regional asymmetries in monetary transmission: the case of south africa

Journal of Policy Modeling 28 (2006) 965–979

Regional asymmetries in monetary transmission:The case of South Africa

David Fielding a,b,∗, Kalvinder Shields c

a Department of Economics, University of Otago, Dunedin 9001, New Zealandb Centre for the Study of African Economies, Oxford University, United Kingdom

c Department of Economics, Melbourne University, Australia

Received 1 February 2006; received in revised form 1 August 2006; accepted 1 September 2006

Abstract

PPP is unlikely to hold instantaneously for all commodities across the different regions of a monetaryarea. It is therefore possible that monetary expansions or contractions will have different effects in differentregions, if there are regional asymmetries in the monetary transmission mechanism. We estimate the size ofsuch asymmetries across the nine provinces of South Africa over the period 1997–2005. There are large andstatistically significant differences in the response of prices to monetary expansions and contractions. Theproblems arising from transmission mechanism asymmetries are not restricted to international monetaryunions.© 2006 Society for Policy Modeling. Published by Elsevier Inc. All rights reserved.

JEL classification: E31; E50; O11

Keywords: Inflation; Monetary union; South Africa; Transmission mechanism

1. Introduction

The majority of papers on monetary policy, both theoretical and applied, are based on repre-sentative agent models in which all agents face the same prices. However, there is a small bodyof literature that explores the possibility that there is some heterogeneity in the prices faced bypeople in different parts of a country. For example, Cecchetti, Mark, and Sonora (2002) use annualcity-level data for 1918–1995 to test for purchasing power parity within the United States. They

∗ Corresponding author at: Department of Economics, University of Otago, Dunedin 9001, New Zealand.Tel.: +64 3 479 8653.

E-mail address: [email protected] (D. Fielding).

0161-8938/$ – see front matter © 2006 Society for Policy Modeling. Published by Elsevier Inc. All rights reserved.doi:10.1016/j.jpolmod.2006.09.001

mailto:[email protected]

dx.doi.org/10.1016/j.jpolmod.2006.09.001

966 D. Fielding, K. Shields / Journal of Policy Modeling 28 (2006) 965–979

do find price convergence among the 19 cities in their data set, but convergence is very slow,with a half-life of around 9 years. The persistence of relative price differences across cities isattributed to transportation costs and differences in the speed of adjustment to supply shocks.Within a shorter time period (5 years, say) relative prices across cities approximate to a randomwalk. In related work, Engel and Rogers (2001) show that the size of deviations from the law ofone price across US cities depends on the distance between them. Ceglowski (2003) finds similarresults for Canadian cities, noting also that price deviations are especially great when the citiesare in different provinces.

Such heterogeneity has potentially important consequences for monetary policy. Just as thetheory of optimal monetary policy is founded on the idea of maximising the welfare of a represen-tative agent, real-world monetary policy targets make reference to the rate of inflation of a nationalconsumer price index, that is, the rate of growth of the cost of living for an “average” consumer.But if consumers in different regions face different prices then concerns about the distribution ofinflation rates resulting from a given monetary policy become relevant. De Grauwe (2000) andGros and Hefeker (2002) show how in theory a monetary policy rule that ignores informationat the regional level may lead to welfare losses when there are asymmetries in the transmissionmechanism.

The papers on regional price heterogeneity in North America do not explore the sources of theshocks that drive such persistent wedges between prices in different locations, so there is no directevidence relating to the theoretical results of De Grauwe and Gros and Hefeker. In particular, wedo not know the extent to which the price deviations are a consequence of asymmetric responsesto monetary policy. Nevertheless, there is some research on the asymmetric regional effects ofmonetary policy on output and employment. Carlino and DeFina (1998) show that in the US, thereare two specific regions (the Southwest and the Rockies) in which output and employment aresignificantly less sensitive to monetary policy than average, and one (the Great Lakes region) inwhich sensitivity is significantly higher. Similar conclusions appear in earlier papers employingless sophisticated econometric techniques (Toal, 1977; Garrison & Chang, 1979; Garrison & Kort,1983). Given the size and statistical significance of regional asymmetries in the effects of monetarypolicy on output and employment, there is reason to suspect that there are also asymmetries inthe response of regional prices.

