Winners and Losers in the Commodity Lottery: Terms of Trade Volatility and Underdevelopment in

the Periphery, 1870-1939

by

Christopher Blattman University of California, Berkeley

Jason Hwang

Harvard University

Jeffrey G. Williamson Harvard University

October 2004 We acknowledge the superb research assistance of Martin Kanz. We are also greatly indebted to Luis Bértola, Michael Clemens, John Coatsworth, and Yael Hadass for their help in constructing the underly-ing data base for other projects with Williamson. We are also grateful to them and to David Albouy, Luis Catao, Brad Delong, Charles Engel, Chang-Tai Hsieh, Michael Jansson, Ian McLean, Edward Miguel, Kevin O’Rourke, Leandro Prados, Martha Olney, and seminar participants at Harvard University and University of California Berkeley for comments. Blattman and Hwang thank the Harvard Center for In-ternational Development and the National Bureau of Economic Research for office space and computing resources. Williamson thanks the National Science Foundation for support under NSF grant SES-0001362. Note that an earlier version of this working paper was published as NBER WP10600, “The Im-pact of the Terms of Trade on Economic Development in the Periphery, 1870-1939: Volatility and Secu-lar Change”.

1

Abstract This paper argues that commodities and the terms of trade need to be more central to our thinking about long term development. We argue that a look at commodities and commodity price fluctuations help us explain the growth puzzle noted by Easterly, Kremer, Pritchett and Summers more than a decade ago: that the contending fundamental determinants of growth—institutions, geography and culture—exhibit far more persistence than do the growth rates they are supposed to explain. Commodity choice, dependence and the associated price trends and volatility seem to be especially powerful determinants of growth in the developing world. The commodity-specialized were not uniformly slow-growing and poor. Neither were their commodities uniformly volatile and decreasing in value. Reconstructing nearly a century of terms of trade experience from 1870 to 1939 and assessing its impact on the economic performance of 35 coun-tries, we see that some commodities proved more volatile in price than others, and those countries with more volatile commodities have grown more slowly than other commodity-specialized nations. Terms of trade shocks, meanwhile, are strong predictors of both slowdowns and accelerations in income growth, and higher volatility is associated with dramatically lower foreign capital inflows (a possible channel of impact on growth). We also observe a striking asymmetry between Core and Periphery. While terms of trade volatility dragged foreign capital flows and domestic output down in the Periphery, this was not true of the Core. Terms of trade trend growth, meanwhile, raised income growth in the Core and depressed it in the Periphery. Christopher Blattman Jason Hwang Department of Economics Department of Economics University of California Harvard University Berkeley, CA 94720 Cambridge, MA 02138 blattman@berkeley.edu jjhwang@fas.harvard.edu Jeffrey G. Williamson Department of Economics Harvard University Cambridge, MA 02138 jwilliam@fas.harvard.edu JEL No. F1, N10, O1

2

1. Introduction

This essay explores an underappreciated pattern of long term growth: (i) most countries in the world

have been specialized in the export of just a handful of commodities for most of their history; (ii) some of

these commodities have proven more volatile in price than others; and (iii) those countries with more

volatile primary products have grown more slowly than other commodity-specialized nations. Looking at

nearly a century of price change, we draw a direct link between commodity price volatility and economic

underdevelopment, explaining a good part of the diversity in economic performance among commodity

producers. In a small open economy, commodity price fluctuations are essentially exogenous, and so we

feel the effect of volatility on growth is likely causal. Our findings are reminiscent of what Carlos Diaz-

Alejandro (1984) called the commodity lottery: each country’s exportable resources, he explained, were

determined in large part by geography and chance, and differences in later economic development were a

consequence of the economic, political and institutional attributes of each product. We argue that the

price volatility of each primary product also mattered, by generating internal instability, reducing invest-

ment, and diminishing economic growth.

Up until now the long-term implications of commodity price fluctuations have remained unexplored

due to a lack of country-specific terms of trade data. We develop new terms of trade for 35 rich and poor

countries for the years 1870 to 1939. We classify twenty-one as “Periphery” countries and the remainder

as the “Core”, a classification based on the level of economic diversification, commodity specialization,

and historical economic relations with the world’s industrial leaders, as explained in Section 3. Together

these countries cover more than 85 percent of the world population and nearly all of world GDP in 1914.

The vast majority, especially the Periphery, are commodity-specialized in their exports. Even by the end

of our study period, twenty-five of our sample had more than three-quarters of their exports in primary

products, and in twenty just two products made up more than two-thirds of the nation’s total exports.

Using the new data, Figure 1 charts income per head in 1939 against volatility in the terms of trade

for the years 1870 to 1939.1 Volatility is measured as the standard deviation of departures from a slow-

moving trend.2 The figure clearly depicts a negative correlation between terms of trade volatility and sub-

sequent level of development, not just in the total sample but also within the subset of primary product-

specialized countries traditionally known as the Periphery. Figure 2 charts 1939 income per head against

the average trend in the terms of trade. Within both the Periphery and the Core, we see a positive correla-

tion between growth in the terms of trade and the subsequent level of development. We argue that the

1 Here and throughout the paper we omit the World War I and World War II decades. For a full account see section 3.

3

causation is much more likely to run from the terms of trade to growth rather than the other way around.

Our terms of trade data are based on world market prices, and the primary producers in our sample are for

the most part price-takers on the world market. Thus, secular changes and volatility in the terms of trade

were exogenous events in the Periphery.

Our results shed light on three longstanding economic growth puzzles. First, although specialization

in the production and export of primary products has proven to be one of the most robust determinants of

poor economic growth, economists have yet to account fully for the power of commodity-specialization

in predicting economic performance. Sala-i-Martin (1997), in his infamous one million regressions, finds

that primary product specialization is one of the most robust determinants of growth.3 We propose that

primary product price volatility is one reason why. The second puzzle is that, in spite of the explanatory

power of commodity specialization, economic performance among the resource-rich has varied enor-

mously. One can clearly see the remarkable diversity in the growth experiences of the resource-dependent

in Figures 1 and 2. In the last 150 years, some have grown remarkably fast (e.g. Canada and Norway),

some have grown very slowly (e.g. Colombia and, at least as of 1950, India), while many lie somewhere

in between (e.g. Turkey and Brazil). Clearly, all primary product exporters were not created equal.

The third puzzle is the observation that while differences in long term economic performance are usu-

ally explained by the contending fundamental determinants of growth—namely institutions, geography

and culture—these fundamentals exhibit far more persistence over time than do the growth rates they are

supposed to explain.4 Part (i) of Table 1 compares the fraction of total variation in five-year panel data

that is due to within-country variation versus between-country variation for the years 1960 to 1999. Look-

ing at similar data, Pritchett (2000) pointed out the fraction of the variance in GDP growth due to within-

country variance is 0.73 of the total variance, while common explanatory variables for growth were in-

stead dominated by between-country variance. In Table 1(i) we see the ratio of within-country variance to

total-variance is 0.29 for population growth, 0.28 for investment rates, and 0.09 for primary schooling in

the years 1960 to 1999. Although not displayed, the persistence of other common explanatory variables—

geography, culture, and political institutions—is of course also very high, with total variation almost en-

tirely between countries.5 Extending Pritchett’s analysis, we note that the same patterns exist in the pre-

WWII data. In parts (ii) and (iii) of Table 1 we look at analogous figures for the forty-year period from

1870 to 1909 (quinquenially) and also for the eighty years from 1870 to 1949 (by decade). Even in this

historic period of divergence between the Atlantic economies and the rest of the world, we see even

2 This trend was calculated using a Hodrick-Prescott filter as discussed in Section 2, though the pattern holds for alternative de-compositions. 3 See also Sachs and Warner (2001). 4 See Pritchett (2000) and Easterly, Kremer, Pritchett, and Summers (1993).

4

greater dominance of within-country growth variation (and comparatively less within-country variation in

population growth, foreign capital inflows, primary schooling, and primary product concentration).

Econometrically, the impact of using low-frequency variables to describe a high-frequency one is to

reduce the accuracy of estimation: country fixed effects sweep out much of the variation in the explana-

tory terms, increasing the share of variation arising from the high-frequency components of growth. More

broadly, it is hard to escape the feeling that an explanation of something as volatile as growth with fea-

tures as invariant as geography or education is somehow unsatisfactory. We suspect that an important de-

terminant of growth has been overlooked in the contest between constitutions, cultures and coastlines.

Observers regularly point to terms of trade shocks as a key source of macroeconomic instability in com-

modity-specialized countries, but they pay far less attention to the growth implications.6 In Table 1 we see

that, unlike other potential determinants of growth, the high-frequency terms of trade are dominated by

within-country variation. The ratio of within-to-between-country variance in the average annual growth

rate of the terms of trade is 0.90 from 1960 to 1999, 0.96 from 1870 to 1909, and 0.90 from 1870 to 1949.

While there are of course other high-frequency determinants of growth, the terms of trade have the advan-

tages of ease of measurement and, in the small-country case, relative exogeneity. In this essay, we exploit

the long-term historical evidence to explore the hypothesis that exogenous shocks in the terms of trade

can account for both the variance around fundamentals and the fundamentals themselves.

Our results support this hypothesis. Using the new terms of trade database we conclude that terms of

trade volatility can not only explain the instability of growth rates, but also differences in long-term

growth. Looking at the Periphery alone (the European Offshoots, Latin America, Asia and the Middle

East), simple OLS estimates strongly suggest that between 1870 and 1939 a one standard deviation in-

crease in terms of trade volatility was associated with a decrease in the rate of growth of GDP per capita

of roughly 0.6 percentage points per annum, an extremely large decline that happens to be robust to

changes in the time period, volatility measure, and nations on the margin assigned to the Periphery cate-

gory. We also analyze the incidence of growth starts and stops, and see that higher volatility and declining

terms of trade reduce the likelihood of an acceleration in growth and increase the likelihood of a decelera-

tion. Looking at the Periphery alone helps us eliminate an important endogeneity concern—that more de-

veloped nations have a more diversified export base, and therefore a less volatile terms of trade. While

our Periphery countries had wildly divergent growth experiences, all were more or less equally dependent

on a handful of commodities, in effect holding export diversification constant.

5 Looking at the post-WWII period, Isham, Kaufman and Pritchett (1997) identify high levels of persistence in civil liberties and democracy measures, and Deninger and Squire (1996) do the same for measures of inequality. 6 For important exceptions, see Mendoza (1997), Deaton and Miller (1996), Kose and Reizman (2001), Bleaney and Greenway (2001), and Hadass and Williamson (2003).

