these checks are almost nonexistent. so that corruption is mostly discovered by chance or through the reporting by outsiders. including the media. In this connection, the role of a free press in controlling corruption cannot be exaggerated. See Brunetti and Weder ( 1 998).

Several countries. including Singapore, Hong Kong. Uganda. Argentina, and others. have created anticorruption commissions or ethics offices expressly charged with the responsibility of following reports on corruption or reducing corruption through the requirement on the part of public officials to report their wealth and in other waysY To be effective. these offices must have independence from the political establishment. ample resources. and personnel of the highest integrity. They must also have the power to enforce penalties or, at least. have others. including the judiciary. enforce the penalties. Unfortunately, in some countries these offices are required to report confidentially to the president or the prime minister of the country rather than, say. openly to the legislative body. This reduces their effectiveness and politicizes the process. In other countries. these commission� do not have the power to impose pena[ties and their reports may not have any following by other institutions.

Transparency of Rules. Laws. and Processes

In many countries, the lack of transparency in rules, laws. and processes creates a fertile ground for corruption. Rules are often confusing, the documents specifying them are not publicly available. and. at time�. the rules are changed without properly publicized announcements. Laws or regulations are

u For the experience of Uganda in the fight against corruption. sec Ruzindana ( 1 997) and Langseth and Stapenhurst ( 1997).


576 VITO TANZI

written in a way that only trained lawyers can understand and they are often conceptually opaque about important aspects, thus leaving grounds for different interpretations. Processes or procedures on policy matters and other actions, for example, competitions for public projects. are equally opaque, so that at times it is difficult to understand or to determine the process that was followed before a decision was reached. This makes it difficult to detennine whether corruption has played a role in some important decisions.

Some countries-for example, New Zeal and-have made great efforts in recent years to bring more transparency to all the accounts and actions of the government. The IMF has recently issued a code on fiscal transparency for its member countries that, if followed, would have the effect of reducing corruption.

Examples by the Leadership

A final contributing factor is the example provided by the leadership. When the top political leaders do not provide the right example, either because they engage in acts of corruption or, as is more often the case. because they condone such acts on the part of relatives, friends, or political associates, it cannot be expected that the employees in the public administration will behave differently. The same argument applies within particular institutions such as tax administration. customs. and public enterprises. These institutions cannot be expected to be corruption free i f their heads do not provide the best examples of honesty.

In some countries, the leadership has been somewhat indifferent to this problem. In an African country, a president refused to fire ministers widely reputed to be corrupt. In an Asian country, a minister that was accused of comtption was simply moved to head another ministry. In a Latin American country. a president who was planning to create an anticorruption commission proposed to appoint as head of this commission an individual widely reported to be corrupt. Examples such as these do not help create the climate for a corruption-free society.

V. Measurement of Corruption

l f corruption could be measured, it could probably be eliminated. In facr, conceptually it is not even clear whar one would want to measureY Simply measuring bribes paid would ignore many COITupt acts that are not accompanied by the payment of bribes. An attempt to measure acts of corruption

"'One could measure acts of corruption on bribes paid.


CORRUP1l0N AROUND THE WORLD 577

rather than the amounts of bribes paid would require counting many relatively unimportant actions and identifying each act-information that is simply not available. While there are no direcl ways of measuring corruption, there are several indirect ways of getting information about its prevalence in a country or in an institution. Useful information can be obtained from:

• reports on corruption available from published sources including newspapers. The lntemet has become a most valuable source.35 Newspapers such as Le Mo11de, The Fi11ancia/ Times, and The New York Times and magazines such as The Economisl and The Far Easlern Economic Review have published many articles on corruption.

• case studies of corrupt agencies such as tax administrations, customs, and police. Unfortunately. while there are many such studies. often the reports are intemal and are kept confidential.

• questionnaire-based surveys. These can be related to a specific agency (for example, Peru's or Argentina· s tax adminiso·ation); or to a whole country. These surveys measure perceplions of corruption rather than corruption per se. The World Bank has been making use of these surveys in its work in Tanzania. Uganda, lrndia. Ukraine. and other places. It has used these surveys to improve the effectiveness of particular programs such as health care.

Countrywide surveys are available from the following organizations: Global Competitiveness Report (Geneva): Political and Economic Risk Consultancy (Hong Kong); Transparency International ( Berlin): and Political Risk Services (Syracuse). The Gallup poll has also conducted a major survey for 44 countries dealing with corruption in particular activities, and the World Bank has conducted a survey for many countries.

The results obtained from these surveys are now widely used by researchers and business people. The best-known of these surveys. the Transparency International index. for example, assesses the perception of corruption on a scale ofO to 10. Ten refers to a con·uption-free country; zero refers to a country where most transactions or relations are tainted by corruption. The variance of these indexes. which reflects how the views are spread among respondents, is also important and has been used by some researchers in their work.36 People may tend to confuse these indexes with actual measurements of corruption. It is important to keep in mind that the

•1� For some countries, the Internet reports tens of thousands of entries on the subject of corruption.

36 See. for example, Wei ( 1997a and 1 997b).


578 VITO TANZI

indexes reflect perceptions and not objectil•e and quantitatil'e measures of actual corruption. One good feature is that the various indexes available are highly con-elated among themselves.

Table I shows the indexes for I 995 to I 998 reported by Transparency International. Comparing the 1995 figures with those reported in 1998 for the countries covered in both years indicates that countries tend to hold their positions although some important changes in particular countries are shown. See Figure 2. How closely changes in these indexes reflect real changes within given countries is an important, open question. A single but widely reported case of corruption may easily change perceptions in a given country and in a given year, and lead to an index that may not correclly assess the extent of corruption in that country.

VI. Economic Effects of Corruption

Review of Theoretical Arguments

The recent fairly broad consensus seems to be that corruption is unqualifiably bad. However. in past years, the views on corruption had been more divergent and some economists had even found some redeeming value in it.J7 Until the I 997 financial crisis, some countries from Southeast Asia seemed to provide support for the view that corruption might promote growth. Indonesia. Thailand. and some other countries were often mentioned as countries growing fast in spite of. or even because of, perceived high levels of corruption. This corruption was associated with a low degree of uncertainty. Js For Indonesia, it was argued that institutionalizing corruption made i t less damaging to economic development than random corruption. One knew where to go and how much to pay for specific services. Some of the arguments in favor of the view that corruption may promote efficiency and even growth are summarized below.39 This survey is not intended to be exhaustive but just to provide a feel for the relevant literature.

Leff ( 1 964) and Huntington ( 1 968) advanced the view that corruption can be efficiency enhancing because it removes government-imposed rigidities that impede investment and imerfere with other economic decisions favorable to growth. Thus, corruption "oils the mechanism" or

37 Even today a few economists argue that. within well confined circumstances, corruption may promote faster growth. See, for example, Braguinsky ( 1996). 18 Since the fal l of 1 997, some accounts have blamed corruption for 1he currency crises caused by unproductive investments and high short-term borrowing.

39 For a review. see Bardhan ( 1997) and Susan Rose-Ackerman ( 1997 ).



Table I. Cormption Percep1ion lndex. /995-98

Country 1995 1996 1 997 1998 Rank in 1998 Rank in 1996

Denmark 9.32 9.33 9.9-1 10.00 I 2 Finland 9. 1 2 9.05 9.48 9.60 2 4 Sweden 8.87 9.08 9.35 9.50 3 � New Zealand 9.55 9.43 9.23 9.40 4 I Iceland 9.30 5

Canada 8.87 8.96 9.10 9.20 6 5 Singapore 9.26 8.80 8.66 9.10 7 7 Norway 8.61 8.87 8.92 9.00 8 6 Netherlands 8.69 8.71 9.03 9.00 9 9 Switzerland 8.76 8.76 8.61 8.90 10 8

United Kingdom 8.57 8.44 8.22 8.70 1 1 1 2 Luxembourg 8.61 8.70 1 2 Australia 8.80 8.60 8.86 8.70 1 3 1 0 Ireland 8.57 8.45 8.28 8.20 1 4 1 1 Germany 8.14 8.27 8.23 7.90 1 5 1 3

Hong Kong 7.12 7.01 7.28 7.80 1 6 1 8 Austria 7.13 7.59 7.61 7.50 1 7 1 6 United States 7.79 7.66 7.61 7.50 1 8 1 5 Israel . . . 7.71 7 97 7.10 19 I-I Chile 7.94 6.80 6.05 6.80 20 2 1

France 7.00 6.96 6.66 6.70 2 1 19 Portugal 5.56 6.53 6.97 6.50 22 22 Spain 4.35 4.31 5.90 6. 10 23 32 Botswana 6.10 24 Japan 6.72 7.05 6.57 5.80 25 1 7

Estonia . . . 5.70 26 Costa Rica 6.45 5.60 27 Belgium 6.85 6.84 5.25 5.40 28 20 Malaysia 5.28 5.32 5.01 5.30 29 26 Taiwan

Province of China 5.08 4.98 5.02 5.30 30 29

Namibia 5.30 3 1 South Africa 5.62 5.68 4.95 5.20 32 23 Hungary 4 . 1 2 4.86 5 . 18 5.00 33 3 1 Mauritius 5.00 34 Tunisia 5.00 35

Greece 4.04 5.01 5.35 4.90 36 28 Czech Republic 5.37 5.20 4.80 37 25 Jordan 4.89 ... 4.70 38 30 Italy 2.99 3.42 5.03 4.60 39 34 Poland 5.57 5.08 4.60 40 24

Peru 4.50 4 1 Uruguay . . . 4. 1 4 4.30 42 South Korea 4.29 5.02 4.29 4.20 43 27 Zimbabwe 4.20 44 Malawi 4.10 45


580 VITO TANZI

Table I . (concluded)

Country 1 995 1996 1997 1998 Rank in 1998 Rank in 1996

Brazil 2.70 2.96 3.56 4.00 46 40 Belarus 3.90 47 Slovak Republic 3.90 48 Jamaica 3.80 49 Morocco 3.70 50

El Salvador 3.60 5 1 China 2. 1 6 2.43 2.88 3.50 52 so Zambia ... 3.50 53 Turkey 4. 10 3.54 3.21 3.40 54 33 Mexico 3. 1 8 3.30 2.66 3.30 55 38

Philippines 2.77 2.69 3.05 3.30 56 44 Ghana 3.30 57 Senegal 3.30 58 Cote d' lvoire 3 . 1 0 59 Guatemala 3. 1 0 60

Argentina 5.24 3.41 2.81 3.00 6 1 35 Thailand 2.79 3.33 3.06 3.00 62 37 Romania 3.44 3.00 63 Nicaragua 3.00 64 Yugoslavia 3.00 65

India 2.78 2.63 2.75 2.90 66 46 Egypt 2.84 2.90 67 4 1 Bulgaria 2.90 68 Bolivia 3.40 2.05 2.80 69 36 Ukraine 2.80 70

Pakistan 2.25 1 .00 2.53 2.70 7 1 53 Latvia 2.70 72 Uganda 2.71 2.60 73 43 Vietnam 2.79 2.50 74 Kenya 2.21 2.50 75 52

Russia 2.58 2.27 2.40 76 47 Venezuela 2.66 2.50 2.77 2.30 77 48 Ecuador 3. 19 2.30 78 39 Colombia 3.44 2.73 2.23 2.20 79 42 Indonesia 1 .94 2.65 2.72 2.00 80 45

Nigeri:1 0.69 1.76 1 .90 8 1 54 Tanzania 1 .90 82 Honduras 1.70 83 Paraguay 1.50 84 Cameroon 2.46 1 .40 85 49 Bangladesh 2.29 5 1

Memorandum i1ems: Number of countries 4 1 54 52 85 Average 5.93 5.35 5.67 4.89 Median 5.62 5.02 5.23 4.20 Minimum 1.94 0.69 1.76 1 .40 Maximum 9.55 9.43 9.94 10.00

Source: Transparency International. Notes: Data refer to perception of corruption ranging from 10 (highly clean} to 0

(highly corrupt). Ranking for 1998 and 1 996 are based on 1998 and 1996 data respec-Lively. The data have been rearranged by the author.


CORRUPTION AROUND THE WORLD

Figure 2. Corruption Ranking: /995 vs. 1998 (53 countries)

I 995 rank

581

"greases the wheel." This reasoning was often used to explain the high rates of growth in some countries of South East Asia.

Beck and Maher ( 1 986) and Lien ( 1 986) have developed models that show that, in bidding competitions. those who are most efficient can afford to offer the highest bribe. Therefore. bribes can promote efficiency by assigning projects to the most efficient firms.

Lui ( 1 985) has argued that time has different values for different individuals, depending on their level of income and the opportunity cost of their rime. Those for whom time is most valuable will offer bribes to public ofticials to be allowed to economize on time by jumping in front of the line, that is, by getting decisions more quickly. Thus, corruption can be efficient because it saves time for those for whom time has the greatest value. In a later paper, Lui ( 1 996) argued that while corruption may improve the allocation of resources in some circumstances. it reduces growth because it provides some individuals the incentive to acquire the kind of human capital that can be used to improve corruption opportunities. This argument is related to those by Baumol ( 1 990) and by Murphy. Shleifer, and Vishny ( 1991) , discussed below.

Corruption can be a useful political glue by allowing politicians to get funds that can be used to hold a country together. The latter outcome may be a necessary condition for growth. See Graziano ( 1 980).


582 VITO TANZI

Bribes can supplement low wages. Thus, corruption can allow the government to maintain a lower tax burden, which can favor growth. See Tullock ( 1 996) and Seeker and Stigler ( 1974). The issue here is whether a lower tax burden is more favorable to growth than a lower degree of corruption. This is a classic second-best problem.

The above theoretical arguments, which seemingly favor corruption. can be countered in many ways. First. rigidities and rules are not exogenous and unmovable feawres of a society; a society is not born with these rigidities. They are created, and, in fact, they may be intentionally created by public officials, to extract bribes. When rules can be used to extract bribes, more rules will be created. Furthermore, these rules are often kept intentionally opaque so that more power will remain on the side of those who enforce them. Knowledge gives power to those who have it.

Second, those who can pay the highest bribes are not necessarily the most economically efficient but rather the most successful at rent seeking. If bribes are seen as investments. those who pay them must consider that they are investments with a high rate of return. Baumol ( 1990) and Murphy, Shleifer, and Vishny ( 1 99 1 ) have advanced related arguments that can be used to argue that. in traditional or corrupt societies, the most able individuals will be diverted. by existing incentives, from pursuing socially productive activities and toward rent-seeking activities. This diversion will impose a high cost for the growth of these countries. If the potentially most socially productive individuals are in scarce supply, as they are assumed to be. the diversion of their talent toward rent-seeking activities and corruption will be particularly damaging to society.

Third, payment of speed money may be an inducement for the bureaucrats to reduce the speed at which most practices are being processed. See Myrdal ( l 968). Bribes may change the order in which public officials perform the process. say. of providing permits, but they may also slow down the average rime for the whole process.

And while corruption and rent seeking may be helpful as political glue or as wage supplements in the shorr run. they may lead to major problems over the longer run, as shown by the experience of za·lre under Mobutu.

Qualitative Effects of Corruption on the Economy

Corruption reduces public revenue and increases public spending. It thus contributes to larger fiscal deficits, making it more difficult for the government to run a sound fiscal policy. Con·uption is likely to increase income inequality because it allows well-positioned individuals to take advantage


CORRUJYriON AROUND THE WORLD 583

of government activities at the cost of the rest of the population.40 There are strong indications that the changes in income distribution that have occurred in recent years in previously centrally planned economies have partly been the result of corrupt actions such as nomenklatura privatization.

Corruption distorts markets and the allocation of resources for the following reasons, and is therefore likely to reduce economic efficiency and growth.

• It reduces the ability of the government to impose necessary regulatory controls and inspections to correct for market failures. Then the government does not satisfactorily perform its regulatory role over banks. hospitals, food distribution, transportation activities, financial markets and so on. When government intervention is motivated by corruption. as for example when the government creates monopolies for private interests, it is likely to add to the exisring market failures.

• It distorts incentives. As already mentioned, in a corrupt environment, able individuals allocate their energies to rent seeking and to corrupt practices and not to productive activities. In some cases, the resulting activities have a negative value added.

• lt acts as an arbitrary tax (with high welfare costs). Corruption's random nature creates high excess burdens because the cost of searching for those to whom the bribe must be paid must be added to the cost of negotiating and paying the bribe. Also, the contractual obligations secured by the payment of a bribe are more likely to be violated when corruption is decentralized.41

• It reduces or distorts the fundamental role of the government in such areas as enforcement of contracts and protection of property rights. When a citizen can buy his or her way out of a commitment or out of a contractual obligation, or when one is prevented from exercisi ng one's property rights because of corruprion, this fundamental role of the government is distorted and growth may be negatively affected.

• It reduces the legitimacy of the market economy and perhaps also of democracy. In fact, the criticisms voiced in many countries, especially in transition economies, against democracy and the market economy are motivated by the existence of corruption. Thus, corruption may slow down or even block the movement toward democracy and a market economy.

4° For a quantitative analysis that establishes a connection between higher corruption on the one hand and higher income inequality and poverty on the other hand, see Gupta, Davoodi, and Alonso-Terme ( 1998).

41 Furthermore, random corruption may also be accompanied by higher penalties if the act of corruption is discovered.


584 VITO TANZI

• Finally, corruption is likely to increase povetty because it reduces the income eaming potential of the poor.

ln many countries (e.g., Ukraine. Russia. and Indonesia). enterprisesespecially small ones-are forced by public officials to pay to make things happen or even to keep bad things from happening. Often these payments must be made i f the enterprise is to remain in business. In Indonesia. there is a term for these payments ("pungli") and. according to a recent report, these payments may raise the costs of doing business for small activities by as much as 20 percent of total operating costs (see Sjaifudian, 1997). This is equivalent. to imposing very high sales taxes on these enterprises. Similar information has been reported for Russia (Shleifer, 1996) and for Ukraine ( Kaufmann. 1997), but the problem may be much more widespread.

Cost-increasing corruption is often coercive for small enterprises, especially for emerging enterprises, which are often bullied by bureaucrats and tax inspectors into making substantial payments. Pressures on new enterprises often come from local government officials, who impose high pecuniary costs-some legal and some not-for licenses and authorizations.�2 These officials also impose high costs in terms of the time d1at the managers of the enterprises must spend to comply with the many requirements imposed on them. The burden of these costs is likely to fall on the small enterprises because they operate in a far more competitive market than large ones, so that they have greater difficulty in passing the costs on to their customers. Since small enterprises are the engine of growth in most countries, obstacles to their creation and growth cause economies to languish, especially in developing countries and increasingly in economies in transition.

Large enterprises can protect themselves more easily from problems of corruption because

they have specialized departments that can deal with aggressive bureaucrats;

• they can use "facilitators"-individuaJs skilled at fighting through the jungle of opaque regulations and tax laws;

• their size makes them more immune to the extortion of petty bureaucrats: and

• they can use their political power to influence relevant individuals in the public administration or to pursue rent-seeking activities not

�2 These small enterprises may be preyed upon by the police, health inspectors, tax inspectors, and the myriad of other individuals presumably representing the government.



available to others. For example. they may use gifts, or bribes camouflaged as political donations, to acquire market power through reduction in competition or to obtain tax incentives. subsidized credit, or other benefits.

Recent reforms, such as trade liberalization, in many countries have removed obstacles to economic growth that had characterized earlier periods. However, these obstacles were imposed mostly by the national governments. They were probably more important for large enterprises, which are more likely to trade abroad and to operate in the whole national territory. These reforms have not done much to reduce the many regulations, controls, opaque taxes, and fees imposed by local governments. unions. professional groups, and so on. These are probably the most important obstacles for the small enterprises.

Effects of Corruption on the Economy: Econometric Results

In the past couple of years, several studies. using cross-sectional analysis and the available corruption indexes, have reported important quantitative results on the effects of corruption on economic variables. These results suggest that corruption has a negative impact on the rate of growth of countries. Some, but not all, of these studies are mentioned below.

It has been found that corruption has the following effects.

• It reduces investment and, as a consequence, reduces the rate of growth. See Mauro ( 1 995). Such reduction in investment is assumed to be caused by the higher costs and the uncertainty that corruption creates. In this analysis, the reduction in the rate of growth is a direct consequence of the decline in the investment rate. In other words, the analysis is based on a production function that makes growth a function of investment.

• It reduces expenditure on education and health, which does not lend itself easily to conupt practices on the part of those who make budgetary decisions.43 See Mauro ( 1 997).

• It increases public investment because public investment projects lend themselves easily to manipulations by high-level officials to get bribes. See Tanzi and Davoodi ( 1997). Conuption also distotts the

41 Of course, this does not mean that there is no corruption in the provision of these services. The provision of health is often distorted by bribes to doctors or other medical personnel to get better or faster service.


586 VITO TANZI

effects of industrial policy on investment. See Ades and Di Tella ( 1 997).

• lt reduces expenditure for operation and maintenance for reasons similar to those that reduce expenditure for education and health. See Tanzi and Davoodi ( 1997).

• It reduces the productivity of public investment and of a country's infrastructure. See Tanzi and Davoodi ( 1997).

• It reduces tax revenue. mainly because of the impact that it has on the tax administration and on customs, thus reducing the ability of the government to carry out needed public expenditure. See Tanzi and Davoodi ( 1997).

• It reduces foreign direct investment because corruption has the same effect as a tax. and in fact operates as a tax. See Wei ( 1997a). The less predictable the level of corruption (the higher is its variance), the greater is its impact on foreign direct investment. A higher variance makes corruption behave like an unpredictable and random tax. See Wei ( 1 997b). Thus, increases in corruption and in its unpredictabiliry are equivalent to increases in the tax rate on enterprises. Wei concludes that raising the index of corruption from the Singapore level to the Mexican level is equivalent to increasing the marginal tax rate on enterprises by 20 percentage points.

VII. The Fight Against Corruption and the Role of the State

The many factors that contribute to corruption tend to be more common in poorer countries and in economies in transition than in rich countries. Thus. at some point in time, economic development reduces the level of corTuption of a country. However, at similar levels of development, some countries are perceived to have more corruption than others.

Some economies (Singapore. Hong Kong, Portugal) have managed to reduce the incidence of corruption significantly. Lindbeck ( 1998, p. 3) has pointed out that even in Sweden "cormption flourished . . . in the second half of the 18th century and in the early 19th century." Thus, governments should not be fatalistic or passive about cormption. With well-focused and determined efforts, con·uption can be reduced, though not to zero. Trying to eliminate corruption altogether would be too costly, both in terms of resources and in other ways. For example, it may require excessively high public sector wages, major legal or organizational changes, excessive limitations in civil rights, or very harsh effective penalties. An optimal theoretical level would be reached where the marginal social costs of reducing cormption further would be equal to the marginal social benefits from that reduction.44



Thus, it is realistic to think that the level of corruption will remain above zero in all count1ies. In fact. probably no country is free of com1ption.

Corruption is a complex phenomenon that is almost never explained by a single cause. If it were. the solution would be simple. Of the many factors that influence it, some can be changed more easily than others. Because or the complexity of the phenomenon, the fight against corruption must be pursued on many fronts. It is a fight that cannot be won in months or even in a few years. The greatest mistake that can be made is to rely on a strategy that depends excessively on actions in a single area, such as increasing the salaries of the public sector employees. or increasing penalties, or creating an anticorruption office. and then to expect quick results.

Any realistic strategy must start with an explicit recognition that there are those ll'ho demand acts o.f corruption on the part of public sector employees and there are public employees willing .for a price to petform these acts. There is thus both a demand for and supply of corruption. And as is the case with all demands and supplies. the price plays a major role. Various incentives determine the elasticities of these supply and demand functions. In the basic case, the briber wants something (a reduction in a cost or an increase in a benefit) from the public official and is willing to pay a bribe for it. The official has something to sell (i.e .. power) and wants to be compensated for the risk and the effort involved.45 However, in the background there is the state in the totality of its actions carried out by the many agencies that constiLUte the public sector. To a large extent it is the state that. through its many policies and actions, creates the environment and incentives that influence those who pay bribes and those who accept or demand them. It is the state that influences the relationship between briber and bribee. See Klitgaard ( 1988).

In the ideal bureaucracies described by Max Weber. the public official (as the agent of the state) is the faithful executor of the mandate and instructions that he receives from the state (the principal). The public official is just a conduit, or a direct and legitimate channel for the relationship between the state and the citizen. He would not distort the state-citizen relationship. In this Weberian world. no principal-agent problems would develop. Unfortunately. in the real world. Weberian bureaucracies are rare46 in pan because the actions and the policies of the state are not always transparent and in part because of characteristics oft he bureaucracies themselves. The citizens may question the

•• In pracrice. of course. these marginal costs and marginal benefits are impossible to measure.

•; Of course. we are ignoring the case� when corruption is coercive and reflects 1he pressures of public officials or individuals.

•6They may be approximated only in a few countries such as Denmark. Sweden. New Zealand, Canada. and perhaps a few others.


588 VITO TANZI

legitimacy of some state actions. attributing them to rent seeking by public officials and not to the pursuit of the public interest. The state may have de facto fractured into several power centers (ministries. public enterprises, independent institutions. subnational governments. and so on), each pursuing somewhat distinct interests. Sometimes the policies of these power centers are not consistent with one another and the instructions that emanate from these public centers are conflicting.47

At times. the instructions passed on to the agents who will carTy out the execution of the policies are not clear because the top policymakers do not have clear ideas or perhaps do not wish their actions to be totally transparent. Total transparency in processes and in policies may imply less power for particular policymakers. in the sense that their discretion in affecting the welfare of particular groups is reduced. Thus, the fight against corruption is not distinct and independent from the reform of the state. because some of the measures to reduce corruption are at the same time measures that change the character of the state. Let us consider a few examples.

It is generally believed that the level of relative wages in the public sector is an important variable in the degree of corruption in a country. Singapore, a country with a good index of corruption and where corruption was much reduced over the years. has some of the highest wages for public employees. Reportedly, its ministers receive the highest salaries in the world. The civil service of Singapore is small and enjoys a high status.

In countries where public sector wages are low with respect to those in the private sector, these wages are often low because public policies have inflated the number of people working for the government. In other words, the governments have traded wage levels against the number of civil servants on their payrolls. The increase in the number of employees has meant lower real wages paid to them. In these sitlrations. it is not realistic or wise to suggest that these countries simply increase real wages without first reducing the size of the civil service. However. for many governments, reducing the number of public employees would nrn counter to the objectives of their government, or at least would be politically difficult. And many governments believe that unemployment can be reduced through public sector hiring.

The same argument applies to wage differentials. Many governments aim to reduce the spread in their employees' wages. In some countries, the ratio between the highest salaries and the lowest has been reduced to three. This is far less than in the private sector. The effect of this wage compression is that the most qualified and the most honest employees tend to leave the

47This is particularly the case of tax officials in some countries with decentralized fiscal systems. In. these situations. the officials may be subject to conflicting pressures from the national governmem and the local governments.



public sector for the private sector. For many governmems, changing this policy would be contrary to their goals and in many cases would be opposed by the labor unions.

It is generally believed that increasing the penalties for acts of con-uption would reduce a country· s COilllption. However. the imposition of higher penalties could run into problems with employees' associations, with trade unions, with the judiciary system. and so on. Also, there is the danger that an unscrupulous government would use this weapon to go after political opponents. In other words. penalties could be used selectively or. worse, they could be used in connection with fabricated accusations. In democratic societies, penalties are imposed only after what is often a lengthy and costly process. The immediate supervisors of the collllpt officials nnay be reluctant to carry the brunt of some of the procedural costs (in terms oftime lost. damaged friendships. etc.), and may prefer to close their eyes to acts of collllption.

It is generally believed that many acts of corruption are stimulated by the existence of regulations. Some kinds of quasi-fiscal regulations at times substitute for taxing and spending actions. It is useful to point out that some of the countries perceived to be the least corrupt (e.g .. Sweden. Denmark, Canada) have some of the highest tax burdens. On the other hand, some of the countries with the highest indexes of corruption (e.g .. Nigeria. Pakistan. Bangladesh, China. Venezuela) have some of the lowe�t tax burdens. In the latter group of countries. quasi-fiscal regulations (i .e .. regulations that substitute for taxing and spending) substitute taxes and public expenditures.4R Thus. to reduce COITuption these countries would need to eliminate these quasi-fiscal regulations and, if necessary. replace them with taxing and spending policies. But quasi-fiscal regulations are predominant in these economies and they would have a difficult time in raising the level of taxation substantially. Once again we come to the conclusion that the fight against corruption and the reform of the state are two sides of the same coin.

Tax incentives. especially when they imply discretionary decisions on the part of public officials. create conditions i n which conuption develops. A simple recommendation would be to eliminate tax incentives and replace them with tax systems with broad bases and lower rates, as often suggested by tax experts. Unfortunately, the roles that some governments want to play require the use of these incentives.49 Therefore, once again, the fight against conuption requires a refo1m in the role of the state.

"8 For a discussion of quasi-fiscal regulations and their power to replace taxing and spending. see Tanzi ( 1995b ).

49Th is has been a major issue in economies in transition where governments want to continue to directly influence the activities in some sectors. Thus, the policymakers have found it hard to accept the approach associated with broad-based taxes.


590 VITO TANZI

Corruption often accompanies the provision by the government of goods and services at below-market prices. This often occurs with credit, foreign exchange, the prices of public utility services, public housing, higher education, health services. and so on. The low or zero prices create excess demand and the need to ration the good or service. Rationing always brings corruption. Thus, raising these prices to equilibrium level whenever possible would elimjnate or reduce corruption. However. it would also change the role of the state in a way that many governments are not willing to accept.

Many other examples could be provided. However. these are sufficient to make the point that the fight against corruption often cannot proceed independently from the reform of the state. In many ways. it is the same fight. Thus. corruption will be reduced only in those countries where governments are willing to substantially reduce some of their functions.

VIII. Concluding Remarks

In this paper I have discussed the phenomenon of corruption. which affects many countries. r have shown the incidence of this phenomenon and the damage that it brings to economies and democracies. When corruption is widespread and especially when it contaminates the actions of the policymakers in democratic. market-oriented economies. it becomes more difficult to argue in favor of such economic and political arrangements. 50 The widespread disillusion among the population of some economies in transition and some developing countries with both market economies and democratic processes is very much caused by the widespread corruption that prevails in these countries and is wrongly attributed to the market economy and the democratic process.

I have also argued that corruption is closely linked to the way govemments conduct their affairs in modern societies, and therefore also to the growth of some of the government's activities in the economy. It is unlikely that corruption can be substantially reduced without modifying the way governments operate. The fight against corruption is, thus. intimately linked with the reform of the state.

In any case, any serious strategy to attempt to reduce corruption wi If need action on at least four fronts:

I. honest and visible commitment by the leadership to the fight against corruption, for which the leadership must show zero tolerance:

50 It should be remembered that many dictators or many potential dictators make the fight against corruption one of the reasons why they should be given Lhe reign� of a country. Some associate democracy with lack of discipline.



2. policy changes that reduce the demand for corruption by scaling down regulations and other policies such as tax incentives, and by making those that are retained as transparent and as nondiscretionary as possible:

3. reducing the supply of comtption by increasing public sector wage . increasing incentives toward honest behavior, and instituting effective controls and penalties on the pubtic servants: and

4. somehow solving the problem of the financing of polilical parties.

Societies can do much to reduce the intensity of corruption. but no single action will achieve more than a limited improvement-and some of the necessary actions may require major changes in existing pol icies.

REFERENCES

Ades. Albeno, and Rafael Di Tclla, forthcoming. "Rents. Competition and Corruption.'' American Economic Re1•iew.

---, 1997, "National Champions and Corruption: Some Unpleasant Interventionist Arithmetic.'' Economic Journal, Vol. I 07 (July), pp. I 023-42.

Bardhan, Pranab, 1 997. "Corruption and Development: A Review of Issues:· Journal of Economic Literature. Vol. 35 (September), pp. 1320-46.

Baumol, William J .. 1990. ··Entrepreneurship: Productive, Unproductive, and Destructive, .. Journal of Political Economy, Vol. 98 (October), pp. 893-921.

Beck. Paul J . , and Michael W. Maher. 1986. "A Comparison of Bribery and Bidding in Thin Markets:· Economic Letters, Vol. 20, No. I , pp. 1-5.

Becker. Gary S .. 1968. ·'Crime and Punishment: An Economic Approach ... Journal

of Political Economy, Vol. 76 (March/ April). pp. 169-2 17.

Becker. Gary S .. and George J. Stigler. 1974. "Law Enforcement. Malfeasance, and Compensation for Employees:· Journal of Legal Swdies (January). pp. 1-18.

Braguinsky, Serguey, 1 996. "Corruption and Schumpeterian Growth in Different Economic Environments." Contemporm:v Economic Policv. Vol. 14 (July). pp. 14-25.

Brunelli, Aymo, and Beatrice Weder, 1 998, ''A Free Press Is Bad ews for Corruption,.. WWZ Discussion Paper No. 9809 ( Basel: Winschaftswissenschaftliches Zentrum der Universitat Base!. April).

Cazzola. Franco, 1988. De/la Corru::,ione: Fisiologia e Patologia di un Sistema Politico (Bologna: 11 Mulino).

Davigo, Piercamillo, 1998, La Giubba del Re: lmervista sui/a Corru::.ione (RomeBari: Editori Laterza).

Ferrero, Mario. and Giorgio Brosio. 1997. "Nomenklatura Rule Under Democracy: Solving the Italian Political Puzzle ... Journal of Theoretical Politics. Vol. 9 (October).


592 VITO TANZI

Galasi. Peter, and Gabor Gertesi, 1987. "The Spread of Bribery in a Centrally Planned Economy:· Acta Oeconomica. Vol. 34 (3-4). pp. 371-89.

Galtung. Frederik. 1997. "Developing Agencies of Restraint in a Climate of Systemic Corruption: The National Integrity System at Work," paper presented at the Third Vienna Dialogue on Democracy. Vienna, June 26-27.

Goldman, Marshal!. 1997, "The Pitfalls of Russian Privatization," Challenge. Vol. 40 (May-June). pp. 35-49.

Graziano. Luigi. 1 980. Cliemelismo e Sistema Politico. 1/ Caso de/1'/talia (Milan: F. Angeli).

Grossman. Gregory, 1982, "The Second Economy in the USSR:' in The Underground Economy in the United States and Abroad. ed. by Vito Tanzi (Lexington. Massachusetts: D.C. Heath). pp. 245-70.

Gupta. Sanjeev, Hamid Davoodi. Rosa Alonso-Terme. 1998. "Does Corruption Affect Income Inequality and Poverty?"' IMF Working Paper 98176 (Washington: International Monetary Fund).

Haque, Nadeem Ul, and Ratna Sahay, 1 996, "Do Government Wages Cuts Close Budget Deficits? Costs of Corruption." Staff Papers. International Monetary Fund. Vol. 43 (December). pp. 754-78.

Hines. James R .. Jr .. 1995. "Forbidden Payments: Foreign Bribery and American Business After 1977:· NBER Working Paper No. 5266 (Cambridge. Massachusetts: National Bureau of Economic Research, September).

H untington. Samuel P .. I968, Political Order in Changing Societies (New Haven. Connecticut: Yale University Press)

Johnston, M ichael. 1997. "Pub! ic Officials. Private Interests, and Sustainable Democracy: When Politics and Corruption Meet:· in Corruption and the Global Economy. ed. by Kimbcrly Ann Elliott (Washington: Institute for International Economics). pp. 6 1-82.

Kaufmann, Daniel. 1997. "The Missing Pillar of a Growth Strategy for Ukraine: Reforms for Private Sector Development." in Ukraine: Accelerating the Transition ro Marker. ed. by Peter Cornelius and Patrick Lenain (Washington: International Monetary Fund), pp. 234-74.

---. and Paul Siegelbaum, 1996. "Privatization and Corruption in Transition Economies," Journal of International Affairs. Vol. 50 (Winter). pp. 419-58.

Klitgaard. Robert E., 1 988, Controlling Corruption (Berkeley. California: University of California Press).

Kopits, George. and Jon Craig. 1998. ·Transparency in Government Operations." IMF Occasional Paper No. 158 (Washington: International Monetary Fund).

Krueger, Anne. 1 974. 'The Political Economy of the Rent-Seeking Society," American Economic Re1·iew. Vol. 64 (June), pp. 291-303.

Langseth, Petter. and Rick Stapenhurst. I 997. "National Integrity System: Country Studies:· Economic Development Institute Working Paper (Washington: World Bank).

Leff, Nathaniel, 1964. "Economic Development Through Bureaucratic Corruption:' American Behal'ioral Scientist, pp. 8-14.



Lien. Da Hsiang Donald. 1986. ·'A ote on Competitive Bribery Games,'' Economic Leuers, Vol. 22. o. 4, pp. 337-41 .

Lindbeck. Assar. 1998. "Swedish Lessons for Post-Socialist Countrie� ..

(unpublished: Stockholm: University of Stockholm Institute for International Economic Studies).

Lui, Francis T., 1985. "An Equilibrium Queuing Model of Bribery:· Journal of Political Econom_, .. Vol. 93 (August). pp. 760-81 .

---. 1996. "Three Aspects of Corruption:· Comempor(//:v Economic Policy. Vol. 14 (July). pp. 26-9.

Manzeui, Luigi. and Charles Slake. 1997. ··Market Reforms and Corruption in Latin America: New Means for Old Way�:· Re1·ieu· of International Political Economr. Vol. 3 ( Winter). pp. 662-97.

Mauro, Paolo. 1995, "Corruption and Growth." Quarterly Journal of Economics. Vol. 1 10 (August). pp. 681-712.

---. I 997. "The Effects of Corruption on Growth. Investment, and Government Expenditure: A Cross-Country Analysis," in Corruption in the Global Economy. ed. by Kimberly Ann Elliou (Washington: Institute for International Economics).

Murphy. Kevin M .. Andrei Shleifer. and Robert W. Vishny. 1991. "The Allocation of Talent: Imp I ication for Growth." Quarterly Joumal of Economics. Vol. I 06 (May), pp. 503-30.

Myrdal. Gunnar, 1968. Asian Drama: An lnquil:r into the PoFerty of Nations ( ew York: Twentieth Century Fund).

Nordio. Carlo. 1997. Giusti::.ia (Milan: Angelo Gorini e Associati).

Noonan, John T. Jr .. 1984, Bribes (New York: Macmillan).

Rauch. James E .. and Peter B. Evans. 1997. ··Bureaucratic Structure and Bureaucratic Performance in Less Developed Countries" (unpublished: San Diego. California and Berkeley. California: University of California at San Diego and University of California at Berkcley).

Rem nick. David. 1994. Lenin ·s Tomb: The Last Days of the Soviet Empire (New York: Random House).

Rose-Ackerman. Susan. 1997. "Corruption and Development. ..

paper prepared for the Annual Bank Conference on Development Economies. Washington. D.C .. April 30 and May I.

Ruzindana. Augustine. 1997. "The I m porta nee of Leadership in Fighting Corruption in Uganda." in Corruption in the Global Economy, ed. by Kimbcrly Ann Elliott (Washington: Institute for International Economics). pp. 133-146.

Shleifer. Andrei, 1 996. "Government in Transition," HIER Discussion Paper o. 1 783 (Boston, Massachuseus: Harvard Institute for Economic Research).

---. and Robert W. Vishny, 1993, "Corruption," Quarterly Joumal of Economics, Vol . 108 (August). pp. 599-617.

Simis. Constantine M .. I 982, USSR: The Corrupt Society. The Secret World of Stwiet Communism (New York: Simon and Schuster).


594 VJTO TANZI

Sjaifudian. Shctifah, 1 997, ··Graft and the Small Business:· Far Eastern Economic Review, October I 6. p. 32.

Tanzi. Vi to. I 995a. "Corruption, Arm's-Length Relationships, and Markets ...

in The Economics of Organised Crime. ed. by Gianluca Fiorentini and Sam Peltzman (Cambridge. Massachusetts: Cambridge University Press), pp. 16 1-1 80.