In this paper, we directly address the question of asymmetries in the effect of monetary policyon regional prices. Our data are taken not from North America, but from the Republic of SouthAfrica. South Africa is at an earlier stage of development than the US or Canada, but transmissionmechanism asymmetries in prices are likely to be larger in economies where markets are lessdeveloped, with less arbitrage across regions. The problem is likely to be at least as relevantto development macroeconomics as it is to the macroeconomics of industrialized countries. Theasymmetries in South Africa turn out to be large, which brings in to question the existing monetarypolicy based on an aggregate national inflation target.

The paper is organised as follows: Section 2 provides a brief introduction to the current mon-etary policy regime in South Africa; Section 3 presents our econometric model and Section 4discusses the policy implications of our results.

2. Monetary policy in South Africa

Since the democratic reforms of 1993, South Africa has been divided into nine provinces: East-ern Cape, Free State, Gauteng, KwaZulu-Natal, Limpopo, Mpumalanga, North West, NorthernCape and Western Cape. These regions are depicted in Fig. 1. South Africa forms a contigu-

D. Fielding, K. Shields / Journal of Policy Modeling 28 (2006) 965–979 967

Fig. 1. South African Provinces. (E) Eastern Cape, (F) Free State, (G) Gauteng, (K) KwaZulu-Natal, (L) Limpopo, (M)Mpumalanga, (NC) Northern Cape, (NW) North West, and (W) Western Cape.

ous geographical area, bounded to the south by coastline, to the north-west by Namibia andBotswana and to the north-east by Zimbabwe, Swaziland and Mozambique. The independentstate of Lesotho lies entirely in the interior of South Africa, bounded by the Eastern Cape, theFree State and KwaZulu.

South Africa constitutes a single monetary area. The Rand circulates in all parts of the country,as it does in many of the countries bordering South Africa. Recognising that South Africa isa relatively small open economy, the Reserve Bank enacts policy ultimately designed to stabi-lize the exchange rate. Since 2000, this objective has been pursued through an explicit inflationtarget:

“The primary objective of monetary policy is to protect the value of the currency in order toobtain balanced and sustainable economic growth in the country. . . It requires the achievementof financial stability, i.e., price stability as well as stable conditions in the financial sector as awhole.”1

Current Reserve Bank monetary policy is based on adjustment of the interest rate on its repur-chase transactions, although in the recent past it has made reference to a wider range of policyinstruments. Because the focus on the interest rate is a recent phenomenon, this paper will beconcerned mainly with the inflationary effects of an intermediate monetary policy variable, thatis, the stock of M1. Although the Reserve Bank does not control M1 directly, it can influencethe money supply through a variety of conventional policy instruments. Our model is designedto identify the impact of changes in M1 on prices without modelling monetary policy explicitly.Given the recent changes in the targets and instruments of the Reserve Bank, modelling monetarypolicy is a complex exercise beyond the scope of this paper.

There is a national money market in South Africa: there are no restrictions on the movement ofcapital between provinces. Indeed, much of the money stock consists of liabilities of banks that arenot specifically located in any one province. No regional disaggregation of M1 is economicallymeaningful, and no direct or intermediate monetary policy instrument operates at a provincial

1 Appendix to the Statement of the Monetary Policy Committee (6 April 2000).


level. Nevertheless, as we shall see, consumer price indices do vary from one region to another. Thisis due partly to regional variation in the weights on different commodities (because, for example,of regional differences in average income levels), but the variation in weights is typically quitesmall. The main reason for the variations in consumer prices indices is the absence of completearbitrage for certain kinds of goods and services. South Africa is a large country: inter-provincialtransport costs for low-value items (such as staple foods) can be substantial, as can regionalvariations in the value of real estate.

Given the absence of arbitrage and the economic heterogeneity of South Africa’s provinces,there is an a priori case for considering the extent to which monetary expansions or contractionshave a heterogeneous impact in different parts of the country. For example, regional variations inthe degree of price inertia might lead to regional variations in the size of the initial response ofinflation to a monetary expansion. It then remains to be seen for how long the variations persist.Large and persistent variations may mean that monetary policy needs to be based on a socialwelfare function that permits a richer characterisation of the economy than one based on a singlerepresentative consumer.

3. The empirical model

In this section we present an empirical model of the factors driving provincial prices in SouthAfrica. Our main interest is in the impact on prices of monetary expansions and contractions.Before we begin to describe the model, a point of clarification is in order. Our sample periodspans a substantial change in the conduct of monetary policy in 2000: it is highly unlikely thatthere is a stable Reserve Bank policy reaction function over the whole of the period.2 Our modelis fitted to the data in a way that allows us to remain agnostic about the factors driving monetarypolicy changes. This model is then used as the basis for simulations, but these simulations arenot conditioned on a certain monetary policy rule specifying feedback from income and pricechanges. Rather, we perform simulations based on a given hypothetical change in policy thatleads to a permanent expansion of the money supply, without any feedback. Our interest is in theway in which purely hypothetical changes in policy impact on prices, rather than in tracing theimpact of policy shocks.