5

Although our principal results come from looking at terms of trade shocks in the commodity depend-

ent Periphery, it will be instructive to compare the experience of the Periphery to that of the Core, (where

we see higher average terms of trade growth and lower volatility than in the Periphery). Doing so, we

identify asymmetric effects: terms of trade trends and volatility were distinctly harmful to growth in the

Periphery, but not so in the Core.

Finally, we investigate one prime channel through which volatility could have impacted growth—

foreign investment. Eichengreen (1996) argues that in much of the Periphery capital inflows shrank as

price shocks reduced the attractiveness of investment. We attempt to quantify this response. We examine

a formal model in which a secular improvement in the terms of trade leads to higher levels of investment,

and hence long-run economic growth, while higher volatility in the terms of trade reduces investment, and

hence growth, because of aversion to risk. We then test the predictions of the model using the only source

of investment data for the period in question—British capital flows from 1870 to 1913. Figure 3 plots the

natural log of average capital flows against average terms of trade volatility for the full period, 1870-

1913. We see a negative relationship between capital flows and volatility, although the variation is ex-

tremely high. We turn to regressions for a more precise analysis, and while we do not find evidence that

trend changes in terms of trade influenced capital flows, volatility mattered. A one standard deviation in-

crease in terms of trade volatility was associated with at least a 25 percent decrease in average capital

flows to the Periphery. Our investment results are mildly sensitive to specification, however, and so we

conclude that while investment was clearly one channel of impact, volatility must have impacted growth

along other channels as well.

Our long-run historical findings complement and extend Ramey and Ramey (1995) who, looking at

the mean and volatility of growth for the years 1962 to 1988, conclude that countries with higher year-to-

year volatility in growth tend to have systematically lower growth rates over time, particularly outside the

OECD. They only identify proximate sources of volatility, however, such as government finances and (to

a lesser extent) investment. We identify an important root source of volatility—commodity choice and the

terms of trade. Commodity-specialization and participation in global markets alone were not sufficient

conditions for high volatility and low growth. Rather, the peculiar nature of each commodity mattered, in

particular its variation in price. The conclusion we draw is that the neglect of commodities and terms of

trade in the modern growth literature has been costly to our understanding of what matters for long run

growth outside of the US and Western Europe. It is time to bring commodities back into the picture.

The remainder of this essay is organized as follows. Section 2 reviews some of the historical, empiri-

cal and theoretical support for the relationship between the terms of trade and development, and makes

the economics explicit. Section 3 presents the data. Our empirical strategy is described in Section 4, and

the results are presented in Section 5. Section 6 concludes.

6

2. Theory and Evidence on the Terms of Trade and Development

In the introduction to his famous 1950 article, Hans Singer proposed that fluctuations in the terms of

trade dramatically affected the funds available to poor countries for capital formation, and hence growth.

Unfortunately for Singer, he missed an opportunity by failing to dwell on this point, and focused instead

on the implications of a secular deterioration in the terms of trade. The literature on this postulated secular

deterioration is long and contentious. It is also, for the most part, immaterial, since from the longer back-

wards view in 2004 little trend appears to exist.7 We will thus ignore long term drift and focus on shorter

term fluctuations in the terms of trade.

Commodity price fluctuations feature prominently in the economic history of developing countries.

Looking broadly at the Periphery, Eichengreen (1996) argues that fluctuations reduced growth through

the investment channel. Noting the destabilizing shifts in international capital flows due to terms of trade

shocks before and after the First World War, Eichengreen makes a case that negative terms of trade

shocks depressed export revenues, and that capital inflows shrank as investment became less attractive.

Current and capital account shocks thus reinforced one another, provoking financial crisis and inhibiting

growth. This relationship seems to have held in the postwar period as well. Using post-WWII data,

Aghion et. al. (2004) present evidence that, for countries at low levels of financial development, negative

terms of trade shocks are associated with lower average growth rates of national income. Like Eichen-

green, the channel they emphasize is procyclical short-term investment.

In this paper we intend to focus on establishing a connection between terms of trade fluctuations and

growth, and focus in detail upon Eichengreen’s suggested channel—foreign investment. Yet after decades

of debate, there is still no well-articulated theory, let alone consensus, on the short run effects of terms of

trade shocks on growth and investment. One set of theories predicts a negative correlation, a relationship

often referred to as the ‘resource curse’. Sachs and Warner (1995, 2001), for instance, have observed that

resource-rich countries tend to grow more slowly than resource-poor countries. Yet no single resource

curse theory is universally accepted. Sachs and Warner prefer the “crowding-out” logic, whereby primary

production crowds out manufacturing activities. A political economy approach offers an alternative, usu-

ally relying on some form of government ineptitude or corruption. Krueger (1974) famously argued that

rent-seeking was a growth-suppressing tendency of resource-owning elites in poor countries.

7 It was possible to look back from 1950 and observe a downward drift in the commodity terms of trade. Along with Raoul Pre-bisch (1950), Singer argued that the fundamental nature of these commodities made it inevitable that primary product prices would fall relative to manufactures in the long run, and so the terms of trade and incomes of commodity exporters would decline over time. The “Prebisch-Singer thesis” has not survived the half century since they wrote, however, since we now think that structural breaks, serially correlated residuals, and unit roots may explain the patterns we see. Grilli and Yang (1988) analyze 20th century commodity price data and find evidence of periodic structural breaks, but no trend. Bleaney and Greenaway (1993) contest this finding, and demonstrate the existence of a downward trend, but one of only -0.5 percent per annum. Such a trend, if

7

Another set of theories claims that terms of trade improvements raise the value of output and the re-

turns to investment in developing countries, and hence predict a positive correlation between those im-

provements and growth. For instance, Basu and McLeod (1992) develop a stochastic growth model in

which imported inputs make production more efficient, but a drop in export prices make such inputs more

costly and reduce output. They confirm that transitory terms of trade shocks have persistent effects on

output levels in a sample of 12 primarily Latin American countries. Empirical evidence from Africa

seems to support this prediction over the resource curse. Deaton and Miller (1996) find that a sudden 10

percent increase in commodity prices results in a 6 percent increase in output. Deaton (1999) also finds

that a 12 percentage point increase in commodity prices in contemporary Africa adds 1.8 percentage

points to the GDP per capita growth rate.

The analysis of such shocks is useful in illustrating the direct and immediate effects on GDP, but it

does not identify the long term impact on development, and it fails to account for the difference between

permanent and transitory changes in the terms of trade on growth. When thinking about fluctuations, it

will be useful to decompose annual disturbances, or shocks, into two parts: the short-run secular trend and

the variance around trend, which we call volatility. The literature has addressed the impact of all three

types of fluctuations on macroeconomic performance, although the distinction is seldom articulated.

Short Run Terms of Trade Trends and Long Run Macroeconomic Performance

Here again theory and evidence are mixed. As above, the resource curse crowd argues that natural re-

source sectors are inherently unproductive, and a rise in the return to those resources will reduce growth.

The alternative view presumes that natural resource investments are productive. For instance, Mendoza

(1997) captures the idea that an increase in the price of the commodity export represents an increase in the

expected rate of return on investment in that sector, thus augmenting investment and growth. Mendoza’s

model, which we review below, seems to be supported by recent developing country experience. Using

cross-country panel regressions of 9 industrial and 31 developing countries from 1970 to 1991, Mendoza

finds that an increase in the growth rate of the terms of trade by 1 percent raises the growth rate of con-

sumption by 0.2 percent. Bleaney and Greenaway (2001) use Mendoza’s model to analyze GDP per cap-

ita growth in 14 sub-Saharan countries from 1980 to 1995, finding that growth and investment increase as

the terms of trade improve.

Do such findings spell doom for the resource curse theory? Not quite. Hadass and Williamson (2003)

find that terms of trade growth between 1870 and WWI reduced growth in a sample of commodity ex-

porters, lending support to the resource curse in the long run. Yet their sample covers few of the develop-

it even exists, is very small relative to short term fluctuations, and so the remainder of this essay expects volatility to matter far more.

8

ing countries that remained poor up to World War II, and they did not explore the influence of volatility.

A larger sample of underdeveloped countries is needed to test for the influence of the terms of trade dur-

ing the period that motivated the Prebisch-Singer debate in the first place.

Terms of Trade Volatility and Economic Growth

Most theories of terms of trade volatility operate through the investment channel.8 The development

literature offers an abundance of microeconomic evidence linking income volatility to lower investment

in both physical and human capital. Households imperfectly protected from risk change their income-

generating activities in the face of income volatility, diversifying and skewing towards low-risk alterna-

tives with lower returns.9 Volatility can also result in suboptimal levels of investment in productive as-

sets.10 Finally, severe cuts in health and education seem to follow from negative shocks to income—cuts

that disproportionately affect children and hence long term human capital accumulation in poor countries.

For example, it has been shown that negative income shocks caused large numbers of households to with-

draw their children from school in Cote d’Ivoire and India, while the recent Indonesian financial crisis has

been shown to have reduced enrollment and health expenditures.11 Poor households cannot smooth their

consumption and investments in the face of shocks because they are rationed in credit and insurance. So

too on a macroeconomic scale—governments in the Periphery often find it difficult to borrow internation-

ally, making it hard to smooth public investment and expenditure in the face of terms of trade shocks.12

Ramey and Ramey (1995), for instance, find that government spending fluctuations and macroeconomic

volatility are closely related, and that this volatility is associated with lower growth.

We have already surveyed models where growth in the terms of trade leads to higher levels of in-

vestment because of higher returns. Likewise, higher volatility in the terms of trade should reduce in-

vestment and growth in the presence of risk aversion. Several simple models exist sharing these features,

and for illustrative purposes we look at Mendoza’s (1997). His model captures both features and it mim-

ics the small, commodity-dependent exporting nation. It is a one-sector, stochastic endogenous growth

model with risk-averse households who produce for export and consume imported goods. The terms of

trade influences both the risk and return on domestic assets, and when risk-averse individuals cannot in-

8 Another smaller literature examines civil conflict as the channel through which terms of trade shocks affect long run growth. Rodrik (1999) offers one example. Miguel et al. (2004), however, find no relationship between the terms of trade and conflict in Africa over the last two decades. We plan to explore this issue in future papers. 9 For a review, see Dercon (2004) and Fafchamps (2004). 10 See, for example, Rosenweig and Wolpin (1993) and Fafchamps (2004). 11 On the effects of schooling shocks, see Jensen (2000) or Jacoby and Skoufias (1997). On the Indonesian crisis, see Franken-burg et al. (1999) or Thomas et al. (2004) 12 While greater volatility increases the need for international borrowing to help smooth domestic consumption, Catão and Kapur (2004) have shown recently that volatility constrained the ability to borrow between 1970 and 2001.