---. 1995b, "Government Role and the Efficiency of Policy Instruments:· IMF Working Paper 95/100 (Washington: International Monetary Fund). Also published in Public Finance in a Changing World. ed. by Peter Birch S�rensen (New York: Macmillan), pp. 5 1 -69.

---, 1 998, ·'Corruption and the Budget: Problems and Solutions:· in Economics of Corruption. ed. by Arvind K. Jain (Boston. Massachusetts: Kluwer Academic Publishers). pp. 1 1 1-28.

---. and Hamid Davoodi. 1997. ··corruption, Public Investment. and Growth." IMF Working Papcr 97/l39 (Washing10n: International Monetary Fund).

Tanzi. Vito, and Ludger Schuknecht. 1 997, "Reconsidering the Fiscal Role of Government: The International Perspective."' American Economic Review, Papers and Proceedings. Vol. 87 (May), pp. 164-8.

Theobald. Robin, 1 990. Corruption. Del•elopmem. and Underdel'elopment (Durham, North Carolina: Duke Unive rsity Press).

Tullock. Gordon. 1 996, ·'Corruption Theory and Practice," Contemporary Economic Policy. Vol. 14 (July), pp. 6-13.

van Tulden. Frank. and Abraham van der Torre, 1997. "Crime and the Criminal Justice System: An Economic Approach."' paper presented at the 53rd Congress of the International Institute of Public Finance. Kyoto. Japan. August.

Van Rijckeghem. Caroline. and Beatrice Weder. 1997. "Corruption and the Rate of Temptation: Do Low Wages in the Civil Service Cause Corruption?" IMF Working Paper 97/73 (Washington: International Monetary Fund).

von Klimo. Arpad, 1 997. "Fra Stato Cenlralmstico e Peri feria. Alti Funzionari Statali i n ltalia e nella Germania Prussiana dal I 870 al 1 9 I 4," i n Centralismo e Federalismo ne/ XIX e XX Secolo. Un COI!(romo tra la Germania e ltalia, ed. by Oliver Janz and Pierangelo Schiero (Bologna: 11 Mulino).

Weber. Max, 1947, The Theory of Social and Economic Orgcmi::.ation (London: The Free Press of Glencoe).

Wei, Shang-Jin, l997a. "How Taxing Is Corruption on International Investors?" NBER Working Paper No. 6030 (Cambridge. Massachusetts: ational Bureau of Economic Research).

--- l997b. "Why Is Corruption So Much More T;1xing Than Tax? Arbitrariness Kills,'' NBER Working Paper No. 6030 (Cambridge, Massachusetts: National Bureau or Economic Research).


IMF Staff Papers Vol. -!5. No. 4 (December 1998)

© 1998 lnwrna1ional Monetary Fund

Countries' Repayment Performance

Vis-a-Vis the IMF

An Empirical Analysis

LYNN AYLWARD and RUPERTTHORNE*

While the literarure on external debt repayment pe1jormance by sovereign debtors is extensil·e. repayment pe1jormance vis-a-vis the lntemational Monetary Fund has not been dealt with separately. Cil•en differences between the !M F and other providers of financial resources, !his paper considers whether it is possible to distinguish lhrough logit analysis between 1he countries that make limely repayments 10 !he Fund and those !hat become overdue. The paperfmds that the inclusion of IMF-specific financial variables and a small number of macroeconomic l'ariables yields a highly significant econometric model of !he probability of a country incurring IMF arrears. [JEL F33, F34]

THE FAILURE of some member countries to repay obligations on time became a matter for serious concern for the IMF toward the end of the

fourth decade of the lMF' s history-around the mid-1980s, a few years after the debt crisis began. At that time. the duration and magnitude of countries' arrears ro the IMF increased, as did the number of countries involved, in contrast to the rare and brief cases of late payments in earlier years.

Total arrears rose from SDR 25 million in the first quarter of 1981 to a peak of SDR 3.7 billion in 1992. Since then, total arTears have declined as a number

*Lynn Aylward is Assistant to the Director in the JMF's External Relations Department. She holds an M.Sc. from the London School of Economics. Rupert Thorne is a Senior Manager in the Foreign Exchange Division of the Bank of England, and holds an M.A. in Mathematics from Cambridge University. Both authors were Economists in the Treasurer"s Department of the IMF when the research for this paper was carried out. They are particularly indebted to James Corr for extensive guidance on this work. Helpful comments were also provided by Michael Wattlewonh. Orlando Roncesvalles. James Blalock. and Krishna Srinivasan.

595


596 L YNN A YLW ARD and RUPERT THOR:--IE

of countries have settled their overdue obligations to the IMF. but they have remained at SDR 2.2-2.3 billion since end-1995. The number of countries making late payments to the fMF in the course of a single year peaked in 1986 at 63. and declined since then. with 32 member countries making late payments to the fMF in 1997. of which 6 were in prou·acred an·ea rs during that year.1

Thus, the pattern of IMF arrears i n the aggregate can be distinguished by three phases. Until 1983. only one member country had experienced protracted arrears to the IMF, and a handful of others, arrears of short duration. Then. every year from 1983 until 1 990. the amount of outstanding overdue obligations to the IMF grew substantially. From 1 991 to 1995. the level of arrears fel l sharply. and from 1995 to present. it has changed little. It should be noted that over 1989-90. the IMF s Executive Board developed a strengthened cooperative strategy to resolve the problem of protracted overdue obligations to the IMF.�

This paper seeks through econometric analysis to identify empirically the financial and macroeconomic variables most closely correlated with IMF member countries· repayment performances. To the authors' knowledge. there have been no previous empirical studies of IMF arrears. There is an extensive related body of l iterature. though. on external debt repayment problems by sovereign debtors to creditors other than the IMF. One question. therefore. is whether it is in fact possible to distinguish econometrically between countries that seek financial support from the IMF and the small subgroup of those countries that eventually encounter difficulties in repaying these and other resources. Since the IMF extends its resource� only to countries that have problems with their balance of payments or international reserves position. both the countries that go on to make timely repayments and those that become overdue would be expected to have exhibited common initial characteristics. which might make distinguishing between the two groups statistically difficult.

A further question this study investigates is whether the determinants of countries· repayment behavior vis-a-vis the IMF are similar to the factors influencing countries· repayment behavior vis-a-vis other creditors. Although the individual circumstances and economic performances of the countries varied widely. those that experienced large and protracted arrears to the IMF have been almost without exception countries with prolonged

1 The IMF considers arrears protracted when their duration i� !.ix month� or longer. The members with protracted arrears in 1997 were Afghanistan. Iraq. Liberia. Somalia. Sudan. and the Democratic Republic of the Congo (formerly Z<l'lre}. One nonmember of the IMF. the Federal Republ ic of Yugoslavia (Serbia/Montenegro). was also in protracted arrears.

2 For a description of the main elements of the strategy and its implememation. �ee IMF. Annual Reporr ( 1991 and onward).


COUNTRIES' REPAYMENT PERFORMANCE VIS-A-VIS THE IMF 597

problems of economic management. including external payment arrears, which were evident. and a source of active concern to the IMF, before their overdue obligations to the IMF themselves became an issue. fn this regard. one might expect a good deal of overlap between the emergence of repayment difficulties to creditors generally and to the IMF in particular. However, although the global economic events leading up to and culminating in the debt crisis of the 1 980s appear to have contributed to the emergence of arrears to the IMF. it does not seem that the IMFs experience with arrears has simply been part and parcel of the debt crisis. One distinction is that the number of countries that have incurred protracted arrears to the IMF is much smaller than the number of countries that have failed to service their debt to other creditors and/or have had to enter into debt-rescheduling arrangements.3 A second distinction is that political instability seems to play a more prominent role among the most protracted cases of IMF arrears than it does among countries with non-IMF-specific repayment problems.4

The IMFs status as a preferred creditor, its role as a catalyst in attracting financing for countries from other creditors, and its character as a cooperative financial institution suggest that a country would likely place relatively greater emphasis on meeting its financial obligations to the IMF than on meeting those to other official bilateral or commercial creditors. Since the IMF generally makes its financial resources available to countries only in the context of a macroeconomic adjustment program, the conditionality associated with this support should, in principle, strengthen a country's capacity-to-repay prospects, and could imply that variations in countries· repayment performance with the IMF are less . ensitive to changes in certain indicators of creditworthiness than are variations rn countries' repayment performance vis-a-vis other creditors.

I. Empirical Evidence on the Determinants of External Debt Repayment Problems

A basic premise of most empirical analyses of external debt repayment problems, country risk. and creditworthiness is that a limited number of financial, macroeconomic, or sociopolit ical indicators can be identified as

' For example, while 23 countries have had protracted arrears on principal obligations to the IMF during its history. 58 countries rescheduled their official debt through the Paris Club in the period 1980-92.

4 As noted in the IMF Working Paper ve rsion of this study (Aylward and Thorne. 1998), if one considers the 12 countries that had protracted arrears in 1 992 (a year chosen because it precedes a period when a significant number of arrears cases were resolved). all but one could be said to have had markedly troubled political environments.


598 L YNN A YLWARD and RUPERT THORNE

the main determinants of debt repayment behavior. Empirical studies of external debt repayment generally assign a value to the likelihood that a country will repay its debt on rime. The studies use the observable event of whether or not a country incurred arrears, rescheduled its debt, or otherwise evidenced external debt repayment difficulties5 as the dependent variable, with a country's probability of evidencing repayment difficulties being the true but unobservable underlying dependent variable of interest. Saini and Bates ( 1984) reviewed the development of empirical estimations of country risk and traced the emergence of probit and legit models as the preferred estimation techniques.6 Subsequent key papers are McFadden and others ( 1985) and Hajivassiliou ( 1 989). the former because it specifies a complete model of loan demand and supply, and both papers because they deal econometrically with country heterogeneity and state dependence. Eaten, Gersovitz, and Stiglitz ( 1986) suggest theoretical extensions of country risk analysis, including consideration of the significance of the international loan contract and its enforcement, and they emphasize the distinction between ability and willingness to pay. Solberg ( 1988) estimates countries' propensity to incur arrears as a policy choice based on the costs and benefits of default. Feder and Uy ( 1 985). Berg and Sachs ( 1 988), and Li ( 1 992) account for the large role of sociopolitical factors in repayment behavior. Lloyd-Eilis, McKenzie, and Thomas ( 1990) attempt to explain both the occurrence and quantity of developing country debt rescheduling.

Table I summarizes information on the variables tested in a number of studies of external debt repayment behavior. including this paper's results, which are formally presented below.7 Avramovic ( 1964) carried out an important systematic study of the factors that influence a country's balance

5Some studies (McFaddcn and others, 1985, and Hajivassiliou. 1989) include the use of higher-tranche IMF arrangements as a sign of external debt repayment problems. Others (Euh. 1979, and Haque and others. 1996) use a creditworthiness index as the dependent variable.

6Probit and log it models allow the analysis of qualitative or binary dependent variables. The probit and logit models. in which the probability of a qualitative outcome is related to the standard normal distribution function or the logistic distribution function, respectively, are better suited for dealing with discrete dependent variables than ordinary least squares regression, but are simiaar to the regression technique in that the probability of the event is related to a vector of independent variables by a functional form that includes a set of nonbinary coefficients. In these cases, the underlying dependent variable is generally the probability of an external debt repayment event such as a rescheduling. versus the alternative outcome of no rescheduling, or the incurrence of arrears. versus the alternative outcome of timely debt servicing. What is actualli' observed, of course. is not the probability of the event. but the yes/no outcome.

In Table I . for all studies, the results are displayed so that a "+" indicates positive correlation of the independent variable with the probability of external debt repayment problems. and a "-··. negative correlation. "NS" indicates the variable wa� tested but found not to be significant au the 95 percent level of confidence. A blank indicates that the variable was not tested in the study.

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

Tab

le I

. Su

mm

ary

Re

sulr

s of

Em

pir

ica

lll

ll·e

stig

ati

on

s of

En

ema

/ De

bt

Rep

ay

mel

ll P

rob

lem

s�

Stu

die

s. I

de

ntifi

ed b

y A

uth

or(

s) a

nd

Ye

ar o

f P

ubli

cati

on

Fed

er.

F

ran

k &

S

aini

&

Just

&

Fe.d

er

}.1cf

adde

n B

erg

&

"'

c

Cli

ne

. S

arg

en.

Bat

es,

Eu

h.

Ro

ss,

Edw

ards

. &

Uy

. &

ol.he

rs.

Sa

chs.

c

V

ari

ab

le

197

1 19

77

1984

19

79

1981

19

8-t

1985

19

85

1988

z

�

�

D

ebi!

GD

P

+

+

m

:r.

Deb

l/e

x.port

s +

+

NS

:;:;

Deb

t se

rvic

e/ex

pon

s +

+

N

S

+

+

+

NS

+

�

)>

R

ese

rYes

lim

port

s N

S

s -<

�

E

xpon

s/G

DP

s

m

z

-i

I mpo

rts/

GD

P

NS

N

S

s +

""

m

C

urrem

acco

uni/G

DP

+

;:;:;

ln,·

est

men

i/G

DP

6

;:;:;

Go

,·ern

mem

ex.pe

nd

imre

JGD

P

s �

>

z

T

erm

s o

f 1r

ad

e

NS

I')

m

el

no

nc

om

mer

cia

l o

r <

n

on

-de

bt-

cre

atin

g i

nll

ow

s c;

;; . >·

P

er c

ap

ita

in

co

me

NS

NS

NS

NS

NS

s • <

G

DP

gro

wth

rat

e N

S

s ,::;

;

Rat

e o

f in

fla

tio

n

1:\!S

::l

+

+

�

Mo

ne

y s

up

ply

gro

wth

rat

e +

+ �

E

xch

ane

e r

ate

ov

erv

alu

atio

n

:-.rs

'!1

+ in

dic

ato

r

His

tory

or

amo

un

t o

f me

+

u.

o

f lM

F re

sour

ces

"'

"'

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

Var

iab

le

Polit

ica

l va

ria

ble

s te

sted

: fo

und

sig

nifi

can

t,.

Co

unt

ry l

lete

rog

en

eiry

tes

ted

: fo

und

sig

nifi

can

t?

Rep

aym

enl

hist

or

y te<>

ted

: fo

urrd

sig

nifi

can

t?

Var

iab

le

Deb

LIG

DP

Deb

lle

'\j)On

:S

Deb

t se

rvic

e/ex

port

s

Re

serv

es/

i m p

on

s

Exp

on

s!G

DP

lmp

on

s/G

DP

Cur

rent

acc

ount

/GD

P

---

ln\

e<;lm

ent

/GD

P

So

lber

g.

198

8

+

+

Ta

ble

I. (

conc

lud

ed)

Stu

die

s.ld

emif

ied

by

Aut

hor(

s) a

nd Y

ear

of

Pu

blic:

uio

n

Fed

er.

Fra

nk &

S

ain

i&

Just

&

Fed

er

�\kF

add

en

Clin

e.

Sar

gen

.

Bat

es.

Euh

, R

os s

. E

dw

ard

s,

&U

y.

& o

the

rs,

197

1 19

77

198

4

197

9 19

81

198

4

1985

19

85

J\o

Y

es

Yes

Ye

s Y

es

Yes

Y

es

A)

I\\ a

rd &

Tho

me

Elm

ore

&

Ha

que

A

rrea

r� to

the

A

rrea

r-. 10

H

aji,

·as�

ilio

u.

Mc

Ke

nz

ie.

Li.

&

oth

en;.

IMF

.L

DV

tile

IM

F.

19&9

19

92

199

:::! 199

6

not

incl

ud

ed

1 L

DV

inc

lude

d

+

+

NS

NS

+

+

:xs

+

x

s

1\'S

NS

NS

+

1\

S

+

NS

NS

N

S

Ber

g&

S

ach�

.

198

8

Ye

s

Ye

s

Ye

s

Arr

ears

to

oth

er

cred

ito

r"'

+ s

NS

0

"'

z

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

Gov

ernme

nt e

xpend

iture

JGD

P

+

'S

+

Tem

1s o

f tra

de

NS

'et

nonc

omm

erci

al o

r no

n-d

ebt-

crea

ting

inflo

ws

Per

cap

ita

inco

me

NS

NS

NS

GD

P g

row

th r

ate

NS

NS

Rat

e of

infl

atio

n l\

S �

+

NS

)..1on

ey s

uppl

y g

row

th r

ate

Ex

ch

ang

e ra

te o

verv

alu

atio

n in

dica

tor

+

xs

N:S

Hi

wry

or a

mo

um o

f us

e of

l�·lf

res

ourc

e.

+

+

+

Polit

ical

mriab

les t

este

d:

foun

d si

gni

fican

t?

Ye."

C

ount

r� h

eter

ogen

eit�

te�

ted

: fo

und

sign

ifica

nt?

Yes

Y

e�

Rep

ayme

nt h

isto

l) t

este

d:

foun

d ig

nilic

am?

Ye

s Y

es

Yes

Y

es

Ye.

Y

es

• F

or

all

stud

ies.

the

resu

Its a

re d

isp

lay

ed s

o th

at a

· · + .

. in

dic

ates

po<; i ti

v·e

corr

elat

ion

of

the

i nde

pend

em \

·ari

ab le

'' it

h th

e pr

oba

bi lit

� o

f ex

tern

al d

ebt

rep

aym

ent

prob

lem

�. a

nd a

. . _ ..

neg

ativ

e c

orre

lati

on

. "N

s·· i

ndic

ate.,

the

\'ar

iab

le w

as t

este

d b

ut fo

und

n01 t

o be

sig

nifi

can

t at

th

e 95

perc

ent

leve

l of

co

nfide

nce.

A b

lank

ind

icat

es th

e ,·a

riab

le w

as n

ot

test

ed in

the

stu

dy.

"Th

at i.s

. re

sults

whe

n t

he la

gged

depe

nd

ent

vari

able

(LD

V).

an i

ndic

ato

r o

f re

pay

men

t hi

sto

ry. w

as n

ot i

nclu

ded

. o

te t

hat

rep

aym

ent

his

torv

to

cre

di.to

rs o

ther

than

the

I !\IF

wa

s te

sted

. T

hat i

s. re

sults

whe

n th

e la

gged

dep

ende

nt '

ari

able

lLD

V)

was

inc

lud

ed

. Th

eo;e

re

sult

s re

late

10

:0.1o

del

lE in

this

pape

r.

�In

the a

nal

ysi

s of

Haq

ue a

nd o

ther

s. a

co

unt!

)'"

rat

e o

f in

flat

ion

was

a d

eter

minan

t o

f it�

cre

dit r

atin

g on

I} i

f it

was

a r

ela

tiv

e!)

IO\\

in

flatio

n co

untr

y.

r,

0

c

z

_,

;o


602 LYNN AYLWARD and RUPERT THORNE

of paymems and the country's related ability to service its external debt. The results of their work-the identification of short-term liquidity factors (the so-called traditional debt or financial ratios: debt-GDP. debt service-exports, reserves-imports) and longer-term indicators of economic health and growth (GDP growth rate. investmem. exports, inRation)-seem to have guided the selection of variables in many of the subsequent empirical analyses of external debt repayment behavior. Even in studies where the independent variables are selected based on formal models of external loan demand and supply or utility maximization. they are, not surprisingly. often similar to those identified by Avramovic ( 1 964).

For presentational clarity. Table I includes only the variables that have been tested most frequently in empirical studies. The analyses reported in the table used different data sets, dependent variables (e.g., arrears. rescheduling, higher-tranche IMF arrangements), and analytical techniques, and most present a number of versions of their respective models with differences in the significance of some variables (only one of which is reported in the table. again due to presentational considerations). Despite these disparities, some summary observations may be drawn from the studies covered in the table. First, a fairly limited set of variables have generally been found to be significanr in empirical analyses of external debt repaymem behavior. Second, it is the so-called financial ratios, such as debt-GDP or reserves relative to imports, that seem most consistently to be significant. Third. for most indicators of economic conditions and policy stance (as opposed to financial ratios), the results are mixed: inflation and indicators of exchange rate overvaluation, for example, have been found to be positively correlated with the incun·ence of arrears in some studies. but insignificant in others.8 Fourth. past repayment history has been found significant in every study in which it has been tested.

These observations go some way toward suggesting variables that might be tested in the present study of correlates of the behavior of repayment to the IMF. Two other aspects of the empirical analysis of external debt repayment problems must also be noted.

First, the use of debt ratios in empirical analyses of the likelihood of arrears raises questions about whether such variables are symptoms or causes of external debt repayment problems. In response, some studies attempt to measure the structural correlates of external debt repayment problems (e.g .. unfavorable terms of trade developments or exchange rate overvaluation) rather than proximate. and obvious, indicators such as the

8 The current account-GDP ratio was significant with the ··wrong'' sign in Edwards ( 1 984), that is. a larger surplus was associated with a higher probability of arrears.



amount of debt service a country owes. Berg and Sachs ( 1988) seek correlates of rescheduling that "are more fundamental than the value of financial variables on the eve of rescheduling;' and find that greater income inequality. lower per capita income. a lower share of agriculture i n GDP, and lesser outward orientation of the trade regime are significant predictors of a higher probability of debt rescheduling. As Table 1 indicates. though, and as might be expected, the proximate debt indicators have yielded more consistent results in explaining differences in countries· repayment performance. While the circumstances leading countries toward the brink of balance of payments crises are diverse. once at the threshold of default most nations enter the common ground of a high debt burden and low reserves.

Second, McFadden and others ( 1985) pose the question whether it is reasonable to expect a macroeconomic pattern that is stable across countries and time to emerge from econometric analysis of repayment problems. Though they find the answer is yes, the results surveyed in Table I indicate the importance of the question, since only a very few macroeconomic variables have consistently been found significant, while the significance of other factors varies from one study to another. Haj ivassiliou ( 1989) points out that persistent heterogeneity among countries in their debt repayment behavior may result from colonial histories. types of government, religious institutions. or other attributes not easily captured with macroeconomic time-series data. In practical terms. country heterogeneity reflects the noteasily-measured factors that explain why Country A may default when, say. its reserves are down to three weeks of imports and debt service is consuming 30 percent of export earnings. while Country B may continue to service its debt even when faced with the same reserves and debt service ratio and otherwise evidencing macroeconomic variables similar to Country A. In econometric terms. persistent country heterogeneity with pooled crosssection and time-series data violates assumptions about the randomness of the error term and casts doubt on the measured coefficients and their significance. McFadden and others ( 1985) and Hajivassiliou ( 1 989) find that a large share of the variation in countries' repayment behavior-forexarnple, 30 percent for Hajivassiliou's database and model-is country specific, rather than attributable to the macroeconomjc variables included in their respective models. It is computationally difficult to test for the presence of fixed country effects with panel data in a logit analysis. and many limited dependent variable analyses of external debt repayment behavior, including this one. do not fully deal with this problem.

The analyses of McFadden and others ( 1 985) and Hajivassiliou ( 1 989) also indicate a strong influence of countries' past record i n timely servicing of their debt in explaining current repayment behavior. The strength of


604 L YNN A YLW ARD and RUPERT THORNE

this impact is such that, in these authors' studie�. inclusion of the lagged dependem variable or other indicators of historical creditworthiness renders insignificant independent macroeconomic variables that otherwise contribute to the explanation of repayment behavior.

II. Methodology

Methodological Issues

For the analysis. a database was constructed covering the 138 developing countries (with the exception of Cambodia) that had payment obligations to the IMF in any year during the period 1 976-93.9

The main dependent variable in this study is a binary indicator (IMFARR,) signifying whether or not a country was in arrears to the IMF in a given year. IMFARR, takes the value I if a country was in arrears to the IMF in year r. and 0 otherwise. 10 A second binary dependent variable EXTARR, is also tested. EXTARR, indicates whether a country was in arrears to creditors other than the IMF and takes the value I i f a country had external arrears in year r to any external creditors, and 0 otherwise. The two dependent variables allow a comparison between the detenninants of the probability of a country incurring arrears to the IMF versus to other external creditors.

Since the dependent variable has only two possible outcomes, 0 or I. we use a bivariate logit model where

K exp L�kx�;,

P. = Prob(!MFARR. = 1) = k= l . u u ( K ) I + exp L�kxk" �=I

�The data were drawn from the IMFs World Economic Outlook database and the Treasurer's Department's records of members· financial obligations and paymems to the IMF. From the 138 countries and 1 8 years studied, the 2.484 possible countryyear observations were reduced to 1.871 by excluding all coumry years for which a country did not have any obligations to the IMF falling due in the year, since in those cases there is no test to make of whether IMF repaymenL� were made on time.

10 Specifically. the variable takes the value I it country i was in arrears to the IMF for a eominuous period of at least three months including at least part of year r. and the value 0 otherwise. Thus. if coumry i was cominuously in arrear!> to the IMF over the period July 1983 to February 1 984. and current in its obligations at all other times, then /MFARR,= I fort = 1983 and 1984.and O forall other t. Note the iMFs standard usage of protracted arrea rs refers to overdue obligations of six momhs or longer duration.



For country i in year 1. there are K explanatory variables x1,, • • • • x411•

Parameters p, . . . . , P4 are then estimated by a standard maximum likelihood procedure, under the assumption that the ob�ervations IMFARR;, are independent of each other. General background on econometric models of this type can be found in Maddala ( 1 983) and Greene ( 1 993).

In fact, consideration needs to be given to the possibility of countryspecific effects and time dependency and autocon·elation between observations. lt is not computationally easy to deal with such considerations within the context of the maximum likelihood logit model; informal tests for the presence of country-speci fic effects were carried out. but were not conclusive. 1 1

The models tested were all of a standard logit form. The explanatory variables. which are described below. were lagged by one year. reflecting the assumed direction of causality (except REPAX, which was lagged by two years for technical reasons).1�

Explanatory Variables

Our selection of explanatory variables was influenced by the literature reviewed in Table I and by the objective of testing for differences between the determinants of repayment perfo1mance to the IMF versus that tO other creditors.

Debt Burden

Two debt-burden variables cover financial obligations to all external creditors including the IMF: the ratio of debt to GDP (identified as EDTGDP) and the ratio of debt service due to exports ( TDSX). As measures of IMF financial obligations, use of the ratio of IMF credit to IMF quota (UFCQU) and the ratio of IMF obligations to exports (REP X) are used. The latter variable comprises total obligations (principal and charges or interest) due from a country to the IMF in a given year, both those obligations newly falling due and those in arrears. lt is also separated into its current and overdue components to form two other variables: the ratio of current

1 1 These results are reported in Aylward and Thorne, 1998. 12 To extract the maximum amount of information available from the IMF'�

repayment history-since. similar to most studies of external debt repayment problems. there were a relatively small number of .. positive .. or arrears observations in our data base, and a large number of "negative" or nonarrears observation�-our variable selection was guided to some extent by the practical considerarions of whether time series of certain variables were unavailable for large numbers of countries or years. and whether other series seemed to be less reliable in terms of accuracy or of comparability across countries.


606 LYNN AYLWARD and RUPERTTHORNE

IMF obligations to exports (REPCX) and the ratio of overdue IMF obligations to exports (REPAX).

Debt Repayment History

As noted above, a country's credit history has been found to be one of the strongest correlates of future performance in repaying external debt. We use as credit history indicators the level of arrears to all external creditors relative to external debt (ARRED1): the amount of arrears to the IMF (REPAX), described above as an IMF-specific debt-service ratio; and IMFARR11_1 , and EXTARR,,_ 1 ,, the lags of the two dependent variables analyzed in our study.

Availability of Financial Resources

Because of the IMF' s preferred creditor status, differences in the explanatory power of measures of the resources available to service external obligations may occur depending on whether the dependent variable is the likelihood of arrears to the IMF, or to other creditors. If the fMF's preferred creditor status were respected, available foreign exchange would tend to be directed toward repaying the Fund even if the debt burden to other external creditors were high. The three measures relating to resource availability used here are the ratio of reserves to imports (RESM): the ratio of exports of goods and services to GDP (XGDP); and the ratio of imports of goods and services to GDP (MGDP).

There are alternative. and opposite. interpretations of the potential impact of the size of exports and imports on the I ikelihood of external debt repayment problems. The more traditional view is that the stronger a country's current account position and the more foreign exchange it has available from the sale of exports, and the less it must expend on noncompressible imports, the lower the probability that it would default on external debt service. Alternatively. it has been argued by Solberg ( 1988) that the weaker a country's current account is, and the more it must rely on imports. the more dependent it is on continued capital inflows, and so the more incentive it has to make whatever financial or political sacrifices are necessary to avoid incurring arrears.

Domestic Economic Conditions and Policy Stance

We use two summary indicators of economic conditions: the log of U.S. dollar real per capita income (PC!)13 and the rate of change in ihis variable

11 The specific series used was the IMF World Economic Outlook purchasing power parity-based per capita income series.


COUNTRIES'

REPAYMENT PERFORMANCE VIS-A-VIS THE IMF 607

(GROWTH). Even though PCI has been found not to be correlated with non-Fund repayment performance in a number of studies, it may have explanatory power for IMF arrears. Such a result could reflect the unique role of the Fund, indicating greater likelihood that poorer countries, with the fewest options for raising resources or cutting expenses when faced with external imbalances, tend to incur arrears to the IMF. Such a relationship would be in contrast to the performance of numerous middle-income countries, for example, the Baker- 1 5 countries, which during the debt crisis incurred arrears to non-IMF creditors, but never to the Fund.

Indicators of macroeconomic policy stance that are examined are the fiscal position and the price environment. We test both the central government spending-GDP ratio (EXPGDP) and the central government revenue-GDP ratio (REVGDP). The measure of inflation is the adjusted log of the rate of change in the consumer price index (INFL).14

External Developments

We investigate whether the shifts in the frequency and magnitude of arrears to the IMF described in the introduction reflect changes in the environment external to the individual debtor countries that could have made it first more likely (between 1982 and 1989) and then less likely (after 1 989) across the board that countries would incur arrears to the Fund. More specifically, we hypothesize that the onset of the debt crisis lowered the cost to countries of arrears to the IMF, perhaps because the loss of international reputation that a country suffered when it failed to meet its financial obligations to the Fund was perceived by the country as being diminished as more countries experienced external debt repayment crises. or because the negative implications of IMF arrears for potential flows from other creditors receded as those flows were curtailed in any event. We also test whether the likelihood of countries incurring lMF arrears fell after 1989. While this could have occuJTed because of the implementation of the Fund's strengthened cooperative arrears strategy, the easing of the debt crisis. or other factors, we do not attempt to distinguish e.conometrically among these. Two dummy variables, one that takes the value I in and after 1984 and 0 before (DUM84). and a second that takes the value I in and after 1 990 and 0 before (DUM90). were tested.15 Since the dependent variable takes the value I if

1• Because a number of observations of negative rates of inflation and hyperinflation had to be dealt with, the series INFL is actually the log of one plus the percentage change in the consumer price index.

1$The earlier dummy could not be constructed to change value in 1982, the year generally regarded as the beginning of the debt crisis, because there were so few IMF arrears observations in the sample before 1982.


608 L YNN A YLW ARD and RUPERT THORNE

IMF arTears occur. and 0 if they do not, it is hypothesized that DUM84 will bear a positive sign, and DUM90 a negative one.

Ill. Results

IMF-Specific Variables

First, a group of models was tested to determine whether the selected explanatory variables are correlated with the probability of a country incurring IMF arTears. and to investigate the impact of the different IMF-specific debt indicators in general and the role of past repayment behavior in particular. Table 2 presents the results of Models I A through I E.

Model I A includes no IMF-specific debt indicators. It thus attempts to explain the incidence of IMF arrears without reference to a country's financial relations and credit history with the IMF. In Model I A. all but three of the independent variables are significant and of the hypothesized sign.•� DebtGDP. the economic growth rate, and the 1 990 dummy variable have no explanatory power for IMF arrears in the specification of Model I A. Both the amount (ARRED1) and the existence (EXTARR, .) of an·ears to creditors other than the IMF are correlated with the likelihood of incurring IMF arrears. The coefficient on impons. the expected sign for which was ambiguous. i negative. indicating that a higher share of import in GDP makes IMF arrears less likely. The regression correctly classifies I ,297 of the I ,307 occurrences of timely repayment to the IMF, but misses about half of the cases of an·ears.

Model l B adds UFCQU and REPX. !MF-specific variations of the debt burden and debt service due ratios. respectively. to the regression}7 These two variables are both highly significant and positively con-elated with the probability of arrears to the IMF. TDSX loses significance when these IMFspecific debt measures are included, and the significance of several of the other independent macroeconomic variables falls. The change in significance of the TDSX variable may mean that, given the IMF's preferred creditor status. along with the relatively small share of total debt and debt service owed to the IMF by most countries. the likelihood of a country meeting its obligations to the IMF in a given year is not greatly dependent on the amount

16 All references to significance in the text are to significance at the 95 percent confidence level. unless otherwise noted. In the tables of results. one asterisk indicates significance at the 90 percent level; two asterisks, significance at the 95 percent level. and three, significance higher than 95 percent.

17 The reader is reminded that REPX includes both currem and overdue obligations to the IMF.



of obligations due to other creditors. Including the lMF-specific variables greatly improves Lhe overall fit of Lhe model, as indicated by the lower Type I error of Model I B (34 percent. versus 50 percent with Model I A).1H

Model I C tests what information is provided by separating IMF debt service into current and overdue obligations by substituting REPCX and REPAX for REPX. Model I C shows improvement over Model I B through the smaller absolute value of its log-likelihood ratio.

Model ID adds the lagged dependent variable. IMFARR,_ 1 • to the measure of arrears, REPAX, to test the hypothesis that it is the existence of past protracted arrears to the IMF. rather than the amount of arrears. that best captures the impact of IMF repayment history on the likelihood of future arrears. The results of Model ID show that IMFARR, 1 is a more powerful explanatory variable than REPAX; its inclusion renders REPAX. as well as REPCX, insignificant. In this fourth specification. many of the variables that had previously been significant are either no longer so (RESM. XGDP, MGDP, REVGDP. INFL) or are less significant (EXTARR, _ 1 • PC!. EXPGDP. DUM84). In contrast. economic growth becomes significant once the likelihood of future IMF arrears is also conditioned on the event of past IMF aiTears. Also. UFCQU"s explanatory power is undiminished. That only one variable related to non-IMF debt (EXTARR,_ 1 ) remains significant may reflect the distinctiveness of IMF obligations from the bulk of a country's external debt.

Model lE indicates that omitting REPCX and REPAX from the model while retaining !MFARR, 1 actually improves the model's predictive power, with the number of correctly classified arrears cases increasing from 79 to 82. A likelihood ratio test of whether REPCX and REPAX provide additional information to IMFARR,_ 1 in determining the likelihood of IMF arrears confirms that REPCX and REPAX may be omitted from the model.

The results shown in Table 2 suggest that the use of TMF credit, repayment history both to the IMF and to other creditors. per capita income, and economic growth are determinants of likelihood of a country" s incurring lMF arrears in the future. The non-lMF-specific debt-service ratio, reserves. exports and imports. fiscal indicators, and inflation appear to be con·elated with the likelihood of IMF arrears. but they do not provide significant explanatory power once information on a country"s past record in meeting lMF financial obligations is included. The IMF repayment history indicator is the most significant of the explanatory variables in Model I E. and use of IMF credit is rhe second-most significant correlate of IMF arrears. Of

18 Type I error i n this analysis is the proportion of actual arrears cases that are incorrectly classified as nonarrear cases. that is, as false negatives. A Type 2 error representS the occurrence of false positives, that is. actual nonarrears cases classified by Lhe model as arrears cases.


6 1 0 L YNN A YLWARD and RUPERT THORNE

Table 2. Tests of IMF-Specijic Debt and Repayment Hist01:r Indicators

Model l A I B I C I D l E

Number o f country-year observations 1 .410 1 .410 1 .410 1.410 1.410

Log likelihood -194 -1 40 -126 -97 -99

Coefficient Variable (Description) ( Statistical significance)

Constant l . l I 0.79 -0.09 -J.J 5 -1.35

EDTGDP (total debt/GDP) 0.15 0.06 0.24 0.24 0. 1 1

TDSX (total debt �ervice/exports) 0.32 0.12 -0.34 -0.35 -0.04 **

UFCQU (use of IMF 0.52 0.59 0.75 0.89 credit/quota) ** *** *** ***

REPX {total IMF 7.70 ob I igations/expom) ***

REPCX (current IMF 4.08 1.53 obligations/exports) **

REPAX (fMF obligations 46.63 1 1 .70 in arrears/exports) ***

IMFARR,_ 1 (binary indicator of IMF 4.16 4.75 arrears. lagged one year) *** ***

ARREDT (total non-IMF arrears/debt) 2.25 2.91 1 .95 1 . 1 9 1 .24 ** **

EXTARR,_1 (binary indicator of non-IMF 3.32 2.19 1 .97 1.48 1 .39 arrea rs. lagged one year) *** ** ** ** *

RESM (reserves/imports) -0.80 -0.45 -0.57 -0.30 -0.29 *** ** ** * *

XGDP (exports/GDP) 5.26 5.27 5.29 2.99 1.95 * * * * *

MGDP (imports/GDP) -5.47 -4.73 -5.73 -4.33 -3.36 *** ** **

PC/ {per capita income) - 1 .00 -0.93 -0.86 -0.75 -0.79 *** *** *** ** **

GROWTH (rate of change of PCI) -0.03 -0.24 -0.73 -6.53 -7. 1 3 * * **

EXPGDP (expenditure/GDP) 4.27 7.84 10.03 8.06 6.81 ** ** *** * * *

REVGDP (revenue/GDP) -7.34 -9.83 -13.32 - 1 0.75 -8.61 ** * * ** * *


COUNTRIES REPAYMENT PERFORMANCE VIS-A-VIS THI:. IMF 6 1 1

Table 2. (concluded)

Model l A I B I C I D l E

Number of country-year observation� 1.410 1 . 4 1 0 1.410 1,410 1 . 4 1 0

Log likelihood -194 -140 -126 -97 -99

Coefficient Variable (Description ) (Statistical signi Acance)

INFL ( inflation rate) 0.64 0.78 0.68 0.44 0.42 ** *** **

DUM84 ( 1984 dummy) 2.82 2.38 2.03 1.73 1 .85 *** *** *** ** **

DUM90 ( 1990 dummy) -0.57 -0.47 -0.99 -0.79 -0.45 * **

Number of' actual arrears observations 103 103 103 103 103 Of which: number of predicted 5 1 68 7 1 79 82

cases of arrears

Number of actual nonarrea rs observations 1.307 1.307 1.307 1 .307 1 .307 Of which: number of predicted 1,297 1.299 1.302 1.296 1 .295

cases of nonarrea rs

Type I error (in percent)" 50 34 3 1 23 20

Type 2 error (in percent)h 0 0

Notes: Asterisks appearing underneath coefficients indicate signifkance: no asterisk indicates the coefficient was not significant; one asterisk. that the coefficient was significant at the 90 percent confidence level; two asterbks. that it wa� significant at the 95 percent level; and three asterisks, that it was significant at the level of 0.998 or higher.

As explained in the text, all independent variables are lagged one year, except for REPAX. which is lagged two years by construction. However. only the two dependent variables are given the '·1- I " subscript. to distinguish that they appear on both the leftand right-hand side of the logit equations.

"Type I error represents the occurrence of false negatives. that is. actual arrears cases classified by the model as nonarrears cases.

hType 2 error represenL� the occurrence of false positives. that i�. actual nonarrears cases classified by the model as arrears cases.

course, the lagged dependent variable is to some extent capturing and summarizing the impact of those macroeconomjc variables that it renders insignificant; the role of members' past performance in repaying the IMF is further addressed below through the formulation of a state-dependent model. It is notewotthy that per capita income has explanatory power in modeling IMF arrears, in contrast to the results of several studies of nonIMF arrears sutru11arized in Table I . The results for the dummy variables suggest that the likelihood of IMF arrears rose across the board-that is. for a given set of outcomes for financial and economic indicators-after 1983, but did not necessarily fall after 1989. even though the actual number of arrears cases declined.