Our underlying methodology is very straightforward. We fit a national money demand equationfor South Africa in which there is long-run price homogeneity: when there is a monetary expansion,we eventually see aggregate prices rising in proportion to the growth in the money stock, ceterisparibus. We then test whether the proportionate growth in aggregate prices embodies regionalasymmetries in inflation, with different inflation rates in different provinces.

Our empirical model of provincial prices is constrained by the availability of relevant datapublished by the Reserve Bank of South Africa and Statistics SA. Monthly and quarterly provincialprice data are available only from 1997. Aggregate price, money supply and real income data areavailable for a longer period, but income is measured only at the quarterly frequency. In order tobuild a long-run money demand function into our empirical model in a coherent way we make useof both the quarterly data, to estimate the long-run elasticity of money demand, and the monthlydata, to estimate the short-run price dynamics, subject to the long-run constraints implicit in themoney demand equation. We first clarify the nature of the data available and then outline thestructure of the model.

2 Aron and Muellbauer (2005) discuss the complexities of South African monetary policy.


The following monthly time-series data are available for South Africa. The first of these, pit ,is reported for the period 1997m1–2005m2. The others are reported for a longer period.

1. The urban consumer price index (pit , i = 1, . . ., 9) for all nine provinces. pit is constructed as aweighted average of the prices of consumer goods sold in urban areas in each province. (Notehowever that the consumers need not be resident in urban areas; many will commute to townto make purchases.) The weights on individual commodities change periodically.

2. The national urban consumer price index (pAGt ). This is a weighted average of the pit ; againthe weights change periodically.

3. The volume of M1 at the end of the last working day in month t (mt). This timing will beimportant in the identification of our model.

4. An index of national industrial production (qt).

Quarterly averages of pAGt , qt and mt, denoted pAGt , Qt and Mt, are reported for a much longerperiod (1982q1–2004q4). So also is real gross domestic expenditure (Yt).

Using standard stationarity tests, the null that ln(Yt) is I(1) cannot be rejected; but neither canthe null that it is I(0). The same is true of ln(Mt) − ln(pAGt ). So we estimate the long-run moneydemand equation on quarterly data using an ARDL specification of the form

�[ln(Mt) − ln(PAGt )] = κ� ln(Yt) − η[ln(Mt−1) − ln(PAG

t−1) − μM]

+ θ(Yt−1 − μY ) + ut (1)

whereμM andμY are the mean values of ln(M) − ln(PAG) and ln(Y), respectively, and ut is a regres-sion residual.3 The regression results corresponding to Eq. (1) are discussed in detail in the Tech-nical Appendix available online at www.business.otago.ac.nz/econ/Personal/df files/index.html;our fitted regression coefficients are κ = 0.30, η= 0.13 and θ = 0.33. The main point to note here isthat our results provide strong evidence for cointegration between the real money stock and realincome, using the test statistic outline in Pesaran, Smith, and Shin (2001). This paper provides arange of critical values for an F-test of the joint null that η= θ = 0, i.e., there is no long-run rela-tionship between the variables. The lower bound of the range of critical values corresponds to theassumption that all variables are I(0); the upper bound corresponds to the assumption that they areall I(1). Our F-statistic (4.28) exceeds the upper bound of the relevant 5% critical value, so there isevidence for some long-run relationship. The implicit long-run elasticity income of money demandis 2.61. Note that there is no interest rate term in this regression equation. The addition of a t-billor deposit interest rate to the regression did not produce any statistically significant coefficients.