9

sure against terms of trade fluctuations, savings and growth are reduced. Formally, households choose a

consumption path that will maximize expected lifetime utility:

( ) ,10,010

1

<<>

−

= ∑∞

=

−

βγγ

βγ

t

tt cECU

where Ct is consumption of the imported good, β is the subjective discount rate, and γ is the intertemporal

elasticity of substitution. Households maximize utility subject to the following period-by-period budget

constraint:

−≤−

t

tttt z

CARA 1 ,

where At is the stock of wealth in units of an exportable commodity in period t, Rt is the stochastic rate of

return realized each period that yields units of the exportable commodity that agents exchange for units of

the importable good, and zt is the relative terms of trade (or the price of exports in terms of imports in

world markets). Rt and zt are non-negative random variables such that the effective rate of return rt =

Rtzt+1/zt follows a log-normal i.i.d. distribution. Hence ln(rt) has mean µ and variance σ2.

Mendoza obtains closed form solutions for consumption and wealth, and demonstrates that consump-

tion growth can be expressed as:

,)1(1t

t

t rC

C λ−=+

where (1-λ) is the saving rate with respect to wealth, and it is a function of the mean and variance of the

terms of trade:

−−−≡ −

γσγβµσµλ γγ

2112 )1(exp)(1),( .

Since it is assumed that agents cannot insure against fluctuations in rt, an increase in the volatility of the

terms of trade (i.e., a mean-preserving increase in σ2) leads to reduced savings and increased consumption

if the coefficient of relative risk aversion, γ, is lower than 2. Growth in the terms of trade (an increase in

µ) has the opposite effect.13

In samples over short time periods, the empirical evidence seems to support a negative impact of

terms of trade volatility and a positive impact of terms of trade growth. Mendoza tests his predictions for

13 As Bleaney and Greenaway (2001) note, the strong predictions of the Mendoza model with respect to the impact of terms of trade trend on consumption growth do not necessarily carry over to output growth. Rather, the desired relationships between out-put growth and terms of trade trend and volatility demand that the coefficient of relative risk aversion γ be less than 1. Thus,

10

40 industrial and developing countries over the period 1971–1991. Using consumption growth as a proxy

for economic growth, he confirms the predicted positive relationship between secular terms of trade im-

provements and economic growth. He also finds a significant negative relationship between growth and

terms of trade volatility.

Mendoza’s empirical specification is vulnerable to endogeneity concerns, however, since industrial-

ized countries with a more diversified export base can be expected to have lower terms of trade volatility.

As we noted in the introduction, one reasonable solution is to limit the sample to developing countries

alone. So long as these countries have equally concentrated export bases but divergent growth experi-

ences, we can estimate the effects of terms of trade trends and volatility conditional on export concentra-

tion. Several studies use late-twentieth century data on developing countries alone and corroborate Men-

doza’s findings. For 61 developing countries between 1975 and 1992, Turnovsky and Chattopadhyay

(2003) find that an increase in the growth rate of the terms of trade has a weak positive effect on the mean

growth rate of output while an increase in volatility has a strong negative effect on output growth. Their

results hold for both high and low volatility sub-samples. Furthermore, for 14 sub-Saharan countries be-

tween 1980 and 1995, Bleaney and Greenaway (2001) find that terms of trade volatility has a significant

negative impact on growth in income per head and investment levels.

Although each of the above studies employs different measures of volatility and different developing

country samples, their results are remarkably consistent. Each is limited to the 20 or 25 years after 1970,

however, and so none of them can really speak to the long term. Short periods also constrain their ability

to isolate volatility from trend. Since our data cover the 70 years after 1870, our analysis does not suffer

that limitation. Our longer time series also makes possible what we think is a better method of decompos-

ing terms of trade shocks into trend and volatility.

3. Discussion of the Data

Our analysis stretches over three twenty year periods—the rapid expansion of global markets from

1870 to 1889, theit maturation from 1890 to 1909, and the tumultuous interwar years from 1920 to

1939.14 These periods were chosen because they represent episodes of global integration and disintegra-

tion, inducing large terms of trade changes and economy-wide responses. The two war decades are ex-

cluded due to the absence of data and to the gross distortions to trade and prices attributable to war de-

mands, blockades and skyrocketing transport costs. The years 1950 to 2000 are also excluded because

technically speaking, the theory is ambiguous regarding the effects of trends and volatility of the terms of trade on growth and investment. 14 We stop at 1909, and omit the years 1910-1913, because the trend and volatility data for are only comparable if they are based on a full decade.

11

they have been explored elsewhere, and because radical changes in the composition of production and

trade in the periphery make the use of a continuous terms of trade series less meaningful.

We have collected new terms of data for 35 countries. Table 2 lists our sample, their GDP per capita,

the share of their exports in primary products, their export concentration, and their export share in GDP.

We divide our sample into Core and Periphery nations, an allocation based on traditions in the historical

literature which rely on geography, endowments, economic structure, commodity dependence, and the

level of development. Our Periphery was mostly poor, commodity-specialized, and highly concentrated in

one or two export products. There are fourteen countries in our Core, including four Industrial Leaders

(France, Germany, the UK, and the USA), five European Industrial Latecomers (Austria-Hungary, Den-

mark, Italy, Norway, and Sweden), and five in what we call the European Periphery (Greece, Portugal,

Serbia, Spain, and Russia). The Periphery numbers twenty-one, including eight in Latin America (Argen-

tina, Brazil, Colombia, Chile, Cuba, Mexico, Peru, and Uruguay), ten in Asia and the Middle East

(Burma, Ceylon, China, Egypt, India, Indonesia, Japan, the Philippines, Siam, and Turkey), and three rich

European Offshoots (Australia, Canada, and New Zealand). Any division between Core and Periphery is

of course somewhat arbitrary, but our results are generally robust to the allocation, in particular the relo-

cation of the European Periphery or the European Offshoots.

Key Variables

GDP per capita. For most countries in our sample, the primary source of GDP per capita estimates (in

1990 $US) is Maddison (1995). Since his developing country GDP estimates often begin only with 1900

or even 1913, estimates for earlier years must be obtained from supplementary sources, in particular back-

casting from the earliest Maddison benchmarks using estimates of real wage or output growth constructed

by Jeffrey Williamson.15 The doubtful quality of the GDP per capita growth estimates for some countries,

and especially for the late 19th century, implies that measured changes in GDP per capita over a decade or

more are almost certainly more reliable than annual ones. They will also give a better fix on long run ef-

fects.16 As long as the measurement errors are random, they should not, of course, bias our econometric

estimates, but rather only raise standard errors.

Capital Flows. Our capital flow data are measured with far more precision.17 They refer to British

capital exports for the period 1870 to World War I, and our country sample received 92 percent of it. We

choose British capital exports as a measure of international capital flows for two reasons. First, it is avail-

15 Full details are available in the data Appendix. 16 Decades were chosen to be the main unit of analysis in part to reduce measurement error in the dependent variable, in part to have a sufficient span of time over which to measure terms of trade trend and volatility, and to ensure there were sufficient ob-servations to give the econometric tests power. 17 The data were originally collected by Leland Jenks and Matthew Simon, and then reported by Stone (1999). They were cleaned recently by Clemens and Williamson (2004).

12

able. Second, Britain was then the world’s leading capital exporter, far exceeding the combined capital

exports of its nearest competitors, France and Germany. Thus British capital was the dominant source of

international capital for more than forty years. Clemens and Williamson (2004) document that this capital

flowed to where it was more profitable—chasing natural resources, educated populations, migrants and

young populations. We explore the impact of terms of trade shocks on such flows.

Terms of Trade. The net barter terms of trade is, of course, defined as the ratio of export to import

prices. The US and six of our European countries have excellent terms of trade data, but this is not true of

the rest. Thus, we constructed new terms of trade series for the remaining twenty-eight.18 Formally,

for product i, country j, and period t. Note that we employ an international market price, p*, for both nu-

merator and denominator, prices which have the advantage of greater accuracy and availability.19 More

importantly, in the small country case (a definition that seems to capture our Periphery exporters) changes

in the world price of the commodity can be treated as exogenous shocks. We will test this identification

assumption in Sections 4 and 5 and find it stands up to rigorous inspection.

Also note that the import index in the denominator is in fact a price index for extensively traded US

manufactured goods. The same import price index has been used for all periphery countries since reliable

country-specific import mix data are not available for most periphery countries before World War I.20

What data do exist, however, suggest that the assumption closely approximates reality. Moreover, focus-

ing our analysis on the purchasing power of a country’s exports in terms of a common basket of manufac-

tured products is actually advantageous in that we can isolate the between-country effect of relative com-

18 Given the extensive previous effort that has gone into the construction of existing US and the six European terms of trade se-ries, we have opted to employ these terms of trade indices. These traditional indices have not necessarily been constructed using the same price sources and weighting schemes, and so are not strictly comparable to our indices constructed for the periphery. But since the primary object of our investigation is the Periphery, and the Core is considered for comparative proposes only, we do not think differences in terms of trade construction pose a serious problem. Moreover, construction techniques seem similar and hence fairly comparable. 19 To the extent that transport costs and tariff rates change significantly over time, a country’s terms of trade measured in home markets might have obeyed somewhat different laws of motion than ours having been measured in international markets. While our indices might not exactly represent domestic prices, we are confident that the two sources would yield very similar trends, especially after 1900 when the decline in ocean transport costs slowed down considerably (Shah Mohammed and Williamson 2004). 20 The import index in the denominator is a weighted sum of the prices of textiles (55%), metals (15%), machinery (15%), build-ing materials (7.5%), and chemicals and pharmaceuticals (7.5%) from the US Department of Commerce Historical Statistics. US data are employed because continuous price indices for comparable goods are unavailable for any other country. We draw com-fort from the fact that the US price index tracks very closely the British Board of Trade index of export prices, and terms of trade series constructed from the UK and US price indices have a correlation coefficient of 0.91.

∑

∑

=

=

⋅

⋅= J

jiit

J

jijijt

jt

wp

wpTOT

1

*

1

*

13

modity price movements on growth.21 Sources and additional information on the index are found in a data

appendix. Our measure of export concentration, or the share of exports from the top two exports, also

come from these data.