6 1 2 LYNN A YLWARD and RUPERT THORNE

A State-Dependent Design for the Model

An implicit assumption i n this and other econometric creditworthiness studies using pooled time-series and cross-sectional data i s that each country-year observation represents an independent event in which the outcome-in this case either the incurrence of arrears or the timely meeting of debt service due-is not correlated with the outcome in previous country years. However. factors related to the likelihood that a country incurs arrears to the IMF in a coming year may differ depending on whether a country has a track record of timely repayments. or instead already has overdue obligations to the IMF (and may therefore either exit or remain in arrears). For example. for countries with no arrears to the IMF. factors relating to current resource availability such as reserves might be important indica10rs of IMF repayment performance. while for countries already in protracted arrears to the IMF. low levels of reserves may be so widespread that there is not sufficient variation among countries and over time for this variable to significantly explain exits from arrears. Such state dependence could also be explainable i n term� of the IMF's strengthened cooperative strategy for dealing with overdue obligations. In particular. countrie� deemed to be cooperating with the IMF i n seeking a solution t o their Fund arrears may benefit from the IMF's rights approach. or other procedures that. inter alia. may allow them gradually to improve their financial and economic situation: in this case. correlation could result between variables that signal cooperation with the IMF and the probability of exiting arrears to the IMF.

With these considerations in mind. the same logit model was run for two different subsets of the database, the first consisting or observation� where the country had not been in arrears to the IMF i n the previous year ( i .e .. where IMFARR,_ 1 = 0). designated the ·'entry" subset, and the second consisting of observations where the country was already overdue to the IMF (IMFARR, 1 = I ). designated the "exit'' subset. The whole database and the two subsets were tested against all valid independent variables. and then variables insignificant in all three of the specifications were sequentially eliminated. The purpose of this process was to derive a common specification. so that a formal test could be made of whether the entry and exit models are i n fact distinct. The results i ndicated that for members that have no recent h istory of IMF repayment problems. the use of IMF credit. reserves, per capita income. growth. and inflation are cor

related with the probability that they will fall into arrears in the coming year. For members that are already incurring overdue obligations to the Fund. only UFCQU and ARREDT are significant i n determining whether they will remain in or exit a state or arrears. A maximum likelihood test


COUNTRIES' REPAYMENT PERFORMANCE VIS-A-VIS THE IMF 6 1 3

indicates with a confidence level of 99 percent that the entry and exit models are distinct.19

The Determinants of Arrears to Other Creditors

Table 3 presents the results of investigating whether the likelihood of two distinct types of external debt repayment problems-arrears to the IMF versus arrears to other creditors-is rooted in the same economic and financial variables. Model 2A is comparable to Model I A: it tests the same independent variables but omits the Jag of the dependent variable, which in this ca�e is EXTARR,_ 1 • Most of the financial and economic variables are significant. the exceptions being REPAX. IMFARR, 1 • GROWTH. DUM84.�0 and DUM90. In contrast to the results for IMF arrears. the export and import variables have the expected signs. that is, higher imports (export�) are associated with a higher (lower) likelihood of arrears to other creditors.

Model 28 adds the lagged dependent variable. EXTARR, 1 • to the regression. Similar to rhe results of Model I D when IMFARR, 1was added to the regression on IMF arrears, the lagged dependent variable is highly signil1-cant. and its inclusion causes the significance of many of the independent variables in Model 2A to either decline (XGDP) or disappear (EDTGDP. TDSX. UFCQU. MGDP. PC!. GROWTH. INFL). Model 2C presents the more parsimonious specification that emerges when insignilkant variable� are sequentially excluded according to the absolute value of their /-statistic. The external debt burden. currem IMF debt service. non-IMF-specific repayment history. reserves, exports. and both government expendirure and revenue are significant determinants of the likelihood of the incurrence of arrears to creditors other than the IMF. It is noteworthy that while UFCQU is not significant in the models i n Table 3. REPCX is. and bears a negative sign. This indicates that the higher the ratio of current IMF obligations to exports in year 1. the less likely are arrears to other creditors in year 1 + I . This result might seem counterintuitive. given that the IMFs preferred creditor status implies that other creditors get paid only after IMF obligations are met. However. it could reflect the small ratio of IMF debt service to total debt service for most countries. which means that in most case� timely payments to the IMF are unlikely to preempt payments to other creditors. Another interpretation is that higher current IMF obligations are

19The full results arc reponed in Aylward and Thornc ( 1 998). 111 As noted earlier. DUM82 was the preferred dummy indicator of the debt crisi!>. but was not used in tests of IMF arrears because of the few incidences of IMF arrears in the database before 1 982. Both a DUM82 and a DUM84 variable were tested against the probability of arrears to other creditors. and both were found 10 be insignificant.


6 1 4 L YNI'\ AYLWARD �nd RCPERT TIIORNI:

Tahle .3. Te.lt.l' Using Arrears lo Other Crediwr.1 ll.\ thl' Dependent Variable

Model :!A 2 B 2C

Number of country-year observations 1.31)8 I .W8 1.398

Log likelihood -502 -248 -255

Cocf'ticicnt Variable C De:-.cription ) (Statistical :-.igni flcance I Constant 0.06 -0.49 - 1 .44

**

EDTGDP (total debt/GDPI 1.08 0.7 1 1 .08 *** l•:tc

TDSX (total debt 'en·icc/e'lports) 1 .72 0.% ""**

UFCQU cu�c of IMF credit/quota) 0 30 0.08 ,,. i'*

REPCX (current IMF obligation�/cxportsl -8.05 - 1 0.77 -U9 *** �* tt'

R£PAX ( IMF obligation:- in arrcar�/expons) -5.07 17.08

IMFARR, 1 (binary indicator of IMF 1.96 1 .06 arTe<u·-.. lagged one yeur)

ARRF:DT ( total non-IMF arrea r,/total debt ) 24.64 -::!.03 *** ;>.

EXTARR, I ( binar� indication or non-ll\IF 5.77 5.71 arrear�. lagged one yeun �*''- f:*;l:

RI: Ski ( rc!'-cn·c,/i m portS) -0.1\) -IUO -O.JO *** .;.:* .,� .:t

XGDP (cxpom/GDPl -3.43 -3.6 1 -2.98 **.;.-. �* **"'

MGDP ( irnpon,/GDPJ 1 .67 1 .47 **

PC! (per capita income) -0.28 -0. 1 5 •::;-

GROWTH (rate of change of PCI) -1 .9::! -O.Yll EXPGDP (cxpcnditure/GDPJ 3.76 4.7::! 5.75

*';, ! ;"; *.0:.•

REI'GDP Crc\'enuc/GDPl --1.-12 -7.37 -8. 1 1 ** :?. �: * * -�

INFL C intlation rate) 1 .2 1 (J.!\1 ***


COUNTRIES'

REPAYMENT PERFORMANCE VIS-A-VIS THE IMF 6 1 5

Table 3 . (concluded)

Model

Number of country-year observations

Log likelihood

2A

1 .398

-502

:m 1.398

-248

Coeflicient

2C

1 .398

-:255

Variable ( Description) (Statistical �igni ficance)

DUM84 ( 1984 dummy)

DUM90 ( 1990 dummy)

Number of actual arrears observations

Of which: number of predicted cases of arrears

Number of acwal non-an·ears observations

Of which: number of predicted ca .. e� of nonruTear�

Type I error (in percent)"

Type 2 error ( in perccnt)1'

0.24

0. 1 2

622

478

776

74

:23

-l

-0. 1 8

-0. 1 9

622 622

572 572

776 776

754 753

8 8

3 3

Notes: Asterisks appearing underneath cocfllcients indicate significance: no asteri�k indicates the coefficient was not significant: one asterisk. that the coefticient was significant at the 90 percent conl1dence level: two asterisks. that it was significant at the 95 percent level: and three asterisks. that it was significant at the level of 0.998 or higher.

As explained in the text. all independent variables are lagged one year. except for REPAX. which is lagged two years by construction. However, only the two dependent variables are given the ··1 - I" subscript. to distinguish that they appear on both the leftand right-hand side of the logit equations.

"Type I error represents the occurrence of false negatives, that is. actual arrear� case� classified by the model as nonarrears cases.

�>Type 2 error represents the occurrence of false positives. that is. actual nonarrears cases classified by the model as arrears cases.

indicative of a past IMF adjustment program. and, all things being equal, the conditionality attached to such a program improves the member's creditworthiness and makes timely debt service to other creditors more likely.

fn contrast to the models of IMF an·ears. both Type I and Type 2 error� in Models 2B and 2C are low. The lower incidence of Type I errors in part reflects the higher proportion of arrears cases in the sample for the models of arrears to other creditors.

IV. Conclusions

The results presented above provide evidence that there are a small number of financial and macroeconomic factors that are leading indicators of the likelihood of a member country f.alling into arrears to the IMF. The


6 1 6 L YNN A YLW ARD and RUPERT THORNE

coefficients on the significant variables are generally �table in terms or sign and magnitude from one model specification to the next. the exception being the distinct entry and exit models. Type I error rates are acceptable and Type 2 error rates are minimal. Thus. even though countries that seek and obtain an IMF-supported financial arrangement tend to exhibit common economic characteristics related to the balance of payments problems that lead them to request IMF assistance. it is possible to distinguish econometrically, ex post, between the financial and economic characteristics of those countries that use IMF financial resources and the much smaller subgroup of countries that subsequently encounter difficultie� in repaying those resources on rime. A larger use of IMF credit relative to quota. a lower level of per capita income. and a slower rate of economic growth are correlated with a greater likelihood of a country fai ling to meet current financial obligations to the IMF.

Three other implications of the results may be highlighted. First. when IMP-specific financial indicators are omitted, a broader set of variables is correlated with the likelihood of a country incurring IMF arrears. and this set bears considerable overlap with the variables identified in other studies as correlates of non-IMF repayment problems. Adding IMF-specific financial variables to the specifications improves the performance of the modeb (i.e .. more an·ears cases are correctly predicted as such. for a lower Type I error). In particular. adding the lagged dependent variable as an indicator of past IMF repayment behavior lowers the Type I error rate. as well as rendering some of the macroeconomic variables insignificant (and allowing for the model's autoregressive properties). This result confirms in particular the relevance for this study of the work of McFadden and others ( 1985) and Hajivassiliou ( 1 989) on external debt repayment to non-IMF creditorsthat is, the best predictor of whether a country will go into arrears to the IMF in the future is whether it has been in arrears to the IMF in the past.

Second, the results provide support for a state-dependent model of repayment petformance. This specificatio111 of the model was ba�ed on the conjecture that once a country enters protracted a1Tears on its obligations to the TMF. it has begun a new phase in its external financial relations. and the likelihood that it will be in arrears to the IMF in subsequent years may be determined by different factors than when it was still current with the IMF.

The third set of results we would like to highlight involves the comparative results for fMF ve rsus non-IMF an·ears. When both lMF-specific and non-fMF credit history variables are included and parsimonious specifications are used, the macroeconomic correlates of IMF mTears found in this analysis tend to reflect more a country's overall economic circumstances and policy stancethat is. per capita income and economic growth (Model I E), while the correlates of non-IMF an·ears seem rather to be indicators relating to the availability


COUNTRIES'

REPAYMENT PERFORMANCE VIS-A-VIS THE IMF 6 1 7

of and claims on the financial resources that could be used to service debt namely, reserves, exports, and government expenditure and revenue (Model 28 or 2C). These results could be indicative of the IMFs preferred creditor status: a country may rep�y the IMF even when its reserves and financial resource flows have diminished to levels where it has decided to slow or cease repayments to other creditors. Instead, countries that fall into arrears to the rMF are distinguished by low income and slow growth, suggesting that the intractability of the macroeconomic imbalance problems may be a major factor.

There are rwo possible explanations for the fact that the amount of debt and debt service due to creditors other than the IMF has no explanatory power for the likelihood of IMF repayment problems, once information on lMF-specifi<.; financial variables is included. It may reflect the IMF s preferred creditor status. Alternatively, it may be a result of the conditionality attached to the use of IMF resources. in that while the probabi lity of a non-IMF creditor being repaid falls as debt or debt service due rise�. the likelihood that the IMF will be repaid is less sensitive to these traditional key creditworthines!-. indicators. Another implication of the results with regard to IMF conditionality may be drawn from the finding that a higher ratio of curTent IMF financial obligations to exports is associated with a lower probability of arTears to non-IMF creditors. As indicated in Table I , some researchers have found that IMF financial support is associated with a lower probability of external debt repayment problems. while others have found the reve rse. These earlier studies tested the impact of the existence or not of a higher-tranche IMF adjustment program as a binary qualitative variable. rather than the impact of the amount of IMF financial obligations, on the likelihood of repayment problems to other creditors. Our result-that additional debt service owed to the IMF makes it more. rather than less. likely that other creditors will be repaid-could reflect the efficacy of IMF conditionality leading to a positive shift in the fundamental creditworthiness of the country.

REFERENCES

A vramovic. Dragoslav. 1964. Economic Grou·tll and Enema/ Debt (Baltimore. Maryland: Johns Hopkins Pre5!>).

Aylwru·d, Lynn. and Rupert Thorn e. 1998. "An Econometric Analy!>i!> of Countries· Repayment Performance to the International Monetary Fund:· IMF Working Paper 98/32 (Washington: International Monetary Fund).

Berg. Andrew. and Jeffrey D. Sachs. 1988. "The Debt Crisi�: Structural Explanations of Country Performance:· Journal of Del'elopme/11 Economics. Vol. 29 (November). pp. 271-306.

Eaton. Jonathan. Mark Gersovitz. and Jose ph E. Stiglitz. 1986. "The Pure Theory of Country Risk:· European Economic Revie11·. Vol. 30 (June). pp. 48 I -5 I 3.


6 1 8 LYNN A YLWARD <tnd RUPERT TI-IORNE

Edwards. Sebastian. 1984. "LDCs' Foreign Borrowing and Default Risk: An Empirical Investigation. 1976-80:· American Economic Review. Vol. 74 (September). pp. 726-34.

Elmore. Catherine. and George McKenzie. 1992. "Predicting LDC Debt Arrear� ..

(unpublished: Southampton, England: University of Southampton).

Euh. Yoon-Dae, 1979. Commercial Btmks and 1he Crediiii'Orlhiness of Le.B De1·e/oped Cowuries (Ann Arbor. Michigan: UMI Research Press).

Fcder, Gershon, Richard Just, and Knod Ros�. 1981 , ''Projecting Debt Servicing Capacity of Developing Countries .

.. Journal of Financial and Quanlilafil·e

Analysis. Vol. 16 ( December). pp. 651-69.

Feder. Gershon. and Lily V. Uy, 1985, "The Determinants of International Creditworthiness and Their Policy Implication�." Journal of Policy Modeling, Vol. 7 (Spring). pp. 133-56.

Frank. Jr., Charles R., and William R. Cline, 1971, "Mea�urement of Debt Service Capacity,

.. Joumal of lmermuional Economics. Vol. I (August), pp. 327-44.

Greene. William H .. 1993, Econome1ric Analysis (New York: Macmillan, 2nd ed.)

Hajivassiliou. V .A .. 1989. "Do the Secondary Markets Believe in Life After Debt?" World Bank Policy Planning and Research Working Paper WPS 252 (Washington: World Bank).

Haque. Nadeem U., Manmohan S. Kumar. Nelson Mark. and Donald J. Mathie!>on. 1 996, "The Economic Content of Indicators of Developing Country Creditwonhines�:· IMF Working Paper 96/9 (Washington: lmernational Monetary Fund).

International Monetary Fund. 199 1-95. Annual Reporl (Washington: IMF).

Li, Carmen A .. 1 992. "Debt Arrears i n Latin America: Do Political Variable� Matter?" Journal ofDel·elopmelll Studies. Vol. 28 (July), pp. 668-88.

Lloyd-EIIis. Huw. George W. McKenzie, and Stephen H. Thoma!>. 1990, "Predicting the Quamity of LDC Debt Rescheduling:· Economics Letters. Vol. 32 (January), pp. 67-73.

Maddala. G.S .. 1 983, Econometrics (New York: McGraw-Hill).

Mcfadden. Daniel, and others, 1985. "Is There Life Afler Debt? An Econometric Analysis of the Creditworthines� of Developing Countrie� .

.. in l111anational Debl and the De1·e/oping Countries. ed. by Gordon W. Smith and John T. Cuddington (Washington: World Bank).

Saini. Krishnan G .. and Philip S. Bates, 1978, "Statistical Technique� for Determining Debt-Servicing Capacity for Developing Countrie�: Analytical Review of the Literature and Further Empirical Result�:· Federal Reserve Bank of New York Research Paper No. 781 8 (New York: Federal Reserve Bank of ew York).

---. 1 984. '"A Survey of the Quantitative Approaches to Country Risk and Analysi� ... Journal o.lBanking and Finance, Vol. 8 (June), pp. 341-56.

Sargen. Nicholas, 1977. '"Economic Indicators and Country Risk AppraisaL" Economic Re1•iew of the Federal Reserve Bank of San Francisco (Fall). pp. 19-35.

Solberg, Ronald L.. 1988. Sovereign Rescheduling: Risk and Porifolio Management (London: Unwin Hyman).


11\JF Staff Papers Vol. �5. No. � !December 19981

@ 1998 International Monetary Fund

Virtual Deficits and the Patinkin Effect

ELIANA CARDOSO*

The paper del'e/ops a model of inflationary .finance thal defines !he fiscal deficit as a funclion of 1he l'irlllal deficit-one thal would be obser11ed at zero inflation. /1 sludies the negmive relationship betll'een the inflalion ra1e and real governme111 expendilures-the Patinkin ef

fect-a poweJ:ful slabi

!i-;_er during megainflcuion. The model outpe1jorms other seigniorage models in explaining the persislence of jour-digil inflmion rcues 1ha1 11ner explode into an open hyperinflation. It also explains how apparenlly expansionist fiscal policies end ill measured real deficils !hat are small and COIIIpatible with the small amount of seigniorage thal can be col/ec1ed a! high i11f1ation roles. [JEL E I 0, E31. E58, E62]

. . . ill a silllation where-because of coalilion considermions-lhe finance minisfer does nof hm·e file poll'er fo force individual millisfries to make adequafe reduclions in fheir respecfive budgewry demands and is fhus CO!�{ronled wilh an overall budge! whose planned expendilllres far exceed ils expec1ed rel'enues, he may seemingly accept 1/use demands, and fhen finance fhe deficil by priming money and felling file resulling inflalioll enforce file necessan· reduclion in real governme111 expendifllr.es.

Don Patinkin ( 1993, p. 1 1 5 )

ECONOMISTS THINK of extreme innation as an unstable process, the instability reinforced by the Tanzi effect-a decline in real tax revenues

as inftarion rises. But empirical evidence suggests a powerful effect that runs in the other direction through declining real spending levels-the Patinkin effect. This paper introduces the concept of a virtual budget

* Eliana Cardoso is a Lead Specialist i n the Poverty Reduction and Economic Management Division of the World Bank's Latin America and the Caribbean Regional Office. When she wrote this paper she was an Advisor in the IMF's Research Department. She thanks Rudi Dornbusch, Stanley Fischcr. Roben Flood. Peter Isard, and Paul Masson for helpful comments.

619


620 ELIANA CARDOSO

deficit (a deficit that would be observed if inflation were zero). develop� a model of inflationary finance. and applies the model to the case of Brazil.

Observed aggregate budget data on nominaL operationaL and primary deficits contain very little information about the true fiscal position of the public sector when inflation exceeds 500 percent a year. The Tanzi effect predicts that real tax revenues decline as inflation rises and thus the budget deficit is higher at higher inflation rates. But there is also a reverse Tanzi effect-referred to here as the Patinkin effect . lf the Patinkin effect dominates at high inflation rates, real expenditures appear lower than they would be if there were no inflation. and real expenditures tend to increase when inflation disappears. Thus, the fiscal adjustment needed once inflation disappears is usually underestimated. Several factors explain this phenomenon.

• Real interest rates decline with increasing inflation rates and usually rise following stabilization. This rise in real interest rates contributes to the increao;e in real government expenditures once inflation disappears.

• During periods of high inflation. local governments usually delay payments of salaries and wages. When inflation exceeds 1 .000 percent a year, this delay produces a substantial decline in real expenditures. When inflation disappears. delaying payments no longer reduces real expenditures.

• Although governments have learned to lessen gaps in tax collections and to index delayed tax payments to inflation, they still program expenditures with a forecast for inflation that is usually lower than observed inflation. As a consequence, realized real expenditures are much lower than programmed expenditures. When inflation disappears. actual expenditures will be closer to their programmed levels.

• The inflationary revenue of state banks can finance credit subsidies that are not recorded. This revenue disappears when inflation disappears. Furthermore, if inflation conceals banks' weaknesses. and these weaknesses are accentuated by the rise in real interest rates that follows stabilization, the government will have to use fiscal revenues to rescue banks. and recorded real expenditures will increase with stabilization.

Because inflation reduces real expenditures but not real taxes when governments fully index taxes and reduce gaps in tax collections, inflation can be used to accommodate conflicting spending programs of different government levels. Thus. inflation produces operational budget deficits consistent with the amount of real seigniorage that the government needs to finance the deficit.


VIRTUAL DEFICITS AND THE PATINKIN EFFECT 621

Section I introduces the concept of a virtual budget deficit and the concept of a Patinkin effect.' It also develops a model of inflationary finance. The section discusses multiple equiiLbria and shows that. even if the virtual budget deficit exceeds the maximum amount of seigniorage that the government can collect. one stable high-intlation equilibrium exists if the Patinkin effect is strong enough. The model outperforms other seigniorage models in explaining the persistence or four-digit inflation rates in countries where innarion never explodes into an open hyperinflation. Furthermore. it explains how expansionist fiscal policies end in measured real deficits compatible with the small amount of seigniorage that can be collected at high inflation rates. The model can also accommodate the traditional analysis of explosive hyperintlations if the Parinkin effect is not strong or if indexing breaks down at extremely high inflation rates. The last parr of Section I discusses the shares in seigniorage accruing ro the central bank and deposit banks.

Section 11 applies the model to the case of Brazil, discussing the banking sector's share of seigniorage, interest rare spreads. and nonperforming loans following the Real Plan. instituted in mid-1 994. The reduction of the banking sector's share of seigniorage immediately after the stabilization was a consequence of the changes in required reserves. The increase in required reserves and the decline in seigniorage of the banking sector in part explains the increase in interest rate spreads. the high active real interest rates. and the increase in nonperforming loans after stabilization. Section l l i offers concluding remarks.

I. Budget Deficits and Inflationary Finance

Tanzi ( I 978) was among the first to explore the impact of inflation on tax revenues. He observed that a rise in inflation could increase or decrease real tax revenues depending on lags in tax collection, built-in elasticity. and indexation. ln general, tax collection lags in developing countries, where real tax revenues are assumed to decline as inflation rises, are thought to be long relative to those observed in industrial countries. A rise in inflation would thus increase the budget deficit in developing countries. a process known as the Tanzi effect.

But inflation also affects real expenditures. Bresciani-Turroni ( 1 937. p. 34 ) . one of the first economists to study the relationship between the inflation rate

1 The concept of the virtual deficit is mentioned by Fisc her ( I 994) in a footnote. He claims that there is a case for calculating a "zero-inflation deficit" (different from the operational deficit) because of the Tanzi effect and because the real interest rate might change if inflation were stabilized.


622 ELIANA CARDOSO

and the budget deficit, observed that as inflation accelerates, the relationship between the budget deficit and inflation can become negative:

. . . the German authors, who have maintained that the depreciation of the mark was the cause of the disequilibrium between income and expenditure (in Germany) because, given the imperfect adaptation of income to the monetary depreciation, the yield was diminished, have not considered that in the period now under examination the depreciation of the mark influenced both income and expenditure in the same direction. Computed in gold marks, the total expenditure also diminished considerably from July 1 9 1 9 to February 1 920 and more rapidly than the income.

Patinkin ( 1 993) shows how pressure among political coalitions can lead to the use of inflation to erode the real burden of conflicting nominal expenditure demands by different ministries, as in the case of Israel before 1985. Guardia ( 1992) reports that during high inflation years in Brazil. realized real deficits were always smaller than the programmed real deficits. According to Bacha ( 1994). programmed real expenditures in Brazil exceeded realized real expenditures because projected inflation was always less than observed inflation, and indexation of expenditures was always avoided.2

Government Spending, Deficits, and Inflation

How are spending decisions actually made. in a high-inflation country such as Brazil? In the early and mid-1990s, Brazil's treasury would collect actual federal revenues I 0 days at a time, allocate the constitutional shares to state and municipal governments, cover current interest on the public debt. meet the payroll for federal employees, and then allocate the remaining balance to investment and other current expenditures in proportion to congressional appropriations. Then, individual ministries would have discretion over which projects or programs to finance. This system created an arena for bargaining between the national administration and politicians. And bargaining became an important element in securing congressional support for legislation catering to pork and patronage interests of congress members. It also meant that actual real expenditures deviated from programmed real expenditures in significant ways.

Of course, in high-inflation countries some expenditures-such as wages-are indexed. Because indexation is imperfect and linked to past

2 Bacha ( 1994) also proposes a deficit finance model in which the budget deficit is represemed by a linear inverse function of inflation.



inflation, rising inflation implies declining real wages. Moreover, local governments in Brazil. for instance, used to postpone wage payments when they were short of cash. With double-digit monthly inflation, a 1 5-day delay in payments implies a significant decline in real expenditures. When the annual inflation rate reaches 4.000 percent a year-as it did in mid-1 994-a 15-day delay in payments reduces real expenditures by 15 percent.3

With some expenditures indexed and more rigid than others, one would not expect inflation to reduce all expenditures equally in real terms but to affect expenditures that are not subject to strict rules, such as investments by both government agencies and public enterprises. It is this negative relationship between high inflation and real expenditures that can be attributed to the Patinkin effect. When inflation exceeds 1 .000 percent a year, observed budget deficits reveal very little about the true fiscal position once inflation is curtailed. If inflation disappears and expenditure commitments remain unchanged. the virtual budget deficit would be much higher than the observed budget deficit at high inflation rates.

One could thus argue that the budget deficit increases with inflation when inflation is low and declines with inflation when inflation is high. At low inflation rates there may be no motivation to index taxes and reduce tax collection gaps, and the Tanzi effect will produce a positive relationship between deficits and inflation. In contrast, when inflation is high, there is a clear incentive to introduce indexation and reduce tax collection gaps. It can also be argued that once arrangements to avoid losses of tax revenues are put in place, they would continue to be used even if inflation were to decline. Thus. in countries with long inflationary histories. we would not observe a positive relationship between inflation and the budget deficit because the Tanzi effect would cease to work. Yet because indexation is perceived as a mechanism that perpetuates inflation, stabilization programs often introduce a clause forbidding indexation. as did the Real Plan in Brazil. In an attempt to eliminate the indexation habit, fines on delayed tax payments were no longer indexed to the price level. This policy could reintroduce the Tanzi effect and the positive association between inflation and the budget deficit at low inflation rates.

At the same time. in a country where political coalitions generate conflicting expenditure demands. inflation can be used to accommodate those demands. and the Patinkin effect becomes operative. Payment delays also start to have a significant impact on real expenditures. In these circumstances, if tax collection cominues relatively well, a rise in inflation will

' Ir payment is postponed by 15 days. real outlays are reduced by 3 percent if inflation is I 00 percent, by 7 percent if inflation is 500 percent. and by I 0 percent if inflation is I ,000 percent.


624 ELIANA CARDOSO

reduce the budget deficit. Nonetheless, under extremely high inflation rates, any indexation scheme would break down, and the possibility of declining real taxes and increasing deficits would reappear.

Formal Model

The inflationary finance model developed in this subsection is general enough to accommodate different scenarios. That is. at low inflation rates the budget deficit can be assumed to be increasing with inflation or it may be constant. At high inflation rates the model assumes that the Patinkin effect is operative and that it could dominate the Tanzi effect. At even higher inflation rates. indexation could break down and the budget deficit could once again increase with inflation. Thus, a cubic function is a natural candidate to express the share of the budget deficit in GDP as a function of the inflation rate in a form consistent with the stylized findings described previously.

Equation I shows the share of the budget deficit in GDP. g. as a function of the inflation rate, rr:

g = g(O) + cm3 + brr2 +err a > 0, b > 0. c > 0. ( I )

where g(O) i s the virtual deficit. The response of the budget deficit to the inflation rate in equation ( I )

depends on how strong the Tanzi and the Patinkin effects are at different levels of inflation. that is. it depends on the relative sizes of a. b, and c. Figure I shows three different possibilities.

l n the first case both effects are relatively modest (a. b, and c are small. that is, the budget deficit does not respond very strongly to inflation). Also. b is big enough relative to a and c to permit the Patinkin effect to dominate the Tanzi effect at annual inflation rates between 1 .500 percent and 4,000 percent. The case of a very strong Patinkin effect (a very big b) that would produce a downward-sloping schedule starting at low inflation rates cannot be ruled out but is not considered here.

In the second case. the Patinkin effect is strong enough to generate a budget surplus at annual inflation rates between 2,500 percent and 4,000 percent when the virtual deficit is 4 percent of GDP. This case is of interest if we consider Brazil's experience. Between 1 990 and 1994, when inflation averaged close to 2,000 percent a year, the primary surplus was 3.5 percent of GDP. It exceeded 5 percent of GDP in 1 994 when inflation reached 2,500 percent. The operational balance was also in surplus in 1993 and 1994. In 1995, when inflation fell to 15 percent. the 1994 operational surplus turned into an operational deficit of approximately 5 percent of GDP. Appendix I discusses in more detail the empirical evidence on the relationship between



Figure I . The Response of the Deficit to Inflation

Deficit as share of GDP a = 0.000005; b = ..{).0004: c: = 0.008 0.15 .----------------------------,

0.10

0.05

g (0) = 8% ------- / - - / _ .. - · - · - · - · - - -... .... · - .

-0 ,. - - - 8 ( ) = 4% _ _,. -----

.... .. - .. -.. g (0) = 0 .,.. .. -O F�--------------------------------��------��--------�

..{).05 L__ ___ _L_ ___ ___J ____ _L_ ___ ___J ____ _J_ _ _J 0 1,000 2,000 3.000 4.000 5.000

Inna1ion ra1e (percent)

Deficit as share of GDP (I= 0.0000 I : b = -0.00065: c = 0.008 0.15 .----------------------------,

0.10

0.05 - · - · - - - .. .... · - g (0) = 8%

-0.05 - • - . 8 (0) = 0 • � . '

I , - · - · - · - · - ·-I ..{). J0 '-------'------'------'------'------U

0 1 .000 2.000 3.000 4.000 5.000 lnnation rate (percenl)

Defici1 as share of GDP (/ = 0.000006: b = -0.0004: c = 0.008 0.25 ,....-----------------------------,

0.20

0.15

0.10

0.05

0 .,. 0

- � ·

8 (0) = 8%

- - - - - - g ( 0 ) = 4% -- ------- · - - - · - · - · - · - · - · - · - - -?!�� - - -

1,000 2.000 3.000 1nna1ion rate (percent)

4.000

Note: a, b, and c are parameters of equation J and g(O) is I he virtual deficit.

5.000


626 ELIANA CARDOSO

inflation. taxes. expenditures. and the different measures of the fiscal deficit in Brazil.

In the third case the Patinkin effect is not strong enough to generate a declining relationship between the budget deficit and inflation. That is, b is not big enough relative ro a and c to produce a downward-sloping schedule at any level of inflation.

Tanzi and Patinkin effects are sho:rt- and medium-run stories about the monetary authorities, the tax authority, and the spending authorities revising their inflation predictions at different speeds or revising them more slowly than the private sector revises inflation expectations. Eventually, all sectors of government will face their imprecise predictions and try to correct them. Appendix l l discusses the government's corrections of its inflation predictions and the implications for a long-run equilibrium.

Equation (2) expresses seigniorage collected by the central bank as a function of the inflation rate:4

6H/Y = j..L:!I•(rr) &1· /orr > o. (2)

where t::.H/Y is the ratio of the increase in the monetary base to income. :: is the ratio of the monetary base to money. and v is velocity, with a Cagantype velocity functional form:5

v = v(O)ekn_ (3)

Since the budget deficit is financed by money creation

g(g(O), rr) = 11�/1·(rr), (4)

it follows that the required money growth to finance the budget is 1-l = g (1'1::). Money growth increases when the budget deficit, g, exceeds the amount of seigniorage, z(!llv), generated by the current rate of money growth:

(5)

• To obtain equation (2). define the share in income of seigniorage collected by the cemral bank as s = t::.HIY. Given l1H = ::.fl.M, where ::. = the inverse of the money multiplier (or the ratio of H to M). substitution yields s = :.fl.MIY. Divide and multiply this expression by M. define t::.MIM (money growth) = J.l. and assume that the money supply i s equal to the demand for money. MIY = 1!1-(n:).

' In the simulations, the parameters of the Cagan function are consistent with those observed in Brazil. where between 1 950 and 1 995 the central bank' s seigniorage averaged about 2 percent of GDP. bull never exceeded 4 percent of GDP. even at four-digit inflation rates. Appendix I l l contains the empirical evidence on velocity i n Brazil between 1950 and 1995.



Figure 2. Money Growrh and lnjlarion

Money growth (percent) a = 0.0000 I: b = -0.00065: c = 0.008: � = 0.18: : = 0.4

2.000 .----------------------------,

1.500

1,000

5.000

Constant money growth ifg (0) = 2%

/ / ,"' """" / ' /

/ / / / /

., ,-" Inflation = money growth

0�----�----�------�----�------�----�----� () 250 500 750 1,000 1.250 1 .500 1.750

lnnation rate (percent)

Note: a. b. and c are parameters or equation (I). k and : are defined in equations (2) and (3).

In Figure 2 the constant money growth schedule, 8j.!l8t = 0. crosses the vertical axis at the point where money growth generates enough seigniorage to finance the virtual budget deficit, g(O). If inflation is low and the Tanzi effect is strong, the schedule slopes upward but declines with inflation once the Parinkin effect becomes strong enough. The schedule would once again reverse its slope at even higher inflation rates (not represented in Figure 2). Above the curve representing constant money growth, 8j.!/81 = 0, money growth exceeds the amount of seigniorage needed to finance the budget deficit and is declining. Below the curve, money growth is not sufficient to finance the budget deficit and is increasing.

ln a model with perfect information and no uncertainty, expected inflation is equal to observed inflation. Inflation inertia exists. nevertheless. as a result of formal and informal indexation mechanisms, and inflation moves slowly to catch up with monetary growth:

8rr:./8t = El(j.! - rr.). (6)

The constant inflation rate. 8rr./81 = 0, is represented in Figure 2 as the 45° line from the origin. Above it the rate of inflation is lower than the rate of money growth, and inflation is rising: below it the rate of inflation exceeds the


628 ELIANA CARDOSO

rate of money growth and inflation is falling. lnflation and money growth are constant and equal at the poim where the two schedules cross. lhar is. where

(7)

Observe that the necessary condition for a stable equilibrium is that the constant money growth schedule cross the 45° line from above. Figure 2

shows three equilibria for lhe set of parameters of the budget function shown in the middle panel of Figure I and a vinual deficit of 2 percent of GDP. The parameters of the velocity function are those of Brazil. an economy that has been demonetized by a long inflationary history. There is one stable equilibrium at an inflation rate equal to 190 percent, one unstable equilibrium at an inflation rate equal to 790 percent. and another stable equilibrium at an inflation rate equal to I ,600 percent. There is also a foul1h unstable equilibrium at an inflation rare in excess of 4.500 percent (not represented in Figure 2).6

Expansionary Fiscal Policy

An expansionary fiscal policy is defined here as an increase in the virtual budget deficit-that is. a policy that increases g(O). lhe difference between expenditures and revenues under a zero inflation rate. If the virtual budget deficit is small relative to the amount of seigniorage that can be collected in the economy. multiple equilibria will atise. As fiscal policy becomes more expansionary. the schedule that shows constant money growth shifts upward (Figure 3). If the virtual budget deficit is higher than the amount of seigniorage that can be raised at any inflation rate. and the Patinkin effect is very strong. the schedule showing constant money growth crosses the 45° line once from above. This equilibrium is stable. If the Patinkin effect is not strong, the schedule showing constant money growth does not slope downward and expansionary fiscal policies would lead to hyperinflation.

If the Patinkin effect is strong. there is one stable equilibrium for a large range of budget deficits even if seigniorage collection is small and the revenue-maximizing inflation rate generates an amount of seigniorage that is less than the virtual budget deficit. Convergence to this stable equilibrium is through oscillations. As fiscal policy becomes more expansionary. money growth increases ahead of inflation to generate more seigniorage. Inflation catches up with money growth and then exceeds it.

6The fourth unstable equilibrium is also important. If a shock. such as a very big devaluation, drives inflation above the unstable equilibrium, a fiscal contraction or reserve requirement shift that is apparently in the right direction could �et off a hyperinflation.



Figure 3. Money Grou·Jh and lnjlmion Equilibria Under Di[ferenl Vimwl Deficits

Money growth (percent)

3.000

2.500

2,000

1.500

1.000

500

Inflation = money groll'th

0��---L----�------��---L--��L------L-L--� 0 500 1.000 1.500 2.000 2500 3.000 3.500

Inflation rat.: (percent)

Note: Each curve assumes constant money growth.

Expansionary fiscal policies. which induce an increase in the virtual budget deficit in excess of maximum seigniorage, increase the steady-state inflation rate (Figure 3). In the new steady state, the share of seigniorage in GDP and the share of the realized budget deficit i n GDP are smaller than in the initial steady state, as both decline with inflation when the Patinkin effect dominates the Tanzi effect. This model describes the experience of megainftation countries, such as Brazil from the 1980s to the mid-1 990s or Israel before 1985. better than other models of seigniorage. in which very expansionary fiscal policies-policies that imply a virtual budget deficit in excess of optimal seigniorage-result in open hyperinflation.7 l n Brazil and in Israel inflation was used to reduce real expenditures and inflation remained at megainftationary levels for long periods without ever exploding into open hyperinflation.

7The literature on seigniorage defines optimal seigniorage Hl> that obtained at the revenue maximizing rate of inOation. If the virtual deficit increases above optimal seigniorage. there are cases where no equilibrium exists. For the parameters in this subsection. there is no equilibrium for virtual deficits in excess of 9 percent.


630 ELIANA CARDOSO

Reserve Requirements and Inflation

The model assumes that the ratio of the monetary base to M I , �. is a constant. This subsection examines this assumption more closely. An increase in the ratio of required reserves to deposits raises :c and should in principle increase the central bank's share of total seigniorage, reducing money growth and inflation.8 If the Patinkin effect is operative. the reduction in inflation would increase the budget deficit. which would be financed by higher seigniorage in the new equilibrium. The higher seigniorage, in return, is a result of the decline in velocity in response to the lower inflation rate. Thus, an increase in the reserves-to-deposit ratio can produce lower inflation even with an unmodified fiscal policy (unchanged g(O)).

The combinations of the long-run equilibrium inflation rate and the central bank· s seigniorage share, that is, equation (7)-given the virtual budget deficit, g(O), and demand for money, v(7t)-are shown in Figure 4. 1f the virtual budget deficit equals 2 percent of GDP and the central bank gets twofifths of the seigniorage revenue (�= 0.4 ). four possible equilibria exist (the fourth. unstable equilibrium at inflation in excess of 4.500 percent is not shown in Figure 4). Among the three equilibria. those corresponding to the low inflation rate and the high inflation rate are stable, while that con·esponding to the average inflation rate is unstable. Starting from a stable equilibrium, an increase in the reserves-to-deposit ratio moves the central bank's share in seigniorage upward and reduces inflation.

If the virtual deficit is very high-for instance, g(O) = 9 percent-the possibility of using the required reserves-to-deposit ratio to reduce inflation practically disappears. Inflation then becomes very inelastic with respect to z because the amount of seigniorage that can be collected at high inflation rates is very low. Thus. increasing the central bank's share of a very small amount of seigniorage (because inflation is very high) will not significantly increase the amount of the budget that can be financed by seigniorage.

If the virtual budget deficit is not so high, but still high relative to optimal seigniorage collection-for instance, g(O) = 8 percent of GDP-the schedule becomes discontinuous. But it is still possible to observe different inflation equilibria at very high required reserves-to-deposit ratios. If the economy is stuck at a high inflation equilibrium, a monetary reform that curtails inflation combined with an increase in the reserves-to-deposit ratio can move the economy from a high-inflation equilibrium to a low-inflation

� Before calculating the tax on cash balances in Austria. Germany. Greece. Hungary, Poland, and Russia after World War I, Cagan ( 1956) observes that institutions other than the government typically receive some of the revenue from issuing money.