Our short-run price dynamics are estimated by fitting a model to monthly data, the unrestrictedversion of which is a conditional VAR in provincial price inflation rates:

B(L)� ln(pt) = st + β1(L)� ln(mt−1)β2(L)� ln(yt−1) + α ECMt−1 + vt (2)

where st is a seasonally varying intercept and vt is a vector of regression residuals. p = [p1, . . .,p9]’ is a vector of the nine monthly provincial price series and y is a monthly interpolation ofthe quarterly real income series using the monthly variation in the industrial production series:yt = 0.25 × [qt/Qt] × Yt. B(L) is a 9 × 9 matrix of lag polynomials; β1(L) and β2(L) are 1 × 9vectors of lag polynomials; α is a 1 × 9 vector of constants. Standard ADF tests indicate that ln(m),

3 Lags of �ln(Y) and �[ln(M) − ln(P)] are statistically insignificant.

http://www.business.otago.ac.nz/econ/Personal/df_files/index.html


Fig. 2. Monthly growth rates of aggregate price, money and real income.

ln(y) and all elements of p are at least difference-stationary. ECM is the equilibrium correctionterm [ln(m) − ln(pAG)* − 2.61 × ln(y)], where ln(pAG)* is an approximation to ln(pAG) with time-invariant weights on the logs of the different provincial price indices, so that we have a modelthat is linear in lags of the provincial price levels. The full set of weights is given in the TechnicalAppendix; the correlation coefficient for � ln(pAG)* and � ln(pAG) is over 96%.

We should not expect large values for the α coefficients. Fig. 2 depicts the three series� ln(pAG

t )∗, � ln(mt) and � ln(yt). It can be seen from the figure that the variation in aggre-

gate inflation is much smaller than the variation in either money growth or output growth. (On theother hand, the variation in inflation rates does not differ greatly from one province to another; allof the provincial mean monthly inflation rates lie between 0.39% and 0.49%, with correspond-ing standard deviations between 0.43% and 0.58%.) This suggests that any response of prices tochanges in the money supply will be marked by a considerable degree of inertia. It remains to beseen precisely how large this inertia is, and how much asymmetry is exhibits across provinces.

Eq. (2) implies that provincial prices will adjust in response to disequilibrium in the moneymarket, captured by ECM, but it does not imply that all prices will necessarily respond at the samerate. Nor does it embody any long-run PPP condition across the different elements of p. Whenstandard ADF tests are applied to the 36 bilateral provincial real exchange rate series ln(pi/pj)across the nine provinces, there is just one, probably spurious case (Eastern Cape—Mpumalanga)in which the ADF statistic exceeds the 5% critical value. Fig. 3 illustrates eight of these bilateralreal exchange rates, in each case depicting prices relative to those in the Free State. In each casethe relative price series appears to be drifting upwards with no apparent mean reversion, at leastwithin our sample period. We do not impose the restriction that changes in m have a uniformlong-run effect on all of the provincial price series. A 1% increase in m will lead eventually to a1% increase in pAG, but some of the provincial elements of pAG may rise more than others.

Despite the absence of PPP, there is a high correlation of inflation rates across the nine provinces.The lowest correlation coefficient is 0.79 (Mpumalanga and the North West); the highest is 0.90


Fig. 3. Monthly provincial consumer price indices relative to the free state, 1997–2005.

(Gauteng and the West Cape). Prices in each region appear to be driven largely by common shocks,despite the fact that they do not converge in the long-run.

Eq. (2) is to be interpreted as part of a reduced-form VAR in y and the elements of p; cor-responding to this reduced-form representation is an unidentifiable structural model embodyinginstantaneous interactions between yt and the different pit . However, it is reasonable to supposethat mt has no impact on pt or yt: remember that pt and yt measure prices and income duringmonth t, whereas mt is a stock variable measuring the money supply at the end of the last working


day of month t. Moreover, mt−1 is weakly exogenous to all period t variables, so the impact ofchanges in m on the system will be identified, and the fitted coefficients can be used to trace outthe effect on prices of a hypothetical exogenous change in m. (If we fit an additional equation for� ln(mt) then we can trace out the effect on the system of structural shocks to m. However, inour case there is no evidence that the impact on prices of shocks to m is different to the impactof anticipated changes; indeed, we will see that m impacts on prices only with a considerablelag. Our policy analysis will not be concerned with the effects of monetary shocks with feedbackfrom p to m; rather, it will be concerned with the effects of a hypothetical change in policy that isdesigned to cause a permanent rise or fall in m by a certain value.)