Institutional quality data are hard to come by for the full sample and period. Most institutional meas-

ures date from the post-war period, and are therefore endogenous. We propose two proxies for initial in-

stitutional quality. One is the percent of the population of European descent in various years, including

1600, 1800 and 1900, and taken from early versions of Acemoglu, Johnson and Robinson (2001).22 Fol-

lowing these authors, we presume that the presence of European settlers will be an (imperfect) indicator

of the quality of institutions. Unlike the settler mortality data, European population data is available for

the full sample. A second set of indicators come from the POLITY IV database23. A democracy indicator

and an autocracy indicator measure on a 10 point scale the competitiveness of political participation, the

openness and competitiveness of executive recruitment, and constraints on the chief executive. The polity

indicator takes on a value between (-10, 10), and is the difference of the democracy and autocracy indica-

tors. While covering the period 1820-2000, POLITY IV data is only available for countries following

their Independence. Hence data for certain country-years (e.g., Australia before 1900) are not available.

Even with these gaps, however, this dataset is one of the most complete for the era.

Our remaining control variables come from a historical database constructed by Jeffrey Williamson

and others, and sources and methods of construction are described in Clemens and Williamson (2002).

Export share is calculated as the ratio of exports to national income. The share of exports in primary

products follows a definition similar to that in Sachs and Warner (1997) and is the ratio of all exports ex-

cept manufactured products and specie to total exports. Population growth rates are straightforward.

Schooling reflects estimates of primary school enrollment rates for those of school age, and are calculated

by dividing reports of per capita primary enrollment rates by the fraction of the population under the age

of 14. Openness is the weighted average of own tariffs in the four or five countries to which the country

in question exported the largest absolute value of goods.

Behavior of the New Commodity and Terms of Trade Data

Looking at the behavior of 42 commodities on world markets, we see the most important feature of

commodity prices is not their long-term drift (as Prebisch and Singer would have had us believe), but

21Kose and Riezman (2001) suggest that the use of such a price index is superior to an import index when examining the effects of trade shocks, and that such an index exhibits similar levels of volatility compared with a pure terms of trade measure. Finally, focusing our analysis on the purchasing power of a country’s exports in terms of a common basket of manufactured products is actually advantageous in that we can isolate the between-country effect of relative commodity price movements on growth. 22 The authors reference McEvedy and Jones (1978) and other sources for the original data. 23 Marshall, M.G. and Jaggers, K. “POLITY IV Project: Political Regime Characteristics and Transitions, 1800-1999,” Integrated Network for Societal Conflict Research (INSCR) Program, Center for International Development and Conflict Management (CIDCM), University of Maryland, www.bsos.umd.edu/cidcm/inscr/polity

14

rather their volatility. In fact, volatility across products varies by more than a factor of ten. To illustrate,

Figure 4 depicts trend growth and volatility in the prices of 9 primary products over the three twenty-year

periods. Commodity prices experienced the highest average growth and lowest volatility from 1890 to

1909, and the highest volatility and slowest growth during the interwar years. Within each period, how-

ever, some primary products were very volatile (such as coffee or tobacco), while others were relatively

stable (like wheat or iron). Note that volatility between commodities often differed by a factor of two or

three, and in some cases by a factor of six or seven. What’s more, while some commodity prices rose

(such as tobacco or wool) others suffered sharp reversals (such as rubber, which was supplanted by

cheaper synthetics in the interwar period). Since most countries specialized in a handful of commodities,

such differences in price behavior translated into diverse country experience, or, as we noted above, what

Diaz-Alejandro called the “commodity lottery” (Diaz-Alejandro 1984).

Figure 5 displays the paths of the terms of trade by region, 1870 to 1939. At first glance it may seem

as though, looking backwards from their vantage point after WWII, Prebisch and Singer were right to ar-

gue a secular decline in the Periphery’s terms of trade. Thus, over our seven decades there seems to be an

upward drift for the Industrial Leaders and the European Industrial Latecomers, and a downward drift for

Latin America and Asia and the Middle East. Such a conclusion, however, is and was premature. First, as

Spraos (1980) noted, such a trend is very sensitive to the choice of time period, and when the series is

extended beyond 1950 the downward in the Periphery’s terms of trade disappears. Second, as we noted

above, downward drift is not itself evidence of a trend. Consistent with much of the literature24, time se-

ries tests (not shown) fail to find evidence of a secular trend. We cannot reject the presence of a unit root

in two-thirds of the countries and, in countries where trends appear, they are generally small and certainly

not universal to all primary product producers.

In short, our country terms of trade exhibit considerable year-to-year fluctuations, as well short term

trend movements or cycles, but little long-term trend—bolstering our prior conjecture that terms of trade

volatility mattered far more in the historical past than did long run trend. This conjecture accords well

with Mendoza’s model, as outlined in the previous section. Growth of the trend in the terms of trade and

decreases in the variance around trend (i.e., volatility) each raise economic growth.

The Decomposition of Terms of Trade Fluctuations

There are several options for decomposing terms of trade movements into trend and volatility. Men-

doza employs the terms of trade growth rate and the standard deviation of the growth rate. There are three

potential drawbacks to this approach. First, the growth rate of the terms of trade over a period of time

(such as a decade) will be overstated if there is a positive shock in the tenth year, and understated if there

15

is a negative shock. More volatile countries will thus be measured with less accuracy, leading to system-

atic measurement error. Second, a structural break or a discrete change in the rate of growth will register

as both a change in trend and a change in volatility, potentially confusing the effects. Third, persistent

shocks away from trend will result in a lower measure of volatility than a shock that returns to trend the

following year. Shocks that persist for more than a year before returning to trend will register as volatility,

since they remain deviations from measured trend. The standard deviation of the growth rate of the terms

of trade, on the other hand, will instead register only the initial shock and its eventual return to trend as

volatility. The more gradual and consistent the return to trend, the lower the volatility measure, and so

shocks that die out slowly will register as less volatile, even though the distortion may be greater.

We would prefer measures of trend and volatility that do not relate in a systematic way to one another

and that minimize the measurement error in the trend. A practical solution is to use a filter that produces a

smooth trend and stationary deviations. The Hodrick-Prescott (HP) filter is a common choice, used by

Basu and McLeod (1992) among others,25 and we employ it here.26 Over the 10-year intervals used in our

analysis, the HP data filter and Mendoza’s method generate very similar results. The correlations of

growth rates and volatility produced by the two methods are .85 and .86 respectively, and our findings are

robust to both. As we will see in Section 5, our results are also robust to changing the speed of adjustment

in the HP filter, or to using a backward-looking moving average filter instead.

4. Empirical Strategy

Our identifying assumption is that in the Periphery terms of trade fluctuations are exogenous, an as-

sumption that we will see stands up to rigorous inspection. Since exogeneity of the terms of trade may not

hold in our Core countries, and in particular because Core countries may have systematically lower vola-

tility because of a diversified export base, we estimate the effects of terms of trade fluctuations on the

Core and Periphery separately. Since we are principally interested in explaining the diversity in economic

performance among the commodity-specialized, we will typically focus on the Periphery results.

24 See Grilli and Yang (1998) or Bleaney and Greenaway (1993) for a review of the literature on declining terms of trade. 25 Similar methodologies have been used by other authors to measure the variability in the terms of trade. Lutz (1994) for exam-ple removes a linear trend (in the log of the terms of trade) and uses the variance of the residuals as a measure of volatility. This strategy has the problem of generating large residuals and therefore large implied volatility when there are changes in slope or structural breaks. Turnovsky and Chattopadhyay (2003) measure volatility as the variance of residuals from estimating a first-order autoregressive process for the terms of trade. The results from this procedure may be sensitive to the exact specification used and it seems unlikely to us that the terms of trade series for different countries in our sample are equally well described by a single autoregressive process. In any event, there is a high correlation (.87) between our measure of volatility and an alternative measure calculated using Turnovsky and Chattopadhyay's methodology. 26 We set the smoothing parameter in the HP filter at 300, which implies a relatively slow-changing trend. A more quickly chang-ing trend (such as that achieved with a smoothing parameter of 100 or lower) does not materially affect our results.

16

The basic empirical specification is a standard regression of the average annual growth rate, GR, on

measures of the growth and volatility in the terms of trade, TOTG and TOTV:

ittiitititit ZTOTVTOTGGR ενµββββ ++++++= 3210 (1)

for country i and time period t, a vector of controls, Z, country and time period dummies, µ and υ, and

residuals ε. Deaton and Laroque (1992) find that the persistence of commodity shocks is fairly high, and

therefore a decade seems like a reasonable unit of observation (although we will see the results hold if we

analyze the data in 5-year increments instead). Thus the dependent variable is the average annual growth

of GDP per capita over some decade. All of our specifications include country and decade fixed effects in

order to control for unobserved fundamentals that were also determining growth performance, fundamen-

tals that are not the focus on this paper. Finally, all standard errors are heteroscedasticity-robust.

Our terms of trade growth measure is the percent change in the trend in the terms of trade over the

decade, while volatility is measured by the standard deviation of departures from this trend. As discussed

in the previous section, we choose an HP filter to decompose the fluctuations into trend and volatility,

although we will see that our result is robust to alternate decompositions. This marks a departure from

much of the cross-country growth literature, which typically estimates the following equation:

ittiititit ZTOTGGR ενµβββ +++++= 210 (1’)

where TOTG is simply the log difference of ending and beginning terms of trade levels. The decomposi-

tion of the terms of trade shock is critical, however. When measuring the impact of terms of trade changes

over a period of time, such as a decade, failure to decompose the shock results in systematic measurement

error and biased coefficients. Consider two countries: in one there is a mild, steady improvement in the

terms of trade over the period, while in the other the terms of trade are highly volatile, but come out

higher at the end of the period than at the beginning. Both would have the same measured terms of trade

growth, but each would have had a sharply different growth experience that decade. In fact, in a regres-

sion of decadal growth on the change in the terms of trade, failure to account for volatility will systemati-

cally underestimate terms of trade shocks when measured over time, biasing the coefficient on terms of

trade growth upwards. Moreover, if commodity price shocks are also skewed, as suggested by Deaton

(1999), then the impact of the measurement error on the coefficients becomes difficult to predict. An an-

nual growth regression avoids both difficulties, but will not capture the medium and long term effects of

terms of trade growth and volatility on economic development.