\'IRTL'AL DEFICITS AI\'D Tllfo P\TI'\KI'I U-H-.CT 6}1

F1gurc -1 . The Ce111ral Bun/... \ !:>hart' in Set-.:11wrage and fllf/alloll

11 = 0.0000 I : ll = 0.00065: , = O.OOX. 1- = 0.1 X

1 .6 ,.-----------------------------, " " "

.. . . . .. .. " " ·. "

1 -l , ,

I �

n.x

0.1

()

()

e4uilibrium. I! i:-- not clear hO\\ long thi.., ne\\ equilibriulll can be -,u-..twnl·d

if the higher re:--ene:---to-depo'>it ratio reduce� the profih or depo-..il hanf..-.. �ub!-.tantiall) . The itl\:rea�e in required re:-.ene-.. '' i l l abo ha\e an impact on intere:-.t rate :-.pread:-. that depend on the a' cragc co:-.h o f f und:-. and the Jc, cl of re\ef\ e requirement),. J f both -..pread-.. and real i ntcre\t rate\ i ncrea-..e \\ 1th �tabi I inllion. nonperforming loam, m a) al'o increa-..e and I urther con1ribu1e to reducing bank•; ' profitabi l i t) .

1 1 . Reserve Requirements, Interest Rates Spreads, and Nonperforming Loans After Stabilization in Brazil

l n mid- 199-L Brazil' ... Real Plan reduced inflation u-.. ing three t�pe-.. of reform-..: a -,hon-li,ed fi:..cal adju),tment. a monetary reform. and the u-..e of the exchange rate a;, a nominal anchor. A temporary monctar\ reform mea,urc linked l'Oiltract�. price'. wage-... and the exchange rate to a -..1 11gk daily e-..calator and unit of account. the unidade real de \'11/or. The adJu\l· ment. \\'hich began on March I . 1 99-.f. Ja,ted four month,. The ccnlral bank e'>tabli�hed a daily parit) between the cru;eiro real and the unidwle


632 ELIANA CARDOSO

real de 1•alor based on the current rate of inflation. as reflected in the three most closely watched price indice�. Since the cruzeiro real and the unidade real de ,·a/or depreciated relative to the U .S. dollar at roughly the same rate, most prices and contracts were implicitly set in U.S. dollars. On July I. 1 994. a new currency. the real, was introduced by converting contracts denominated in unidades real de l'alor into reals at a rate of one to one. Brazil's success in bringing clown inflation was associated with real exchange rate appreciation. This pattern. similar to that observed in other countries where the exchange rate was used as a nominal anchor runs as follows: there is a real exchange rate appreciation. a rise i n real wages. a deterioration in external accounts. an economic boom. and then a slowdown.

Brazil's stabilization was supported by tight monetary policy that was based on an increase in reserve requirements. The increase in required reserves and the decline of inflation led to a substantial decline in the inflationary revenues of deposit banks (Tables I and 2). The required reservesto-deposit ratio increased from an average of 26 percent during January-June 1 994 to an average of 64 percent during November 1 994-Apri l 1995.'� With the increase in reserve requirements following implementation of the Real Plan. the share in total seigniorage seized by the central bank (:::. the inverse of the money multiplier) increased from an

average of 60 percent during January-June 1 994 to an average of 84 percent in the period January-June 1995. As a consequence, the �hare in GDP of seigniorage seized by deposit banks fell from 2 percent to close to zero (Figure 5).10 This decline is consistent with estimates by the lnstituto Brasileiro de Geogrqfia e Estatfstica ( l BGE. 1997) calculated using a different methodology (Table 2). IBGE calculated banks· inflationary revenue in two steps. First. the difference between the monthly average stock of non-interest-bearing-liabilities and the non-interest-earning assets. NNL. was multiplied by the monthly inflation rate of the general price index (!GP-DJ) to obtain the inflationary revenue. Then monthly revenues were added and the annual sums were divided by GDP. Although IBGE found higher inflationary revenues in deposit banks. the magnitude of the change in the seigniorage revenue of deposit banks between 1993 and 1 994 was 2 percent, as it is here.

?Jn the second half of 1994 the required reserve�-to-loans ratio increased from 0 percent to 15 percent and the required reserves-to-savings deposit� rose from 20 percent to 30 percent. and in May 1995 required reserves-to-time deposits also increased from 20 percent to 30 percent (source: Brazil's central bank). 10 The share in GDP of deposit banks' seigniorage is: ( I - ;.)D.M IIGDP = [ { I - RID)I( l + CID)]D.M 1/GDP = (D.M I - D.H)IGDP. where RID is the reservesto-deposit ratio and CID is the currency deposit ratio. Thi� share was calculated using changes i n the average money balances during the year.


Institution

Central bank Deposit banks Total

VIRTUAL DEFICITS AND THE PATINKIN EFFECT

Table l . Seigniorage in Bra:.il. 1950-95

Average (percentage of GDP)

2 . 1

1 .8

3.9

Standard deviation ( percem)

7.5

7.5

1.2

Source: Banco Cemral.

633

Notes: Total seigniorage is calculmed as the increase of the annual average M I relative to the annual average M I in the previous year. Seigniorage collected by the central bank is equal to the increase of the annual average monetary base inclusive of all noninterest-bearing required reserves relative to Lhe same variable one year before. The seigniorage collected by deposit banks is equal to the difference between total seigniorage and seigniorage collected by the central bank.

Table 2. lnjfarional)' Re1•enue of Pril•are and Public Banks in Bra:) I. 1990-95

(Percentage of GDP)

Deposit banks 1990 1991 1992 1993 1994 1995

Private banks 1.4 1 .4 1 . 7 1 .6 0.7 -0.0

Public banks 2.6 2.4 2.3 2.7 1 .3 0.0

Total 4.0 3.8 4.0 4.2 2.0 0.0

Source: I nstituto Brasi leiro de Geogratia e Estatistica ( 1997).

Figure 5. Bra:ilian Banks · Seigniorage Re1·enue as Percemage oJCDP. 1950-96

Share of GDP

0.08 .---------------------------,

0.07

006

0.05

0.04

O.O:l 0.02

0.01

0 U---------------------------------------------------� 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995

Source: Banco Central do Brasil, Boleti111 do Banco Cenrral (monetary base from Table 1 1 . 2 and non-interest-bearing required reserves from Table 1 1 .4)

Notes: Total seigniorage = change in annual average M I. Central bank'� seigniorage = change in annual average monetary base + non-interest-be<lring required reserve�. Deposit bank's seigniorage = difference between total and central bank's �eigniorage.


634 ELIANA CARDOSO

The reduction of the deposit banks' share in total seigniorage is explained by the increase in required reserves. Following the srabilization. the rise in required reserves not only contributed to the increase in seigniorage collection by the central bank, but also explains in part the increase in spreads between passive and active rates. This spread increased from 4 percent a year in January-June 1994 to 86 percent a year in January-June 1995 (Campelo. 1 997). Required reserves were gradually reduced, but other factors contributed to keeping the spreads high-such as taxes on financial transactions and the increase in nonperfonning loans motivated by the increase in real interest rates. Nonperforming loans doubled from an average of 7.8 percent of total loans during July-September 1 994 to an average of 15.6 percent during February-August 1 997.

Seigniorage collected by banks did decline with stabilization but seigniorage collected by the central bank did not-at least. not immediately. In 1 993, the pea.k inflation year, seigniorage collected by the central bank was 1 .8 percent of GDP. It increased to 3 percent in 1 994, the year of the Real Plan. and was 2 percent in 1 995-the level of average seigniorage during the high-inflation years. 1 1 This evidence supports the view that the decline in the inflation rate was achieved through the monetary reform, the fixing of the exchange rate, and tight monetary policy. Stabilization was not achieved through a tightening of fiscal policy, which would have reduced financing of the deficit through seigniorage collected by the central bank. A more balanced policy would not have transferred the revenues from money creation so drastically from deposit banks to the central ba11k and would have avoided the increase in interest rate spreads and nonperforming loans.

Ill. Concluding Remarks

The Patinkin effect contributes to the understanding of sustained extreme inflation rates. Using parameters of Brazil"s velocity function between 1950 and 1 995, and evidence from the relationship between inflation and fiscal deficits in Brazil, the paper simulates an inflationary model in which extremely high inflation rates are stable and do not explode into open hyperinflation.

The paper also argues that in analyzing inflation stabilizations. attention should be paid to the virtual deficit-an estimate of what the deficit would

11 Because a decline in total seigniorage collection was matched by a decline in seigniorage collection by the commercial banks. leaving seigniorage r.:ollected by the central bank unchanged. there was not a wealth effect from the decline in inflation. but only a transfer between the banking sector and the nonbanking sector. In 1996, though, the cemral bank's seigniorage did decline to I percent of GDP.



be if inflation were reduced to zero. The virtual deficit is differem from the operational or inflation-adjusted deficit, which deducts the decline in the real value of government debt caused by inflation from the nominal deficit. because it takes into account both the Tanzi and Patinkin effects and because the real interest rate mjght change if inflation were stabilized.

Following stabilization, fiscal adjustment may have to be more severe than projected. since inflation clouds structural fiscal problems if expenditures are not indexed and the Patinkin effect is strong. Moreover, the high real interest rates that follow stabilization expose banks· weaknesses. which demand fiscal resources for restructuring. If many public banks have accumulated bad loans to local governments. sustainable reform wi 11 require an even harsher fiscal effort. By mid-1 997, a fiscal adjustment that could sustain recently achieved low innation had not yet been undertaken in Brazil.

APPENDIX I

The Budget Deficit and Inflation in Brazil

In the model developed in the paper. two imporrant empirical relationships play a role in determining equ i l ibria: the effect of inflation on the budget deficit and the response of veloci ty to inflation. This Appendix examines the empirical evidence on the relationship between the budget deficit and inflation, and then studies the relationship between vdocity and inflation in Brazil during 1 949-95.

An analysis of Brazil" s public finances relies on three concept� of fi�cal balance: the public sector borrowing requirement (PSBR), the operational balance, and the primary balance. The PSBR is equal to total revenues less total expenditures of the public sector, which includes all government levels. the central bank. and public en terprises but excludes state and federal banks. Traditional analysis uses the PSBR-which peaked at 83 percent of GDP in 1989 (Table A I )-to assess the impact of the government" s actions on aggregate demand and inflation. But the PSBR may not be the appropriate measure of the deficit in countries with high inflation and a high ratio of domestic public debt to GDP (see, for instance. Blejer and Cheasty, 1993). l merest payments rise with the increase in the inflation component of the nominal interest rate on the domestic debt. But these increased payments represent compensation for the erosion of the real value of the debt principal. Payment of the innation component of the nominal interest rate i s thus equivalen t to a financing item used to amortize the public debt. l t follows that a more meaningful measure of the deficit should exclude the payment of the inflation component of the nominal interest from the PSBR. Exclusion of this component yi elds the operational balance, which uses the real interest rate in calculating interest payments. Because real interest rates are sensitive to monetary policy and the level of activity in the economy and because interest payments are the result of deficits incurred i n previous years, a narrower definition


636 ELIANA CARDOSO

Table A I . Bra::.il: Public Sector Balance. 1983-96 (Percentage of GDP)

Public sector Primary balance Operational balance borrowing (PB) Real interest <OB= PB - RIP)

Year requirement ( - = deficit) payments (R/ P) (-= deficit)

1 983 19.9 1.7 4.7 -3.0 1 984 23.3 4.2 6.9 -2.7 1 985 28.0 2.6 7.0 -4.4

1 986 1 1 .3 1.6 5.2 -3.6

1987 32.3 - 1 .0 4.5 -5.7

1988 53.0 0.9 5.7 -4.8 1 989 83.1 -1.0 5.9 -6.9

1990 29.6 4.6 3.3 1.3 1991 27.2 2.8 2.8 0.0 1992 44.2 2.3 4.5 -2.2

1993 58.1 2.6 2.4 0.3 1994 43.8 5 . 1 3. 8 1 . 3

1995 7 . 1 0.4 5 . 1 -4.8

1996 6.1 -0.1 3.8 -3.7

Source: Funda�iio Getulio Vargas, 1 997.

Figure A I. Operational Budget Deficit as Share of COP and ll�flation. Bra�il. 1981-96

Operational budget dclicit/GDP CpcrccntJ

8 .--------------------------------------------------,

• • -1 L--------L------�---------L------�--------�------�

0 500 1 .000 1.500 2.000 2.500 3.000 lnnmion rate (percent)

Source: Banco CenLral do Brasil.


VIRTUAL DEFICITS AND THF PATINKI:-.l EFFECT 637

of fiscal balance. which exclude� interest payment� from expenditures. could rellect more clearly the discretionary budgetary �tance. Thi� mea�ure i� the primary balance.•�

The relationship between the primary budget deficit and inflation in Brazil during 1983-96 wa� negative: at very high inflation rates. the primar) balance wa� in <;urplus. The relationship between the operational budget deficit and inflation abo seem� LO be negative (Figure A l l. There b no con�istcnt information that would allow the calculation of operational deficits before 1981. ruling out observation� for periods of low inllation. Among the 1 5 observations for the operational budget deficit, only 5 correspond to inflation rate:-. below 200 percent. The 15 observation:-. arc too few to permit meaningful empirical results. In a <;implc ordinary lea't square!> regression. the relationship between the budget deficit <tnd inflation i� negative and thus consi�tcnt with a <,trong Patinkin effect.

Even though data for the operational deficit do not exist before 198 I, data for some components of expenditure� and taxe!> are available. To '>hm\ that the exi\ting information is con�istent with the hypotheses in Section I of the paper. empirical tests should reject the hypothesis that there is an inver5e relation!>hip between taxes and inllation and reject the hypothesis that there is a positive relation'ihip between investment �pending and inflation. since investment expenditures are easier to cut than wages and �alaric�. The relation!>hip between con!>umption expenditures and inllation is trickier: part of these expenditures (such a' wage<; and salaries) were indexed umil recently. Furthermore. there could be a po�itivc relation�hip between expenditures and in nation with cau-;ality running from expenditures to inflation.

Table A2 show� the re�ult� of unit root te�t� for the �hare� in GDP of income taxes. �ales taxc�. government con�umption expenditure�. and inve�tment by public enterprise;,. 1' The hypothesis that the <.hare in GDP of income taxes. \ales taxe�. and government·� consumption expenditures have a unit root cannot be rejected. The Dickey-Fuller statistic for the share of investment by public cntcrpri-;cs i n GDP reject� the unit root hypothesis at the I pcrcelll and 5 percent lcvcb.

Table A3 shows the results of cointegration tesh for income taxe.,. �ale-, taxe,. and inflation. The tests reject any cointegration at the 5 percent significance level. The evidence doe:-. not 'upport the exi;,tence or a T::uvi effect during the highinflation years i n Brazil.

11 A difficult issue. which the fbcal figure� in Table A I do not reflect. concern\ the quasi-fiscal deficits i n federal and state banks. which could be �ub\l<lntial. For instance. the federally owned Banco do Brasil has traditionally subsidized credit to agriculture. and the National Bank of Development ( BNDES} :-.ubsidizcs credit to exporters. In 1 996 the treasury rccapitalizcd Banco do Bra�il by 7.9 billion reals (more than I percent of GDPJ. This rccapitali.�ation ha;, contributed to the increase of total net public debt. e�timated to have risen from 30 percent of GDP in 1 995 to 35 percent in I 996. The costs of restructuring the banking 'ector and the impact or these changes on the fi�eal budget are not yet dear. And with the end of high inflation. bad loan:, from ;,tate bank� to ;,tate government\ have abo emerged as a serious problem.

1' Data for taxes and government conwmption expenditure-. arc from Bnllil'-. national accounts for the years between 1965 and 1994. except where noted. Income taxes are collected by the central government. The ;,ales taxe� arc valueadded taxes collected by ;,tale governments ( ICMS) and by the central go\'ernment (IPI). Data for investment by public enterprise� between 1980 and 1996 are from the Treasury Department of the Finance Ministry.


638 ELIANA CARDOSO

Table A2. Bra::.il: Unit Root Tests on Income Taxes, Sales Taxes. Govemment Consumption Expendi!Ures. and Public Enterprises · lnves/ment

Augmented Dickey-Fuller Test (equation includes intercept, change

Income taxes/ GDP

in lagged variable. and trend) -2.2365 I percem critical value -4.3226 5 percent critical value -3.5796 Phi lli ps-Perron Test

Sales taxes/ GDP

- 1 .6858 -4.3226 -3.5796

Public Government enterprises·

consumption/ investment/ GDP GDP

-1 .2540 -3.9658 -4.3082 -4.73 1 5 -3.5731 -3.761 I

(equation includes intercept) -2.8305 -2.4995 0.2289 -0.8000 I percent critical value -3.6752 -3.6752 -3.666 I -3.9228 5 percent critical value -2.9665 -2.9665 -2.9627 -3.0659

Notes: Sales taxes and governmem expenditure taxe� data from 1965-95. and pub I ic enterprises investmem data from 1 980-96.

Table A3. Bra�.il: Johansen Coimegration Tes1 S{(llislics for Taxes and ir!fiaiion. 1965-94

Like I i hood 5 percent I percent Eigeravalue ratio critical value critical value

Hypothesis: There is no coimegration 0.241 between income taxes and inflation

Hypothesis: There is no cointegration 0. I I 04 between sales taxes and inflation

I 1.25

6.026

Notes: The log likelihood is 43.33. The lag interval is I to I .

I 5.4 I 20.04

15.41 20.04

On the consumption expenditure side, the Johansen cointegration test indicates one cointegrating equation at the 5 percent significance level. and the relationship between consumption expenditures and inflation is positive (Table A4). But considering wage indexation and the reversed causality between expenditures and inflation. this result, even if i t does not support the Patinkin effect. is not surprising. A Granger causality test gives mixed results: with a one-year Jag. the test cannot reject the hypothesis that government spending does not cause inflation. But with a one-year Jag and a two-year Jag, the test rejects the hypothesis that government spending does not cause inflation. With a two-year Jag. the test indicates that there is a 22 percent probability that inflation does not cause government spending (Table A5).

There are fewer observations for investment by public enterpri�es than for the other variables, and thus the tests are weaker. StilI they reject the hypothesis of a unit root for the share of investmelll by public enterprises in GDP. The ordinary leastsquares regressions reported in Table A6 show that an increase in inflation reduces this share. The coefficient is significant and robust to different specifications.

The empirical findings are broadly i n line with the hypothesis of a negative relationship between real budget deficits and high inflation rates i n Brazil. Such



Table A4. Bra::.il: Johansen Coimegraiion Tesi for Govemmellf Consumption Expenditures and lnjicuion. 1965-95

Coimegration Test

Likelihood 5 percent I percent Eigenvalue ratio critical value critical value

Hypothesis: There is no cointegration Hypothesis: At most one cointegration

equation exists

0.686

0.044 1 .30

Normalized Cointegrated Coefticients (One cointegrating equation)

Government consumption expenditures

1 .00

Inflation

-0.0046

(0.0003)

19.96

9.24

24.60

1 2.97

Constant

-0.100

Notes: Standard erro rs are in parentheses. The log likelihood is 58.82. The Jag interval is I to I .

Table A5. Bra-:.il: Granger Causality Tests for Govemmellf Consumprion

Expendirures and Inflation, 1965-95

Null hypothesis Lags Observations F-statistic Probability

Inflation does not cause I (one year) 30 2.75 0. 1 1 0

government consumption expenditures

Governmem consumption I (one year) 30 0.44 0.510

expenditures do not cause inflation

Inflation does not cause 2 (two years) 29 1.59 0.220 government consumption expenditures

Government consumption 2 (two years) 29 7.81 0.002

expenditures do not cause inflation

a rel ationship stems from the interaction of two forces. First. Brazir s tax system has been continuousl y adjusted to protect real tax collections; collection lags are small, and until mid-1 994 late payments and fines were indexed. As a consequence. the ratio of tax revenues to GDP varied little despite enormous oscillations in in nation. The share of cenLral government revenues in GDP remained around 1 5 percent of GDP between 1 986 and 1993. when inllation peaked. I n contrast. not all expenditures were indexed and realized real expenditures were less than programmed real expenditures. Furthermore. the share of investment by public enterprises shows a significant negati ve relationship with inllation.


640 ELIANA CARDOSO

Table A6. Regressio11 A11alrsis

(Dependent variable: ratio of public enterprises· investment to GDP. 1980-96)

Constant

Inflation

Pub I ic enterprises· investment (- 1 )

Dummy for period after 1985

Adjusted R!

3.39 (5.69)

-0.07 (-1.89)

0.14

0.74 (2.5 1 )

-0.03 (-4.18)

-0.60 (-4.40 1 )

0.80 (7.74)

0.87

Note: /·Statistics in parentheses corrected by Ncwcy-West standard errors and eo variance.

APPENDIX 1 1

Government Predicted l11flation and the Long Run

Equation ( I ) leaves out inflation predictions by the government. If we define government-predicted inflation as A, eqlllation { I ) can be written as

(8)

The government revises its predicted inflation slowly:

(9)

In the long-run steady state. rr = A and g = g(O). In that case, the constant money growth schedule. o�.tRi1 = 0. is g(O) = ).1.::/F(-rr) and is represented by the upward-sloping schedule labcled as long-run constam mo11ey growth in Figure A2. If the virtual budget deficit is 2 percent of GDP and the simulation uses the same parameters used in the fonnal model Section I. there are two long-run equilibria corresponding to the intersections of the long-run constant money growth schedule and the 45° line in Figure A3. These two equilibria correspond to the two equilibtia found in seigniorage models such as those discussed in Mundell ( 197 1 ) or Bruno and Fischer ( 1986). for instance. In models where the inflation rate does not jump. the high-inflation long-run equilibrium is unstable. In the model in this paper, as the government's predicted inflation rate approaches the actual inflation rate. it becomes impossible to reconcile planned expenditures and revenues. and the economy moves into hyperinflation.

Using the same parameters used in the other simulations in the paper. Figure A3 il lustrates a situation in which the virtual budget deficit is 8 percent of GDP. As long as the government's predicted inflation rate is below the actual inflation rate and the Patinkin effect is strong. there is a high-inflation medium-run equilibrium. But there is no long-run equilibrium if the government's predicted inflation matches the observed inflation rate. as shown in Figure A3, where the long-run constant money growth does not intercept the 45° line.



Figure A2. Con sWill Money Gr011'1h in rhe Medium and Long Run

Money growth 2.000 .------------------------------.,

800

Medium·mn COIISIWII IIIOney grOII'Ih if g(O) = 2<k �--.,.

1.000

.·

.·

Long-run con.wam money growrh if g(O) = 2o/r

1.200 1,.100 1 .600 lnllntion rate (percent)

:

1.800

Figure A3. Comranr Money Grou·rh in rhe Medium and Lonf: Run

Money growth

1.200 r----------------,.-----------., Long-mn

1.000 growrh ifg(O) = 8% .·

.·

800

600

400

200 -- --- - ---

·.

Medium-nm COIISIWII money gmwrh

\ lfg(O) = 8%

. . .

45• line_ -·.-. -- - - .

0��---L------L-----�----�------�----�----� 0 500 1.000 1 .500 2.000 2.500 3.000 3.500

lnllation rnte (pert·em)


642 ELIANA CARDOSO

APPENDIX Ill

Velocity and Seigniorage

Velocity-defined as the ratio of income to Ml-increased in Brazil from 4.5 in 1950 to 32 in 1993, when annual inflation reached 2,700 percent. At the same time. the ratio of income to high-powered money peaked at 52. Basic descriptive statistics for velocity and inflation in Brazil between 1950 and 1995 appear in Table A 7 .

Figure A4 shows the very strong positive relationship between the growth rate of velocity and the growth rate of inflation during that period.

Unit root tests on velocity and inflation cannot reject the hypothesis that velocity and inflation have unit roots (Table A8). Velocity and inflation are endogenous variables in models in which the demand for money depends on expected inflation and inflation depends on money growth. These models predict that velocity and inflation are cointegrated. Cointegration tests strongly reject the hypothesis that there is no cointegration of velocity and inflation-that is, that velocity and inflation do not have an equilibrium condition that keeps their proportion constant in the long run. (Table A9) The logarithm of velocity and inflation are also cointegrated. and the normalized cointegrating vector shows a coefficient of -0.18. a value consistent with the theoretical prediction of a positive relationship between the logarithm of velocity and inflation, as in Cagan' s function for the demand for money (Table A I 0).

Seigniorage

The share in GDP of seigniorage collected by the central bank in Brazil has averaged 2 percent of GDP during the past 47 years. It remained unchanged in 1994-95 after the Real Plan succeeded in sharply reducing inllation. Seigniorage collected by deposit banks declined.

Total seigniorage. the revenue from money creation collected by the central bank and deposit banks. is defined as

TS, = M, - M,_1, ( lO)

where TS is seigniorage and M i s currency and non-interest-bearing demand deposits. The portion of seigniorage that accrues to the central bank corresponds to the change in high-powered money (currency and reserves). and the portion of seigniorage that accrues to deposit banks is the change in non-interest-bearing demand deposits minus reserves.

The ratio of total seigniorage to income is denoted by IS. In the steady state.

_ _!_[ ( I + .r)(l + rr) - 1 J Is - ,, (l + x)(l + rr) · ( 1 1 )

where v is velocity and x is the growth rate of real income.'� If the growth rate of real income were zero in the steady state. then, in the steady state. with constant velocity. total seigniorage would be equal to the inflation tax. 't = ( 1/v)[rr/( I + rr)j.

'•To obtain equation ( I 0). divide equation (9) by current income, Y,. then divide and mulliply by lagged income. In steady state v, = "' 1 = ''· and ( I + J..l) = ( I + rr)( I + x). where J..l is the growth rate of money, rr is the in nation rate, and x is the growth rate of rea I income.



Figure A4. Growth Rate <>(Ve/ociry and Grmrth Rate oflnjfatiOit in Bra:il. /950-95

0.3

0.2 0

O.I

0 -I

-O.I -2

-0.2 -J

-0.3 -4

-0.5 L...L..l....L...l-L...L..JL...L...J-L.l....L...l-L...l-I......__J-L.l....L...l-L...l-IL...L...J-L.l....L...l-L...l-IL...L...J-L.l....L..J...J.....L.....I--1...;L...J -5 I950 I955 I96() 1965 I970 I975 I980 I985 I990 I995

Table A 7. Bndl: Descriptil•e Statistics of Velocity and !l�flation. /949-95 (Number of observations = 47)

Velocity levels Inflation rate Measure ( Income/M I ) (Average percent per year)

Mean 1 1 .0 259

Median 7.0 39 Minimum 4.5 7 Maximum 32.0 2.700

Sources: Central bank and author's calculation�.

Table A8. Bra::il: Unit Root Tests on Velocity and Inflation. 1949-95 (Number of observations = 47)

Augmemed Dickey-Fuller test (equation includes intercept and change in lagged variable)

Phillips-Perron Test (equation includes intercept}

(equation includes intercept and trend)

" 1 percent critical value. h 5 percent critical value.

Velocity

-0.358

0.373

- 1 .649

Inflation Critical values

-3.332 -3.581"

-2.927h

-2.721 -3.578" -2.926/,

-3.3 1 9 -4.1 68" -3.5Q9i'


644 ELIANA CARDOSO

Table A9. Bra:il: Johansen Cointegrarion Test Staristicsfor Velocity and il!flation. 1949-95

Hypothesis:

Cointegration Test

Eigenvalue Likelihood 5 percent

ratio critical value I percent

critical value

There is no cointcgration 0.4977 33.03 1 9.96 24.60

Hypothesis: There is at most one cointcgrating equation

Velocity

1.0000

0.0444 2.04

Normalized Cointegrated Coefficients (One cointegrating equation}

lnlflation

-2.279 (0.354)

9.24

Constant

-6.4946 ( 1 . 1 16)

1 2.97

Notes: Standard errors are in parentheses. The log likelihood i� - 1 57.51 17. The Jag interval is I to I .

Table A 10. Bra:.il: Jolumsen Co;/llegration Test for Log (Velocity) and li!flarion. /949-95

Hypothesis:

Coimegration Test

Eigenvalue Likelihood

ratio 5 percent

critical value I percent

critical value

There b no cointegration 0.3909 26.94 19.96 24.60

Hypothesis: There is at most one cointegrating equation

Log ( Velocity}

1 .0000

0.0978 4.63

Normalized Cointegrated Coenicients (One cointegrating equation}

Inflation

-0. 1 8 1 3 (0.0370)

9.24

Constant

- 1 .842 (0.1319)

1 2.97

Notes: Standard errors arc in parentheses. The log likelihood is -39.63252. The Jag interval is I to I .


VIRTUAL DEFICITS AND THE PATINKtN EFFECT 645

Figure AS. Seigniorage and lnjlwion TtLr ofBra::.ilian Banks as Share ofGDP. /950-95

t950 t955 t960 t965 1970 t975 t980 1985 1990 1995

In the short run. money growth is different from the inflation rate becaul.e velocity moves and the growth rate of real income is different from zero. Thus. as expected. a comparison of short-run total seigniorage to GDP ratiO!. and short-run inflation tax to GDP ratios between 1950 and 1995 shows that the two ratios were never identical because the growth rate of money and the inflation rate differed. But during 1950-95 both short-run seigniorage and the inflation tax for the entire banking system, including the central bank, never exceeded 8 percent of GDP in any year (Figure A5). Total seigniorage collected by the central bank and deposit bank!> together averaged 4 percent of GDP.

REFERENCES

Bach a. Edmar. 1 994. ··o Fisco e a lnfla�ao: Uma I nterpreta<;ao do Caso Bra.�i leiro:· Rel'is1a de Economio Pof(Iica. Vol. 14 ( I ), pp. 5-17.

Blejer. Mario 1.. and Adriennc Cheasty. 1993. "How to Measure lhe Fi!.cal Deficit: Analytical and Methodological Issues·· (Washington: International Monetary Fund).

Bresciani-Turroni. Costarllino. 1937. The Economics of fl(f/aiion. translated by Millicent E. Sayers (London: Alien and Unwin).

Bruno. Michael. and Stanley Fischer. 1986 . .. Israel's Inflationary Process: Shock5 and Accommodation:· in The l.�raeli Economy: Maturing Through Crises. ed. by Yoram Ben-Porath (Cambridge. Massachusetts: Harvard University Press ).


646 ELIANA CARDOSO

Cagan. Phillip. 1956. "The Monetary Dynamics of Hyperinflation:· in Studies in rhe Quanriry Theory of Money. cd. by Milton Friedman (Chicago: University of Chicago Press).

Campelo. Jr.. Alofsio. 1997, "Por Que os Juros Nao Caem:· Conjumura Economica, Vol. 5 1 (June). pp. 35-6.

Fischer. Stanley. 1994, "Modern Central Banking.'" in The Future of Ce111ral Banking: The TercenrenaJ)' Symposium of rhe Bank of England. ed. by Forrest Capie and others (New York: Cambridge University Press).

Funda<;ao Getulio Vargas, 1997. Conjumura Economica (Rio de Janeiro: Funda<;ao Getulio Vargas. May).

Guardia, E.R .. 1992. ·'Or<;amento Publico e Polftica Fiscal: Aspectos lnstitucionais e a Experiencia Rccente" (unpublished M.A. dissertation: Sao Paulo: Universidade de Campinas).

lnstituto Brasileiro de Geografia e Estatfstica. 1997. Sisrema Financeirn. Uma Antilise a Parrir das Comas Nacionais. 1990-1995 (Rio de Janeiro: IBGE).

Mundell. Robert A .. 1971. "Deficit Finance and Growth." Chapter 4 in Monerary Theory: lnjlarion. Interest. and Growth in the World Economy (Pacific Palisades, California: Goodyear Publishing Company).

Patinkin, Don, 1993, ''Israel's Stabilization Program of 1985, Or Some Simple Truths of Monetary Theory," Joumal of Economic Perspectil·es. Vol. 7 (Spring). pp. I 03-28.

Tanzi. Vi to, J 978, "Inflation. Real Tax Revenue. and the Case for Inflationary Finance: Theory with an Application to Argentina," Sraff Papers. International Monetary Fund. Vol. 25 (September). pp. 4 1 7-51.


lMF Staff Papers Vol. 45, No. 4 (December 1998) <!:> 1998 International Monetary Fund

Anticipation and Surprises

in Central Bank Interest Rate Policy

The Case of the Bundesbank

DANIEL C. HARDY*

Market reaction to a change in official interest rates will depend on the extent to which the change is anricipated. and 011 how it is imerprered as a signal of jwure policy. In this paper, a technique is developed to separare the anticipated and unanticipared components of such changes, and applied to esrimare rhe response of euro-deutsche mark interest rares ro adjusrmenrs in the Bundesbank's Lombard a11d discou11t rates. LJEL E43, E47]

GOVERNMENT OFFICIALS. financial market participants. and agents in the economy at large attach importance to official central bank interest

rates. What are termed official rates typically comprise the rates applied at one or more central bank standing facilities and in some cases at which the central bank operates a regular tender. In most industrialized countries. as in a number of developing countries, the central bank determines these rates both to define the range within which it manages short-term interbank rates through on-going open market operations. and to signal its medium-term policy stance (see Borio, 1997, for a recent survey). A change in official rates can thus affect expectations that are reflected in longer-term interest rates and other financial market prices, and hence initiate the monetary policy transmission process. ft is therefore important that policymakers be able to predict the market response to such changes. Yet market participants have an incentive to anticipate policy shifts. and insofar as they succeed, market prices should largely adjust in advance of the implementation of a

* Daniel C. Hardy is a Senior Economisr in the Middle Eastern Deparrment. He thanks R. Flood, H. Herrmann, 0. lssing. J. Reckwert, S. Schich. K.-H. Todter. J. Zeuelmeyer, and participants at a seminar at the Deutsche Bundcsbank for their helpful comments.

647


648 DANIEL C. HARDY

change. Therefore, predicting the market response to changes in official rates requires that these changes be decomposed imo their anticipated and unanticipated components. Such a decomposition may reveal what aspects of a change in official rates influence expectations over different time horizons, and whether the central bank can achieve different ends depending on the nature and degree of forewarning that has been given of the change. These considerations were the motivation for this paper.

A number of past studies have looked at the impact effect of changes in official interest rates by the U.S. Federal Reserve (for instance, Lombra and Torto, 1977; Thorn ton. 1 986 and 1 994; Cook and Hahn. 1 988 and 1 989; and Radeck i and Reinhart, 1994 ), the Bank of England (Dale. 1 993 ), the Bank of Canada (Paquet and Perez. 1995), and recently the Deutsche Bundesbank (Nautz, 1995: Hardy, 1996). In most such studies the change in money market rates on the days surrounding a change in an official rate is simply regressed on the change itself. However. changes in market rates ought largely to reflect changes in expectations, based presumably on new inforn1ation. so it is important to distinguish between the anticipated and unanticipated actions by the central bank. These studies also usually limit their focus to the relatively rare instances when central bank rates were actually changed and neglect occasions when a change was thought possible but did not materialize.

One common approach to identifying anticipated changes in official rates, suitable for the United States and followed by Smirlock and Yawitz ( 1985) and subsequently others. is to categorize the changes as either .. policy·· and therefore unanticipated, or "technical" and anticipated. on the basis of published explanations of the central bank's actions. Even if the necessarily somewhat subjective evaluations are accepted, it may be inappropriate to regard actions as falling exclusively into one or other category. Roley and Troll ( 1 984) use ordinary least squares (OLS) estimation to predict changes in the U.S. discount rate. but they do not take into account the censored nature of the sample and achieve very low explanatory power. Skinner and Zettelmeyer ( 1 997) resort to the assumption that the change in the three-month interbank rate is a good proxy for the unanticipated policy change. Favero and others ( 1 996) calculate implicit forward rates, which they use in conjunction with the assumptions that the pure expectational model of the term structure holds and that the overnight rate is that controlled by the authorities to estimate market expectations of policy changes and reactions to surprises. The statistical properties of these estimates are obscure, in part because the errors cannot be taken to be symmetrically distributed: even when. say. some reduction in official rates is deemed very likely, the probability of a zero change remains positive.

Assessment of the effect of anticipated and unanticipated changes in official interest rates must start from a recognition that realized changes are


ANTICIPATION IN CENTRAL BANK INTEREST RATE POLICY 649

discrete and relatively rare events. However. there may be many occasions when market participants attach some probability to a change occurring in the near future. Market participants· expectations are unobserved. but information concerning them ought to be contained in market prices. In these circumstances. a limited dependent variable technique is appropriate to the estimation of the probabi lities attached by market participanrs to an increase or decrease in official rates. and the expected magnitude of the change. In this paper an appropriate technique is developed and applied. and indeed one of the aims of the paper is to examine how financial market variables retlect (short-run) expectations about central bank policy.' Anention focuses here on the relationship between interbank rates and the interest rates on the two standing facilities of the Deutsche Bundesbank. namely the Lombard and discount rates, but the technique could be applied to other financial market prices and other central bank rates (such as that on repurchase operations) and official rates in other countries.2

I. The Bundesbankts Monetary Instruments and Operating Procedures

The Bundesbank has long maintained both a Lombard and a discount facility (for details see Deutsche Bundesbank. 1 994).1 At the Lombard facility banks may obtain very short-term liquidity at relatively high interest rates, and at the discount facility banks may obtain a limited amount of funds for up to three months at a lower interest rate. The Bundesbank also conducts regular repurchase operations (''Pensionsgeschafte''), which are currently implemented through a tender every Wednesday. In addition. the Bundesbank occasionally organizes ad hoc "Schnelltender'' repurchase operations. and has i n the past issued securities to absorb liquidity.

Since 1985, repurchase operations have been the Bundesbank· s main vehicle for active liquidity management and the control of shon-term money market interest rates. Nonetheless. weight is still attached to the '·official rates ..

at the discount and Lombard facilities. Changes in these official rate5 are viewed. by the Bundesbank and others. as signals of its policy stance:

1 This paper is an extension of Hardy ( 1996). which concentrates more on the reaction of a wide variety of interest rates and exchange rates to overall changes in the Bundesbank' s Lombard, discount. and repurchase rates, and does not examine closely which variables reflect market expectations.

1 For instance, instead of interest rates. the exchange rate or a stock price index could be used to estimate the reaction in those markets.

� In the past the Bundesbank operated various other specialized facilitie�. but the rates offered on them were not generally regarded as indicative of the policy stance. The phrase "official rates" will be reserved for the Lombard and discount rates.


650 DANIEL C. HARDY

··Interest rate policy longer-run adjustment provides longer-run guidelines for interest rates in the money and credit markets. This applies particularly to changes in the discount and Lornbard rates" (Deutsche Bundesbank, 1 995, p. 97). The spread between the Lombard and discount rates effectively forms a band or corridor within which short-term money market rates fluctuate.4 However. the constraint is not rigid because borrowing at these facilities are not perfect substitutes for interbank borTowing. In particular. access to the discount facility is limited by quota, and in practice banks are reluctant to make very heavy and frequent use of Lombard loans, which ''should be extended only to bridge temporary liquidity needs and only if the size and maturity . . . seems appropriate and warranted" (Deutsche Bundesbank. 1994, p. I 02).

The discount and Lombard rates are reviewed by the Bundesbank Council in its morning meetings every other Thursday, and decisions on any change are announced that afternoon or early the next morning.5 Meetings are normally preceded by public discussion of the Bundesbank's likely actions. Changes have tended to be rare (with only 33 during the period 1985-95), with long periods of no change being interspersed with series of small changes all in one direction spaced over several years. Official rates are always changed in multiples of 1/, percentage point, and often the Lombard and discount rates are moved together. The width of the spread between the two rates varies but is typically about 2 percentage points. These characteristics of evolution of the Lombard and discount rates are apparent from Figure I . which also illustrates the path of one typical money-market rate.