It turns out that there are many nuisance parameters in the unrestricted monthly model, so wealso fit a parsimonious version of Eq. (2) that sets coefficients on some individual lags of moneygrowth, income growth and provincial inflation equal to zero, so as to minimize the Schwartz-Bayesian Information Criterion4; this restricted model is discussed in more detail in the TechnicalAppendix. There is a marked symmetry in the restricted model. No lag of income growth issignificant in any of the provincial inflation equations. The only lag of money growth that issignificant in any equation is � ln(mt-2), and this term is significant in all nine equations: it takesover a month for monetary growth to feed through to prices. The only lagged inflation rate thatis significant in any of the equations is � ln(pt−1

KwaZulu), and again this term is significant inall nine equations. ECMt−1 is significant only in the equation for inflation in KwaZulu, whereit is has a predictably small coefficient (0.0034): prices take a long time to adjust in response toa change in m. Since the interactions of the different prices are captured only in reduced form,one should be cautious in placing any economic interpretation on this structure. Nevertheless, theresults are consistent with the interpretation that KwaZulu is in some sense the “lead” region, inthat the effects of global price shocks are typically felt there before they have an impact in otherparts of South Africa.

The main asymmetry that appears in the fitted model is in the size of the coefficients on laggedmoney growth in each of the provincial inflation equations. The differences in the� ln(mt−2) coef-ficients imply provincial asymmetries in the immediate effect of changes in the money supply oninflation and (in the absence of inter-provincial PPP) on prices in the long-run. Such asymmetriesare best illustrated graphically, so Figs. 4–7 illustrate the impact on provincial inflation and pricesof a permanent, exogenous unit increase in ln(m).

Fig. 4 illustrates the effects on inflation in the two regions that are most and least affected bythe monetary expansion, that is, Mpumalanga and the Free State, respectively. The figure showshow inflation can be expected to evolve over the 12 months following the monetary expansion(months t + 2 to t + 13), with two-standard-error bars around the point estimates. It can be seenthat there is a significant difference in the inflation responses in the two regions in the first coupleof months, although beyond this point the typical “butterflying” of the error bars renders thedifferences insignificant. Table 1 shows the effect of the monetary expansion in month t + 2 inall nine provinces. It can be seen that there is a group of “high inflation” provinces (Gauteng,Eastern Cape, Mpumalanga) whose responses are all significantly higher than another group of“low inflation” provinces (Free State, Northern Cape). The inflation response in Mpumalangais almost twice as large as that in the Free State. The relative magnitudes of all the inflationresponses are shown in Figs. 5 and 6, which depict the inflation in each province in excess of that

4 The nine equations in the model are estimated simultaneously by maximum likelihood. These estimates differ slightlyfrom OLS estimates, because the ECM term appears in only one equation.

D.F

ielding,K.Shields

/JournalofPolicyM

odeling28

(2006)965–979

973

Table 1Provincial statistics

Free state N. Cape N. West Limpopo W. Cape KwaZulu Gauteng E. Cape Mpum-alanga

(i) Initial responses of inflation to a unit increase in M1, in percentMean 5.12 5.16 6.40 6.42 7.37 7.45 7.97 8.18 8.64Two standard errors above the mean 7.23 7.24 9.08 8.90 9.72 9.94 10.52 10.67 11.48Two standard errors below the mean 3.01 3.08 3.71 3.94 5.02 4.96 5.42 5.70 5.79

(ii) Basic economic statisticsa

Real per capita income growth rate 2000–2004 −0.3% 1.4% 1.6% 0.8% 0.5% 1.6% 2.0% 1.3% 0.7%Per capita income in 2004 (1000 Rand) 19.6 19.1 21.4 9.3 44.0 19.0 51.4 7.0 26.7% of Population below the SA poverty line 61.6% 61.0% 57.0% 55.1% 22.6% 56.9% 29.0% 75.8% 79.5%Of all SA poor, the % living in this province 7.3% 2.4% 7.6% 11.7% 4.1% 22.5% 11.7% 20.5% 12.2%

a Source: DTI.


Fig. 4. Twelve-month inflation response to a unit increase in M1 in Mpumalanga and the free state (in percent withtwo-standard-error bars).

in the Free State over the first 12 months following the monetary expansion. In several provinces(Gauteng, Eastern Cape, Mpumalanga, KwaZulu, Western Cape), the inflation rate exceeds thatof the Free State by over 2% at t + 2, and remains over 1% higher for several subsequent months.These differences are large relative to some international differences in the monetary transmissionmechanism, for example, those illustrated in Haug, Karagedikli, and Ranchhod (2003, Fig. 3.1).

The differences in inflation correspond to long-run differences in price levels: the monetaryexpansion changes all of the provincial real exchange rates. Fig. 7 shows the magnitude of thechanges in the real exchange rates. In the provinces where there is most inflation (Mpumalanga,

Fig. 5. Twelve-month inflation response (relative to the Free State) with a unit increase in M1.


Fig. 6. Twelve-month inflation response (relative to the Free State) with a unit increase in M1.