Following Mendoza (1997), we could employ a very parsimonious empirical model without controls,

regressing average GDP per capita growth rates on average trend growth and volatility in the terms of

17

trade alone.27 The assumed exogeneity of the terms of trade implies that adding additional control vari-

ables should not change the estimated impact of the terms of trade. We find this to be the case, whereby

the addition of conventional determinants of growth as do not alter our results. Adding such variables can,

however, help reduce standard errors even when terms of trade fluctuations are exogenous, and so in gen-

eral we control for the initial values of income per capita, population growth, and schooling.28 However,

we might expect that the effect terms of trade shocks on output is conditional on some country character-

istic, X, and we can explore this hypothesis with the following specification:

ittiitititititit ZXTOTVXTOTGTOTGGR ενµββββββ ++++++++= 543210 )*( (2)

We investigate the interaction of terms of trade shocks with both export share (testing whether the effects

of trade fluctuations on output is an increasing function of dependence on trade) and our measures of in-

stitutional quality (investigating whether ‘good’ institutions moderate the negative impact of a shock on

the economy).

As Hausmann, Pritchett and Rodrik (2004) suggest, an alternative way to investigate the sources of

variation in GDP growth is to look at instances where trend growth experiences a clear shift. Using post-

WWII data they identify several episodes of rapid growth that mark a clear break from the past. An indi-

cator variable for a large positive terms of trade shock turns out to be a large and significant predictor of a

growth acceleration in their model. We perform a similar analysis, estimating probit models of both the

probability of a slowdown in growth, Pr(S), and the probability of an acceleration, Pr(A):

ittiittiittiitit ZTOTVTOTVTOTGTOTGS ενµββββββ ++++++++= −− 51,331,210)Pr( (3)

ittiittiittiitit ZTOTVTOTVTOTGTOTGA ενµββββββ ++++++++= −− 51,331,210)Pr( (4)

for country i and time period t, where we include current and lagged values of terms of trade trend growth

and volatility. Following the logic of our theory on fluctuations and growth, we expect that the probability

of a slowdown is decreasing in current terms of trade growth and increasing in current volatility. We

would also expect the signs on the lagged variable to be the opposite of the current ones. If terms of trade

growth is positively correlated with GDP growth, then a positive growth shock in the previous period will

have raised growth in that period, increasing the likelihood of a slowdown in this period. Likewise, we

expect the probability of an acceleration to be increasing in current terms of trade growth and decreasing

in lagged growth. In the Periphery, we expect the probability of an acceleration to be decreasing in terms

27 As noted above, Mendoza actually employed per capita consumption growth as a proxy for output growth. We do not have data on consumption, and so follow other authors in analyzing the effects on per capita GDP growth. 28 Our results will prove robust to the inclusion of openness and institutional variables as well. Conventional growth regressions would also include a measure of physical capital, such as investment share. We do not have such data for our period, but do not feel this to be a problem for two reasons: (i) as discussed , the terms of trade shocks are exogenous and should not be vulnerable

18

of trade volatility. For the Core, we might expect the same, but not necessarily so. Aghion et. al. (2004)

argue that countires with a high level of financial development are likely to benefit from volatility be-

cause long-term investments (such as R&D spending) are countercyclical. Thus the sign on current and

lagged volatility measures will provide an indication of whether or not this relationship holds true for the

period under investigation.

We examine slowdowns and accelerations by comparing average annual GDP per capita growth rates

over a five-year period to the growth rate in the preceding period. Figure 6 displays the frequency distri-

bution of changes in growth rates from one five-year period to the next in bins 0.5 percentage points wide,

for the period 1870-1939. Since we do not have any data on GDP before 1870, and since we do not want

to capture the effect of war years, we have eliminated three periods: 1870-74, 1915-19, and 1920-24. For

our sample of 35 countries, this gives us 385 observations. In Figure 6 we see an amazing degree of sym-

metry around a zero mean. More than half of the changes in growth rates are between -1.0 and 1.0 per-

centage points. In roughly fifteen percent of the cases, however, there was a reduction in the GDP per

capita growth rate of 2 percentage points or more. Likewise, in fifteen percent of the cases there was an

increase in the growth rate of 2 percentage points or more. Using these cutoffs for growth slowdowns and

accelerations, we obtain 61 slowdown and 59 acceleration episodes.29

Table 3 describes the incidence and magnitudes of our slowdown and acceleration episodes. Sixty

percent of our sample is in the Periphery, and they experience roughly sixty percent of the slowdowns and

accelerations. Thus there is little asymmetry between Core and Periphery. The Industrial Leaders and the

European Offshoots experience a relatively greater number of slowdowns than the average country, and

the European Periphery relatively few (although all regions are well-represented). Growth accelerations

are relatively evenly distributed across all regions, although again the European Periphery appears to have

relatively few. Looking across time periods, the highest number of slowdowns occurs in the period 1930-

34, and highest number of accelerations in 1935-39, as a consequence of the Great Depression. We are

not worried about bias from these observations, however, since period-specific shocks like these will be

captured in the time fixed effects, and the coefficients on our terms of trade fluctuations will be estimated

off of between-country variation in these periods. To the right in Table 3 are frequency diagrams of the

magnitudes of slowdowns and accelerations. Roughly a third of the slowdowns are reductions in growth

of -2 to -3 percentage points, with the second third between -3 and -5 percentage points. We observe a

similar pattern in the accelerations, though this upper tail declines somewhat more smoothly.

to omitted variable bias; and (ii) inclusion of foreign capital flows as a percent of GDP (as a proxy for investment share) does not alter our results, but unfortunately is only available for 1870-1909, so that we cannot use it in our 1870-1939 regressions. 29 . Hausmann and his coauthors, using postwar data, are able to identify the break points in growth rates more precisely, and note that doing so improves the power of the estimation. As discussed in the previous section, however, our growth data are estimated

19

Finally, as noted in Sections 1 and 2, terms of trade fluctuations are frequently thought to affect

growth through the investment channel. These hypotheses can be tested by estimating equations (1) and

(2) using investment levels as the dependent variable. Accordingly, we investigate the impact of terms of

trade fluctuations on the only reliable and available investment data for the period in question, the natural

log of capital flows from the UK, ln(I):

ittiitititit ZTOTVTOTGI ενµββββ ++++++= 5210)ln( (5)

ittiitititititit ZXTOTVXTOTGTOTGI ενµββββββ ++++++++= 543210 )*()ln( (6)

Alternative measures of UK capital flows can be employed (such as the percent of capital inflows in

GDP) but we did not find them to be significant.

Addressing Identification Concerns

There are two main sources of concern in our identification assumption: (i) endogeneity of the terms

of trade shock in cases where countries are not international price-takers, and (ii) the presence of an in-

visible third force (such as the quality of political institutions and governance) that result in both poor

growth and the choice of export concentration in a volatile commodity.

The first concern is that some countries in our sample will be large enough producers of a particular

commodity to affect the world price. There are two reasons why we do not believe this to be a problem

when looking at our Periphery. First, we would expect violations of this assumption (e.g. Chile and phos-

phates) to cause our results to understate the predicted positive impact of the terms of trade on growth.

For instance, if a negative shock to the supply of copper in Chile caused the world market price to rise

just as copper output (and hence GDP) in Chile fell, there would be a negative correlation between the

terms of trade trend and output growth, biasing the coefficient on trend growth downwards. Second, we

will see that our results are very robust to the exclusion of Periphery countries with a large share of world

exports in their principal export commodities. In Appendix Table A1 we tabulate estimates of country

export shares by major commodity in 1900. We initially singled out a country for exclusion if its export

share represented more than one third of the world market for that commodity. Since these are export

shares only for major exporters, since such measures ignore the importance of local production satisfying

local demand (especially in big economies like Russia, the US and China), and since cross-price elastic-

ities between some of these products are high (e.g. hemp versus jute, or cotton versus wool), we feel that

one third is a reasonable lower bound on world market share before a country begins to have a significant

impact on world price. According to this criterion, countries to be excluded from the panel include Brazil,

with less precision than postwar data, and so we feel that the comparison of 5-year means is a more appropriate method of analy-sis with the historical data.

20

China, India, the Philippines, Australia, and Chile (representing roughly a quarter of our observations).30

Gratifyingly, the growth regression results in the following section will prove to be very robust to their

exclusion—point estimates are virtually unchanged, and statistical significance actually increases to 99

percent. Our results are still robust if we reduce the lower bound further. Excluding all countries with

more than a quarter share of world exports in a commodity, we drop Japan from the panel and (although

we do not have full information on their share in world markets) drop Canada and Argentina to err on the

side of caution.31 Again, our point estimates for the full period are almost unaffected, although (with more

than forty percent of our observations dropped) some of the results are significant only at the 90% confi-

dence level.32

The second concern is that a third force, such as the quality of political institutions, is driving both the

choice of primary commodity and the pattern and rate of development. It might be suspected, for instance,

that countries with particularly poor governance or ‘extractive’-type institutions might systematically

choose more volatile commodities, and so volatility would be negatively correlated with growth because

it acts as a proxy for poor institutions, rather than for its own sake. There are at least three reasons not to

be concerned. One, to the extent that initial institutional quality influenced the choice of commodity, it

should be captured in a regression by country fixed effects. The reason is that, in virtually every Periphery

nation we consider, the choice of commodities was made before 1870 and persisted until the end of our

period. Whether we consider Canada or Colombia, each Periphery country’s development over the late

nineteenth and early twentieth centuries was not one of steady export diversification, but rather expansion

in the production of the same handful of two or three commodities. Any impact of institutional quality on

commodity choice is therefore largely time-invariant. The use of fixed effects will sweep out any such

time-invariant differences between countries, and estimate the impact of trend and volatility in the terms

of trade from the time-varying components alone.

Two, the historical and empirical evidence strongly support the “commodity lottery” view: the choice

of which commodity to produce and export was not a choice at all, but rather was an outcome determined

by geography, factor endowments, and international demand, not institutional quality. In simple regres-

sions of terms of trade trend and volatility on our institutional proxies (percent of population of European

30 We do not have data on world export shares of gold and silver, and in fact several countries (like Mexico) are major producers. These countries need not be dropped, however, because the prices of these precious metals were generally fixed so no feedback effects were possible. The silver price was pegged in world markets by the bi-metallic core Franco-countries during most of our period. As for gold, it too was pegged by the gold standard except for wars and the 1930s. 31 Canada may have had more than a quarter share of world exports in lumber before 1900, and Argentina may have had more than a quarter share in wool. 32If we reduce our lower bound to twenty percent of world exports (and hence drop Ceylon and Russia) our point estimates re-main roughly unchanged, but lose their statistical significance. The loss of statistical significance, however, seems to be a func-tion of the loss of observations (eleven of the twenty-one ‘Periphery’ countries have now been dropped) rather than because these observations are particularly influential, since it does not particularly matter what combination of countries are dropped if we only drop nine countries.