Il. Model Specification and Estimation

Official Rates and the Term Structure

How would one expect market rates to react to a change in official interest rates or information about a forthcoming change? To obtain an intuition. suppose that on each day -r. -r = 0. I , 2 . . . . the central bank announces an

• A number of count.ries besides Germany share the approach of using two official rates to form a corridor for short-term rates; the decision has been taken that the European Central Bank will also have two standing facilities.

There is an analogy with an exchange rate floating within an adjustable band. The Bundesbank steers rates within the interest rale corridor with its repurchase operaLions. Similarly. a central bank can steer the exchange rate within a band with intramarginal intervention. Typically the market rate must get quite close to one or the other edge of the band before there is significant expectation of a revaluation.

�There were 266 Bundesbank Council meetings during the period 1985-95, which will fom1 the sample period. Occasionally the meetings are held on other days of the week or are missed due to holidays. The dates of meetings arc published in advance. Bundesbank Council meetings immediately preceded all changes in official rates.

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

Perc

ent

10 8 6

/

2

Janu

ary

1985

Fig

ure

I. L

om

bard

, Dis

coum

. and

Mar

ket I

nter

est R

ates

Lomb

atrl ra

te 3-

month

imer

est r

me

I..,

I I

I L-

, �-

--

�-

1 I I I

�--

-' _

__

__

__

r

-.J

I _

I I

Janu

ary

1990

Disc

mmtm

te

Janu

ary

1995

"0 0 r n -<


652 DANIEL C. HARDY

official interest rate s, that exactly determines the current overnight interbank interest rate. which will be de11oted r( I ),. The interest rate on interbank loans of longer maturity M is assumed to be determined simply by the expected average of the overnight rates over that time span. which equals the expected average level of the official rate:

( I )

where the information set available to market participants at time -r is represented by .0,, and E is the expectations operator.

If at date -r = 1 the central bank unexpectedly increases the official rate by an amount t)s and this change is believed to be permanent, all market rates should immediately increase by the same amount. If the change is expected to last for one period only, then the overnight rate r( I )1 should increase by t)s from period -r = 0 to 1 = I . but the change in the M-period rate should be only /1s( 1/M). If instead the central bank announces at -r = I that it will increase its official rate pem1anently by t)s at -r = 2. then the M-period rate should increase by /:::,s(M- I )/M from 1 = 0 to -r = I . and a further .0.s( 1/M) from 't = I to 't = 2, when the anticipated change is realized. The overnight rate will increase first in period -r = 2, but then by the full amount .0.s. It is easy to construct other scenarios where the change in the official rate is more or less anticipated and expected to be more or less permanent.

This illustrative model shows that if a change in official rates is expected to be relatively permanent. then the reaction of longer-term market rates will be relatively large. Longer-term rates should also react more to new� about future changes in official rates, and their movements in the period leading up to a possible change should reflect the accumulation of information on the central bank's intentions. Longer-term rates should react correspondingly less to the realization of anticipated events: the immediate, one-day response of longer-term rates should be caused almost entirely by the unforeseen component of a change in official rates. This differential response to anticipated and unanticipated events may be represented by the equation

[r(M),. - r(M),] (2)

= b0 + b1£(.0..s,I.O,) + b2[ E( .0.s, ln,.) - E( .0..1·, 1.0,)] + e, ..

where [r(M),. - r(M),] is the change in the M-period market interest rates between (the morning of) day 't and some other day -r': .0.s, is the change in official rates announced during a particular day 1 = r: and et is an error term. The immediate reaction of market rates to a change in official rates is given by the change from -r = r to -r' = r + I . For -r. t' < r, the equation is meant to capture



the effect of shifting expectations as information is released in the days before a possible change in centraJ bank rates. This inforrnation might include. for example. economic data relevant to predicting the policy reaction and also statements from officiaJs. For 't' > 1. market participants are assumed to have teamed the actual change in official rates. so (E(L�s, I n .. ) - E(!J.s, I flr)] = lM,- £([).s, l Qr)]. that is. the unanticipated component of the realized change. The equation is then meant to capture the effect of learning by market participants as they reflect on the central bank· s announcement of a change and pronouncements thereafter. and thereby assess the likely persistence of the change. Equation (2) is the main regression specification used in this paper.

According to the illustrative model, one would expect b0 = 0. One would also expect that b1 will be very small for longer maturities. Thus, expected changes in central bank rates should not significantly affect market rates except for very short maturities. The magnitude of the coefficient b! should reflect the market's interpretation of policy signals sent through the unexpected component of changes in official interest rates and other innovations in expectations. For example. a large estimate of b� for changes in rates on long-term assets from 't = 1 to 1' = 1 + I could indicate that the unexpected component of an increase in official rates is viewed as the start of a long period of higher interest rates. A large estimate of b2 for change� in market rates in the days before a possible change in official rates (that is. when -r. r' < 1) could indicate that shifts in expectations are important. and that the central bank can influence market rates strongly by releasing information on its intentions during this period. The magnitudes of both b1 and h2 should vary with the difference between 't and 1' and their relationship to time 1 when the official rates are changed. The greater the difference (-r' - 't). the longer market prices have to incorporate news. so generally one would expect b1• the coefficient on the unanticipated change, to increase.

Under the institutional arrangements established by the Bundesbank (see above), these qualitative relationships should still hold, but may be weakened. The Lombard and discount rates do not exactly determine market interest rates at any maturity. first because these bounds on market rates are not usuaiJy binding. and second because banks cannot freely and without limit arbitrage between these facilities and the money market. Furthermore. banks must meet reserve requirements as an average of daily positions. and therefore have considerable flexibility in managing their short-term liquidity. Hence. for example, they may try to build up their liquidity positions when they anticipate that interest rates will rise in the near future. Then even the overnight rate will be strongly intluenced by expected future rates. and the simple term structure model that was posited may not hold. All the�e factors would tend to decrease the market reaction to changes in official rates and especially to anticipated changes.


654 DANIEL C. HARDY

Estimation

Expectations and surprises are not observed, but under the assumption of rational expectations the realized decisions of the central bank should differ from expectations about them by no more than an uncorrelated, zeromean error term. An instrumental variables technique, implemented through multistage regressions, caru be used to deal with this "error-invariables" problem. The explanatory variables used in the initial stages need not reflect all available information perfectly. nor need financial markets be informationally fully efficient for this procedure to be valid, because the errors in the estimates of expectations will be orthogonal to other error terms by construction. Moreover, since changes in the central bank rates are discrete events, the dependent variable comes from a truncated sample, which demands application of nonlinear estimation techniques (see Maddala, I 983, for a survey). These procedures have recently been applied to study an analogous problem concerning expected realignments of exchange rate bands (see Edin and Yredin, 1993: and also Bertola and Svensson, 1 993). I n the case of German official rates the task is simplified by the fact that changes in the Lombard and discount rates occur only after Bundesbank Council meetings, the dates of which are known; the probability of a change on other days is zero. An extra difficulty, compared with the study of most exchange rate realignments, is that both increases and decreases must be considered.

Estimation proceeds in three stages (see the Appendix for details). First, an "ordered probit" model of changes in official rates is estimated by maximum likelihood. The dependent variable can be thought of as a set of dummy variables that identify when official rates were increased, decreased, or left unchanged following a Bundesbank Council meeting. Candidate right-hand side variables are those that may contain relevant information to the formulation of market participants' beliefs about the probability of the central bank increasing or decreasing official rates and are known at the time, or that reflect these beliefs. The results of this stage may themselves be of interest insofar as they suggest what variables indicate market sentiment and reveal a pattern in central bank behavior. The fitted values from the first stage are treated as the market's assessment of the probabilities of a forthcoming increase or decrease in official rates.

In the second stage, the estimated probabilities are appropriately combined with other candidate informational variables (with compatible dating) in a linear regression to generate a forecast of the magnitude of any change. The additional informational variables, or instruments, are meant to capture market expectations concerning the path of interest rates in the coming months and perceptions of the intentions of the Bundesbank. The series of


ANTICIPATION IN CENTRAL B.ANK INTEREST RATE POLICY 655

fitted values from the second-stage regression are taken as a proxy for the expected magnitude of any movement in official rates. The unanticipated component is simply the difference between the estimated market expectation and the realized change.�

In the third stage. the change in market rates is regressed on the estimates of the anticipated and unanticipated changes in official rates. a� in equation (2). The resulting coefficient estimates can be shown to be unbiased. and that of b2• to be efficient.

In the work that follows. the logarithms of interest rates are used instead of levels. This somewhat unusual specification was chosen to respect the restriction that interest rates cannot become negative. and the supposition that an interest rate change from, say, 3 to 3.5 percent might be more important for the economy at large than one from 8 to 8.5.7 The equations were also estimated using a simple linear specification and the results were not qualitatively different from those reported below.

Data Sources

Daily data were taken from Bundesbank publications on the Lombard and discount repurchase rates, rates on euro-deutsche mark deposits with maturities of I day. and of I . 3. 6. 12. and 24 months. and Frankfurt money-market rates for I day and 1- . 3-. 6-. and 1 2-month maturity interbank loans. from January 1 985 through January 1996 or as far back as available.H The dates of Bundesbank Council meetings were obtained and changes in market rates over surTounding days calculated. In particular. the changes from four days before a meeting to the day of a meeting or five days afterward (1 = 1 - 4 to -r' = 1 or 1' = 1 + 5, respectively), and from the day of a meeting to the next day (1 = I to 1:' = 1 + I ) will be reported; these time spans cover from one working week before Bundesbank Council meetings to one week thereafter.

Data are generally recorded at I :00 p.m. in Frankfurt except for the eurocurrency deposit rates, which are measured at I 0:00 a.m. by the BIS. Thus. a change in an official rate announced on a Thursday afternoon or Friday morning ought to act as a "surprise" affecting the difference between market prices

6 By construction. the expected and unexpected components are mutually orthogonal. Except for the need to estimate the probabilities of a change in official rate5. the procedure is simi lar lO instrumental variables estimation implemented through two-stage least squares. l t would be possible to estimate al l stages jointly. but the properties of such estimates would be less straightforward to establish.

7 The absolute level of interest rates and spreads may be of primary importance to financial market participants such as commercial banks.

s Hardy ( 1996) provides evidence that tJ1e reaction of market rates to changes in the Lombard and discount rates was significantly different before 1985 when open market operations were much less important.


656 DANIEL C. HARDY

recorded on Thursday and those recorded on Friday. Therefore. the corresponding information sets are dated 't = 1 - 4 and -r = 1.'1

Ill. Anticipated and Unanticipated Changes in Official Rates and Market Response

Overall Response lo Changes in Official Rates

To provide a benchmark with which to compare the effects of anticipated and unanticipated changes in official rates. changes in market rates were simply regressed on the total changes in official rates using the specification10

(3)

In the estimates reported here, the dependent variable is the log change in the euro-deutsche mark interest rate with maturities between I day and 24 months. 1 1 As mentioned above, the discount and Lombard rates are often moved together. which makes it difficult, given the sample. to distinguish the possibly different effects of changes in the two rates. Therefore. the average of the Lombard and discount rates is used as the explanatory variable.12

Results are presented in Table I . The constant term was always insignificantly different from zero and is not reported. The estimates of the parameter b1 suggest that the announcement of a change in official rates (at the end of day 1) usually had a highly significant effect on market rates for all marurities below two years. However, anticipation of such a change usually had a larger total effect over the days leading up to Bundesbank Council meetings (from 1 - 4 to r). Almost no effect is observed in the days following a decision. as indicated by the fact that the estimates of b1 for the change in market rates from t - 4 to t + 5 are all almost equal to the sum of the respective estimates for t - 4 to r. and t to 1 + I , the former. However. the change in market

� Estimates were performed for changes across a number of other time spans and based on other information sets. including information lagged one day. The results corroborated those reported here.

10 All estimation was performed using TSP Version 4.2 econometric software. 1 1 Similar results were obtained in estimates for Frankfurt interbank interest rates.

The technique has also been applied to estimate the effect of changes in the Lombard and discount rates on yields on German government securities, implied forward rates. exchange rates, the DAX share price index from the Frankfurt stock market. and interest rates outside Germany (see Hardy, 1996. for related results).

12 An alternative would be the average when the rates are moved together. and the change in the Lombard or discoum rate when one or the other alone was changed. Estimation results did not differ qualitatively when this construct was used as the explanatory variable.



Table I . Reaction of Euro-Deutsche Mark Rates to Chanf?eS in Ofjicial Rates

Change from Change from Change from 1 - 4 tO / I tO ! + I 1 - 4 to 1 + 5

"· R� b, R� b, R� s.e. s.e. s.e.

I day" 0.3266 3.651 0. 1035 1 . 2 1 7 0.44 1 5 2.2 13 (0.1035)** 0.'0464 (0.0574)• 0.0257 (0.181 0)** 0.08 1 1

I month" 0.2472 7.639 0 . 1 243 8.248 0.3059 6.582 (0.0556)** 0.0242 (0.0268 )** 0.0 1 1 6 (0.0746)** 0.0324

3 months" 0.2378 9.807 0 . 1 000 5.035 0.3066 7.148 (0.0466)** 0.0203 (0.0281 )** 0.0122 (0.0715)** 0.03 1 0

6 months• 0. 1 903 6.868 0.0980 7.322 0.2390 4.867 (0.0493)** 0.0203 (0.0245)** 0.0101 (0.0744)** 0.0305

1 2 months•' 0.1302 3.283 0.095 1 4.802 0.2338 4.877

(0.05 18)* 0.0206 (0.03 1 0)** 0.0123 (0.0757)** 0.0301 24 months• 0.122 1 2.258 0.0082 0.2 1 1 0.2380 6.683

(0.0637)* 0.0226 (0.0445) 0.0 1 58 (0.0712)** 0.0251

Notes: OLS estimation of [r(M), - r(M), I = b0 + b,t::.s, + e,.. Estimated coeflicient b,. standard errors in parentheses. percentage R�. and equation standard error reponed. Two asterisks indicate sign ificance at I percent; one asterisk indicates significance at 5 percent: a plus sign indicates significance at 10 percent.

"Number of observations = 266. "241 observations. ' 205 observations. '' 189 observations. r 163 observations.

rates from 1 - 4 to 1 + 5 tends to be marginally less than the sum of previous changes. suggesting that rates tend to revert slightly after the announcement.

The announcement day effect is nearly the same for interest rates of all maturities, but over longer time spans (from r- 4 to 1 or 1 + 5) the effect tends to decrease with maturity. The market for 2-year euro-deuL<>che mark deposits was reportedly not very active during the sample period, which may explain the slow responsiveness of these rates to changes in official rates.

Estimated Probabilities of Changes in Official Rates

Estimation of the anticipated and unanticipated components of changes in official rates and their separate effects on market rates can now proceed along the lines laid out in Section n. The first step is to estimate the probabilities of a change in official rates, for which purpose suitable explanatory or informational variables need to be found. These variables must represent information relevant to the prediction of the Bundesbank's actions that is publicly available at the appropriate point in rime. One consideration is that when a change in official rates is expected, money market interest rates should tend to move


658 DANIEL C. HARDY

in advance toward the new level. However. these rates are to some extent constrained by the operation of the standing facilities where the old rates still apply. Moreover. the Bundesbank tends to lead up to changes in the Lombard or discount rate with changes in the repurchase rate. which in turn steers money market rates in the appropriate direction. Therefore, the convergence of short-term rates to one or the other boundary of the interest rate "band'' may indicate that a corresponding shift in the band is expected. Another consideration is that the Bundesbank has tended to change official rates in "runs" of small changes fairly close together, so a change in one direction should make another in the same direction more likely. For the same reason also, the time elapsed since the last change may be informative.

After some experimentation it wa found that the differences between the three-month interbank rate and the discount rate and the Lombard rate, all in logarithms. were useful as informational variables to capture market sentiment concerning the likelihood of an official rate change (denoted by 1n(R3MID1SC) and ln(R3MILOMB), respectively).13 The difference of logarithms was chosen over levels to capture a nonlinear phenomenon, namely, that short-term market rates (and the repurchase rate) can fluctuate in a middle range between the discount and Lombard rates without signifying expectations of a change in the band. The logarithm of the last change in the average official interest rate. In LCHC, and the logarithm of the time in days elapsed since the last change, In LAPS, were included as a way to represent the tendency for rate changes to be positively serially correlated but spaced some weeks or months apart. Several other financial variables. for example, capturing the slope of the term strucliUre. were deleted from the list of informational variables because their influence did not seem to be robust enough to warrant the loss of parsimony. In principle, macroeconomic variables such as price and money supply developments could also have been used as informational variables. It is, however, difficult to determine exactly when these data became available, and insofar as they influence the expectations of market participants, prices should already reflect the information. An extension of this paper could consider the information contained in exchange rates and. were they available. quantity variables such as the stock of lending at the Lombard and discount windows. Variables observed at time 1 were used as instruments for the change from t to 1 + I ; the same variables dated 1-4 were used as instruments for the change from 1 - 4 to t or 1 + 5.

The results of the first-stage estimation are presented in Table 2 and illustrated in Figure 2, which depicts the estimated probabilities al each date t of an increase or a decrease in official rales (the latter shown as a negative

D The one-month interbank rate. the repurchase rate. and the lhree-month euro-deutsche mark rate were found to be about equally good instruments.



Table 2. Estimation of the Probability of Changes in Official Rates

Constant

ln(R3M/D/SC)

ln(R3M/LOMB)

lnLCHG

In LAPS

Constant

ln(R3M/DISC)

ln(R3MILOMB)

lnLCHG

In LAPS

Log likelihood

Based on information at time t - 4

Ba�ed on information at timet

Probability of a decrease -1.9834 -2.6990 (0.7140)** (0.9032)**

-8.5210 -10. 1404 { 1 .847 1 )** (2. 1 240)**

-4.5854 -6.3617 {2.2454)* (2.3889)*

2.3293 3.6356 ( 1 .7478) ( 1 .945 1 )+

0.3307 0.4765 (0. 1708)+ (0.2 1 1 Ol*

Probabili ty of an increase -4.5474 -5.4543 ( 1 .4 1 23)** ( 1 .7854)**

4.47 1 7 3.2692 (2.2091 )* (2.476 1 )

1 0.7469 1 7.0490 (4.5526)* (5.8442)**

-2.5418 -4.1598

( 1 .6970) ( 1 .8999)*

0.4222 0.67 1 1 (0.2079)* (0.2700)*

-81 .3872 -72.1383

Notes: Based on 266 observations. Estimated standard errors in parentheses. Two asterisks indicate significance at I percent: one asterisk indicates signitkance at 5 percent: a plus sign indicates significance at 1 0 percent.

number). and occasions when rates were in fact changed. The results seem plausible. For prolonged periods no change is expected. Expectations or an increase or decrease in rates tend to build up from one Council meeting to the next, peaking on the occasion of a realized change, and then falling to near zero.14 On only a few occasions did a rate change come as a complete surprise or a firmly expected change fail to materialize (lhe probability attached to a rise in rates in 1987 is due to turbulence following the stock market collapse of that year).

14 Estimated probabilities of exchange rate realignments, as reported in the articles cited. tend to display a similar pattern of asymmetric peaks.

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

Probabi

lity

1.0

-1.0

Ja

nuat)

198

5

Figu

re 2

. E

stim

ate

d P

rob

abil

itie

s of

Ch

ang

es i

n O

ffic

ial

Ra

tes

Incr

ease

A �

�A

Probabi

fil) of

incr

east'

� ,,

•

Janu

af) 1

990

"""

!

• �,

.. �

1\ �

••

........

\ • •

f

• �

• �

' ••

J •

,

• •

� ;)· ,

, r r

.: I', '

-..!,

, ll

I I

J I

lo . ··"'

,,..

: i �

I

• 11 �·

·• ,

I loo

�I I

� ;�

�;

...

. ; �

. ..

I

� �!

� �;

I "\

' 1

I I

' ""

\} I

'I Proba

bilif)

· of d

ecn!'

ase Ja

nuaf)'

1995

u

>

z


ANTICIPATION IN CENTRAL BA K INTEREST RATE POLICY 661

The estimated parameters are largely as expected (due to the estimation procedure. separate parameters are estimated for interest rate increases and decreases). When the three-month interbank rate is close to the discount rate (so ln(R3M/D/SC) is small or negative) a decrease in official rates is likely. When it is close to the Lombard rate (so ln(R3MILOMB) is large) an increase is likely. When the interbank rate is far from either official rate, neither an increase nor a decrease is likely. Therefore. ln(R3MILOMB) is significant in explaining the occurrence of reduction in official rates and ln(R3M/DISC) is significant in explaining increases. The sign of changes in official rates tends to be persistent. especially for increases. and the likelihood of a change increases with the time elapsed since the last change. but at a declining rate. The predictions made using the information set available at time 1 - 4 are not as reliable as those that can be made with the information set available at time I, but estimated parameters are similar in sign and only slightly less significant. The qualitative results are not very sensitive to the inclusion or exclusion of patticular informational variables, or to the use of lagged data (i.e .. observations dated 1 - I instead of those dated 1. where 't = 1 or 1 - 4 ). The results were also qualitatively unaffected when a dummy variable was included to capture the "surprising'' interest rate reductions in 1987, or when only post- 1987 data were included in the sample.

Estimated Magnitude of Changes in Official Rates

In the second stage the magnitude of the expected change is estimated. The dependent variable is the actual change in official rates (if any). To account for the discrete nature of the dependent variable, the constant and the right-hand side informational variable(s) are multiplied by the estimated probability of a reduction or increase in official rates (termed W I and WJ. respectively). and the value of the density functions (Xi and XJ) are also included as explanatory variables. Candidate informational variables were suggested by the consideration that market interest rates should depend on both the actual and the expected future level of official rates. Therefore. the slope of the term structure could contain relevant information. In particular. the difference in the (log) overnight and three-month interbank rates was chosen.15

Estimation results are presented in Table 3. and Figure 3 shows that the equation yields estimates of expected changes (based on information available on day 1) thar are of plausible amplitude and variability. The interpretation of

1 5 Rudebusch ( 1995) contains a discussion of the relationship between the term strucLUre and central bank imerest rate policy. lt might have been preferable to use, say. the difference between the 7 -day and the 1 - or 3-month rates. but the necessary data for this were unavailable.


662 DANIEL C. HARDY

Table 3. Estimation o.f the Expected Magniwde of Changes in OJ]icial Rates

Based on in formation Based on information at time 1 - 4 at time 1

Wl -0.1568 -0.1333 (0.0522)** (0.0298)**

Wl · ln(RI D/R3M) 0. 1 9 1 8 0.2033 (0.35 1 2 ) (0.1738)

XI 0.0697 0.0579 (0.0393)" (0.0258)*

W3 0.1 208 0.1330 (0.0848) (0.0455)**

W3·1n(RI D!R3M) 0.3586 0.6331 (0.1876)+ (0. 181 3)**

X3 0.0007 -0.0025 (0.0526) (0.0433)

Standard error of regression 0.0234 0.02 1 8

R! 0.2907 0.3851

J?! 0.2770 0.3733

Notes: Based on 266 observations. Estimated standard errors in parentheses. Two asterisks indicate significance at I percent: one asterisk indicates significance at 5 percent: a plus sign indicates significance at 1.0 percent.

the positive coefficient on the ln(R I O/R3M) term may be as follows: market participants may have a sense of the trend in interest rates, so that they believe that in the next few months rates are likely to change by, say, P percentage points, and this expectation will be built into the level of d1ree-month rates. If they believe that the central bank intends to .. front-load" this move in rates with a big change in d1e near future, very short-term money market rates will move approximately P percentage points so as to anticipate the change, and the term structure wiiJ flatten. If the central bank is expected to effect the adjustment only slowly, then overnight rates will move less than P: the term structure will be relatively steep. Hence, the closer the overnight rate is to the three-month rate, the larger the change in official rates expected in the next few days. Jn addition, the Bundesbank may have intervened through its repurchase tenders or other open market operations preceding a change in the Lombard or discount rates to ensure that very-short-term rates do not jump too abruptly when the change is announced. The effect again would be to flatten the yield curve in advance of large changes.

lt was difficult to find other variables that performed reliably as informational variables, perhaps because the magnitudes of changes in official

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

Pe!Cefl

t

0.20

0.15

0.10

0.05 0

--0.0

5 •

..{).10

•

•

-0.15

Ja

nuary

19&5

Figu

re 3

. Est

imm

ed F

ore

cast

s of

Chang

es

itz O

ffici

al

Rm

es

• •

••

• •

Janu

llr) 1990

Acru

al ch

ange

•

•

•

• Ja

nuary

1995


664 DANIEL C. HARDY

rates lie in such a narrow range. Again. the results are not very sensitive to the exact specification of the informational variable or to changes in sample size.

IV. Estimated Reaction to Anticipated and Unanticipated Changes in Official Rates

Finally, the estimated expected and unexpected components of the change in Lombard and discount rates taken from the second stage are used as explanatory variables in OLS estimation of equation (2). the results of which are shown in Table 4.

It is clear from Table 4 that only the unanticipated component of the change in official rates has a systematic positive effect on market rates between day 1 and day 1 + I . The estimated coefficients on the .. surprise" component are consistently significantly different from zero, albeit relatively small. and are almost the same for all maturities up to about one year. This stability could indicate that market participants interpret a surprise change in official rates as signaling a policy shift that will persist over this time horizon. The overnight rate. which displays much more volatility than the other series, reacts si ightly less than the one-month rate, possibly because borrowing at the discount facility is typically for a term of at least several weeks and thus a relatively close substitute for one-month interbank borrowing. The two-year rate. which is determined in what is reportedly a rather thin market and for which fewer observations are available. seems again to react more sluggishly. The reaction to the anticipated component of the change is always close to zero and sometimes even negative. These results can be compared with those obtained when the change in official rates is not decomposed into its anticipated and unanticipated parts (reported in Table I) . The unanticipated change is found to affect market rates more strongly than does the total change. and decomposing the change yields notably higher explanatory power. as indicated by the R� statistics.

The results for the change in market rates from 1 - 4 to 1 are in some ways quite different from those for the change from t to 1 + I . The advance reaction to shifts in expectations about movements in official rates is generally much larger than when the reaction is measured starting on day 1. 16 The Bundesbank normally seems to give considerable forewarning of its decisions whether or not to adjust the Lombard or discount rates. and this news is clearly given considerable weight by market participants. Perhaps news that becomes available in the days leading up to Bundesbank Council meetings is

16 However. the changes in estimated expectations are mostly fairly �mall.

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

Ta

ble

4. R

eac

tio

n o

f E

rtro

-D

ewsc

he

Ma

rlc R

ate

s 10

Am

icipa

ted

m1d

Una

mic

ipare

d C

ha

nges

i11

Off

icia

l R

are

s

Cha

nge

from

r -

4 to

r

Cha

nge

from

t 10

1 +

I C

hang

e fro

m .1

-4

to r

+ 5

Depe

nden

t va

riab

le

Ant

icipa

ted

U

nan

tici

pated

R

2 A

nti

cipa

ted

U

nan

tici

pat

ed

R2

Ant

icip

ated

U

nam

icipa

ted

R

l >

z

(b

y m

atur

ity

) ch

ange

ch

ang

e s.

e.

chan

ge

chan

ge

s.e

. ch

ang

e

chan

ge

s.e

. -1

()

1

day"

0.2809

1.

811

9

10.4

26

0.0

173

0

.187

6

1.7-t

7 0

.140

8 0

.56-t

5 2.

626

�

>

(0.1

876

) ( 0

.3447)

**

0.

044S

(0.09

24)

(0.07

32)

*

0.02

57

(0.3

379)

(0

.215

3)*

0

.08

11

-1

0

1 m

om

tf

02

607

1.2

176

17.2

65

-0.0

482

0.2

176

16

.509

0

.054

7 0

.395

7 8.

200

z

(0.1

02

1 )*

(0

.18

36

)**

0

.022

9

(0.04

38)

(0.0

320)*

*

0.0

111

(0

.14

32)

(0.086

1 )*

* 0

.032

2 z

n

i"l!

3

mont

h�

0.2

48

1 1.

0045

16.9

75

-0.00

05 0

.15

44

7.

677

0.

1067

0.3

781

8.

256

z

-l

(0.0

868 )

**

(0

.15

61 )

**

0

.019

5 (0

.0475

) (0

.03

47

)**

0.

012

1 (0

.13

78}

(0.0

82

7)*

*

0.03

09

::>::

>

6 mo

nths

'" 0

.20

80

0.6

950

10

.59

2 0

.012

3 0

.154

8 10

.746

0

.12

06

0.2

.87

�

5.37

1 ::c

(0

.0890

)*

(0.1

600}*

* O.ol

99

(0.0

391)

(0

.0309

)**

0.009

9 (0

.136

5)

(0.0

878)

**

0.

0305

>

z

;;<:

12

mon�

0

.214

6 0

.363

4

5.090

-0

.0005

0.1

621

8.09

8 0.

174

1 0

2606

5.

022

3 (0

.0997)

*

(0.1

740

}*

0.0

204

(0.0

480

) (0

.040

0)**

0.0

121

(0.1

35

5)

(0.09

11)*

0

.030

1 �

2-t

mont

hS'

0

.39

16

0.0

740

7

.74

9

-O.OM

I 0

.054

2 1.0

26

0.3

703

0.1

769

7.73

9

m

:r.

(0.10

75)*

* (0

.258

2)

0.0

220

(0

.072

0)

(0.0

572)

0

.015

8

(0.1

221 )

**

(0

.084

5)*

0

.07

74

-1

�

Not

e: O

LS es

tim

atio

n o

f [li

M)r

-f1

M),

] = b,)

+ b

1£(ru

,l !!

,) +

h:![E

(ru, I

!lr)

-E

(ru,l

il,)

] +

e,_-.

Est

imat

ed c

oeff

icie

nts

b1

and

b1•

est

imate

d >

-1

st

anda

rd e

rrors

in p

aren

thes

es.

perc

enta

ge R

2. a

nd e

qu

atio

n s

tand

ard

erro

r rep

oned

. Tv.

·o a

ster

isks

ind

icat

e ig

nific

ance

at

I perc

ent;

one

ast

er-

m

"0

isk

ind

icat

es s

igni

fican

ce a

t5

perc

ent.

0

r

• Num

ber o

f obs

ervat

ions

= 2

66.

()

6 24

1 o

bse

rvat

ion

s.

-<

r2

05

observ

atio

ns.

" 18

9 o

bser

vati

on

s.

r 1

63

observ

atio

ns.

a- a- Vl


666 DANIEL C. HARDY

considered especially relevant to forecasting the policy stance over the longer term: it is during this period that the Bundesbank may be signaling how large and permanent a shift in interest rates it envisages. The .. news .. contained in the announcement on day t of the precise magnitude of a change in official rates may be of less relevance. The reaction to the evolution of expectations before day r is more pronounced for shoner maturities.

Equally remarkable is the magnitude and significance of the estimated coefficients on the anticipated component of changes in the Lombard and discount rates. For the longest maturity rates, the estimated coefficient is significantly larger than that on the unanticipated change.17 Several (not mutually exclusive) explanations for this result can be suggested.

As explained in Section II, the operation of the standing facilities themselves may in large part account for the gradual reaction of market prices to expected changes in the rates charged on these facilities. At each point in time, the cunent Lombard and discount rates constrain short-term money market rates from above and below, respectively, although the constraint is not absolute. When a large change in official rates is expected in the near future, short-term market rates will .. hit .. one or other boundary of the interest rate band and will therefore not necessarily move all the way to the new expected level until the change is realized. The effect on short-term rates of the Lombard and discount rate bounds may then be transmitted along the yield curve. As the date of the expected change approaches, uncertainty about the magnitude of the change and the advantage of bringing forward or delaying a transaction decreases, so the effectiveness of the boundaries should diminish.

However. it could also be that the euro-deutsche mark market is not perfectly informationally efficient, so anticipated changes in official rates are not fully discounted in advance. A related possibility is that. when a change in official rates is deemed likely. panicipants adopt a "wait and see·· approach and activity in these markets dries up. The recorded prices may then not represent those at which most agents are willing to trade and so they fail to reflect expectations. These two hypotheses are perhaps most plausible for the longer maturities, where indeed the effect of anticipated changes is greatest.

It might be asserted that the instrumental variables technique in fact uses more information than was available to market panicipants: for each observation, the prediction of the change in official rates is based not only on individual data that were publicly available at the time, but also on parameters in the auxiliary regressions that are estimated from the full sample up to the

17 Estimates were run for the change from t -11 to 1 + I . with various values of n between 0 and I 0. The effect of the anticipated component is larger as n increases (i.e .. the longer the time interval).



end of 1995. The technique may therefore identify as anticipated what was in fact a "surprise" to market participants. The estimated coefficient b1 will be biased downward and b2 biased u pward.1s This argument cannot, however. account for why the phenomenon is much more pronounced for changes in market rates from t - 4 to t than for changes from 1 to t + I .

The separation of the anticipated and unamicipated components of changes in official rates has increased considerably the explanatory power of the estimates for this time horizon; the relevant R2 statistics are up to I 0 percentage points higher in Table 4 than in Table I. The estimated coefficients on the total change are less than those on the unanticipated component for all maturities less then two years. and also less than those on the anticipated component for longer maturities.

Table 4 also shows that the estimated coefficients for the reaction from t - 4 to 1 + 5 are approximately equal to or slightly below the average of those for the reaction from t -4 to t and from t to t + I . While there may be some overshooting. it seems that almost the full reaction to the realized unexpected change occurs by the day of change itself: there is little sign that the markets need several days to ·'digest"' the news or that they rely on explanations after the fact by the Bundesbank. ln thi regard the market may be fairly informationally efficient.

V. Conclusion

The official interest rates applied at central bank standing facilities serve as bounds and guideposts for short-term money marker rates. The relationship between marker rates and these bounds is therefore an important indicator of marker sentiment concerning the probability of a forthcoming shift in the interest rate ··band" and the central bank's operational target range for short-term money market rates over the coming period. This and other available information can be used to estimate the extent to which market participants can foresee the timing and magnitude of changes in the central bank·s official interest rates. and to what extent the changes come as a surprise. fn this paper, such estimates are generated for changes in the rates applied at the Bundesbank"s Lombard and discount facilities. The estimates are used to gauge how far the market response depends on the degree of anticipation.

The reaction of market rates (especially but not exclusively for maturities between one month and one year) to unexpected changes in official rates was

18 In principle. it would be possible to mitigate this difficulty by using a recursive estimation technique. which. however. would exhaust many degrees of freedom and perhaps make use of too little information at the start of the available sample.


668 DANIEL C. HARDY

found to have been sharp but of moderate magnitude. Tn addition. the accmal of information as the central bank signals its intentions in advance of a change in official rates strongly influences market rates. However. the anticipated component is shown to influence market interest rates in the days leading up to a decision. The Bundesbank has relied primarily on open market operations in the implementation of policy since 1985. but even a largely anticipated change in official interest rates on standing facilities can still be effective in confirming and clarifying the public's understanding of a shift in the policy stance.

APPENDIX

Estimation of Anticipated and U nanticipated Changes in Official Rates

The behavior of the Lombard and discount rates can be treated as an instance of an ordered response, limited dependent variable model: if certain conditions obtain, then one or both official rates increase in relatively large steps: if other conditions obtain they decrease: and under intermediary conditions they remain unchanged. The standard ordered response model will be generalized tO allow the explanatory variables to affect the probability of an increase or a decrease in different ways. It is then possible to go on to predict the magnitude of any positive or negative change (see Heckman. 1974: and Maddala, 1 983. pp. 46-9 and Chapter 8). The predictions and the residuals are taken as the anticipated and unanticipated components of the changes in official rates, respectively.

Let the dummy dl take the value of I when an official rate decreases, and zero otherwise. Similarly, let d2 equal I only when rate are unchanged, and let d3 equal I only when rates are increased. it is assumed that there exists a set of explanatory variables Z, which predict the direction of changes in orticial rates. and another {possibly coincidental or overlapping) set of explanatory variables X, which predict through some linear equation the magnitude of the change. The average log change in ofticial rates will be denoted by y.19 The scheme can be �ummarizcd a� follows:

ify,·z + u < o dl = I. c/2 = d3 = 0

ify1 'Z + u > O > yJ'Z + u d2 = l . d1 = d3 = 0

if y;Z + u > 0 d3 = I. dl = d2 = 0

y = �1 'X + u1 (A I )

y = O

y = �,'X + u,,

where y1• y,. �1• and �' are parameter� to be estimated. and 11. 111• and u, arc correlated. homoscedastic random variables with a joint normal distribution.211 The

19Time subscripts are omitted where no ambiguity results. The instrumental variables in X and Z must be known before r is realized.

10The conditions y1'Z + u < 0 and y3'i + u > 0 should not be fulfilled �imultaneously. The method used here does not impose this constraint. but in the application no difficulties result.



variable 11 is standardized to have mean zero and variance of unity. Let.f and F denote the density function and the cumulative distribution function of the �tandard normal. respec1ively. With n observations indexed by i. and recalling that I - Few) = F(- u-). the likelihood function can be written a�

(A2)

The first stage of !he regression procedure consists of maximizing the logarithm of equation (A2) with respect to 1he parameter� g,. i = I , 2, 3. Starting value\ can be obtained by first estimating standard probit models for dl and d3 separately. To estimate 1he predicted magnitude of changes in official rates. note that

E(y,) = Prob(y, < 0) • £(r, I y, < 0) + Prob(y, = 0) • E(y, I y, = 0) + Prob(r, > 0) • £(r, I y, > 0),

which can be shown based on equation ( A I ) to imply !hat

I n the second stage. estimates of p,. f3,. cr1,. and cr,, are obtained by replacing E(y,) in equation (A3) with the realized value of y, and then applying OLS, where use is made of the estimates of y1 and y, obtained in the first stage. The �tandard errors are hetcroscedastic. but the estimated standard errors can be corrected using the procedure from White ( 1 980). Homoscedasticity had to be assumed in equation (A I ) so cr1, and cr1, can be taken to be constants.

The predicted value _v, from the ;.econd-stage regression equation (A3) i:, treated as the expected change £(/ll, I Q,). The residual lr, - _v, l is the unexpected component. which by construction is orthogonal to the fitted value and the instruments.11

The two are then used to estimate equation (2) in the specification

(A4)

Pagan ( 1 984) and MeA leer and McKenzie ( 1 9 9 1 ) di!>cuss the propertie� of regression output with such constructed regre�sors. Under reasonable conditions the OLS coefficient estimates arc unbiased, and that of b! will be efficient. The OLS /-statistic on the estimate of b1 may be biased upward by an amount that varie;. po;.itively with the product of (b1)1 and the variance of the residuals from the auxiliary regression equation (A3 here). and negatively with the variance of the residual� from the regression of interest equation (A4). The estimate of b1 was at most 1 . 8 and often much smaller (mostly in the range 0. I 5 to 0.4 ); the estimated variances of both the auxiliary regre�sion and equation (A4) applied to intere;.t rates were about 4 x I 0-4 (see Tables 3 and 4, respectively). Therefore. the bias is likely to be small relative to the /-statistics achieved.

11 When estimating the effect of changes in expectations, as between day 1 - 4 and day 1. the "surprise·· term is the difference between the expectations based on the two information . cts.


670 DANIEL C. HARDY

REFERENCES

Bertola. Giuseppe. and Lars E.O. Svensson. 1 993. ··stochastic Devaluation Risk and the Empirical Fit of Target-Zone Models."' Review of Economic Srudies. Vol. 60 (July), pp. 689-7 12.

Borio. Claudio E.V .. 1 997. "Monetary Policy Operating Procedures in Industrial Countries." BIS Working Paper No. 40 (Basle: Bank for International Settlemems).

Cook. Timothy, and Thomas Hahn, 1989, "'The Effect of Changes in the Federal Funds Rate Target on Market Interest Rates in the 1970s:· Journal of Monetary Economics. Vol. 24 (November). pp. 331-5 1 .

---. 1988. "The Information Content of Discount Rate Announcements and Their Effect on Market Interest Rates ... Journal of Money. Credit and Banking, Vol. 20 (May), pp. 167-80.

Dale. Spencer, 1 993, "'The Effect of OfFicial I merest Rate Changes on Market Rates Since 1987."' Manchester School of Economic and Social Studies. Vol. 6 1 (Suppl.. June). pp. 76-94.