Fig. 7. Real exchange rates with the free state over 20 months following a unit increase in M1.

Eastern Cape and, in the long-run, KwaZulu), the real exchange rate relative to the Free State canbe expected to rise by over 10%.

4. Policy implications

4.1. Monetary growth and national policy goals

Before discussing the differences in the response of provincial prices to a monetary expansion,we should first point out some similarities. In all provinces, the initial response of prices is quite


small: the figures in Table 1 imply that a one percent increase in the money stock can be expectedto lead to monthly provincial inflation rates of less than 0.1%. Transition to the new steady state(in which the average elasticity of p with respect to m across the provinces is equal to unity)takes a great deal of time: less than one half of one percent of any deviation from the steadystate is eroded each month. This inertia is reflected in Fig. 2 above, in which aggregate inflationvolatility is much smaller than the volatility of monetary growth. Inflation is stationary, but witha substantial degree of persistence.

The phenomenon of inflation persistence is by no means confined to South Africa. Forexample, Angeloni et al. (2004) present evidence for inflation persistence in the Euro area. Theexistence of inflation persistence has important consequences for monetary policy. A simpleTaylor Rule is unlikely to be socially optimal, and the central bank loss function should explicitlyincorporate inflation growth and lagged output (Levin & Moessner, 2005). For example, in thetheoretical model of Amato and Laubach (2003), persistence arises because some firms use rulesof thumb to set their own prices based on past aggregate inflation. The standard Taylor Rule lossfunction

Lt = [πt − π∗]2 + λx2t (3)

is replaced by

Lt = [πt − π∗]2 + λx2t + θ�π2

t (4)

where πt is the actual inflation rate, π* the target rate and xt is the output gap. In Steinsson (2003),firms’ rules of thumb also incorporate past levels of demand, and the loss function becomes

Lt = [πt − π∗]2 + λx2t + θ�π2

t + φx2t−1 + ψxt−1�πt (5)

These differences are likely to be quantitatively important in South Africa, because the differentcomponents of Eq. (5) are not highly correlated. Using our monthly data, with an HP filter tocalculate the output gap, no pair of elements in Eq. (5) has a correlation coefficient greater than0.2, except x2

t and x2t−1, which only have a correlation coefficient of 0.35. Steinsson (2003)

notes that a central bank operating with the extended loss function will tend to put less weighton output stabilization than under the standard Taylor Rule. Currently, the Reserve Bank isoperating under an inflation targeting regime, with no explicit weight on output stabilization. Thehigh degree of inflation persistence might be seen as a rationale for putting less weight on outputstabilization, although the socially optimal weight on output stabilization is still greater thanzero.

4.2. Monetary growth and regional policy goals

Our results also inform regional economic policy. Although all the initial responses of provin-cial inflation rates to an increase in the money stock are quite small, there is substantial heterogene-ity in the long-run price responses across the nine provinces, as indicated in Fig. 7 above. Thesedifferences should be of serious concern to monetary policy-makers. If some positive supply shockcommon to all provinces induces the Reserve Bank to enact a stabilizing monetary expansion,relative prices across the provinces will be permanently affected. Such monetary expansion hasbeen a feature of the South African economy in the recent past. The average growth rate in thequantity of money in our sample period is 1.1% per month, compared with 0.03% for real GDE,


and this is part of the explanation for the drift in the relative price series appearing in Fig. 3.5

Given the asymmetries in the response of provincial prices to monetary growth, there is an a prioricase for adjusting province-specific fiscal instruments in order to redress the change in relativeprices. In order to see what form such adjustments might sensibly take, we should first review thewider dimensions of economic heterogeneity within the South African economy, and how thisheterogeneity relates to regional policy goals.

Production in South Africa is highly geographically concentrated: approximately two-thirds ofGDP is associated with the three main metropolitan areas, which account for around 40% of thecountry’s population. They are, in order of size, Pretoria/Johannesburg in Gauteng, Cape Town inthe Western Cape and Durban/Pietermaritzburg in KwaZulu. Economic activity in these areas isroughly evenly divided between manufacturing, commerce, tourism, finance and public services.A little over 20% of GDP, and about 25% of the population, is located in what the South AfricanDepartment for Trade and Industry (DTI) terms “secondary core regions”. These are mainly the“diversified mining regions” in the north (mostly in Mpumalanga and North West provinces) andthe coastal strip between Cape Town and Durban, specialising in transport and communication,tourism and some manufacturing. The rest of the country accounts for less than 15% of GDP andabout 35% of the population.