21

descent in 1800, and the Polity index) we do not find any statistically or economically significant rela-

tionship.33 This result is precisely what the historical evidence would lead us to believe: regardless of their

institutional makeup, Periphery countries generally exported the product in which they had the greatest

natural advantages. A look at the New World provides concrete examples. Where minerals were pro-

duced—whether phosphates in Chile or silver in Mexico—the choice of commodity was essentially a

function of what was under the soil and not prohibitively far from the coast. The range of agricultural

products available to export was likewise a function of the local production advantages and transport

costs. The mountainous Andean countries could never compete in wheat production with the Canadian

Prairie or the Argentine Pampas—the fertility and lay of the land provided clear cost advantages in pro-

duction of these goods, and hence Canada remained specialized in wheat (alongside lumber) while Argen-

tina focused on wheat and beef. Aside from any considerations of soil fertility, transport costs alone lim-

ited the profitability of many agricultural products in the Andes. In Colombia the cost of road and rail

construction over three mountain ranges alone meant that the cost of overland transport was almost pro-

hibitively high,34 and was justified only for high value-to-weight export products, such as gold, or rela-

tively lightweight and high value products such as tobacco and coffee.35

Where institutions and commodity choice are related, such as in the plantation-style production of

sugar or cotton, and the associated forced labor, both poor institutions and high price volatility seem to

have sprung from the same source—local factor endowments. Engerman and Sokoloff (1997) argue con-

vincingly that the commodity choice, the organization of production, and later institutions in the New

World were both functions of factor endowments, including climate and the availability of forced labor. It

was not the case that initially poor institutions themselves necessarily resulted in the production and ex-

port of volatile commodities. As we saw in Figure 3, while some of these plantation-style products were

volatile, many were not. This is not to say that there is no relationship between terms of trade volatility

and institutional quality. While institutions might not cause volatility, there may be important interac-

tions. Within the Periphery we might expect countries with better institutions to respond to shocks less

severely because of a greater ability to borrow and smooth investment and government finances, or be-

33 Results are not shown. A simple regression of average terms of trade volatility on % of the population that was European in 1800 (or EURO1800) and initial GDP per capita gives a coefficient of -0.0002 and a standard error of 0.02 on EURO1800. Simi-larly insignificant results arise from a regression of volatility on the Polity IV measure of democracy/autocracy. A variety of specifications, sample sizes, and controls result in similarly insignificant results. 34 35 to 60 cents per ton-mile, as compared to 2 to 4 cents in the US over the same period (Safford & Palacios, p.34). 35 The tobacco case is particularly instructive. In the 1870’s plantations in Java began producing cheap, high quality tobacco, permanently lowering the world price and shutting Colombia out of the market. Colombia scrounged for new products for several years, cycling through indigo and cinchona bark (for quinine) before eventually finding a profitable niche in the rising world demand for high quality coffee, which their mountains became famously known for cheaply producing. Such a trend was not uncommon in the Periphery; when export products and mixes did change, the pattern was typically one of substitution, not diver-sification, with any new export product replacing the old. Peru, for instance, focused on cotton and sugar export only after guano and silver stocks depleted. Brazil failed to find a real replacement for natural rubber until almost a generation after the invention of a synthetic substitute that destroyed the market.

22

cause of a reduced likelihood of civil conflict. In the following section, however, we will see that the in-

clusion of our institutional proxies in our regressions do not change our results. Within the Periphery we

see virtually no relationship between institutional quality and later volatility, reinforcing our identifying

assumption.

5. Discussion of the Empirical Results

Our results, displayed in Tables 4 through 7, support the predictions of our theory: terms of trade

growth and volatility are distinctly harmful to income growth in the Periphery, while helpful to income

growth in the Core. Specialization in commodities with highly volatile prices does not simply explain

poor economic performance in comparison to the developed Core, however; more importantly it explains

poor relative performance within the commodity-specialized Periphery.

The Terms of Trade and Growth of per Capita Income

The results from estimating equations (1) and (1’) by OLS, with and without standard controls, are

presented in Table 4. Results are displayed for the full seven decades (1870-1939), with the World War I

decade omitted throughout for the reasons discussed earlier. The results are reported separately for the

Core and Periphery largely because the export diversification of the Core confounds the terms of trade

volatility measure, but also in order to test for asymmetry. The top half of the table reports the point esti-

mates and standard errors for the terms of trade (henceforth called TOT) effects. The bottom half reports

the quantitative and economic importance of these TOT effects, including sample means and standard

deviations of the independent variables, and the impact of a one-standard-deviation increase in TOT

growth and volatility.

Comparing column 1 to 2 and 3 to 4 we see that the point estimates for both Core and Periphery are

virtually unaffected by the inclusion of controls but standard errors are improved, as predicted.36 Accord-

ingly, in the remainder of the paper we will focus on results that include the controls unless otherwise

specified. Now, looking at the second column of Table 4, we see that trend TOT growth is positively and

significantly related to income growth in the Core—a 1 percentage point increase in TOT trend growth is

associated with an increase in the average annual GDP per capita growth rate (henceforth ‘income

growth’) of 0.42 percentage points. Since the standard deviation of TOT trend growth is roughly one for

the Core, in this case the marginal effect happens to be about equal to the impact of a standard deviation

increase. There is no statistically significant relationship between TOT volatility and income growth in

36 The addition of other controls (including our measures of openness, institutional quality, and capital inflows as a percent of GDP) do not affect the results.

23

the Core, however. Looking at column 4, we see the opposite impact in the Periphery: there is no statisti-

cally significant relationship between TOT trend growth and income growth, but there is a large and

highly significant negative relationship between volatility and growth. Recall that volatility is measured

as the standard deviation of departures from the TOT trend, and we see that in the Periphery the mean

volatility is 9.5 (with a standard deviation of 5.5). The marginal effect of an increase in TOT volatility is

to decrease income growth by 0.1 percentage points. Our results also indicate that a country with a TOT

volatility a standard deviation greater than the mean will grow 0.59 percentage points more slowly per

annum than a country at the mean. Given that the average annual rate of growth in the Periphery was only

roughly a percent per year, these volatility effects are extremely economically significant.

Columns 5 to 8 estimate equation (1’), and do not decompose TOT movements into trend and volatil-

ity. We can see that decomposition of the fluctuations is of critical importance. The effects of TOT

growth is positive in both the Core and Periphery, but not statistically significant. Interacted with exports

as a share of GDP (not shown) TOT growth would appear large and significant in the Core, but not in the

Periphery. The failure to find robust coefficients on TOT growth has led many empirical studies to dis-

count its importance. As discussed in the previous section, however, the failure to find statistically sig-

nificant results is largely a consequence of measurement error, a problem that is in large part overcome

through the decomposition into trend and volatility. In the remainder of the paper we will look exclu-

sively at decomposed TOT fluctuations, and the results seen in this table will be robust to changes in time

period, horizon, method of decomposition, and definition of the Periphery.

Table 5 estimates equations (1) and (2), where in the latter TOT trend growth is interacted with export

share to see whether the terms of trade impact was contingent upon the level of openness and export de-

pendence. It seems reasonable that more export-oriented countries would respond more forcefully to ex-

ternal shocks. Export shares are taken from the first year of the decade to avoid problems of endogeneity.

OLS results are displayed for the full seven decades (1870-1939) as well as for two sub-periods: the first

global century from 1870 to 1909 and the interwar autarchic disaster from 1920 to 1939.37 The use of an

interaction term necessitates an addition to the bottom half of the table—the marginal impact of a change

in TOT growth and volatility. Marginal impact is simply the predicted change in output growth from an

increase of one in the independent variable. For terms of trade volatility, the marginal impact is just the

coefficient estimate. Marginal impact is defined the same way for trend growth when there is no interac-

tion term. When we introduce the interaction term, marginal impact is the sum of the coefficient estimates

on TOT Trend Growth by itself and the interaction term, the latter multiplied by the mean of export share.

37 The Periphery consists of 21 countries and we have data for every country and every decade, except for one country-decade observation, giving us a sample of 125 (=21*6-1). There are a few more missing observations from the interwar Core, leaving us with 79 observations instead of 84 (=14*6) that would be available in a complete dataset.

24

Columns 1 and 2 in Table 5 repeat the results from Table 4. The asymmetry between industrial-

exporting Core and primary-product-exporting Periphery strengthens when we introduce an interaction

term between TOT trend growth and export share in columns 3 and 4. The net effect of trend growth will

be sorted out below in the marginal impact calculations, but it is interesting to note the signs. In the Core,

income growth is increasing in TOT growth, but at a decreasing rate as the country become more export

oriented. The marginal effect at the mean implies that a one percentage point increase in TOT growth

raises income growth by 0.42—almost identical to the estimate reached without the interaction. In the

Periphery, income growth is now decreasing in TOT growth, although the positive sign on the interaction

term suggests that the negative effect was mitigated, perhaps entirely undone, by having a more open

economy exporting a larger share of output.38 At the mean, a one point increase in TOT trend growth

lowers annual income growth by -0.04 percentage points, a small but significant amount. An increase in

export share, holding constant concentration, may have acted as a foil to rent-seekers, or exerted a posi-

tive influence on output growth through various channels, such as efficiency gains or the development of

better institutions. The point estimates and standard errors on volatility in the Periphery barely change.

When we restrict our attention to the period before World War I in columns 5 through 8 of Table 5,

our main findings persist. Improvements in long-run trends in the terms of trade affected output growth

positively in the Core, but not in the Periphery, while volatility diminished growth in the Periphery, but

not in the Core. In fact, there is some evidence that volatility was good for growth in the Core—in column

7 the coefficient on TOT volatility is positive and significant at the 90 percent level. A one standard de-

viation increase in volatility in the Core is associated with a 0.25 percentage point increase in average

annual income growth! When we restrict our attention to the interwar period in columns (9) through (12),

however, our results lose their robustness. We are left with a much smaller sample (23 observations in the

Core and 41 in the Periphery) due to the observation of only two decades. As a result, standard errors are

large and statistical significance low, but we note that the point estimates are generally consistent with

those found for the pre-war period. It seems reasonable, therefore, to conclude that the same forces were

at work both before and after the war.