Deutsche Bundesbank, 1 994, The Monetary Policy of the Bundesbank, Sonderdruck der Deutschen Bundesbank (Frankfurt: Deutsche Bundesbank).

Edin, Per-Anders, and Anders Vredin. 1993, "Devaluation Risk in Target Zones: Evidence from the Nordic Countries,

.. Economic Journal. Vol. 103 (January), pp. 16 1-75.

Favero, Carlo A., Fabrizio 1acone. and Marco Pifferi. 1996, "Monetary Policy, Forward Rates and Long Rates: Does Germany Differ from the United States?" Center for Economic Policy Research Discussion Paper No. 1456. (London: CEPR).

Hardy. Daniel C .. 1 996. "Market Reaction to Changes i n German Oflicial 1nterest Rates," Deutsche Bundesbank Discussion Paper No. 4/96 (Frankfurt: Deut che Bundesbank).

Heckman, James J .. 1974. ''Shadow Prices. Market Wages. and Labor Supply," Econometrica. Vol. 42 (December), pp. 679-94.

Lombra. Raymond E .. and Raymond G. Torto. 1 977, "Discount Rate Changes and Announccmem Effects,

.. Quarterly Journal of Economics. Vol. 9 1 ( February).

pp. 1 7 1 -76.

Maddala, G.S .. 1983. Limited-Depende/11 and Qualitative Variables in Econometrics (Cambridge: Unive rsity Press).

May, Don 0., 1992, "'A Reexamination of Market Returns, Discount Rate Changes, and Market Efficiency .

.. Journal of Macroeconomics, Vol. 14 (Summer), pp. 545-53.

McAieer, Michael, and C.R. McKenzie, 1991 "When Are Two Step Estimators Efficient?" Econometric Reviews. Vol. 10. No. 2. pp. 235-52.

Nautz, Dieter, 1995 "Zur Feinsteuerung des Geldmarktcs durch die Wertpapierpensionsgeschafte der Bundesbank,'' Zeitschrifl fiir Wirtschaftsund So:.ialwissenschafren. Vol. 1 1 5, No. 4. pp. 623-44.



Pagan, Adrian, 1984. "Econometric Issues in the Analysis of Regressions with Generated Regressors .

.. International Economic Re1•iew, Vol. 25 ( February).

pp. 221-47.

Paquet, A lain, and Thicrry Perez, 1 995. "La reaction du marche financier face a diffcrentcs sources de signal de la pol itiquc monetairc au Canada ... Canadian Journal of Economics. Vol. 28 (November). pp. 808-2 1 .

Radecki, Lawrence L and Vincent Reinhart. 1 994. ·The Financial Linkage� in the Transmission of Monetary Policy in the United States.'' in National D{[ferences in llllerest Rate Transmission (Base!: Bank for International Settlements), pp. 291-337.

Roley, V. Vance. and Rick Troll. 1 984, "The Impact of Discount Rate Changes on Market Interest Rates.'' Federal Resen·e Bank of Kansas City Economic Revie11· (January). pp. 27-39.

Rudebusch, Glenn D., I995. ''Federal Reserve Interest Rate Targeting, Rational Expectations. and the Term Structure.'' Joumal of Monetary Economics. Vol. 35 (April). pp. 245-74.

Skinner. Tom. and Jeromin Zettelmeyer. 1 997. ·'Long Rates and Monetary Policy: Evidence from the U.S., Germany and the U.K.'' (unpublished: Wa�hington: IMF. December).

Smirlock. Michael. and Jess Yawitz, 1 985. "Asset 'Returns. Discount Rate Changes. and Market Efficiency:· Joumal o.f Finance. Vol. 40 (September). pp. 1 14 1-58.

Thornton. Daniel L., 1986, ''The Discount Rate and Market lnterc�t Rates: Theory and Evidence:· Federal Resen·e Bank of St. Louis Rel'ie11·, Vol. 68 (August/September), pp. 5-21.

---. 1994. "Why doT-Bill Rates React to Discount Rate Change�·)'' Joumal o.f Money. Credit and Banking, Vol. 27 (November), pp. 839-50.

White, Halbert. 1980. "A Heteroskedasticity-Consistent Covariance Matrix and a Direct Test for Heteroskedasticity," Econometrica, Vol. 48 (May), pp. 8 17-38.


IMF' Staff Papers Vol. 45. No. 4 (December 1998) © 1998 International Monetary Fund

Warning: Inflation May Be Harmful

to Your Growth

ATISH GHOSH and STEVEN PHILLIPS*

While few doubt that very high inflation is bad for growth, there is less agreement about the effects of moderate inflation. Using panel regressions and allowing for a nonlinear specification, this paper finds a statistically and economically significant negative relationship between inflation and growth, which holds robustly at all but the lowest inflation rates. A "decision-tree " technique identifies inflation as one of the most important determinants of growth. Finally, short-run growth costs of disinflation are only relevant for the most severe clisinflations, or when the initial inflation rate is well within the single-digit range. [JEL E3l, 040]

RAPTD OUTPUT GROWTH and low inflation are the most common objectives of macroeconomic policy. It is rather surprising, therefore, that a con

sensus about the relationship between these two variables is yet to emerge. While early studies by Phi !lips ( 1958) suggested an exploitable trade-off between output and price stability, the stagflationary experience of the industrialized countries in the 1970s belied this finding and showed that, beyond the short run, any such trade-off is illlllsory. More recent cross-country studies, particularly those that include middle- and low-income countries in their samples, suggest a negative relationship between growth and inftation. 1 Even

* Atish Ghosh is an Economist and Steven PhWips is a Senior Economist, both in the Policy Development and Review Department. The authors would like to thank Kadima Kalonji for research assistance, and Alan Taylor and Maurice Obstfeld for making available their computer programs. Hugh Bredenkamp, Matthew Canzonieri, Sharmini Coorey, Peter Doyle, Martin Evans, StanJey Fischer, Robert Flood, Manuel Guitian, Javier Hamann, Timothy Lane, Michael Sarel, Susan Schadler, and Tsidi Tsikata provided helpful comments on an earlier draft, as did participants in seminars at Georgetown University and at the Graduate Institute of International Studies (Geneva).

1 Fischer ( 1993), Bruno and Easterly ( 1995), Judson and Orphan ides ( 1996), Sarel ( 1996), Barro ( 1995), and IMF (forthcoming).

672


WARNING: INFLATION MAY BE HARMfUL TO YOliR GROWTH 6 TJ

among these studies. however. there is little agreement on whether the empiric<tl association of lower inflation with faster growth is statistically and economically significant. let alone causal.�

If prices exhibit downward rigidity. then very low inflation rates may ossify the structure of relative prices. impeding adjustment to real shocks. A little inflation. therefore. might help to .. grease

.. the economy. On the other hand.

high inflation rate!>, by confounding relative ptice signals and making efficient resource allocation more difficult. could result in more sluggish economic growth. But whether these or other negative effects begin at single-digit inflation rates. or only at much higher ra�es. remains a controversial question. Moreover. it is not clear that a rise in inflation causes a proportional worsening of the country's growth performance: it might be that. once chaotic inflation rates have been reached. relative prices cease to have much meaning anyway. making fur1her increase� in inflation less important.

In a multivariate context. the inflation-growth relationship becomes yet more complicated. Obviously, growth-inflation regressions must include other plausible determinants of growth. Several issues then arise. First. the inflation-growth findings may not be robust once .. conditioning

.. variables

are included in a regression analysis. Levine and Zervos ( 1993). for example. find that inflation does not survive Leamer's extreme bounds tests in growth regressions. Second. the conditioning variables may themselves be functions of the inflation rate. For instance. investment affects GDP growth. but may itself be affected by inflation. To the extent that inflation influence� growth through such indirect effects. inclusion of these variables in a growth regression reduces the apparent effect of inflation. Third. there may be rich and important interactions between inflation and the other determinants of growth. For example, the marginal effect of inflation on growth may differ according to the level of physical and human capital in the country. With growth having many possible determinants. it may be difficult to model such interactions, especially since theory provides little guidance on the appropriate specification. Fourth. inflation is not under direct policy control: e�pecially in the shor1 run. it is affected by shocks that can influence both inflation and growth. possibly resulting in spur-ious correlations. Finally. even if low inflation is generally associated with faster growth. it does not necessarily follow that disinflation is always good for growth. l n particular. rapid disinflation may result in lower growth. at least in the short run.

These considerations suggest that. i r a relationship between inflation and GDP growth exists. it is not likely to be a simple one. The bivariate relationship will not be monotonic. let alone linear: there may be imponant interaction effects between inflation and the other detenninarm, of growth: and

1Ciark ( 1 993).


674 ATISH GHOSH and STEVEN PHILLIPS

the correlation between disinflation and growth may be quite different from the steady-state inflation-growth relationship. Perhaps the lack of a consensus about the effects of inflation on growth is not so surprising after all.

In this paper. we try to address these various methodological problems and examine the relationship between inflation and disinflation and output growth. We employ a large panel data set. covering TMF member countries over 1960-96. Our primary analytical tool is a panel regression. in which our main contribution is to combine a non linear treatment of the inftationgrowth relationship with an extensive examination of robustness. Complementing this analysis is our use of a decision-theoretic ('"tree'") technique that is more robust to outliers and nonlinearities than is standard regression analysis. Throughout, the emphasis is on examining the stillconrroversial question of whether there is any robust inflation-growth relationship. rather than pinning down the dynamics of such a relationship or identifying specific mechanisms through which inflation (or the policy choices it reflects) might influence growth.

Tn general. we find a negative relationship between inflation and growth that is statistically significant and of an economical! y interesting magnitude. This finding survives a battery of robustness checks. While we cannot rule out the possibility that part of this negative relationship stems from effects of growth on inflation. we still find a statistically and economically significant relationship between inflation and GDP growth when we use several sets of instruments to control for such simultaneity. But even if low inflation is associated with more rapid outpm growth, it is possible that the process of disinflation may-at least i n the short run-depress GDP growth. Our results here are striking. Disinflation tends to reduce growth only if the starting level of inflation is already very low. or if the pace of disinflation is severe.

Our more derailed results may be summarized briefly. First, there are two important nonlinearities in the inflation-growth relationship. At very low inflation rates (around 2-3 percent a year, or lower), inflation and growth are positively correlated. Otherwise. inflation and growth are negatively correlated. but the relationship is convex, so that the decline in growth associated with an increase from I 0 percent to 20 percent inflation is much larger than that associated with moving from 40 percent to 50 percent inflation. Taking both these nonlinearities into account, we find that the negative inflation-growth relationship is evident in both the time and crosssection dimensions of the data, and that it is quite robust. Excluding high inflation observations, time-averaging the data. or using various subsamples (defined according to time period or the degree of inflation) does not alter the basic findings. We also find that inftation is a robust regressor in Learner"s extreme bounds sense. and that allowing for nonlinear


WARNING: INFLATION MAY BE HARMFUL TO YOUR GROWTH 675

relationships between the other regressors and GDP growth does not diminish the inflation-growth association.

To allow for threshold effects and nonlinear interactions. we use a technique known as binaty recursi1·e trees. The key advantage of this technique is its robustness to alternative specifications and to ourliers. Indeed. the results are invariant to any monotone transformation of the variables. Importantly, this decision-theoretic analysis identifies inflation as one of the most important determinants of GDP growth (second only to physical and human capital).

Turning to the short-run consequences of rapid disinflation. we find that starting from inflation rates above 6 percent, only the most drastic disinflations (at least halving the inflation rate in a single year) are associated with any negative impact on growth (which itself is largely offset by the higher growth associated with the new lower level of inflation). Starting from lower inflation rates, however, a rapid disinftation (halving the inflation rate) is associated with a fall in GDP growth.

I . Basic Statistics and Correlations

Our complete data set consists of 3.603 annual observations on real per capita GDP growth, and period average consumer price inflation, corresponding to 145 countries. over the period 1960-96.1

As a first step in exploring the bivariate relationship between inflation and growth. Figure I graphs the joint frequency distribution of inflation and growth. It is noteworthy that, while there are relatively few observations with inflation above 20 percent a year. they occur predominantly with negative per capita GDP growth rates. Indeed. two-fifths of the observations with inflation

·1 Each observation corresponds to an individual country for a given year. Of course. with the wide range of countries in the sample. the quality of the underlying data probably varies enormously. For this reason. the results are presented both in aggregate and with breakdowns by per capita income.

The original data et actually consists of 3.772 observation�. but some observations were excluded at the outset. First, observations with GDP growth above 30 percent per year (5 observations) or below -30 percent a year (5 observations) were excluded because such extreme values may be unreliable or occur under exceptional circumstances (e.g., civil war) so that their relevance for economic policymaking is suspect. Their inclusion in the data set does not alter the basic conclusions (see Section Ill). Second, cases of negative in Ration ( 159 ob�ervations) were excluded, not only for being outside the range of interest here. but for the practical reason that the analysis below requires taking the logarithm of the inAation rate. Including such cases by replacing negative in Ration rates either by a small positive number (Sarel, I 996) or by their ab&olute value (IMF, forthcoming) does not alter the basic conclusions of this study.



Figure I . Joiflt Frequency Distribution rJf Inflation and COP Crou·tfl Rate.' (In percent a year)

0. 3

-:; 10.20 '?. '�...; 10.40 �

40.80 1.5. 5

0. 1 . 5 \\ , s o��'' - - · · . f? '15 -10. -2.5 G'V -

10+

Note: Number of observations shown on vertical �calc: total observations: 3.603: maximum: 247: minimum: I.

above 20 percent show up at negative GDP growth rate�. compared to only one-fifth of the cases with inflation below 20 percent. Alternatively. of the observations with positive GDP growth. more than three-fourths occur at inflation rates below 20 percent a year. Thus, grouping the data in this way suggests a negative association between inflation and growth.

Table I presents much the same information, but in tabular form. and for several different samples (the ··consistent sample" consists of the 2,231

observations, for I 03 countries. for which data on all of the conditioning variables used below are available). Again, the bivariate evidence suggests a negative relationship between inflation and growth. This relationship appears to break down. however. somewhere in the very low inflation range.

Figure 2. which is central to our results. gives a more direct view of the inflation-growth association by plotting the median GDP growth rate against the median inflation rate (for each of 20 equal-sized subsamples defined



Table I. Basic Statistics ( In percent a year)

Inflation GDP grow th Number of

observations Mean Median Mean Median

Large Sample

All observations 3.603 39.1 8.3 1.8 2.2 0 < rr < 3 628 1 .7 1 .8 2.6 2.7 3 < rr < 5 525 3.9 -l.O 2.8 2.9 5 < rr < 10 913 7.4 7.3 2.4 2.6 1 0 < rr < 20 843 14.0 13.3 1.8 1 .8 20 < rr < 40 394 27.3 26.1 0.4 0.9 40 < 7t < 80 142 56.7 54.6 0.9 1.3 7t > 80 158 635.4 166.9 -3.8 -2.9

Consistent Sample

All observations 2.231 ..J2.0 9.3 1.9 ") ") 0 < n < 3 321 1 .8 1 .9 2.6 2.7 3 < rr < 5 303 4.0 4.0 2.8 2.9 5 < n < 1 0 570 7.5 7.4 2.6 2.7 1 0 < n < 20 568 1 3.9 13.2 1.6 1.5 20 < 1t < 40 272 27.3 26.2 0.6 1.0 40 < n < 80 104 56.6 55.0 1.0 1.4 n > 80 93 7 1 5.7 163.4 - 1 .9 -1.0

Consistent Sample Upper- and upper-middle-income countries

All observations 937 36.3 6.7 2.7 2.7 0 < rr < 3 180 2.0 2.1 3.6 3.2 3 < n < 5 183 3.9 3.9 3.5 3.5 5 < rr < l0 244 7.2 7.1 2.8 2.9 1 0 < n < 20 177 14.0 13.5 2.0 2.0 20 < 1t < 40 66 26.0 25.0 2.1 2.2 40 < 1t < 80 37 56.6 56.6 2.5 2.4 n > 80 50 497.1 168.2 -0.7 0.1

Consistent Sample Lower- and lower-middle-income countries

Al l observations 1 .294 46.2 10.8 1 . 3 1.7 0 < rr < 3 141 1 .5 1 .7 1.4 1 .8 3 < Jt < 5 120 4.0 4.1 1.8 2.3 5 < n < 10 326 7.6 7.8 2.4 2.5 1 0 < 71: <20 391 13.8 13.0 1.5 1.4 20 < n < 40 206 27.7 26.6 0.1 0.7 40 < n < 80 67 56.5 54.5 0.2 0.5 7t > 80 43 969.9 161 .0 -3.5 -3.5



Table I . (concluded)

Number of Inflation GDP growth

observations Mean Median Mean Median

Consistent Sample Post- 1973 observation�

All observations 1 .786 50.1 10.6 1.5 1 .9 0 < 7t < 3 204 1.7 1 .8 2.0 2.3 3 < n < 5 1 95 4.0 4. 1 2.7 2.7 5 < 7t < 1 0 442 7.6 7.6 2.3 2.6 1 0 < 7t < 20 5 1 3 13.9 13.3 1.5 1.4 20 < 7t < 40 252 27.4 26.3 0.4 0.9 40 < 7t < 80 93 56.5 54.8 1 .2 1.4 7t > 80 87 754.6 1 7 1 .7 -2. 1 - 1 .6

according to degree of inflation).4 Again, the concentration of inflation ob. ervations in the 0-20 percent range is evident. but this data-smoothing technique also makes two key features of the data immediately apparent. First. at the very lowest inflation rates. for which there are quite a few observations. inflation and growth are positil·ely associated. Second. at all other inflation rates, the relationship is negative and clearly convex-implying. plausibly. that an increase in inflation from 5 percent to 25 percent impairs growth more than an increase from I 00 percent to 120 percent. The slope is quite flat over the highest inflation ranges; such observations are a small part of the sample. but as oulliers their effective weight in a regression analysis may be considerable. Indeed. Figure 2 suggests that ignoring these nonlinearities and regressing growth linearly on the inflation rate would impart a downward bias to the estimated slope over the range of greatest policy interest. This bias may. at least in part. account for the failure of previous studies to detect a significant and robust negative relationship between inflation and growth.

ll. Conditioning Variables: Multivariate Inflation-Growth Regressions

Findings of a negative correlation between inflation and growth suggestbut obviously do not prove-the notion that lower inflation promotes faster growth. Causality aside, it is natural to suspect that part of the correlation may be spurious, reflecting the effects of third factors. This section checks

J Using group means rather than medians gives a similar picture. except that the points for the groups with the highest inflation observations shift to the right, makmg the plotted curve even more convex.


GDP growth 4

3

2

0 0


Figure 2. Inflation and Per Capita GDP Growth1 (In percent a year)

20 40 Inflation

60 80

1 Median inflation and growth rates in equal-sized subsamples, defined according to range of inflation (right-most poim not shown).

whether an inflation-growth relationship appears also in multivariate regression analysis. The intem is not to develop an explanatory model of GDP growth, but rather to determine whether the inflation-growth correlation is robust to including a set of conditioning variables. The analysis also checks for nonlinearity of the inflation-growth relationship.

A first step is to include annual dummies in regressing per capita GDP growth on inflation.5 More generally, the regression is augmented with other proposed growth determinants; for this, the empirical literature pullulates with possibilities. Neoclassical theory stresses capital accumulation as the engine of (pre-steady state) growth. More recent growth theories also emphasize the importance of human capital. Various measures of human capital, such as school enrollment rates, average years of primary and secondary education completed. and l ife expectancy. have been proposed. These tend to be highly correlated. so, rather than include them individually. we use the first principal component of primary and secondary school enrollment rates and life expectancy as a measure of human capital (HK).

5 Here, we do not include country dummies since these are highly correlated with some of the additional regressors. In the robustness section. we show that adding such dummies actually strengthens the inflation-growth findings.



Beyond the physical and human capital variables suggested by theory, a largely ad hoc smorgasbord of factors that might affect "'productivity

.. is usu

ally included in growth regressions. To control for "catch-up"' effects, the log of the ratio of U.S. per capita income to country j"s per capita income i n 1960. and measured in international prices. is used (CAP). A large tax burden on the economy or a large share of public consumption might depres:-. economic growth: we include the ratios of revenues to GDP (t!CDP). public consumption to GDP (CICDP), and the fiscal balance (8/CDP).

A number of studies stress the importance of openness to international trade, both as a means of effecting the transfer of technical progress and as an engine of growth: we use the sample average of the ratio of exports plus imports to GDP. ((X+ M)ICDP). The (log of the) black market exchange rate premium (BLK) provides a measure of the overvaluation of the real exchange rate and, i n at least some instances, of economic mismanagement more generally. The terms of trade volatility. cr,.,-. is used as a measure of the imp01tance of external shocks.<> Finally. we include indicator variables for cataclysmic events such as drought (DROUCH1), or cases where there are war-related deaths (DEATH). By controlling for these types of supply shocks. these regressors should reduce the chances of picking up spurious (negative) inflation-growth comovements.

Some of these variables. such as drought or war, are clearly exogenou. with respect to inflation. But other variables, most notably the investment rate. are likely to be influenced by inflation. To the extent that inflation affects growth by influencing these conditioning variable� (and they. in turn, affect growth), their inclusion in the regression could diminish the measured effects of inflation on growth. Since there is no easy way around this problem, we report results both including and excluding the investment ratio. (In the robustness section. we undertake a more systematic analysis of the effects on the inflation-growth relationship of including and excluding the various other regressors.)

The next step is to model the evident nonlinearity of the inflation-growth relationship. From Figure 2, the positively sloped part of this relationship ceases at inflation rates somewhere around 2-3 percent a year. To deal with this non linearity. we follow Sarel ( 1 996) and use a spline technique. allowing the relationship ro have a "'kink"' or turning point where rr = 2 1/2 percent.7 (As shown in Section Tl l . the exact location of this kink turns out not to be important for the questions on which we focus.)

�current and lagged term� of trade changes were found lO be insigni ficant: their inclusion in the regression would change the inOation-growth resulls only slightly (see Section Ill).

1 It turns out that 2 1/: percent inflation i!. also the placement of the kink that yields the best fit of a multivariate growth regression (see Section Ill).


WARNING: INFLATION MAY BE HARMFUL TO YOUR GROWfl-1 68 1

This leaves the question of how to capture the convexity of the negative inflation-growth relationship. We consider a number of possibilities. Our first model (denoted model ( I ) in Table 2) simply ignores the convexity, specifying a linear relationship (beyond the kink at 2 1/� percent). Model (2) uses the real rate of depreciation of the currency. defined as rrl( I + rr), as the measure of inflation.� Model {3) uses the log of the inflation rate. and model (4) generalizes this by replacing log(rr) with ( I - y)- 1rc11 Y1, where y is estimated via nonlinear least squares. This specification collapses to the linear specification as y approaches zero. and to the logarithmic specification as y approaches unity.9

Estimates for the inflation-related parameter� in these four specifications are reported in Table 2. and the im[plied GDP growth rates at various inflation rates are illustrated in Figure 3. The coefficient on (the various) measures of inflation (when inflation is above 2 1/2 percent a year), given by p,, is always negative and statistically significant, with heteroscedasticconsistent /-statistics ranging from about 3 to over 10.

In specification (I )-which is linear beyond the kink at 2 1/2 percent inflation-the inflation coefficient, though statistically significant, i.!> economically paltry. Indeed. the negative slope for this model i� barely discernible in Figure 3. The linear model suggests that raising inflation from I 0 percent a year to 20 percent a year would be associated with a mere 0.0 I percentage point reduction in annual growth. (It is easy to see that even weaker results would appear if the kink at 2 1/� percent were not allowed, and complete linearity imposed.)

In contrast. the nonlinear models-the real rate of depreciation (2), the logarithmic (3), and the more general nonlinear variant (4)-are all suggestive of economically important effects over the inflation range of greatest policy interest. According to these models. an increase in annual inflation from 10 percent to 20 percent a year would be associated with a reduction of per capita GDP growth by about 0.3-0.4 percentage points. while an increase in inflation from I 0 percent to 40 percent a year would be associated with about 0.8 percentage points slower growth.10 Figure 3 also shows that these three non linear models tend to give relatively similar predictions about the apparent effect of inflation on growth-which are far greater than the predictions implied by the linear model.

8 See. for example. Cukiennan ( 1 992) for use of this specification. A log-based specification has been used by Sarel ( 1 996), IMF (forthcoming). and othe rs.

q We do not. however. use log( I + rr)-a specification !>uggested by some authorsbecause this function is close to being linear over the range in which most inflation observations l ie.

10The predicted changes would be even larger if one assumed that inflation would be reduced in part by raising 8/GDP (cutting the budget deficit). Of course. one wants to be careful about applying causal interpretations to growth regression results; this issue is discussed later.

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

"'

00

tv

Tab

le 2

. Alr

erna

riw!

infl

mio

n-G

row

rh R

eg

ress

ion

s

( 1)

uy=

a,,+

[l. «

"":'.'

• (rr

-0.

025) +

�� n

(2

) 8

y =

a,,+

�C<

(II1:5

1 ([l

rf(l

+IT

)]-

0.0

26) +

[l, (

Jr/( I

+ n:

))

(3)

8y =

a,,+

[l..«

""�s

. [lo

g(Jt

) -lo

g(0.

025)

] + 1}

1 lo

g(n:

) (4

) uy

= o:,

, + fl.

-.«IIC

5• (rr

-0

.025

) + 13

1 (I

-Yt

'rr'1

·f

Infla

tion

meas

ure

a.,

1-�ta

tisl ic:

f3,,

1-s

ta1 is

1 ic

!},

r-sta

tistic:

y

1-

tatis

tic

R�

>

Full

Sam

ple

j

"J:

:I:

No

cond

ition

ing

vari

able

� ::1

;:

:; (l

) 1t

0.0

198

20.2

9 0.

0767

0

.27

-4).00

21

-3.5

5**

0.

02

:i! (2

) [rrl

( I

+ n:

)] 0.

029

4 ::?.2

.72

0.6

106

2.22

-4)

.077

9

-10

.59*

* 0

.06

(3) l

og(

rr)

-4).00

50

-2.1

2 0

.022

9

5.78

-4)

.010

8

-11

.75*

* 0

.07

5.. H

l (I -

yr1(n'

1 T')

0.

093

5 2.

39

1.52

92

-U9

-4).0

127

-

11.2

5**

0.8

700

15.09

0

.06

;n

....,

,-\nn

ual d

umm

ie.

� :r

z

(I) rr

0.

0200

2

1.19

0

.34

61

1.25

-4)

_0020

-

3.59

**

0.0

8 �

m

tru< 1

+!il

l 0

.028

7 22

.20

0

.777

1 2.

88

-4).0

704

-9.

39**

O

.ll

E

(3)

log

(rrl

-4).00

26

-1.04

0

.022

5 5.

90

-4).009

8 -

10.0

3**

0.1

1 -.:

1-t

l (I-

y)-'(

n" .

.,,

0.09

35

2.39

1.

5292

-t

-19

-4).0

127

-11

.25*

* 0

.8700

15

.09

0.10

:r.

Co

ndit

ioni

ng '

ariab

le� e

:..c:ep

t 1/G

DP

(I) n:

0

.0199

22

.72

0.5

305

2.20

-4)

.0013

-2

.99**

0.

21

(::!) l

rtlt 1

+ It

l] 0

.025

9

19.3

4

0.7

517

3.18

-4)

_0-187

-5

.86**

0.

22

(3) 1

ogHt

l 0

.0039

1.

49

0.0

176

5.

-10

-4).006

9 -6

.-+0**

0.

22

•-tl 1

1 -

-o '<rr

')

0.

0706

2.

05

1.35

30

4.15

-4)

.0094

-8.3

9**

0.8

625

11

.36

0.

22

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

All

cond

itio

nin

g v

ariab

les

(I) Jt

0

.019

8

�1

.87

0

.390

7

1.6

1 -

0.00

13

-3

.07

**

0

.�3

12

) 1nl

< 1

+ rr

>J

0.0

25

2

19.1

4

0.60

20

2

.54

-

O.O

.t-17

-

5.4

6**

0

.24

(3

) lo

g(n

) 0

.0049

1.

91

0.0

157

-

U4

-

0.006-t

-

6.04

**

0

.24

(-t

) C

l -

'f)-1(

lf ,_

.,.,

0.060

5

2.2

1 1.

0164

3

.23

-

0.00

87

-

7.9

8*

*

0.&-H

J 10

.52

0

.24

�

U

ppe

r-an

d U

pper-

:\1i

dd

le-l

nco

me

Co

untr

ie'

>

;;l

z

Co

ndit

ion

ing

va

ria

ble

s ex

cept

//G

DP

z

::

:l (

I) Jt

0

.03

20

1

1.06

-0

.279

4

-0

. 72

-0

.0018

-

2.9

4*

*

0.3

7

z

(2)[

rtl{l

+n)

] 0

.03

84

9

.19

--

0.1

57

2

-0.4

3

-0

.03

61

-2

.85

**

0

.38

:!;

(3

) lo

g{rt)

0

.02

0-t

5.5

3

0.006

3

1.18

-

0.00

59

-

3.7

0*

*

0.3

8

>

-!

r4>

n-

yJ 'c

.rr•-

1"•

0.06

97

3

.10

0

.16

17

0.4

5

-0

.008

1 -

5.5

0*

*

0.8

110

7

.15

0

.38

;:

::: z A

ll c

ond

itio

nin

g \

an

able

.> �

>

( l

lrr

0.0

27

7

9.5

4

-0

.35

71

--0.9

1 -<

-

0.00

18

-2

.96*

*

0.-t

l c:

(2

) [ lt/

( I +

rrl)

0.0

32

3

7.6

8

-0

.25

03

-

0.6

8 -0

.02

79

-

2.2

5*

0

.--tl

•·

(3)

log

ln)

0.0

188

5

.15

0

.0033

0

.61

-0

.00t5

-

2.7

9**

0

.41

>

;;>:::

(41(

1-

yl '

<rr•-

r·•

0.0

49

2 -

U9

-

0.2

509

--

0.7

7

-0

.007

1 -

5.15

*"

0

.71

49

5

.68

0

.-tl

�

Lo

wer

-a

nd

Lo

wer

-:'>.li

dd

le-l

nco

me

Cou

mne

<, p

-l

C

ond

itio

ning

var

iabl

e .. e

"l:cep

t IIG

DP

0

-<

0

(I)

n

0.0

177

8

.78

1.

06

28

3

.49

--0

.0012

-

2.4

3*

0.1

9

c

;:o

C2l

[lt/(

I +

re))

0

.0264

10

.48

1.

42

11

4.6

4 --0

.068

5

-5

.61""

* 0

.21

a

IJ)

Iog

(nl

--0.00

19

-{).4

7

0.0

25

6

6.-t

l --0

.008

8

-5

.38

*"

0.2

1 ;:o

0

14

1(1

-y

l '(

rr'

'' )

0.0

899

1.50

::!

.19-t-t

-U

S -

0.0

118

-

7.0

8*

" 0.

874

I 10

.09

0

.19

�

.o\11

co

ndit

ion

ing

va

ria

ble

'

Uln

0

.017

1 8

.70

0

.939

5

3.0

3

-0.00

1�

-2

.51"

0

.21

0>

t:!ll

nl( I

+ rrl

] 0

.02

55

10

.3-1

1.

::!89

-t

4.1

7

-0.06

57

-5

.-1-t�

0

.22

:x>

·�

t3

l lo

g(n

) -0

.0019

-0

.-16

0.0

2-l:!

5

.93

-0

.008

5

-5.3

5'"

o.n

1-t

I (I -

·o 1(

lf •

q

0.0

856

L-1-t

1.

97

37

3

.98

-

0.0

11�

-6

.80

"·

0.8

7-1-t

9

.7"2

0

.�1

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

Tab

le 2

. (co

nclu

ded)

o-

Infl

atio

n m

easu

re

a,

1-<;t

atis

tic

�.

/-st

atis

tic

oc

p,

/-st

atis

tic

R

� ""'-

Pre-1

973

Ob

serv

atio

ns

Con

diti

oni

ng ,-

aria

bles

ex

cept

1/G

DP

(I)

n:

0.0

3&6

6

.&8

-0.4

603

-1.

03

-0.0

260

-3.

23**

0

.19

(2)

[rr/(1

+ !t

)]

O.O-U

7 6

.59

-0

.337

1 -0

.78

-0.06

86

-2.

50*

0.1

9

(3) l

og

(Jtl

0.0

176

2

.15

0

.004Q

0.7

9

-0.007

1 -

2.3

1*

0.1

9

(4) (

l -

y)-'C

�' _

.,)

0.0

4-t9

5.9

:5 -0

.34

74

-0.7

7 -0

.023

3 -

2.24

* 0

.337

3 0

.96

0

.20

>

-'!

A

ll co

ndit

ioni

ng v

aria

bles

i:i

i

(I)

n:

0.0

370

6.:5

2 -0

.450

5 -

1.0

3 -0

.02::!

9 -

3.00

*"

0.2

1 0

0

(2) [

!V( I

+ n

)] 0

.039

4

6.0

8 -0

.356

5 -

0.8

3 -0

.057

0

-2.

05*

0

.21

"'

(3)

log

(r.)

0.0

201

2.4

4

0.00

27

0.4

6

-0.00

:55

-1.

75

0.2

0

(4J c

1 -

·f)-·c

�· -

l'J

0.0

-+12

7.

35

-0.4

390

-1.

0-+

-0.0

228

-2.

31*

0

.199

9 OA

9 0

.21

5 0.

Po

st-1

973

Obs

erva

tions

"'

;1 C

ond

itio

ning

var

iabl

es e

xcep

t 1/

GD

P

::::

z

U I

ll

0.0

1:53

14.0

7 0

.86

76

3.0

3 -0

.0012

-

2.9

1**

0.�

1 3!

=

f :!

)[n/

(1 +

TI)]

0

.02

12

13.7

:! 1.

11

10

3.9

9 -0

.04

77

-5.

49

**

0.2

2

c

(3)

1og(

n:)

-0.00

12

-0.4

3 0

.023

2 6

.28

-0

.0072

-6

.20

**

0.2

3

"C

:r.

(4) (

I -

y)-'

(rr'

-T')

0.0

442

2.34

1.

6066

4.49

-

0.00

8�

-7.

29**

0.8

053

8

.59

0

.1:!

All

cond

itio

nin

g \•

aria

ble

s

ll ) 1[

0

.016

1 14

.97

0.7

09

5 2A

6

-0.00

11

-2.

96**

0.2

3 (2

) [rrl

( I +

n:)J

0

.02

16

14.2

0

0.9

450

3.

37

-0.0-+

46

-5.

21*

* 0

.24

( 3

) lo

g(r

r)

0.00

04 0

.14

0

.02

14

5.6

7 -0

.0068

-6.0

3**

0.2

4 (4

)( 1

-y)

-'(Jt'

1-Y')

0

.0-+7

7 2.

05

1.34

07

3.70

-0

.0078

-

7.0

5**

0.8

259

&

.55

0.2

3

No

te: O

ne a

ste

risk

and

two

a'>t

eris

ks in

dica

te s

tati

stic

al s

ign

ific

ance

at t

he

5-an

d !

-perc

ent

I e \

eL r

espe

ctiv

e 1 y.

Sta

ti�t

ica I

sig

nifi

canc

e sh

own

only

fo

rcoe

ffici

eml3

,(es

tim

ated

�lo

pe f

or in

flati

on

abo

ve 2

11! pe

rcen

t a

)ea

r).


WARNING: INFLATION MAY BE HARMFUL TO YOUR GRO\VTH 685

Figure 3. Implied Per Capita GDP Grml'lh Rates Under Altematil·e Models 1 (In perccnl a year)

GDP grow1h

3.0 .------------------------------------------------.

2.0

1.0

0

Linear - - - - - - - - - - - - - - - - - - - - - - - - - -

20 40 lnfia1ion

60 80

1 Bivarime infiation-growth relmionship. holding constanl all other explanatory variables at mean values.

How best to choose between the models? Note that in model ( 4 ), the estimated value of y is significantly different from zero. thus rejecling I he linear model. The estimates of y tend to be somewhat smaller than (but not significantly different from) unity. at least in the variant that includes all the conditioning variables. Thus the log model cannot be rejected.

We conclude that the simple logarithmic model-with a low-inflation kink-provides a reasonable characterization of the inflation-growth relationship. Our base model then becomes:

Ay = 0.004 + 0.015 D25 llog(rr) - log(0.025)) - 0.00641og(rr)

( 1 .91) (4.74**) (6.04**)

+ 0.019 a,., - 0.027 r!GDP + 0.002 G/GDP 0.325 MOP

(2.12**) (2.66**) (0.04) (4.25**)

+ 0.008 HK + 0.1 ('f} 8/GDP - 0.009BLK - 0.02 DROUGHT

(4.46**) (5.39**) (3.24**) (7.06**)

- 0.07 DEATH + 0.0 I 0 GAP + 0.010 (X+ M)/GDP + 0.086 1/GDP ,

(2.30**) (5.40**) ( 1.85*) (6.04**)

( I )



where 025 is a dummy variable equal 10 unity when inflation i!> less than 21/1 percent. and where the coefficients on the annual dummy terms (about half of which are significant) are not reponed.1 1

All but one of the coefficient estimates shown above are statistically significant.12 and all except the one associated with terms of trade volatility have the expected signs. Of greater interest, the coefficient on log inflation is negative and significant by a wide margin. Moreover, the positive coefficient on the spline tem1 is significant. rejecting the hypothesis of no break in the relationship.D Despite the dilemma noted above, and somewhat surprisingly. there is little change in the inflation coefficient if the investment renn is dropped.1�

We examine the robustness of these base model results in the next section, but before proceeding. several points on interpretation may be useful. Inflation is of course not under direct policy control; especially in the shon run. it is more of an outcome determined by both macroeconomic policy choices and various shocks and is therefore probably best thought of as an indicator of those policy choices. We will use several methods to determine whether the inflation-growth correlation found in annual panel data is mainly spurious. being driven for example by shon-run shocks or policy responses. On the other hand. we do not attempt to identify the exact mechanisms or channels through which inflation-or the related policy choices it reflectsmight hinder output growth. Still. several points can be noted in this connection. The inflation variable in equation ( I ) evidently picks up the influence of policy choices other than high government consumption. high budget deficits. or high black market exchange rare premiums. since these are also included in the regression. Also, as shown later. the inflation variable captures something other than the effects of the inflation volatility associated with higher inflation. Finally, in whatever way inflation or its correlates influence growth. it does not seem to be mainly an indirect effect through investment.

Ill. Robustness

The question of robustness is of panicular interest in the empirical analysis of growth. since economic theory provides little guidance on the 'true''

11 Even with annual dummies. the R� of the regression is only about 0.25. With time-averaged data. however. the R" rises to as much as 0.70 (see Section Ill).

1� The exception is GIGDP; we choose to retain this regressor because it is statistically significant in the variant excluding 1/GDP.

1.1 The coefficient on the spline term implies an estimated slope of +0.086 in the low inflation range (i.e .. 0.01 5 greater than the slope elsewhere).

14 The main effect of dropping the investment term is to increase the coefricient on human capital (with which investment is highly correlated).


WARNING: INFLATION MAY BE HARMFUL TO YOUR GRO>VTH 687

specification. Here we examine the robustness of the negative inflationgrowth association.

Robustness of the Negative Inflation-Growth Association

The robustness of this association has already received some attention in the literature. Perhaps best known are the results of Levine and Zervo� ( 1 993), suggesting that inflation-growth findings can depend on a very few countries with high inflation. Simi larly. Bruno and Easterly ( 1 995) show that excluding from a growth regression all countries with inflation above 40 percent can make inflation lose statistical significance. In a more general study, Sala-i-Martin ( 1 997) finds a number of variables to be robustly associated with growth. but not inflation: however. he notes that his methodology presumes a linear relationship.

Indeed. we suspect that fai lure to allow a nonlinear association between inflation and growth is responsible for such negative results.15 As noted, the likely consequence of imposing linearity is a very large downward bias in measuring the inflation-growth slope. Accordingly. we use as our base model the growth regression reported in equation ( I ) above (with all of the conditioning variables, including annual dummies).