As Table 1 shows, per capita value added is far higher in Gauteng and the Western Cape (theprovinces incorporating the two largest metropolitan areas) than in the rest of the country. Theincidence of poverty is much lower in these two provinces, although Gauteng’s total populationis so large that it still accommodates a large percentage of the nation’s poor. Mpumalanga andthe North West, home to the most economically active mining regions, also have a per capitavalue added higher than the national average. Gauteng has shown the highest growth rates inrecent years, and there is little evidence of income convergence across the different provincesin recent years. The econometric analysis of Naude and Krugell (2003), using data for a muchlonger period, suggests that different parts of South Africa converge only very slowly on a commonlevel of per capita income, and that this convergence is conditional on geographical factors and onpersistent differences in the stock of human capital. There are large and persistent macroeconomicasymmetries across different parts of the country.

Regional convergence is a stated policy goal for the DTI.6 It is anticipated that lagging regionswill catch up by building “competitive capabilities” in areas of existing comparative advantage.Another goal is a substantial reduction in poverty rates across South Africa, noting that there arealready large regional variations in the incidence of poverty. The DTI envisages specific microe-conomic interventions in order to achieve these ends, but also notes that a stable macroeconomicframework, including monetary policy, is a necessary condition for the success of microeconomicinitiatives.

Our econometric model indicates that monetary growth causes a particularly large realexchange rate appreciation in KwaZulu, Mpumalanga, Eastern Cape and Gauteng. (Theseprovinces are indicated by lighter shading in Fig. 1; see also Fig. 7 and Table 1.) The rela-tive competitiveness of these provinces is eroded by monetary growth. Gauteng and Mpumalangahave above-average levels of per capita value added, so falling relative competitiveness in theseprovinces is unlikely to hinder the goal of regional convergence. However, KwaZulu ranks seventh

5 At the same time, it is important to recognise that idiosyncratic provincial shocks are also part of the story: thecorrelation of innovations across the inflation equations in our model is less than 100%.

6 DTI Draft Regional Industrial Development Strategy, July 2006.


out of the nine in terms of per capita value added, and the Eastern Cape ranks ninth. To the extentthat economic activity is a function of relative competitiveness, a period of high monetary growthis likely to cause two of the country’s poorest regions to lag even further behind.

Quantifying the impact of falling competitiveness in low-income provinces on regional diver-gence is beyond the scope of this paper. However, it is likely that part of the explanation whyprovinces such as KwaZulu and the Eastern Cape are struggling to catch up relates to losses ofcompetitiveness due to monetary expansion. Moreover, national seigniorage revenue from thisexpansion is equal to a substantial percentage of provincial government revenues. For example,in 2005 the South African money base (currency in circulation plus commercial bank reservedeposits) grew by R8.6bn; the projected 2006/7 revenue of the Eastern Cape provincial govern-ment is R27.9bn (Nel, 2006); the equivalent figure for KwaZulu is R37.2bn (Mkhize, 2006). Ourresults provide a justification for allocating seigniorage revenue to individual provinces accordingto the amount of inflation generated there, perhaps conditional on the overall economic positionof the province. In this case, there would be a substantial increase in provincial governmentrevenues in KwaZulu and the Eastern Cape. The extra revenue would provide provincial govern-ments with the funds to finance fiscal policy adjustments to redress the provincial real exchangerate appreciation, through, for example, increased subsidies or reduced taxes for capital andlabour.

One might also worry about the impact of short-term volatility in the rate of growth of the moneystock on short-term inflation volatility in the different provinces. A higher standard deviation inthe monetary growth rate translates into a higher standard deviation in inflation rates, but theeffect will be particularly marked in those provinces where prices are particularly sensitive tomonetary growth: again, KwaZulu, Mpumalanga, Eastern Cape and Gauteng. In these provinces,any failure to achieve monetary stability is likely to have relatively large effects. However, theoverall magnitude of these effects is probably quite small, because there is so much inertia in theresponse of prices to a change in M1. Provincial variations in the effects of monetary instabilityare unlikely to be as economically important as provincial trends in competitiveness due to trendmonetary growth.