The economic effects of our estimates are big. A one-standard-deviation increase in trend growth is

approximately equivalent to replacing the experience of France or Austria (which saw their terms of trade

deteriorate modestly at -0.2 percent per annum) with that of Germany (which saw its terms of trade im-

prove dramatically over the period at 0.8 percent per annum). It is also reassuring that the empirical mag-

nitudes are very similar across all our specifications. The economic effect of TOT Volatility in the Pe-

riphery was even bigger. To illustrate the impact, consider an example from the primary product-

38 Note that we are holding fixed volatility in the terms of trade so we have in effect controlled for export concentration.

25

exporting Periphery. Per capita income in Canada grew faster than in Indonesia by about 1 percent per

annum. The difference in terms of trade volatility between the two countries was just under one half of

one standard deviation. Our estimates imply that if, through good fortune, Indonesia had experienced the

smaller terms of trade volatility of Canada, then Indonesia would have grown faster by about 0.3 percent-

age points, reducing the growth rate gap between the two by a third.

These magnitudes suggest that terms of trade shocks were an important force behind the big diver-

gence in income levels between Core and Periphery. The gap in growth rates in per capita income be-

tween Core and Periphery in our sample was 0.4 percentage points. If the Periphery had experienced the

same terms of trade volatility as the Core (leaving the observed growth rate of the terms of trade un-

changed), this would have added 0.2 percentage points to average GDP per capita growth rates in the Pe-

riphery. This alone erases half of the output per capita growth gap. If, in addition, the Core had experi-

enced no secular improvement in the terms of trade, instead of the observed 0.36 percent per annum

growth rate, this would have reduced output growth in the Core by 0.15 percentage points. Combined,

these two eminently plausible counterfactuals – the proposed changes are less than one half of one stan-

dard deviation away from the means, would have eliminated nearly the entire gap in growth rates between

Core and Periphery.

We have checked and confirmed the robustness of our results with respect to the following: alterna-

tive period lengths39, alternative methods of decomposing terms of trade series into trend and fluctua-

tions40, alternative Core-Periphery definitions41, and alternative specifications including additional inter-

39 Table A2 reproduces Table 5 using five-year periods instead of decades as our unit of analysis. The same patterns persist al-though the magnitudes are somewhat smaller, suggesting that the terms of trade have larger long-run effects that short-run ones. The standard errors have not increased, but because the coefficients are slightly smaller our statistical significance suffers some-what, so that in some cases the results are only significant at the 90 percent level. 40 Tables A3 and A4 reproduce Table 5 using alternative decompositions of terms of trade growth and volatility. In general the results are the same, if only because alternative measures are very highly correlated. Table A3 checks to make sure that our re-sults are robust to the use an HP filter with a faster-moving trend (a smoothing parameter of 100 instead of 300) and a moving average (MA) filter with a 7-year window. Volatility in both cases is measured as standard deviations of departures from trend. The MA filter may be preferable because many of the arguments about why volatility matters have to do with uncertainty. If markets and investment decisions are backward-looking, then measuring the trend with the HP filter (which looks forward as well as backward) might be misleading.40 We will see that our results are more or less unchanged, however. Our point estimates are nearly identical, and while statistical significance suffers very slightly using the MA filter, the alternative HP filter is highly significant. Table A4 uses Mendoza’s (1997) definitions of TOT growth and volatility—the growth rate of TOT and the standard deviation of the growth rate of TOT. The findings are remarkably similar and higher TOT growth and lower TOT volatility are both associated with improved economic growth. The estimated impact of TOT growth is much greater than that from our HP-filtered trend. As we noted above, Mendoza’s decomposition confounds trend with volatility somewhat, and it seems reasonable to presume that this confounding drives the strength of the TOT growth impact observed in Table A3. Consider that, in the event of a single-year positive terms of trade shock in the last year of the decade, the TOT growth rate by Mendoza would rise, but that using the HP-filter would not. To the extent that there is an immediate and direct increase in measured GDP in the year of the shock (because of the price effect), Mendoza’s measure will be positively correlated with growth. To the extent that such shocks have little or no effect on long-term GDP growth, we should observe little correlation between it at the the HP filter measure of TOT growth. 41 Since any choice of the Periphery is necessarily arbitrary, Tables A5 and A6 re-estimate Table 5 using alternative definitions of the Periphery. We see that whether the European Periphery or the European Offshoots are excluded from the Periphery sample, our central findings stand. Our point estimates actually strengthen our case when the European Offshoots (Canada, Australia and

26

actions and controls. We have also confirmed the robustness of our Periphery results to dropping any

countries that may have violated the price-taking assumption.42 Finally, we note that our findings are ex-

tremely robust to the inclusion of controls and interaction terms for institutional proxies. With both the

Core and Periphery, interacting TOT growth and volatility measures with our two measures of institu-

tions—percent of the population of European descent in 1800, and Polity—yield economically and statis-

tically insignificant interaction terms.43 All of these results are presented in Appendix tables.

Terms of Trade Shocks and Breakpoints in the Rate of Growth

Results from estimating equations (3) and (4) in a probit model are reported in Table 6. The coeffi-

cients reported are marginal effects of a one percentage point increase in TOT growth or a one point in-

crease in volatility. The bottom half of the Table indicates mean values and the impact of a standard de-

viation change in current TOT growth and volatility. In looking at the effect of terms of trade fluctuations

on the probability of a growth acceleration or deceleration, we examine both the marginal impact of TOT

shocks (not decomposed into trend and volatility) as well as the effect of the decomposed shocks. Our

coefficients generally have the expected signs.

Looking at the Periphery, we have 36 slowdowns and 38 accelerations. In columns 2 and 4, we see

that an increase in current TOT reduces the probability of a slowdown, probably because of wealth ef-

fects, and an increase in the TOT in the previous period increases the likelihood of a slowdown in the cur-

rent period, as expected. This relationship is strongest when the shock is decomposed. A one standard

deviation increase in the trend growth of the TOT (an increase of roughly 1.5 percentage points) reduces

the likelihood of a slowdown by 11%. Likewise, a fall in the terms of trade increases the likelihood of a

slowdown. We see no effect of volatility on the likelihood of a slowdown, however. Our point estimates

are essentially zero. This may be a consequence of not estimating the breakpoint precisely, a limitation

difficult to overcome with our data. Upon reflection, however, our result is not necessarily counter-

New Zealand) are removed. Our results lose statistical significance in the full period (although not in the period 1870-1909) once the European Periphery (Greece, Spain, Portugal, Russia, and Serbia/Yugoslavia) are included. With lower export dependence, more wealth, and a more diversified production base than much of the Periphery, these countries probably behaved more like the Core. 42 As discussed in the previous section, in Table A7 we drop countries that may not have been ‘price takers’ on the world market. Using one third of world export share (not world production share) we drop nearly 30 percent of our Periphery observations and obtain almost exactly the same results. Using one quarter of world exports as a lower bound and dropping nearly half of our ob-servations, the point estimate and statistical significance of TOT volatility on income growth in the Periphery suffer somewhat, but are still robust. Only when we drop our threshold further, losing more than half of our observations, do we lose statistical significance (although the signs and rough magnitudes of the coefficients do not change). 43 See Appendix Table A8 for full results. The Table also pools the Core and Periphery results, and here the TOT volatility meas-ure is strongly negative, but the interaction between volatility and % European is positive and significant, suggesting that coun-tries heavily populated with Europeans or those of European descent do not experience a negative impact of volatility. While this could be an indicator that institutions matter, the insignificant results within the Core and Periphery groups suggest otherwise. Rather, the % European is probably just acting as a proxy for ‘rich and developed with a diversified export base’.

27

intuitive. If certain products are consistently volatile across time, they will lower mean growth rates in all

periods, but not necessarily help us predict slowdowns. Only trend changes will do so reliably.

Looking at accelerations in the Periphery, we see the expected signs again. The probability of accel-

eration is increasing in the growth of TOT. Looking at the shock decomposed in column 8, a one standard

deviation increase in trend TOT growth (again, an increase of 1.5 percentage points) increases the likeli-

hood of an acceleration by 13%. Previous period TOT gains reduce the likelihood of an acceleration by a

roughly equal amount, as we would expect. These estimates are highly statistically significant. We now

also see a negative effect of volatility, although it is only significant at the 90 percent level. An increase in

volatility reduces the likelihood of an acceleration—increasing volatility by a standard deviation reduces

the likelihood of speedier growth by 8%. The smallness of this result is again a little surprising, but con-

sistent with the explanation provided for growth slowdowns—if persistently high-volatility countries tend

to have persistently lower mean growth throughout our period of study, then we would expect their likeli-

hood of an acceleration to decrease. The coefficient on current volatility is thus capturing a between-

country effect rather than a within-country effect.

Meanwhile, we see 25 slowdowns in the Core, most of which (we have seen) occurred in the Indus-

trial Leaders—France, Germany, the US and the UK. We do not obtain statistically significant results for

terms of trade shocks on slowdowns in column 3 in Table 6. Again, this may be because of inaccuracy in

calculating breakpoints, or it may be because slowdowns in highly developed, less export-dependent

countries are not nearly as contingent on the terms of trade. While this seems like a reasonable belief, we

do see a mild negative relationship between slowdowns and terms of trade shocks in the un-decomposed

column 1, suggesting that mis-measurement may be at least part of the story. Looking at accelerations in

the Core (of which there are 21) we get more robust results and some interesting asymmetry between

Core and Periphery. Volatility seems to be strongly positively related to growth in the Core—a one stan-

dard deviation increase in volatility is associated with an increase in the likelihood of an acceleration of

18%. This finding is supportive of the hypothesis extended by Aghion et. al. (2004), that richer countries

are better able to take advantage of high volatility because of the counter-cyclical nature of long-term

productive investment. A comprehensive test of this hypothesis, however, will have to await better data

on investment rates and financial development in the Periphery in this era.

Terms of Trade and Foreign Capital Inflows

What were the channels of terms of trade impact? Here we investigate one possibility, capital accu-

mulation financed by capital inflows, leaving further research on this question for future work. We have

already noted that Britain was the paramount source of development capital during these years. We find

that gross capital flows from Britain to Periphery countries were decreasing in TOT volatility. In Table 7,

28

the dependent variable is log of the average capital inflows a country received from Britain over 10-year

and 5-year periods between 1870 and 1909. We estimate equations (5) and (6) with and without controls

for both the 10-year and 5-year periods. Asymmetry is confirmed once more, although our results are nei-

ther as strong or as significant as they were for income growth.44

Looking at the odd-numbered columns, neither long-run changes nor volatility in the terms of trade

seems to have been important in attracting British capital to other countries in the Core. In the Periphery,

however, greater volatility reduced capital inflows from Britain. Looking at the even-numbered columns,

the point estimate on TOT volatility is roughly 0.45, corresponding to a 25 percent decrease in average

capital flows from a one standard deviation increase in volatility. The point estimate is consistent whether

we include or exclude controls and interaction terms, and whether we look at decades or half-decades.