In examining robustness. we consider questions in four categories. First. is the estimated coefficient on (log) inflation stable across various alternate samples? One specific concern here is the role of high inflation outliers. Second, does the relationship found in annual panel data derive from both the time-series and cross-sectional dimensions? Here. the concern is that the panel results might be spurious. driven either by fixed country-specific factors or reflecting mainly shocks that induce shon-run correlations. Third. does the coefficient on log inflation remain stable when the specification of the conditioning variables is changed in various ways? Finally. we also check whether the results are sensitive to the exact placement of the low inflation kink allowed in the inflation-growth relationship.

Stability Across Samples

There is some suspicion in the literature that the apparent negative effect of inflation on growth arises mainly from a small number of outlying cases, that is, countries with unusually high innation and weak growth. For example. Levine and Zervos ( 1993) demonstrate that merely

1� While the Bruno and Easterly specification is not strictly linear. their log( I + Jt) regressor is essentially linear over the range in which most inflation observations lie. Moreover. they allow no kink in the low inflation range.



Table 3. OLS Estimatesji·om Altemmh·e Specifications and Samples

Coenicient on log inflation

Regression variant Estimate r-stati�tic Sample size

(0) Base model, full sample -0.00639 -6.04** 2231

( I ) Excluding inflation > 40 percent -0.00572 -3.79** 2034

(2) In dirterenced form 5-year changes -0.0108 -6.93** 1685 I 0-ycar changes -0.00926 -6.85** 1250 I 5-year changes -D.00837 -5.32** 871

(3) Tn differenced form, excluding inflation > 40 percent

5-year changes -{).00996 -5.07** 1459 I 0-year changes -0.00768 �.16** 1067 1 5-year changes -0.00822 �.12** 744

(4) Adding coumry dummies -0.0083 -5.78** 2231

(5) Adding change in log inflation cunent change -0.00629 -5.76** 2 2 1 8 cunent change and 2 lags -0.00597 -5.27** 2 1 8 2

(6) Base model with annual data, I 967-96 only -0.00664 -6.19** 2 1 39 excluding inflation > 40 percent -{).00635 �.1 8** 1947

(7) Pre-averaged data. 1967-96 5-year averaging -{).0050 1 -3.69** 360 I 0-year averaging -0.00303 -2.37* 150 1 5-year averaging -0.00423 -2.59* 74

(8) Pre-averaged data, 1967-96. excluding inflation > 40 percent

5-year averaging -{).00659 -3.20** 323 l 0-year averaging -{).00229 -{).80 132 15-year averaging -D.00606 -2.00 63

Note: One asterisk and two asterisk.� indicate statiMical significance at the 5- and I -percent level. respectively.

dropping Nicaragua and Uganda from a large cross-section regression can cause the observed relationship to break down. A simi lar, but somewhat broader, suspicion is that the apparent effect of inflation only becomes serious at rates above some fairly high threshold, perhaps 40 percent a year.16 Thus, while no one is likely to argue that hyperinflation is good for growth, there is much less agreement on whether inflation in the 1 0-40 percent range has any deleterious effects on growth. In contrast.

16Such an interpretation is sometimes given to the Bruno and Easterly ( I 995) study.



Figure 4. Coefficient Estimates from Regressions on lnjfmion Subsamples

Coefficient on log inflation

0.005 ....,...,_ --. ------------------------, . '

0 �-��-------------------� • +2 standard errorJ �- ... . . .... ... ..... . - - · · · · - ·· · · · ... .... ...

L�--- .- ·- - - - - - --

-- - - - - - - - - - - -· Coejjiciemestimate

-O.OOS

I j ,.- - - - -- - · -�------· -- - - - -· · - - - ----- -- -_:£;,;;;Ja�d·;;r��--

-O.OIO - : • • • • -,

. -O.QIS -:

-0.020 -+

-0.025 I 0

I I I I 20 40 60 80

Maximum inflation rate used to define subsample (in percent)

-

-

I 100

the results presented above suggest that the slope of the inflation-growth relationship is steeper in the I �0 percent range than in the range above 40 percent.

The issue is disposed of readily, by reestimating the base model in a restricted sample. Rather than excluding only select high-inflation outliers, we subject the base model to a more comprehensive and severe test, excluding all observations with inflation greater than 40 percent. n The result is reported in Table 3 (regression ( l )). The estimated coefficient changes only from -0.00639 to -0.00572, and it remains statistically significant (despite a much-reduced variation of inflation in this truncated sample).

Turning from the role of high inflation to more general questions of stability across samples, we examine what happens to the estimated coefficient as the range of inflation rates allowed in the estimation is varied systematically. We start with a very restricted sample, consisting of inflation rates in the 0-5 percent range, and then gradually expand the upper bound of the sample in small increments (2 1/2 percentage points of inflation). Figure 4

shows the coefficient estimate (±2 standard errors) in each sample. The point estimates are always negative, and the sample need include only the 0-17 1/2

17 Results excluding just Nicaragua and Uganda were even closer (essentially identical) to the full sample results.



percen£ inflation range before statistical significance is found. Of particular interest. as the sample is extended to include inflation of greater than 40 percent, the estimated coefficient does not grow in absolute value. More generally. the estimated coefficient appears fairly stable across all but the smallest samples, and the width of the standard error bands never flares but instead slowly tapers as larger. more diverse samples are considered. These are signs of a well-specified model.

Similarly, we examine stability over time by segmenting the data into time periods. Figure Sa shows the coefficient estimates, starting with a subsample consisting of observations through 1966 only. and then adding one year of data at a time. Of the 3 1 point estimates. all but one is negative (the exception occurs in the smallest subsample). While these results seem to imply that almost 20 years of data are required before a finding of statistical significance can be assured. this could reflect the lesser information in the earlier years (missing observations and lower variation of inflation). When the same procedure is run '·backwards" over time (i.e., starting with 1996 data only and progressively expanding the sample to include earlier years), it takes only two years of data to find statistical significance (Figure Sb). Note that neither Figure Sa nor Figure Sb suggests a structural break occurring at any time during the sample period. As in Figure 4. the standard eJTor bands nan·ow as the sample is increased.

The Roles of the Cross-Sectional and Time Dimensions

The above results all suggest that the log-based model (with a kink for the low inflation observations) is well specified and the negative inflationgrowth relationship robust. But do these panel data results arise from both the sample's cross-sectional and time dimensions? Results coming from only one of these dimensions would be suspect.

The panel results imply that, comparing two countries with different inflation rates, the country with lower inflation may be expected to have higher growth. But this is not necessarily the same as saying that an individual country that achieves lower inflation is likely to achieve faster growth (even ignoring any possible short-run contractionary effects of disinflation). It could be that the panel data results are driven entirely by cross-country variation in inflation or in unmeasured country-specific factors associated with inflation.

We tackle this problem f irst by focusing on the effect of changes in the inflation rate on the change in the growth rate ( the regression includes changes of all the independent variables that vary over time). This allows us to examine whether a country changing its inflation rate can expect a shift in its growth rate. while still pooling observations. l n taking changes of the



Figure Sa. CoejJicie/11 Esrimares ji·om Regressions on Time Sub.mmples (Samples stan in 1963 and end in final year or rcgrc�'ion �ub�amplc)

Coeffkiem on log inll<ttion

0.012 .-----------------------------------------------------,

0.009

0.006

0.00:1

0

-0.003

-0.()()6

-0.009

-0.0 12

-0.01 5

� +2 swntlard error:, . . # ..... ,

.. ' � ' - - -- - -

.-

"· · -' . . .· . -2 stwulartl errors .... .. ... . . . .

66 68 70 72 74 76 78 80 82 84 86 88 90 9� 94 Final )Car of regre;.;.ion ;.ub;.ample

Figure Sb. Coefjicie/11 Esrimare.1 .fi·mu Regression\ on Time Suh.,wnple.l (Samples start in llrst year of regrcs�ion subsample and end in 1996)

96


0.005 ....------------------------------,

0 �--------------------------------------------�

-{).005 - .. - .. - .. ... - · - · - · .... .. - · - · - · - - - - - - - - - - - · · +2 standard errors

·. ·,

CoefficienT estimme

- - - - - - - .. .... _ .. - .. _ .. -0.010 -2 standard errors

-0.015

-0.020 L-1-'--'-..L_L..l_J_-'-..L_L..l-L__L_.,L_.L...J-L--L_L_.L...JL...J.--L_j_..L..JL...J.--L_j_.L....J 96 94 92 90 88 86 84 8:! 80 78 76 74 72 70 68 66

Fit'l't year of rcgrc"ion 'ub;.amplc



variables, we want a fairly long horizon. since results for short horizonssay. one-year changes-might be influenced by spurious short-run correlations induced by supply (or demand) shocks. Returning to Table 3, regressions (2)-(3) repon the results for 5-. I 0-. and 1 5-year changes, including and excluding high-inflation observations.

The inflation coefficient is not even slightly dimjnished by this transformation of the base model; indeed, it becomes considerably larger in absolute value. Moreover. this result cannot be attributed to outliers (i.e., a few cases of countries moving out of, or into. very high inflation) since it holds even when observations for which either current or (long-) lagged inflation exceeds 40 percent are excluded.

The fact that the negative inflation coefficient becomes steeper when the base model is specified in terms of changes may reflect fixed. countryspecific effects: by purging the data of such effects, differencing the data could be correcting for the base model's omission of country dummy variables.'� Indeed, adding country dummies also makes the inflation-growth slope steeper (Table 3. regression (4)).

These fixed-effects results are striking. since it is all too easy to imagine that unmeasured characteristics of some countries (e.g., weak institutions, or political polarization) somehow drive them to have both low growth and high inflation, inducing at least some degree of spurious correlation in cross-sectional or panel analysis. Were this the case. however. we would expect that adding country dummies to a panel regression would diminish-rather than steepen-the inflation coefficient.

Having confirmed that the negative inflation-growth relationship is apparent in the time dimension of the data, we now address the possibility that the panel data results might reflect only or primarily this dimension. For example, supply shocks and policy responses of various kinds could induce negative short-run correlations between inflation and growth. The question is how much our panel estimate might be spuriously influenced by such short-run comovements (keeping i n mind that demand shocks may also be at work).

As a first check. we add to the base model the change in (log) inflation. This augmentation, however. as well as one also adding lagged changes, fails to move the estimated coefficient on log inflation more than slightly (Table 3, regression (5)). Moreover. none of the inflation changes terms are statistically significant. While a comprehensive analysis of inflation-

IM Country dummies were not included in the base model because some of the conditioning variables are constant or nearly constant over the sample period, causing colinearity problems. It is. however, possible to estimate the inflation coefficient despite this colinearity. (The standard errors for certain other regressors jump. but this need not concern us here.)



growth dynamics is beyond the scope of this paper, some basic results can be noted briefly. When the base model is augmented with three lagged levels of Jog inflation. or with only two such lags. none of these terms is statistically significant. When just one lagged level is added. however, it is significant (at the 5 percent level). Interestingly. the coefficient on this lagged level is positive. but the coefficient on the current level of inflation becomes more negative. It turns out that the sum of the (current and lagged) coefficients on log inflation is -0.6 10, quite similar to the base model's single coefficient of -0.639. This result suggests that it may be reasonable to think of the base model results as indicating the long-run relationship between inflation and growth.

Pre-averaging the data over time is a traditional way to reduce the potential influence of any spurious short-term correlations, although it does have its own difficulties. One of these is that the choice of dataaveraging horizon is arbitrary. 19 Another problem is that time averaging may be inappropriate when the relationship is nonlinear. Indeed, in the case of the inflation-growth association-which seems in the annual data to be negative and convex for moderate and higher inflations, but concave within the low inflation range-the potential for a bias toward zero is clear. With these caveats in mind, we report in Table 3 the coefficient on inflation using the time-averaged data. For comparability, we first repeat the annual panel data regression over 1 967-96, the data set's most recent 30 years (Table 3, regression (6J); we then run analogous regressions within this same period after averaging the data at 5-, 10-, and 15- year horizons, both with and without inflation values exceeding 40 percent (Table 3, regressions (7) and (8)).

Our main interest is the magnitude of the estimated coefficient on log inflation: in particular, does this drop off substantially as the data averaging period is extended beyond one year (i.e., annual data)? We see no clear drop off in moving from annual to 5-year data. Note also that the IS-year estimates lie between the 5- and 1 0-year estimates. Still. some of the estimates (especially those at the 1 0-year horizon) are considerably smaller than their corresponding estimates in annual panel data. As regards statistical significance, the results are mixed, as the estimated standard erTors tend to be much larger with time-averaged data. As with the annual data, the results using 5-year averaged data are significant at the I percent level. in both samples. At the two longer horizons, however, only the results in the unrestricted samples are significant (at the 5 percent level).

I? Choosing the longest available horizon for data averaging is one response to this problem, but this makes it impossible to control for country-specific ··fixed effects."'


694 ATISH GHOSH and STEVE:--J PIIILLIPS

Thus we find that researchers' arbitrary choice of data-averaging horizon can have nontrivial implications. for inflation-growth regression�.�� However. the finding of a significant inflation-growth relation!>hip i!> not limited to annual panel data. and it does not appear that the results in annual panel data are driven mainly by spurious short-term correlations. (Further support for this conclusion can be forund in Section V. where we discuss the simultaneity issue in depth and apply instrumental variables to the problem.)

In sum, the negative inflation-growth association is evident in both the cross-section and time-series dimensions of the data.

Other Augmemations. and an Extreme Bounds Test

Our base model includes a wide variety of conditioning variables: we now extend the robustness analysis by considering alternative specifications of such variables. Such analysis is important because some researchers (e.g., Levine and Zervos. 1 993; and Sala-i-Martin. 1 996) have found that inflation-growth correlations are not robust to changes in the conditioning variables. at least not when a linear relationship is imposed. We consider a number of augmentations to the base model. nonlinear specifications of the conditioning variables. and a Learner extreme bounds test.

Fischer ( 1993) argues that the inclusion of changes in the terms of trade as a regressor goes a long way toward dealing with the problem of a spurious inflation-growth correlation. In our base model we do not include tl7T because it is not significant (and considerably reduces the sample size). Nonetheless. Table 4. regression ( I ), reports the inflation coefficient when the change of the terms of trade is included in the regression. (Since many of the tl7T data are missing, the sample size shrink!>: to aid comparison. the base model estimated over this smaller sample is also reported.) We see that adding the change in the terms of trade (and its Jag) barely changes the coefficient on inflation.

Nevertheless, it is possible that oi I price shocks are responsible for the negative inflation-growth association (although the base model's annual dummies serve to control for common global shocks in any given year). Table 4. regression (2). shows the effect of adding the change in real oil prices (the average spot price of crude oil in dollars deflated by the U.S. wholesale price index) to the base model; the inflation coefficient is virtually unchanged. Since oil price changes affect countries differently. however. their effects would not be perfectly captured in this specification. To get around this complication. we also try re�tricting the regression sample to

20 Note that the base model i:-. at some disadvantage in the time-averaged regre�sions. As noted, averaging the data seems likely to bia� the inflation coefficient toward zero. Also, the regression ·s kink is still imposed at 21/� percent inflation. rather than letting the data determine a po�sibly better fit.


WAR.J\liNG: INFLATION MAY BE HARMFUL TO YOUR GROWTH 695

Table 4. Estimates Ji-mn Augmemed Regressions and Other Regression Variants


Regression variant Estimate /-statistic Sample size

(0) Base model. full sample -0.00639 -6.04** 2.231

( I ) Adding change in term), or trade current change -0.00541 -3.66** 1 .5 1 1 current and lagged change -0.00534 -3.59** 1 .5 1 1 (base model. in identical sample) -0.0054 1 -3.66** 1 .51 1

(2) Adding change in real oil prices cun·en1 and lagged change" -0.00742 -7.39** 2.231 (base model. w/o annual dummies) -0.00743 -7.60** 2.231

(3) Excluding 1973-75. 1979-8 1 . and 1 990-92 -0.00589 -4.69** 1 .570

(4) Adding institutions index (BERI) -0.00628 -5.89** 2.167

(5) Adding inflation volatility -0.00625 -5.76** 2.055

(6) Non linear conditioning variables adding squared term� -0.00683 -6.43** 2.231 adding log terms -0.00694 -6.42** 2.175 log terms replacing linear terms -0.00666 -6.20** 2.175

(7) Extreme bounds test (4.096 regression).) weakest estimate -0.0057 1 -6.25** 2.231 strongest estimate -0.01020 - 1 1 . 12** 2.231

Note: One asterisk and two asterisk� indicate Matistical significance at the 5- and 1 -percent level. respectively.

" Becau�e of colinearity between the oil price index and the annual dummies. it is necessary to exclude the latter in this case.

exclude all observations from 1 973-75. 1 979-8 1 , and 1 990-92. The resulting inflation coefficient remains statistically significant at the I percent level, with its magnitude only slightly diminished (Table 4, regression (3)).

We take account of political and economic institutions using an index created by the Business Environment Research Institute (BERI).21 When this regressor is added to the base model, the inflation coefficient and its !-statistic are virtually unchanged (Table 4, regression (4)). Indeed. the BERI index does not enter the growth equation significantly.

Another possibility is to add a term for the volatility of inflation. By making it more difficult for economic agents to discern and respond to shifts in

21 We use a composite index measuring ( I ) the degree of bureaucratic de lays: (2) the enforceabil ity of contracts; (3) the risk of national ization or expropriation: and (4) the quality of communication and transportation infrastructure. The data were kindly provided by the IRIS Center, Universi ty of Maryland, with the permission of Ted Haner. President of BERl.



relative prices. inflation volatility might be expected ro negatively affect growth. Since the average level of inflation and its standard deviation tend to be positively correlated,22 it is possible that the base model's inflation term is picking up this channel. However, adding the volatility of inflation (measured as a 3-year moving standard deviation of log inflation) to the base model reduces the estimated coefficient on log inflation only slightly (Table 4, regression (5)). Still, it is interesting to note that the coefficient on the inflation volatility term is negative (-0.0048. with /-statistic of -2.03).

While we have allowed inflation to enter the growth regression in a nonlinear manner, we have not given the conditioning variables the same degree of attention, specifying all but one in a linear fashion.23 If other non linear relationships were allowed, would the coefficient on inflation be much affected? To check, we add squared terms for each regressor (other than inflation and the dummies). Interestingly. 6 of these I 0 new regressors turn out to be statistically significant, but their inclusion fails to diminish the inflation-growth relationship, with the coefficient on log inflation actually growing slightly (Table 4, regression (6)). We also try using the logarithms of the independent variables (for those variables that do not have numerous negative values). both adding these to the base regression and using them to replace the linear terms. Again, the effect is to slightly increase the absolute value of the inflation coefficient, which remains statistically significant by a wide margin.

Going beyond such augmentations of the base model, we also perform a Learner extreme bounds test on the inflation term. An extreme bounds test determines whether the inflation term is a/wavs significant regardless of which combination from a (finite) set of conditioning variables is included as regressors. Thus we run all possible regressions based on the 1 2 conditioning variables in the base model ( l ). All regressions include the annual dummies, the log of inflation, and the low-inflation kink term. This gives 212 possible combinations (ranging from no additional variables to all 1 2 variables). In contrast to others' results. we find that inflation does enter robustly: in over 4,000 regressions. the inflation coefficient is signi ticantly negative in all cases.24 lndeed, the coefficient estimates range from -0.0057 to -0.0 102, with the associated /-statistics ranging from -6.25 to -1 1 . 1 2 (Table 4). Moreover, limiting the data set to observations with inflation below 40 percent does nor alter this finding.

22 Ball ( 1992) discusses why the average level of inflation and its volatility tend to be correlated. Judson and Orphan ides ( 1 996) use intra year volatility of inflation and find a significant effect. A problem with using intrayear volati!ity is that much of it may be seasonal.

2' An exception is the exchange rate premium, for which the logarithm is used. 24 As Sala-i-Martin emphasizes. his test is less "extreme," requiring only that a

weighted average of the r-statistics be significant.



What accounts for the difference between our results and those of others? The negative extreme bounds results of both Levine and Zervos and Sala-i-Martin are based on a strictly linear specification of the inflationgrowth association. As we have seen. the linear model is misspecified and subject to a severe downward bias.

The Role, and Interpretation, of the Kink at 2112 Percent Inflation

The need to allow sorne kjnk in the low-inflation range was first emphasized by Sarel ( 1 996). Our placement of this kjnk at 2 1/2 percent inflation is suggested first by visual inspection of the (bivariate) inflation-growth relationship (recall Figure 2); it also happens to be the placement that yields the best fit of the multivariate regression.25 The basic results are not sensitive to this placement, however. For example, specifying the kink at possible arbitrary definitions of ·'Jow inflation," such as 5 or even I 0 percent, yields similar results for the estimated slope to the right of the kink.26

We would not interpret the results of this study as indicating precisely 2 1/2 percent as an optimal or growth-maximjzing rate of inflation. Rather, our interest is in whether a robust negative inflation-growth relationship is limited only to the high inflation range-say, above 40 percent-or whether it extends down much further, say to the single-digit range. Since all our findings point to the latter, it is natural to wonder exactly how far down the negative relationship extends, but we leave this more precise and therefore more difficult question to other researchers. For the record, in a likelihood ratio test, we cannot reject the alternative specification of a kink at 3 percent, but we can reject the alternative of a kink at 5 percent inflationY Agrunst this apparent precision. however, one should consider others' recent results. based on somewhat different samples and regression specifications: Sarel ( 1 996) found that a kink at about 8 percent inflation gave the best fit. while IMF (forthcoming) found a best fit at about 5 percent inflation.

IV. Thresholds and Interactions: A Decision-Tree Technique

Just as there are threshold effects of inflation on growth, there may be threshold effects of the other determi111ants of GDP growth. For instance,

25That is, the R� of the multivariate regression has a maximum when the kink is at 2 1/2 perce!ll (searching in 1/2-point steps between 1/2 and 20 percem inflation).

26The negative coefficient becomes somewhat larger in absolute value, and the associated t-statistic remains clearly significant at about -6.

27 The X2 ( I ) statistic for the likelihood ratio test is 1 .62 for the alternative of a kink at 3 percent, and 6.92 for the alternative of a kink at 5 percent.



even if most marginal increases in school enrollment rates have only small effects on growth, there may be some threshold level below which growth suffers greatly because of a lack of sufficient human capital. Moreover. the inreraCiion between inflation and other growth determinants may be nonlinear and complex. For example, perhaps having "low

.. human capital

essentially determines slow growth for some countries, almost regardless of their inflation rate. while countries with "high" human capital have a potential for either average or very high growth. with their inflation rate largely determining their position within this range.

In principle. a regression analysis could deal with such complications by including enough interactive dummy variables. Thus. there would be a dummy variable for low human capital and high inflation and severe terms of trade shocks. another dummy variable for high human capital and high in Ration and severe terms of trade shocks. and so on. In practice, this is quite infeasible. since theory provides little guidance and the number of potential interaction specifications is vast. At best. a few arbitrarily chosen dummy variables could be included.

Fortunately, more systematic methods are available. Recently, Ghosh and Wolf ( I 998) have proposed the use of binary recursil'e trees as means to identify the most important determinants of economic growth.�8 This technique, while less familiar than standard regression analysis, is actually much simpler and perhaps more intuitive. A binary recursive tree begins from observations being classified as either "high growth'' or ·'low growth."29 After a researcher proposes a set of possible determinants of growth performance. a search algorithm creates a hierarchal decision ''tree'' by sequentially splitting the sample observations imo (predicted) high and low growth groups, based on the values of the explanatory variables. Thus. at each branch of the decision tree, the algo1ithm finds the explanatory variable (and the associated threshold point of that variable) that best separates the high-growth observations from the low-growth observations.

For example, suppose that human capital is positively con·elated with high growth. Of course the correlation will not be perfect, and there will be some countries that have plenty of human capital but low growth (a type l error), or that have little human capital but high growth (a type II error). The algorithm would search over all observed values of the human capital variable until it finds the threshold value at which the number of such errors is mjnimized.

�RA non linear discriminator technique, recursive trees are often used in the medical sciences. for example, to analyze the determinants of patient mortality.

29 For case of interpretation. Jhis type of analysis is usually done on binary variables. Here. we define "high .. -growth observations as those in the top third of the data set, and ··1ow .. -growth observations as those in Lhe bol!om third. The middle third is excluded from the analysis.



The algorithm then repeats this process for each of the proposed determinants of growth. The variable (and its associated threshold) that minimizes the number of errors is chosen to form the first branch of the tree (with the sample now split into two). The process then continues, generating from each branch further subbranches until a terminal point is reached. To restrict the tree to a sensible (and interpretable) size, a stopping rulesomewhat like an adjusted R2 statistic-eventually stops the tree from splitting into further subbranches 30

Such an exercise has several advantages over standard growth regression analysis. First, it allows for general complementarities between the different regressors-thus, the effects of inflation on growth. for instance, are allowed to vary according to the value of the other variables. Second, the branch level at which an explanatory variable appears provides an intuitive measure of its importance in determining growth. Third, and perhaps most important for our purposes, the results tend to be robust to outliers and are invariant to any monotone transformation of the variables.JI

Figure 6. following Ghosh and Wolf ( 1 998), illustrates such a decision tree applied to our data set, with all of the regressors in equation ( I ) offered to the search algorithm as potential explanatory variables. The first branch turns out to be based on the investment ratio: countries with investment ratios below 22 percent have only a 0.36 probability of high growth, whereas those with investment rates above 22 percent have a 0.65

probability of high growth. For the countries with low investmenr, the second branching depends

upon the level of human capital. Countries with low human capital have a 0.32 probability of high growth (conditional on being at that node. i.e .. being a low-investment country) versus 0.5 I probability for countries with high human capital. The third branch depends upon the inflation rate, with countries that have less than 1 5 percent inflation a year almost doubling their chances of high growth, from 0.37 to 0.65.

On the right-hand side of the tree in Figure 6 (countries with high investment). the second branching depends on the inflation rate: countries with inflation below 14 percent have a 0.75 (conditional) probability of high growth. while countries with higher inflation have only a 0.45 conditional probability of high growth.

10 lt would. of course. be possible to continue subdividing the tree until every observation is in its own branch. This would be akin to including a� many variables as observations in a regression and thus auaining a perfect fit. 11 The procedure's focus on classifying cases according to their position above or below threshold levels is similar to analysis focu ing on variable�· medians rather than means.


700 ATISH GHOSH ;�nd STEVEN PHILLIPS

Figure 6. Binw:,· Recursirl' Tree of High- Versu.v Low-COP Growth

Ye, 591

� G

175 1

� " "

160 0.51

t

No /).65 70

-1

1

55

I v., t 17 1

0.5.1

G ! v��

.----- 161 ------c

Tree continues

(not drawn)

11.57

1\'o

10

t

0 Note: All decision criteria in percent. unless otherwise noted. Figures in italic� are the proba

bility of high growth. conditional on being at the current node. Source: Ghosh and Wolf ( 1997).


WARNING: INFLATION MAY BE HARMFUL TO YOUR GROWTH 70 I

The tree continues with further subbranches (the tree reported in Ghosh and Wolf. for instance. has a total of 8 nodes).l2 For our purposes. however. it suffices to note that-of the various explanatory variablesonly physical and (to a lesser extent) human capital are beuer able to discriminate between low- and high-growth countries than the country's inflation rate.

These trees draw a fairly complex picture of the interaction between the various determinants of output growth, thus highlighting the limitations of regression analysis. Of course. this technique has its own limitations. but it provides an interesting complement to the regression analysis discussed earlier. As before. the basic finding is that lower inflation is associated with faster growth. Moreover, we again see no sign that the negative effects of inflation only begin, or begin to pick up, after inflation has become rather high.

V. Simultaneity

The results reported above suggest no reason for skepticism about the existence of a robust negative inflation-growth relationship, but this correlation should be interpreted with some caution. A particular concern is that if growth somehow negatively influences inflation. then the inflationgrowth findings presented above could. at least in part, reflect simultaneity bias. In the absence of a methodology to tell us whether inflation causes lower growth. here we pursue a more modest goal: to check that the negative inflation-growth correlation does not disappear once an effort is made to remove simultaneity bias using instrumental variables. In fact. several authors have already shown that this correlation survives this kind of test (see, e.g .. Barro, 1 995, and Cukierman and others. 1993).

In choosing instruments, it is helpful to consider how growth might negatively affect inflation. One potential channel can be seen by considering a simple money demand function, with real money demand as a function of real income. Taking logs and first differences:

full - !J.p = cx.!J.y.

where ex. is income elasticity of money demand. If !J.m is not immediately adjusted to growth shocks and if ex. i s fixed, this money demand function would imply a negative correlation between inflation and growth. As Barro

31 Based on the algorithm. the full tree chooses only 1/GDP. HK. n, 6.POP. G/Y, and GAP as explanatOry variables.



( 1996) argues. however, the simultaneity bias arising from this channel is probably not very important. since for plausible values of a. shocks to output growth are too small to account for much of the observed variation in inflation rates.33 In turn, if there is a large variation in inflation rates. then the component of the shock to inflation that i s correlated with the shock to GDP growth must be small. Moreover, outside the short run. one might expect policymakers to adjust !:!.m in response to. and in the same direction as, changes i n trend !:ly: this also suggests that simultaneity bias might not be much of a concern. To the extent that !:!.m were adjusted negatiFely i n response t o short-term changes in real growth. however. simultaneity would become more of an issue. 3�

Note that the above discussion centers on within-country variation. primarily around short-term responses to output shocks; in this context, the potential for simultaneity bias seems clear. On the other hand, it is difficult to see why moving from one steady-stare growth rate to another might itself lead policymakers to pursue a different steady-state inflation rate.'' It is therefore not clear what instruments would adequately deal with such "long-term'' channels of simultaneity.

Our method is two-stage least squares (2SLS): we first regress log inflation on a set of instnunents, each entered as both linear and squared terms. We then use the fitted log inflation values i n a growth regression (again, the base model used in Sections II and III).36 As always, the validity of potential instruments is an issue. Thus. such variables as the ratio of the fiscal deficit to GDP. lagged money growth, and lagged inflation might be expected to be correlated with inflation, but their validity as instruments is suspect.37 We use instead instruments in several other categories. The first is the nominal exchange rate regime: for example, Ghosh. Guide. Ostry. and Wolf ( 1996) show that pegged exchange rate regimes are associated with lower inflation. Second. we consider three measures of legal central bank independence. as well as the centTal bank governor turnover rate (a proxy for independence): these are

·'' For example. 0: equal to 0.5 in Cagan 's formulation and ranging up to the unit elastic case.

'�This might be the case if tax revenues were countercyclical and policymakcrs used seigniorage (and the inflation tax) to complement conventional tax receiptl>.

'� Possibly. considerations of optimal seigniorage would lead governments with small tax bases to compensate by choosing a higher "inflation tax ..

. However. while

it seems plausible that weak tax bases could be produced by a low leJ•e/ of per capi ta output. it is not clear why they would be correlated with low growth. (Recall that our growth regression controls for countries· initial level of output.)

'6Thus, we do not instrument for the model's regressors other than inflation. As usual. we view these merely as conditioning variables; we arc therefore not concerned with any simultaneity that might affect their associated coefficients.

'7 1n principle. (sufficiently long-) lagged version� of these variables might be valid instruments.



Table 5. Two-Stage Least Squares Results

Coefficient on log i nflation R� of Sample Instrument set" Estimate I·StatiMiC tirst stage size

I . Exchange regime -0.00622 -2. 1 9* 0.17 2,130

2. Legal central bank independence -0.00781 -2.35* 0.13 1.906

3. Central bank governor turnover 0.00621 1 .63 0.20 1.955

4. I + 2 + 3 -0.00001 0.00 0.30 1.602

5. I + 2 -0.00704 -2.69** 0.18 1.816

Memorandum item: OLS regression -0.00639 -6.0-1** 2.23 1

Note: One asterisk and two asterisks indicate statistical significance at the 5- and !-percent level. respectively.

"All 2SLS variants include annual dummies in the first-stage regression.

reported by Cukierman ( 1 992). Finally, we also use the base model's Lime dummies as instruments. Reflecting our interest in assessing robustness, we use a number of different combinations of these instruments.

The 2SLS findings, shown in Table 5. turn out to be sensitive to the choice of instruments. Thus, the results based on the exchange regime indicators, or on the measurements of legal central bank independence, or on both sets together, would suggest that essentially no part of the correlation between inflation and growth reflects a growth-to-inflation channel. That is. the magnitudes of the inflation coefficient estimates are very nearly as great as, or are greater than, the negative ordinary least squares (OLS) estimate. Moreover. although their standard errors are much larger than in the OLS case, these three estimates are statistically significant, at least at the 5 percent level. In contrast. when the central bank governor turnover rate is used as an instrument. the estimated coefficient on log inflation is positive, albeit not statistically significantly different from zero. Alternatively, using the turnover rate together with all the other instruments. the inflation coefficient estimate is negative but extremely small.

Thus. the 2SLS estimates paint a somewhat mixed picture. Results using several sets of instruments suggest that the strong OLS results in Sections I l and I l l are not even slightly influenced by simultaneity bias. On the other hand. using the central bank governor turnover rate as an instrument upsets this result. However, a key shortcoming of this instrument is that it is available only as an average rate over 1950-89 (Cukierman. 1 992); without any time variation. it is probably a poor instrument for a panel regression.


704 ATISH GHOSH <llld STEVEN PHILLIPS

VI. DisinHation and Growth

If inflation is bad for growth. is disinflation good? Not necessarily. ln particular, the process of disinflation may lower GDP growth, at least in the short run. To the extent that high inflation is bad for growth. of course. any such negative effects of disinflation may be offset, at least partly, by the benefits of lower inflation. ln this section. we focus on the simple contemporaneous association between growth and changes in inflation rate.-1x

Figure 7 provides a first pass at this issue. Plotted along the x-axis is the current inflation rate (rr). along the y-axis the percent (not percentage point) change in the inflation rate since the last period (D.rr/rr J ) , and along the vertical axis is the GDP growth rate (D.y). Along the rr dimension. the response surface is downward sloping: higher inflation is again seen to be associated with lower GDP growth. Along the (D.rr/rr_d dimension, things are more complicated. For low cun-ent inflation rates (rr < I 0), growth is decreasing with disinflation. At higher inflation rates. however, the surface flattens, until at 20 percent inflation growth is increasing with disinflation (the surface slopes downward for D.rr/rr 1 > 0).

Table 6 reports the results of a simple regression intended to capture the impact effect of severe disinflation. The base model is augmented to include four dummy variables: (rr_1 < 0.1 0, and D.rr/rr_1 < -0.5 ). (rr_1 < 0.1 0. and -0.5

< D.rr/rr_1 < -0.2), (rr_1 > 0 . 1 0. and D.rr/rr_1 < -0.5) and (rr_1 > 0. 1 0, and -0.5

< D.rr/rr 1 < -0.2). in addition to the usual explanatory variables. Results including and excluding 1/GDP are given.

The results suggest that. when the initial level of inflation is above I 0 percent a year. even severe disinflations (at least halving the inflation rate) do not have a negative impact on output growth. More moderate disinflations, indeed, are associated with 0.8-0.9 percentage points higher GDP growth (/-statistic: 2.62 and 2.41 ).

On the other hand. when the initial level of inflation is below I 0 percent a year, severe disinflations are associated with a fall in GDP growth of about I percentage point (with the effect statistically significant at the 5 or I percent level). More moderate disinflations are also associated with lower GDP growth. by about 0.5 percentage points, except for the upperincome and upper-middle-income countries, where growth picks up with moderate disinflation.

·18 Thus we do not anempt to disentangle short-run and long-run effects of moving from one inflation rate to another, nor do we consider whether possible contractionary effects of disinflation on the level of output might be permanent or temporary.


WARNING: INFLATION MAY BE HARMFUL TO YOUR GROWfH 705

Figure 7. Per Capita GDP Grml'lil Versus b(/iarion and Di�il(/farion (In perce/11 a year)

GDP growth

5

4

Of course. the I 0 percent inflation cutoff. and the definition of ··severe'" and ··moderate'" disinflations are chosen arbitrarily on the basis of Figure 7. A more methodical approach is to maximize the likelihood function. where the two dimensions are the initial level of inflation and the degree of disinflation:

(2)

-:.iJ = I if !1TCj TC . < Mr.j TC < !1rcj TT.; __ ,

I n these regressions, we control for current inflation. but we exclude the investment ratio because some of the adverse growth effects associated with

©In

tern

atio

nal M

onet

ary

Fund

. Not

for R

edis

tribu

tion

Ta

ble

6. E

ffe

cts

of D

isin

jfm

ion

on

GD

P G

mw

rll

--.1

rr

<0

.10

n

_,

> 0

.10

0

:J'

!lrUFI

< -

-().5

-

-0.5

< 0.

1tht <

...{I

. 2

�rJ

rr. <

--0.5

-

t>.5

< !l

rrln

< --0

.2

Co

effic

ien

t /-

sta

tist

ic

Coe

ffic

ient

r-

s1a

1 i st

ic

Co

effic

ient

t-

tati

stic

C

oeffi

cie

nt

t-st

atis

tic

R:

Fu

ll S

amp

le

Inc

lud

ing

1/G

DP

--0

.013

3 -

2.6

7*

*

--0.00

86

-

3.0

5*

*

0.00

51

I . ().

.I. 0

.0090

2.

62"

*

0.2

-l

Exc

lud

ing

JIG

DP

-

{).0

168

-

3.32

**

--0.009

7

-3

.34*

*

O.OO-l

3

0.8

7

0.00

83

2

AI

"

0.2

2

Up

per-

lno

om

e an

d U

ppe

r-1\

tid

dle

-lnc

ome

Co

um

rie

s >

Inc

lud

ing

1/G

DP

-{)

.005

8

-2

.20

*

0.0

22

9

2A

5*

0

.012

2

2A

5*

-t>

.0003

-0

.09

O.·

B

::::?

tr.

E:-.

.clu

din

g 1

/GD

P

-{)

.007

0

-2

.57

*

0.0

219

2

.30

* 0

.010

5

2.0

1*

0.00

29

0

.81

O . .W

c:

L

ow

er-a

nd

Lo

wer

-\1i

dd

le-l

nco

me

Co

un

trie

s c

lnc

lud

in!!

1/G

DP

-t>

.02

09

-2

.98

**

-

0.0

123

-

2.0

8*

-0

.0018

-

0.3

2

0.006

7

1.5

1

0.2

1 (10

Ex

clu

dir{

g 1

/GD

P

-t>.0

24

0

-3..1

1**

-0

.013

6

-2

.30

*

--0.00

2-l

-

O.-l

2

0.006

1 1

.38

0

.20

=

"-P

re-

197

3 O

bser

. at

ion

� "'

-l

Inclu

din

g II

GD

P

-t>.O

107

-

1.0

8

-0

.015

9

-2

.09

*

--()

.0092

-

0.6

1 0

.01

15

0.9

8

0.2

3 :r

<

Ex

clu

din

g 1/

GD

P

--0.0

124

-

1.2

6

-0

.0166

-

2.10*

-O

.oon

-

O.-l

9

0.0

110

0

.95

0

.2

1 z

Post-

197

3 O

bse

rvat

ion�

::

::

Inc

lud

ing

fiG

DP

--0

.0 1

7�

-2

.86

*"

-

0.00

78

-

2.6

2"

*

0.006

2

1.1

7

0.00

87

2

.. -l-+

*

0.2

4

c

Exc

lud

ing

1/G

DP

--0

.02

1 I

-3

.38

**

-

0.00

87

-

2.8

9"'

*

0.00

53

0

.98

0

.007

9

2.2

2*

0

.22

,;,

Peg

ged

Rat

es

Incl

ud

ing

fiG

DP

--0

.0 I

19

-

1.9

7*

-0

.01

13

-3

.12

*"

0

.0065

0

.95

0

.0100

1 .