It is worth noting one further aspect of the spatial pattern of price responses. The DTI iscommitted to reducing the incidence of poverty across South Africa. The four provinces withthe greatest sensitivity of prices to monetary growth (KwaZulu, Mpumalanga, Eastern Cape andGauteng) are also the four containing the largest fractions of the nation’s poor, as shown in Table 1.We do not know whether the prices faced by poor in these provinces are more or less sensitiveto monetary growth than the prices faced by the rich: prices disaggregated by both provinceand income group are not published. However, the relatively high incidence of poverty in thoseprovinces where prices are most sensitive suggests that the publication and analysis of moredisaggregated prices would greatly inform the process of poverty reduction.

5. Conclusion

We have seen that while inflation rates across South African provinces are highly correlated,because the different provinces face common shocks, there is no discernable long-run convergenceof price levels; that is, there is no inter-provincial PPP. Moreover, there is marked heterogeneityin the immediate response of provincial prices to monetary expansions and contractions. (Indeed,these asymmetries are large relative to some international transmission mechanism asymmetries.)This means that changes in monetary policy lead to substantial and persistent changes in relativeprices. It is important that these effects are not ignored by the monetary policy-maker. Monetary


policy based solely on targets for national macroeconomic aggregates, without any compensatingadjustment in provincial fiscal policy, is unlikely to be optimal.

The underlying causes of the heterogeneity are a pressing topic for future research. The greatestsensitivity of prices to monetary policy appears in South Africa’s coastal provinces and aroundthe national capital, but we do not yet know what specific provincial characteristics have led tothis relative sensitivity. It also remains to be seen whether the magnitude of the heterogeneity thatappears in South Africa is replicated elsewhere. If it does, then we should begin to view nationstates as monetary unions, rather than as homogeneous macroeconomic entities.

References

Amato, J., & Laubach, T. (2003). Rule-of-thumb behaviour and monetary policy. European Economic Review, 47, 791–831.Angeloni, I., Aucremanne, L., Ehrmann, M., Gali, J., Levin, A., & Smets, F. (2004). Inflation persistence in the Euro area:

preliminary summary of findings. manuscript, European Central Bank.Aron, J. & Muellbauer, J. (2005). Review of monetary policy in South Africa: 1994–2004: Part 1: the transparency,

credibility and predictability of monetary policy (CSAE working paper). Oxford University.Carlino, G., & DeFina, R. (1998). The differential regional effects of monetary policy. Review of Economics and Statistics,

80(4), 572–587.Cecchetti, S., Mark, N., & Sonora, R. (2002). Price index convergence among United States cities. International Economic

Review, 43(4), 1081–1099.Ceglowski. (2003). The law of one price: intra-national evidence for Canada. Canadian Journal of Economics, 36(2),

373–400.Engel, C., & Rogers, J. (2001). Violating the law of one price: should we make a federal case out of it? Journal of Money,

Credit and Banking, 33(1), 1–15.Garrison, C., & Chang, H. (1979). The effects of monetary forces in regional economic activity. Journal of Regional

Science, 19, 15–29.Garrison, C., & Kort, J. (1983). Regional impact of monetary and fiscal policy: a comment. Journal of Regional Science,

23, 249–261.De Grauwe, P. (2000). Monetary policies in the presence of asymmetries. Journal of Common Market Studies, 38, 593–612.Gros, D., & Hefeker, C. (2002). One size must fit all: national divergences in a monetary union. German Economic Review,

3, 247–262.Haug, A., Karagedikli, O., & Ranchhod, S. (2003). Monetary policy transmission mechanisms and currency unions: a

vector error correction approach to a Trans-Tasman currency union. (Reserve Bank of New Zealand discussion paper2003/04).

Levin, A. & Moessner, R. (2005). Inflation persistence and monetary policy design: an overview. (Working Paper 539.)European Central Bank.

Mkhize, Z. (2006) Province of KwaZulu-Natal Budget Speech 2006–7. Pietermaritzburg, 26 February.Naude, W., & Krugell, W. (2003). An inquiry into cities and their role in sub-national economic growth in South Africa.

Journal of African Economies, 12, 476–499.Nel, B. (2006). Eastern Cape Provincial Treasury Budget Speech and Policy Statement 2006/07. Bisho, February 20.Pesaran, M. H., Smith, R. J., & Shin, Y. (2001). Bounds testing approaches to the analysis of level relationships. Journal

of Applied Econometrics, 16, 289–326.Steinsson, J. (2003). Optimal monetary policy in an economy with inflation persistence. Journal of Monetary Economics,

50, 1425–1456.Toal, W. (1977). Regional impacts of monetary and fiscal policies in the post-war period: some initial tests. (working

paper) Federal Reserve Bank of Atlanta.

regional asymmetries in monetary transmission: the case of south africa

Documents