. 97

*

0.2

2

EAc

lud

ir{g

IIG

DP

--0

.015

2

-2

.50

*

-0

.01

24

-

3.3

8*

*

0.006

1 0

.89

0

.()()9.;

1.86

0

.20

Flo

ati

ng

Ra

tes

Inc

lud

ing

1/G

DP

-

0.0

156

-

IAO

-t>

.0003

-0

.07

0

.002

8

0.4

3

0.00

51

1.1 I

0

.35

Exc

lud

i.i""g

1/G

DP

-

0.0

178

-

1.61

-0

.0001

-0

.02

0

.0009

0.1

3

0.00

39

0

.81

0.3

3

No

te: O

ne a

�teri

.,k a

nd t

wo

a:-.t

eri.,

k�

ind

ica

te ;,

tati

-..tic

al

sig

nil

ican

ce a

t th

e 5

-a

nd 1

-per

cen

t le

' e

l. r

espe

ctiv

e]�.


WARNING: INFLATION MAY BE HARMFUL TO YOUR GROWfH 707

disinfiation may arise from contractionary effects on investment. 39 The cor-responding estimates are as follows:

For 7t_1 < 0.063 {0.023 +0.0 14 log(n:/n: 1 ) if n:/n:_l < -0.48 (2.25) (2.16)

d\'= { 0.021 +0.0 1 7 log(1t/1t 1) if -0.48 < 7t/7t_ l < 0. 70 ( 16.89) (4.02) { 0.027 -0.005 log(n:hr_1) if n:/7t_l > 0. 70. (5.08) ( 1 .58)

For 7t_1 > 0.063 { 0.046 +0.0 1 5 log(n:/n:_1 ) if 1t/1t 1 < -0.63 (5.52) (3.93)

ily = { 0.019 -0.01 1 log(n:/n:_1 ) if -0.63 < n:/n:_l < 1 .28 ( I 7.0) (3.65) { 0.029 -0.03 1 log(n:/n:_J) if n:/n:_l > 1 .28. (0.49) (0.54)

Thus, the procedure segments the data according to whether the initial level of inflation is above or below about 6 percent (in Table 6. the cutoff was chosen at an inflation rate of l 0 percent a year). When the initial inflation rate is above 6.3 percent a year. only the most severe disinflationscutting the inflation rate by more than 63 percent (not percentage points)-is associated with lower growth.�0 Except for these severe disinflations. however. an increase in the rate of inflation is associated with lower GDP growth.

When the initial inflation rate is below 6 percent a year, severe disinflations are again associated with lower GDP growth-as are increases in inflation by more than 70 percent (not percentage points).

l t bears emphasizing that these effects are conditional on the current inflation rate (the current inflation rate is included among the regressors).

w Controlling for investmem actually makes li11le difference to the result�. Starting from low inflation (1t_1 < 0.063), increases in inflation are associated with higher growth (and disinflations with lower growth). If the inflation rate rise� by more than 70 percent. however, there i� again a negative impact on growth. When the starting inflation rate is above 6 percent. however. only disinflations of more than 70 percent (-C11t/1t.1 < -0.70) are associated with lower growth. Again. these regressions are conditional on the current inflation rate.

"0 Notice that the coefficient on log(tln:/Jt) is positive here, meaning that-over this range-an increase in the inflation rate would raise growth.



Thus. even the effect� of severe disinflation wi 11 be partly offset by the (positive) effects of lower cunent inflation.

While one should not take these results too literally, especially without an examination of their robustness. they are at least consistent with the idea that. starting from even moderately high inflation rates, all but the most severe disinflations are beneficial for growth. even in the short run.�1 When the starting inflation rate is already low. however. greater caution may be required. In all likelihood, it is not the fact of disinflation itself that matters for short-run growth-rather. that rapid disinflation will generally be associated with tightening monetary conditions. Thus the disinflation variable can be replaced by. say, the change in real money or real credit growth, with broadly similar results.�l

VII. Conclusions

There are several reasons why governmenb might want to achieve low in Ration. perhaps the most compelling being the potential for faster output growth. Indeed. of the various factors that might affect growth. perhaps none is as readily changed in the shott run as the inflation rate. Few would doubt the negative growth effects of high inRation-say, above 40 percent a year-but there has been much less consensus on the effect of less severe inflation. Yet from a policy perspective it is lhe moderate or intermediate inflation range-perhaps 5 to 30 percent a year-that is of greatest interest.

The results presented here suggest a negative relationship between inflation and growth that is both statistically and economically significant. The relationship is nonlinear, in two senses: first. at very low inflation rates. the relationship is positive; second. at all other inflation rates. the apparent 111arginal effect of inflation on growth becomes less important as higher in nation rates are considered. Failure to take account of both these nonlinearities can seriously bias results toward finding only a small effect. giving the misleading impression that inflation must become quite high before its cumulatil'e effect becomes important.

We cannot of course claim to have shown that inflation causes lower growth: indeed. it is difticult to conceive of any methodology that would

41 Focusing on cases with �ubstantially higher initial inflation. Bruno and Easterly ( 1995) find that growth resumes almost immediately after disinflation.

4�Starting from inflation below 8 percent. contractions of the real money -;upply of greater than 7 percent are associated with lower growth-allhough the effect is not statistically significant (/-statistic: 1.0 I): when the starting level of inflation is above 8 percem. real contractions of the money supply of greater than 1 2 percent are a�sociated with lower GDP growth. but again the effect is not statistically significant.



decisively prove causality from inflation to growth. Rather. this study's more modest contribution is its failure. despite a battery of tests. to find any evidence that casts doubt on the idea that inflation (or the policy choices i t

reflects) reduces growth. Of course, inflation i s not under direct policy control: especially in the short run. it is an outcome of both macroeconomic policy choices and exogenous shocks. Inflation is therefore probably best thought of as an indicator of those policy choice�. Sti 11. we rind no sign that the inflation-growth association found in annual panel data is spurious, arising only from short-run correlations induced by shocks. Moreover. while we have not sought to identify the particular mechanisms or channels through which inflation (or its associated policy choices) might hinder growth, it is interesting that a statistically and economically significant inflation-growth association is found even controlling for such likely policy correlates as government consumption. fiscal deficits. and black market exchange rate premiums.

Finally. it bears emphasizing that this study does not claim to precisely locate a "growth-maximizing·· rate of inflation (any such rate might be expected to differ. at least somewhat. across countries). Rather. our focus is on the more basic question of whether the negative inflation-growth relationship occurs only at very high inflation rates, or whether it extends down much further, perhaps to the single-digit range. All our findings suggest the latter. Exactly how far this negative relationship extends. however, remains an open and difficult question-and one worthy of future research.

REFERENCES

Ball. Laurence. 1992. "Why Does High Inflation Rai�e Inflation Uncertainty?" Joumal of Mone/al)' Economics, Vol. 29 (June). pp. 371-88.

Barro, Roben J., 1995. "Inflation and Economic Growth," Bank of England Quanerfy Bufle1i11. Vol. 35 (May). pp. 166-76.

---. 1996. "Determinants or Economic Growth: A Cross-Country Empirical Study," NBER Working Paper No. 5698 (Cambridge. Ma�sachusem: National Bureau of Economic Research).

Bruno. Michael. and William Easterly, 1995. "Inflation Crises and Long-Run Growth." NBER Working Paper No. 5209 (Cambridge. Massachuscus: National Bureau of Economic Research).

Clark. Todd E., 1993. "Cross-Country Evidence on Long Run Growth and Inflation." Federal Reserve Bank of Kansas City. Research Working Paper No. 93-05; also published in Eronomic lnqui1:r. Vol. 35 (January). pp. 70-8 1 .

Cukierman. Alex. 1992, Ce111raf Bank S1ra1egy. Credibifil.\', and Independence: TheOt:r and E1•idence (Cambridge. Massachusetts: M IT Pres�).


7 1 0 ATISH GHOSH and STEVEN PHILLIPS

---. and others. 1 993. "Central Bank Independence. Growth. Investment. and Real Rates," Carnegie-Rochesrer COI!f'erence Series on Public Policy, Vol. 39 (December). pp. 95-140.

Easterly, Williarn, 1996. "When Is Stabilization Expansionary? Evidence from High Inflation." Economic Policy (April). pp. 67-107.

Fischer, Stanley. 1993. "The Role of Macroeconomic Facwrs in Growth." Journal of Monerary Economics. Vol. 32 (December). pp. 485-5 12.

Ghosh. Atish, Anne-Marie Guide. Jonathan Ostry. and Holger Wolf, 1996. ''Does the Exchange Rate Regime Matter?" NBER Working Paper No. 5874 (Cambridge, Massachusetts: National Bureau of Economic Research).

Ghosh. Atish, and Holger Wolf, 1998, ·'Thresholds and Context Dependence in Growth," NBER Working Paper No. 6480 (Cambridge. Massachusetts: National Bureau of Economic Research).

International Monetary Fund. forthcoming. Economic Adjusrmenr cmd Reform in Low-Income Cowuries: Srudies by rhe Sraff of the IMF, ed. by Hugh Bredenkamp and Susan Schadler (Washington: IMF).

Judson, Ruth. and Athanasios Orphanides, 1996, "Inflation, Volatility and Growth.'' Finance and Economics Discus�ion Paper No. 96- 19 (Washington: Board of Governors of the Federal Reserve System).

Levine. Ross. and Sara Zervos. 1993. "Looking at the Facts: What We Know About Policy and Growth from Cross-Country Analysis." World Bank Policy Research Working Paper o. 1 1 1 5 (Washington: World Bank).

Phillips, A .. 1958. "The Relation Between Unemployment and the Rate of Change of Money Wage Rates in the United Kingdom. 1861-1957.'' Economica. Vol. 25 (November). pp. 283-99.

Sala-i-Martin, Xavier. 1997, "I Just Ran Two Million Regression� ... American Economic Review. Papers and Proceedings, Vol. 87 (May). pp. 1 78-83.

Sarel, Michael, 1996, ·'Non linear Effect� of Inflation on Economic Growth," Staff Papers, International Monetary Fund, Vol. 43 (March), pp. 1 99-215.


ll\1 F Staff Papers Vol. 45. No. 4 (D�ccmbcr 19981

© 1998 lntcrmuional Moncwr·y Fund

IMF Working Papers

Staff Papers draws 011 IMF Working Papers. which are research studies by members ofrhe Fund's staff. A list of Working Papers issued in /998:3 follows:

"The Impact of Economic Security on Bank Deposits and Investment" by Michael Fabricius [98/98]

"Explaining Investment in the WAEMU'' by Athanasios Vamvakidis [98/99]

"From Inflation to Growth: Eight Years of Transition•· by Peter Christolfersen and Peter Doyle [98/100]

"Inflation and Money Demand in Albania" by Sanjay Kalra [98/ 10 1 1 "How Big Js the Brain Drain?" by William J. Carrington and Enrica Dctragiache

[9811021 ·'Policy Responses to External Imbalances in Emerging Market Economies

Further Empirical Results" by Chorng-Huey Wong and Luis Carranza [98/1 03 J

"Determinants of Growth in an Error-Correction Model for El Salvador" by Armando Morales [981104]

"Developing Countries and the Globalization of Financial Market�" by Malcolm Knight [9811051

··Disinflation in Spain: The Recent Experience" by Nicolas Sobczak 198/1061 "Exchange Rate Fluctuations and Trade Flows: Evidence from the European

Union" by Giovanni Dell' Ariccia [9811 071 ··structural Reforms in Government Bond Markets" by Mark De Broeck,

Dominique Guil laurne, and Emmanuel Van der Stichele [98/ I 081 "Considerations in Reducing Inflation From Low to Lower Levels" by Michael

Leidy and Stephen Tokarick 19811091

"Interest Spreads in Banking: Costs, Financial Taxation. Market Power, and Loan Quality in the Colombian Case, I 974-96" by Adolfo Barajas. Roberto Steiner. and Natalia Salazar [98/1 10]

·'Welfare Cost of (Low) Inflation: A General Equilibrium Perspective" by .1-Iowell H. Zee [98/1 I I ]

"Liability-Creating Versus Non-Liability-Creating Fiscal Stabilization Policies: Ricardian Equivalence. Fiscal Stabilization, and EMU" by Tamim Bayourni and Paul R. Masson [9811 12]

7 E I


7 1 2 IMF WORKING PAPERS

"Market-Based Policy Instruments for Systemic Bank Restructuring·· by Claudia Dziobek 198/1 1 3 ]

"The Transmission o f Monetary Policy i n Israel" by Fiorella De Fiore [98/1 14]

··consequences of the Economic and Monetary Union for the Coordination of Tax Systems in the European Union: Lessons from the U.S. Experience" by Vito Tanzi and Howell H. Zee [981 1 15 1

"Impact of European Union Association Agreements on Mediterranean Countries" by Henri Ghesquiere 198/1 16]

"Deviations from Uncovered. Interest Parity: A Global Guide to Where the Action

Is" by Evan Tanner 198/1 17 ]

"Macroeconomic Performance Under AILCrnative Exchange Rate Regimes: Doe� Wage Indexation Matter?" by Esteban Jadrcsic [98/1 18]

"Search Unemployment with Advance Notice·· by Pietro Garibaldi 198/1 19 ]

"The Demise of the Nation State?" by Vi to Tanzi [ 98/1201

"Capital Structures and Portfolio Composition During Banking Crisis: Lessons from Argentina 1995" by Alberto M. Ramos 198/ 1 2 1 ]

"Tax Smoothing in a Financially Repressed Economy: Evidence from India" by Paul Cashin. Nilss Olekalns, and Ratna Sahay 198/ 1221

"Does the Long-Run PPP Hypothesis Hold for Africa?: Evidence from Panel eointegration Study" by Jun Nagayasu 198/123]

"Self-Fulfilling Risk Predictions: An Application to Speculative Attacks" by Robert P. Flood and Nancy P. Marion [981124]

"Fixed Investment and Capital Flows: A Real Options Approach" by Jorge A. Chan-Lau and Peter B. Clark 198/1251

"Central Banking in Transition Countries" by Helmut Wagner [98/126]

"'Financial Crisis and Credit Crunch a� a Result of Inefficient Financial Intermediation-With Reference to the Asian Financial Crisis" by Jorge A. Chan-Lau and Zhaohui Chen 198/1271

"The East Asian Crisis: Macroeconomic Developments and Policy Lessons" by Kalpana Kochhar, Prakash Loungani. and Mark R. Stone 198/ 1281

·'Increasing Dependency Ratios. Pensions, and Tax Smoothing" by Efraim Sadka and Vi to Tanzi [98/1291

"Perspectives on the Recent Currency Crisis Literature" by Robert P. Flood and Nancy P. Marion 198/130]

"Can Shon-Tcrm Capital Controls Promote Capital Inflows?" by Tito Cordclla 198/131 J

·'Output Decline and Recovery in Uzbekistan: Past Perfom1ance and Future ProspectS" by Gunther Taube. and Jeromin Zettclmeyer [98/132]

"The Uzbek Growth Puzzle ..

by Jeromin Zettelmeyer 198/133]

'The Role of Allocation in a Globalized Corporate Income Tax" by Jack M. Mintz [98/1341

'Tax Revenue in Sub-Saharan Africa: Effects of Economic Policie� and Corruption" by Dhaneshwar Ghura 198/135]


IMF WORKING PAPERS 7 1 3

··1s the United States CPI Biased Across Income and Age Groups?"" by Nuri Erbas and Chera L. Sayers 198/1361

··East Asian Growth Before and After the Crisis'' by Nicholas Crafts 198/1371

'·Managing Corporate Distress in the Philippines: Some Policy Recommendations ..

by Cheng Hoon Lim and Charles Woodruff [98/1381 .. Can the Neoclassical Model Explain lhe Distribution of Foreign Direct Investment

Across Developing Coumries?'' by Harm Zebregs 198/1391

··From Autarky to Integration: Imitation. Foreign Borrowing, and Growth'' by Rachel van Elkan [98/140]

.. Recovery and Growth in Transition Economics 1 990-97: A Stylized Rcgrc�sion Analysis

.. by Oleh Havrylyshyn, lvailo lzvorski, and Ron van Rooden 198/141 J

.. Contagion: Monsoonal Effects. Spillovers. and Jumps Between Multiple Equilibria

.. by Paul Masson [9811421

.. Crises. Contagion. and the Closed-End Counu·y Fund Puzzle ..

by Eduardo Levy Yeyati and Angel Ubide [98/143]

.. Monetary Operations and Government Debt Management Under Islamic Banking" by V. Sundararajan. David Marston. and Ghiath Shabsigh [98/1441

.. Explaining the Recent Behavior of Inflation and Unemployment in the United States

.. by Vincent Hogan [98/1451 .. Optimal Fiscal Policy and the Environment'· by Jenny E. Ligthan [98/1461


IMF Staff Papers Vol. 45. No. 4 CD�cember 1998) €> 1 998 lntcrnatoonal Monetary Fund

Papers on Policy

Analysis and Assessment

Papers 011 Policy A11alysis a11d Assessme111 are i111ended to make sraff 1\'0rk in !he area of policy desig11 available to a ll'ide audie11ce. A list of all PPAAs issued i11 1998:3 .folloll's. These Papers IIW\' also be co11sidered.for illclusioll in the journal.

"Sequencing Capital Account Liberalizations and Financial Sector Reform·· by R.

Barry Johnston [98/81

·'Systemic Banking Distress: The Need for an Enhanced Monetary Survey" by Olivicr Frecaut and Eric Sidgwick [98/9]

''What Should Be Done with a Fiscal Surplus?" by igcl Chalk and Richard Hemming [98/10]

··capital Account Liberalization in the Southern Mediterranean Region'' by Saleh M. Nsouli and Mounir Rached 198/1 1 ]

"Inflation, Credibility, and the Role of the International Monetary Fund" by Carlo Cottarelli and Curzio Giannini [98/12]

7 1 4


INTERNATIONAL MONETARY FUND

S T A F F P A P E R S

VOLUME 45, 1 998

WASHINGTON, D.C.


I 1\1 F Staff Paper.< Vol. -15. No . .J tDcccml:>er 19981 ID 1998 lntcrn:nional Monetary Fund

Volume 45 Index

Volume 45 ( 199R) comprise� four issues. as foliO\¥�: March. pages 1-206

June. pages 207-400 September. pages 40 1-55R

December. pages 559-728

Authors

Agenor. Pierre-Richard. and Joshua Aizenman. Con1agion and Volatilily with Imperfect Credit Markels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

Aizenman. Joshua. and Pierre-Richard Agenor. Contagion and Volalility with Imperfect Credit Markets . . . . . . . . _ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

Aylward. Lynn. and Rupert Thorne. Countries· Repayment Performance Vis-it· Vis the IMF: An Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 595

Brixiova. Zuzana. and Pietro Garibaldi. Labor Market ln�titution� and Unemployment Dynamics in Transition Economic� . . . . . . . . . . . . . . . . . . 269

Cardo�o. Eliana. and llan Goldfajn. Capital Flows to Bralil: The Endogeneity of Capital Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 6 1

Cardoso. Eliana. Virtual Deficits and the Potinkin Effect . . . . . . . . . . . . . . . . . . . . 6 1 9

Cordella. Tito. and Eduardo Levy Ycyati. Public Disclosure and B:mk Failure� . . 1 1 0

Demirgli�·Kunt. Asli. and Enrica Detragiache. The Determinams or Banking Crises in Developing and Developed Countries . . . . . . . . . . . . . . . . . . . . . . . 8 1

Detragiache. Enrica. and Asli Dernirgii�·Kunt. The Determinams of Banking Crise� in Developing and Developed Countries . . . . . . . . . . . . . . . . . . . . . . . 8 1

Gable. Jeffery A . . and Eswar S . Prasad. lnternatiomll Evidence on the Determinants or Trade Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401

Frankel. JetTrey A .. and Shang-Jin Wei. Open Region;tlism in a World of Continental Trade Blocs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . -1-W

Garibaldi. Pietro. and Zuzana Brixiova. Labor Market Institution� and Unemployment Dynamics in Transition Economies . . . . . . . . . . . . . . . . . . 269

Ghosh. At ish. and Steven Phillips. Warning: Inflation May Be Harmful to Your Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672

7 J 6


VOLUI-1E -15 INDEX 7 1 7

Gordon. Roger H. Can High Personal Tax Rates Encourage Entrepreneurial Activity? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . -19

Goldfajn. llan. and Eliana Cardoso. Capital Flows to Brat.il: The Endogeneit) of Capital Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 6 1

Hardy. Daniel C. Anticipation and Surpri:.es in Central Ban" Interest Rate Policy: The Case of the Buntlesban" . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647

HofTmaister. Alexander W.. Jorge E. Rold6:-.. and Peter Wic"ham. Macroeconomic Fluctuations in Sub-Saharan Africa . . . . . . . . . . . . . . . . . 13:!

Jochum. ChriMian. and Laura Kodre!.. Doe:- the Introduction of Futures on Emerging Market Currencies De,tabilize the Underlying Currencie,·> . . . . . -186

Kaminsky. Graciela. Saul Lizondo. and Carmen M. Reinhart. Leading Indicators Of' Currency Crises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I

Keen. Michacl. Vertical Tax Externalities in the Theory of Fbcal Federali'rn . . . . -154

Kodres. Laura. and Christian Jochum. Does the lmroduction of Futures on Emerging Mar"et Currencies Destabilize the Underlying Currencie''! . . . . . -186

Krajny;ik. Kornelia. and Jeromin Zeuelmeyer. Competitivene'' in Transition Economic'>: What Scope for Real Appreciation? . . . . . . . . . . . . . . . . . . . . . 309

Levy Yeyati. Eduardo. and Tito Cordella. Public Disclosure and Ban" Failures . . 1 1 0

Li7.0ndo. Saul. Graciela Kaminsky. and Carmen M. ReinharL Leading Indicators of Currency Crises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I

Mackenzie. G.A. The Macroeconomic Impact of Privatization . . . . . . . . . . . . . . . . 363

Pauillo. Catherine. Ill\ estmcnt. Uncenai nty. and lrrcvcr,ihility in Clhana . . . . . . . 52�

Phillips. Steven. and At ish Ghosh. W<trning: Inflation M<t) Be Harmful to Your Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . on

Pra�ad. Eswar S .. <llld Jeffery A. Gable. International Evidence on the Determinant� of Trade Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . -Hll

Rama'>wamy. Ramana. and Tor�ten Sl0k. The Real Effect' ot Monetar) Policy in the European Union: What Arc the Difference,·! . . . . . . . . . . . . . . . . . . . 37-1

Reinhart. Carmen M .. Graciela Kamin�ky. and Saul Liwndo. Leading Indicator� of Currency Crise� . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . • I

Rold6�. Jorge E.. Peter Wickham. and Alexander W. Hollmai�tcr. Macroeconomic Fluctuation� in Sub-Saharan Africa . . . . . . . . . . . . . . . . . JJ�

Senhadji. Abdelhak. Time-Series Estimation of Structural Import Demand Equations: A Cro�s-Country Analysi� . . . . . . . . . . . . . . . . . . . . . . . . . . . . . �36

Sl0k. Torstcn. and Ramana Ramaswamy. The Real Effects of Monetary Policy in the European Union: What Arc the Difference�? . . . . . . . . . . . . . . . . . . . J?-1

Tanzi. Vi to. Corruption Around the World: Cau�e�. Consequence�. Scope. and Cure� . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559


7 1 8 VOLUME 45 INDEX

Thorne. Rupen. and Lynn Aylward. Countries· Repaymem Performa nee Vis-a-Vis the IMF: An Empirical Aflalysis . . . . . . . . . . . . . . . . . . . . . . . . . . 595

Wei. Shang-Jin. and Jeffrey A. Frankel. Open Regionalism in a World of Continental Trade Blocs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440

Wiekham. Peter, Alexander W. Hoffmaister. and Jorge E. Rold61>. Macroeconomic Fluctuations in Sub-Saharan Africa . . . . . . . . . . . . . . . . . 132

Zeuelmeyer. Jeromin. and Krajnyak. Kornelia. Competitiveness in Transition Economies: What Scope for Real Appreciation? . . . . . . . . . . . . . . . . . . . . . 309

Titles

Anticipation and Surprises in Central Bank Interest Rate Policy: The Case of the Bundesbank. By Daniel C. Hardy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647

Can High Personal Tax Rates Encourage Entrepreneurial Activity? By Roger H. Gordon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

Capital Flows to Brazil: The Endogeneity of Capital Controls. By Eliana Cardoso and llan Goldfajn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 6 1

Competitivenes� i n Transition Economies: What Scope for Real Appreciation? By Kornelia Krajnyak and Jeromin Zeuelmcyer . . . . . . . . . . . . . . . . . . . . . . 309

Contagion and Volatility with Imperfect Credit Markets. By Pierre-Richard Agenor and Joshua Aizenman . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

Corruption Around the World: Causes. Consequences. Scope. and Cures. By Vi to Tanzi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559

Countrie� · Repayment Performance Vis-a-Vis the IMF: An Empirical Analysis. By Lynn Aylward and Rupert Thorne . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 595

The Determinants of Banking Crises in Developing and Developed Countrie�. By Asli Demirgli�-Kunt and Enrica Detragiache . . . . . . . . . . . . . . . . . . . . . . 8 1

Doe� the Introduction of Futures on Emerg.ing Market Currencies Destabilize the Underlying Currencies'? By Christian Jochum and Laura Kodres . . . . . . 486

International Evidence on the Determinants of Trade Dynamics. By E�war S. Prasad and Jeffery A. Gable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 I

Investment. Uncertainty. and Irreversibility in Ghana. By Catherine Pauillo . . . . . 522

Labor Market Institutions and Unemployment Dynamics in Transition Economic . . By Pietro Garibaldi and Zuzana Brixiova . . . . . . . . . . . . . . . . . 269

Leading Indicato r!> of Currency Crise,. By Graciela Kaminsky, Saul Lizondo. and Carmen M. Reinhart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I

The Macroeconomic Impact of Privatization. By G.A. Mackenzie . . . . . . . . . . . . . 363


VOLUME 45 INDEX 7 1 9

Macroeconomic Fluctuations in Sub-Saharan Africa. B y Alexander W. Hoffmaister. Jorge E. Rold6s. and Peter Wickham . . . . . . . . . . . . . . . . . . . . 132

Open Regionalism in a World of Cominental Trade Bloc�. By Jeffrey A. Frankel and Shang-Jin Wei . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440

Public Disclosure and Bank Failures. By Tito Cordella and Eduardo Levy Yeyati . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I 10

The Real EfTects of Monetary Policy in the European Union: What Are the Differences? By Ramana Ramaswamy and Torsten Sl0k . . . . . . . . . . . . . . . 374

Time-Series Estimation of Structural lrnpon Demand Equations: A Cross-Country Analysis. By Abdelhak Senhadji . . . . . . . . . . . . . . . . . . . . . . . . . . . 236

Vertical Tax Externalities in the Theory of Fiscal Federalism. By Michael Keen . . 454

Vinual Deficits and the Patinkin Effect. By Eliana Cardoso . . . . . . . . . . . . . . . . . . 6 19

Warning: Inflation May Be Harmful to Your Growth. By Atish Ghosh and Steven Phillips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672

Subjects

To faci litate electronic storage and retrieval of bibliographic data, Staff Papers has adopted the subject classification scheme of the Journal of Economic Li1era1ure (Nashville. Tennessee).

C Mathematical and Quantitative Methods

C22 Econometric Methods: Single Equation Models. Time-Series Models

Does the lmroduction of Futures on Emerging Market Currencies Destabilize the Underlying Currencies? By Christian Jochum and Laura Kodres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486

Time-Series Estimation of Structural Import Demand Equations: A Cross-Country Analysis. By Abdelhak Senhadji . . . . . . . . . . . . . . . . . . 236

C24 Econometric Methods: Single Equation Models. Truncated and Censored Models

Investment, Uncertaimy. and Irreversibility in Ghana. By Catherine Pattillo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552

C32 Econometric Methods: Multiple/Simultaneous Equation Models. Time-Series Models

Macroeconomic Fluctuations in Sub-Saharan Africa. By Alexander W. Hoffmaister. Jorge E. Rold6s. and Peter Wickham . . . . . . . . . . . . . . . . 132


720 VOLUME 45 INDEX

0 Microeconomics

D81 Criteria for Decision-Making under Risk and Uncertainty

Investment. Uncertainty. and Irreversibility in Ghana. By Catherine Pauillo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522

D82 Asymmetric and Private Information

Public Disclosure and Bank Failures. By Tito Cordella and Eduardo Levy Yeyati . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 0

D92 lntertemporal Firm Choice and Growth, Investment, or Financing

Investment. Uncertainty. and Irreversibility in Ghana. By Catherine Pattillo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522

E Macroeconomics and Monetary Economics

EIO General Aggregative Models. General

Virtual Deficits and the Patinkin Effect. By Eliana Cardoso . . • . . • . . . . . 6 1 9

E21 Consumption: Saving

Time-Series Estimation of Structural Import Demand Equations: A Cross-Country Analysis. By Abdclhak Scnhadji . . . . . . . . . . . . . . . . . . 236

E24 Employment: Unemployment: Wages

Labor Market Institutions and Unemployment Dynamics in Transition Economies. By Pietro Garibaldi and Zuzana Brixiov<� . . . . . . . . . . . . . 269

E31 Price Level: Inflation: Deflation

Virtual Deficits and the Patinkin Effect. By Eliana Cardoso . . . . . . . . . . . 6 1 9

Warning: Inflation May be Harmful to Your Growth. By Ati�h Ghosh and Steven Phillips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672

E32 Business Fluctuations; Cycles

International Evidence on the Determinants of Trade Dynamics. By Eswar S. Prasad and Jeffcry A. Gable . . . . . . . . . . . . . . . . . . . . . . . . . . 40 I


E43 Determination of Interest Rates: Term Structure of Interest Rates

Anticipation and Surprises in Central Bank Interest Rate Policy: The Case of the Bundesbank. By Daniel C. Hardy . . . . . . . . . . . . . . . . . . . . 647


VOLUME 45 INDEX

E44 Financial Markets and the Macroeconomy

The Determinants of Banking Crises in Developing and Developed

721

Countrie,. By A�li Demirgi.i<;-Kunt and Enrica Detragiache . . . . . . . . . . 8 I

Contagion and Volatility with Imperfect Credit Markets. By Pierre-Richard Agenor and Joshua Aizenman . . . . . . . . . . . . . . . . . . . . . . . . . 207

E47 Forecasting and Simulation

Anticipation and Surpri�es in Central Bank lntere!.t Rate Policy: The Case of the Bundesbank. By Daniel C. Hardy . . . . . . . . . . . . . . . . . . . . 647

E5 Monetary Policy. Central Banking. and the Supply of Money and Credit

The Real Effects of Monetary Policy in the European Union: What Arc the Differences? By Ramana Ramaswamy . . . . . . . . . . . . . . . . . . . . . . 374

E52 Monetary Policy (Target�. Instruments. and Effects)

The Real Effects of Monetary Policy in the European Union: What Are the Differences? By Ramana Ramaswamy . . . . . . . . . . . . . . . . . . . . . . 374

E58 Central Banks and Their Policies

Virtual Deficits and the Patinkin Effect. By Eliana Cardoso . . . . . . . . . . . 6 1 9

The Real Effects of Monetary Policy i n the European Union: What Are the Differences? By Ramana Ramaswamy . . . . . . . . . . . . . . . . . . . . . . 374

E62 Fiscal Policy: Public Expenditures. Investment. and Finance: Taxation

Corruption Around the World: Causes. Consequences. Scope. and Cures. By Vito Tanzi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559

The Macroeconomic Impact or Privatization . By G.A. Mackenzie . . . . . . 363

Virtual Deficits and the Patinkin Effect. By Eliana Cardoso . . . . . . • . . . . 6 I 9

F International Economics

FIO Trade. General

International Evidence on the Determinant� or Trade Dynamics. By Eswar S. Prasad and Jeffrey A. Gable . . . . . . . . . . . . . . . . . . . . . . . . . . 40 I

Fl4 Country and Industry Studies of Trade

Competitiveness in Transition Economics: What Scope for Real Appreciation? By Kornelia Krajnyak and Jeromin Zeuclmcyer . . . . . . 309

Time-Series Estimation of Strucwral Import Demand Equations: A Cross-Country Analysis. By Abdelhak Senhadji . . . . . . . . . . . . . . . . . . 236


722 VOLUME 45 INDEX

F 15 Economic Integration

Open Regionalism in a World of Continental Trade Blocs. By Jeffrey A. Frankel and Shang-Jin Wei . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440

F21 International Investment; Long-term Capital Movements

Competitiveness in Transition Economies: What Scope for Real Appreciation? By Kornelia Krajnyak and Jeromin Zeuelmeyer . . . . . . 309

F3 I Foreign Exchange

Leading Indicators of Currency Crises. By Graciela Kaminsky. Saul Lizondo, and Carmen M. Reinhan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I

F32 Current Account Adjustment: Short-term Capital Movements

Capital Flows to Brazil: The Endogeneity of Capital Controls. By Eliana Cardoso and llan Goldfajn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 6 1

F33 International Monetary Arrangements and Institutions

Countries· Repayment Performance Vis-a-Vis the IMF: An Empirical Analysis. By Lynn Aylward and Rupen Thorne . . . . . . . . . . . . . . . . . . 595

F34 International Lending and Debt Problems

Countries' Repayment Perfo1mance Vis-a-Vis the IMF: An Empirical Analysis. By Lynn Ay I ward and Rupert Thorne . . . . . . . . . . . . . . . . . . 595

F36 Financial Aspects of Economic Imegration

Contagion and Volatility with Imperfect Credit Markets. By Pierre-Richard Agenor and Joshua Aizenman . . . . . . . . . . 207

F4 Macroeconomic Aspects oflnternational Trade and Finance

F41 Open Economy Macroeconomics

Capital Flows to Brazil: The Endogeneity of Capital Controls. By Eliana Cardoso and Ilan Goldfajn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 6 1

Competitiveness in Transition Economies: What Scope for Real Appreciation? By Kornelia Krajnyak and Jeromin Zettelmeyer . . . . . . 309

International Evidence on the Determinants of Trade Dynamics. By Eswar S. Prasad and Jeffery A. Gable . . . . . . . . . . . . . . . . . . . . . . . . . . 401


Time-Series Estimation of Structural Import Demand Equations: A Cross-Country Analysis. By Abdelhak Senhadji . . . . . . . . . . . . . . . . . . 236


VOLUME 45 INDEX

F47 Forecasting and Simulation

Leading Indicators of Currency Crises. By Graciela Kaminsky. Saul

723

Lizondo. and Carmen M. Reinhart . . .. . . . . . .. . . . . . . . . . . . . . . . . . . . I

G Financial Economics

G 14 Information and Market Efficiency: Event Studies

Public Disclosure and Bank Failures. By Tito Cordella and Eduardo Levy Yeyati .. .. .. .. . . .. .. . . . . . . .. . . . .. . . . .. .. .. .. .. . 110

G 15 International Financial Markets

Does the Introduction of Futures on Emerging Market Currencies

Destabilize the Underlying Currencies? By Christian Jochum and

Laura Kodres ............................................. 486

G21 Banks; Other Depository Institutions; Mortgages

Public Disclosure and Bank Failures. By Tito Cordella and Eduardo

Levy Yeyati . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

The Determinants of Banking Crises in Developing and Developed

Countries. By Asli Demirgli�-Kunt and Enrica Detragiache ......... 81

G28 Government Policy and Regulation

Public Disclosure and Bank Failures. By Tito Cordella and Eduardo

Levy Yeyati .............................................. 110

H Public Economics

HI Structure and Scope of Government

Corruption Around the World: Causes. Consequences. Scope. and Cures. By Vito Tanzi ....................................... 559

H25 Business Taxes

Can High Personal Tax Rates Encourage Entrepreneurial Activity'? By

Roger H. Gordon . . . . . . . .. .. . . .. . . . . .. . . . . . ... . . . . . ... . . . . . . 49

H21 Efliciency; Optimal Taxation

Vertical Tax Externalities in the Theory of Fiscal Federalism. By Michael Keen ............................................. 454

H3 Fiscal Policies and Behavior of Economic Agents

Vertical Tax Externalities in the Theory of Fiscal Federalism. By

Michael Keen ............................................. 454

Corruption Around the World: Causes. Consequences. Scope. and

Cures. By Vito Tanzi . ... . .. . . . . . . . . . . . .. . . . . .. . . . .. . . . . . . . . 559


724 VOLUME 45 INDEX

H7 State and Local Government: Intergovernmental Relations

Venical Tax Externalities in the Theory of Fiscal Federalism. By Michael Keen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454

Health, Education, and Welfare

I 31 General Welfare: Basic Needs: Quality of Life

Contagion and Volatility with Imperfect Credit Markets. By Pierre-Richard Agenor and Joshua Aizenman

.I Labor and Demographic Economics

J63 Turnover: Vacancies

......................... 207

Labor Market ln5titutions and Unemployment Dynamics in Transition Economies. By Pietro Garibaldi and Zuzana Brixiova ............. 269

K Law and Economics

K4 Legal Procedure. the Legal System. and Illegal Bchavior

Corruption Around the World: Causes. Consequences, Scope. and Cures. By Vito Tanzi . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . .. . . . . . 559

L Industrial Organization

L 11 Production and Market Structure: Size Distribution of Firms

Can High Personal Tax Rates Encourage Entrepreneurial Activity? By Roger H. Gordon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

L33 Boundaries of Public and Private Enterprises: Privatization: Contracting Out

The Macroeconomic Impact of Privatization. By G.A. Mackenzie ...... 363

N Economic History

N4 Government. Law. and Regulation

Corruption Around the World: Causes, Consequence�. Scope. and Cures. By Vi to Tanzi . .. . . . . . . ... . .. . . . . . . . . . . . . . . . . . . . . . . . . 559

0 Economic Developmen t. Technological Change, and Growth

031 Innovation and Invention: Processes and Incentives

Can High Personal Tax Rates Encourage Entrepreneurial Activity'? By Roger H. Gordon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49


VOLUME 45 INDEX

040 Economic Growth and Aggregate Productivity. General

Warning: Inflation May Be Harmful to Your Growth. By Ati�h Gho�h

725

and Steven Phillips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672

P Economic Systems

P20 Socialist Systems. General

Competitivene�s in Transition Economies: What Scope for Real Appreciation? By Kornelia Krajnyak and Jerornin Zeuelmeyer ...... 309

?50 Comparative Economic System�. General

Competitiveness in Transition Economic�: What Scope for Real Appreciation? By Kornelia Krajny;lk and Jeromin Zeuelmeycr . . . . . . 309


The Twelfth World Congress

of the

International Economics Association

August 23-27, 1999

Buenos Aires, Argentina

The Twelfth World Congress of the lEA will be organized by the Asociacion Argentina de Economla Polltica. The five-day program includes plenary sessions, invited sessions, and presentation of contributed papers. Invited lectures will cover two topics: "Inequality in the World Today," with a program arranged by Professor Richard Freeman (Harvard, USA) and "The Current State of Macroeconomics," arranged by Professor jacques Dreze (CORE, Belgium).

Call for Papers

Submissions of papers related to relevant economic areas are welcomed. The Program Committee will screen the contributed papers and authors will be notified by April 15, 1999, whether they are accepted. Three volumes of proceedingson macroeconomics, inequality, and Latin American issues, respectively-will be published. Those wishing to submit a paper should send three copies by February 14, 1999, to:

Professor David de la Croix IRES-Universite Catholique de Louvain,

Place Montesquieu 3, B-1348 Louvain-la-Neuve, Belgium

Tel.: 3210473453 • Fax: 3210473945 e-mail: [email protected]


In statistical mauer throughout this issue.

dot�( ... ) indicate that data are not available;

a dash (-) indicates that the figure is zero or les� than half the final digit shown. or that the item does not exist:

a single dot(.) indicates decimals:

a comma(.) separates thousands and millions:

"billion" mean:- a thousand million. and "trillion" mean� a thousand billion:

a short dash (-) is used between years or months (for example. 1997-98 or January-October) to indicate a total or the years or months inclusive or the beginning and ending years or months:

a stroke (/) is used between years (for example, 1997/98) to indicate a fiscal year or a crop year:

a colon (:) is used between a year and the number indicating a quarter within that year (for example, I 998: I):

components of tables may not add to totals shown because of rounding.

15811-9781455221950.pdf - imf elibrary

Documents