Statistical significance is not as consistent, however. The results are significant at the 95 percent level

only in the 5-year case when the standard controls are omitted (columns 10 and 12), is only 90 percent

significant in the 10-year case without controls (columns 1 and 4) and the five-year case with controls

(columns 14 and 16). In the 10-year case with controls, the specification most consistent with our previ-

ous tables, we lose statistical significance altogether. Ironically, our point estimates and statistical signifi-

cance increase when we change our measure of volatility to one calculated using an HP filter with a

faster-moving trend, or a moving average filter, as seen in an Appendix.45 Our results also increase in size

and significance when we include the European Periphery in the Periphery (although not when we ex-

clude the European Offshoots).46 The varying robustness of our capital flows results is probably a conse-

quence of the data. We only have data to World War I, and not beyond, so there are fewer observations.

We only have capital inflows from the UK, and not from the US or Germany, as to our knowledge such

data do not exist. Thus the inflow data is incomplete. Finally, capital inflows were often small and vari-

able, and likely driven by many competing forces. Accordingly, we are satisfied with the strength of our

results.

In the absence of data on capital stocks, we can only conjecture whether the terms of trade impact on

domestic capital accumulation was big or small, but assuming a capital-output ratio of roughly 3:1 sug-

gests that the impact via capital inflows was fairly small. A reasonable conjecture would be that domestic

savings were even more powerfully influenced by terms of trade shocks than were foreign capital flows.

44 We also looked at the effects of TOT growth and volatility on capital inflows as a share of GDP, but did not find significant results. The growth rate of capital flows was not attempted because capital from the UK is a flow not a stock, and growth rates of volatile flow variables are sometimes not well defined. There are many zero years, where a country recorded no UK capital in-flows, and a growth rate from zero to a positive number is not well-defined. This and similar problems are common in the inflw dataset. 45 See Appendix Table A9. 46 See Appendix Table A10.

29

6. Conclusions

Our analysis suggests that commodities and the terms of trade need to be more central to our thinking

about long term development. Like measures of income growth (and unlike most other determinants of

growth) variation in the terms of trade is largely within-country, not between-country, and should enhance

our understanding of levels of growth as well as starts and stops. Commodity choice, dependence and the

associated price trends and volatility seem to be especially powerful determinants of growth in the Pe-

riphery. In focusing upon the forces that led to industrialization and diversification in Western Europe and

the US, scholars of economic history, growth and development may have forgotten that the rest of the

world—rich and poor—took a different path. Some of the primary-product producing nations did quite

well by this strategy—witness Canada, Australia, or Norway. Some, like India or Colombia, did quite

poorly up to 1950. Most others were somewhere in the middle. The commodity-specialized were not uni-

formly slow-growing and poor. Neither were their commodities uniformly volatile and decreasing in

value. In reconstructing nearly a century of terms of trade experience from 1870 to 1939 and assessing its

impact on the economic performance, we see that some commodities proved more volatile in price than

others; and those countries with more volatile commodities have grown more slowly than other commod-

ity-specialized nations. Terms of trade shocks, meanwhile, are strong predictors of both slowdowns and

accelerations in income growth. Countries with just one standard deviation difference in volatility, more-

over, grew on average more than half a percentage point per annum differently. The same difference in

volatility seems to be associated with 25 percent lower capital inflows from foreign sources. Such large

magnitudes go a long way towards explaining the divergent growth experiences in the Periphery. While

our capital flows results are not nearly as large or as robust as our growth results, they still point to a large

and important channel of impact. Other channels are likely to be of importance, however, such as the ef-

fect of terms of trade shocks on the incidence of civil conflict (Rodrik, 1999). These alternative channels

await further investigation.

What is also notable about our results is the striking asymmetry between Core and Periphery. Where

terms of trade volatility was present, it created a significant drag on output growth in the Periphery. This

was not true of the Core—where it experienced the same high price volatility, it did not experience the

same drag on growth. In addition, while the Core benefited greatly from a small but positive long-run

terms of trade trend, positive trends—when they did appear—did not translate in to more growth in the

Periphery, but rather less. Moreover, when we investigate one channel of terms of trade impact—the flow

of investment funds from Britain—we find evidence that capital inflows were negatively influenced by

terms of trade volatility in the Periphery, but not in the Core. This asymmetry is consistent with three

growth narratives. One, positive price shocks should have reinforced comparative advantage, and so they

30

could have induced more industrialization in the Core and less in the Periphery. Two, it may be that rich

countries with more sophisticated institutions and markets were better able to insure against price volatil-

ity than poor countries, so that terms of trade instability is likely to have had a far bigger negative impact

in the Periphery than the Core. We await better data on institutional quality before we can test this hy-

pothesis, however. Three, as recently argued by Aghion et. al. (2004), in countries with well developed

financial markets, volatility tended to enhance growth in the postwar period while reducing growth else-

where, largely because where financial markets are poor (i.e., the Periphery) investment is shorter-term

and more procyclical. We saw some evidence to support this hypothesis in our results on growth accelera-

tions: volatility was positively related to growth in the Core while negatively related in the Periphery.

Each of these questions will have to await future data collection and research, but we hope that it will

emphasize the inherent volatility of growth, the unique impact of commodity, and the individual country

impact of commodity price changes volatility—the direction taken here.

31

References

Acemoglu, D., Johnson, S. and Robinson, J.A., “The Colonial Origins of Comparative Development: An Empirical Investigation,” American Economic Review 91 (December 2001), pp. 1369-1401.

Aghion, P., Angeletos, G.M., Banerjee, A. and Manova, K., “Volatility and Growth: Financial Development and the Cyclical Composition of Investment,” working paper provided by Aghion, September 2004.

Basu, P. and McLeod, D., “Terms of Trade Fluctuations and Economic Growth in Developing Economies,” Journal of Development Economics 37 (1992): 89-110.

Bidarkota, P.V. and Crucini, M.J., “Commodity Prices and the Terms of Trade,” mimeo, 1998.

Blattman, C., Hwang, J.J. and Williamson, J.G., “The Terms of Trade and Economic Growth in the Periphery 1870-1938,” NBER Working Paper 9940, National Bureau of Economic Research, Cambridge, Mass. (August 2003).

Bleaney, M. and Greenway, D., “Long-Run Trends in the Relative Price of Primary Commodities and in the Terms of Trade of Developing Countries,” Oxford Economic Papers 45 (1993): 349-63.

Bleaney, M. and Greenway, D.,“The Impact of Terms of Trade and Real Exchange Rate Volatility on Investment and Growth in Sub-Saharan Africa,” Journal of Development Economics 65 (2001): 491-500.

Bulmer-Thomas, Victor, “The Economic History of Latin America Since Independence” Cambridge: University of Cambridge Press, 1994.

Catão, L. and Kapur, S., “Missing Link: Volatility and the Debt Intolerance Paradox,” unpublished (January 2004).

Clemens, M.A. and Williamson, J.G., “Why Did the Tariff-Growth Correlation Reverse after 1950?” NBER Work-ing Paper 9181, National Bureau of Economic Research, Cambridge, Mass., September 2002.

Clemens, M.A. and Williamson, J.G., “Wealth Bias in the First Global Capital Market Boom 1870-1913,” Economic Journal 114 (April 2004): 311-44.

Coatsworth, John H. and Jeffrey G. Williamson, “The Roots of Latin American Protectionism: Looking Before the Great Depression” National Bureau of Economic Research: Working Paper 8999, June 2002.

Deaton, A., “Commodity Prices and Growth in Africa,” Journal of Economic Perspectives 13 (Summer 1999): 23-40.

Deaton, A. and Laroque, G., “On the Behaviour of Commodity Prices,” The Review of Economic Studies 59(1), January 1992: pp.1-23.

Deaton, A. and Miller, R.I., “International Commodity Prices, Macroeconomic Performance and Politics in Sub-Saharan Africa,” Journal of African Economics 5 (1996): 99-191, Supplement.

Deninger, K. and Squire, L., “A New Dataset Measuring Income Inequality,” World Bank Economic Review 10(September) (1996): 565-91.

Dercon, S., Insurance Against Poverty, Oxford: Oxford University Press, 2004.

Diaz-Alejandro, C., “Latin America in the 1930s,” in R. Thorpe (ed.), Latin America in the 1930s, London: Macmil-lan, 1984: pp. 17-49.

Easterly, W., M. Kremer, L. Pritchett, and Summers, L.H., “Good Policy or Good Luck? Country Growth Perform-ance and Temporary Shocks,” Journal of Monetary Economics 32 (1993): 459-83.

Eichengreen, B., Globalizing Capital: A History of the International Monetary System, Princeton: Princeton Univer-sity Press, 1996.

Engerman, S.L., and Sokoloff, K.L., “Factor Endowments, Institutions, and Differential Paths of Growth Among New World Economies: A View from Economic Historians of the United States.” In Stephen Haber, ed., How Latin America Fell Behind. Stanford, Calif.: Stanford University Press, 1997.

32

Fafchamps, M., Rural Poverty, Risk and Development, Chattenham: Edward Elgar Publishers, 2004.

Frankenberg, E., Beegle, K., Sikoki, B., and Thomas, D., “Health, Family Planning and Well-being in Indonesia during an Economic Crisis: Early results from the Indonesian Family Life Survey,” Santa Monica, CA: RAND Labor and Population Program Working Paper Series 99-06, 1999.

Grilli, E. R. and Yang, M.C., “Primary Commodity Prices, Manufactured Goods Prices, and the Terms of Trade of Developing Countries: What the Long Run Shows,” World Bank Economic Review 2 (1998): 1-48.

Hadass, Y. and Williamson, J.G., "Terms of Trade Shocks and Economic Performance 1870-1940: Prebisch and Singer Revisited," Economic Development and Cultural Change 51 (April 2003): 629-56.

Hausmann, R., Pritchett, L. and Rodrik, D., “Growth Accelerations,” draft working paper, April 2004.

Isham, J., Kaufman, D., and Pritchett, L., “Civil Liberties, Democracy, and the Performance of Government Pro-jects,” World Bank Economic Review 11(May) (1997): 219-42.

Jacoby, H.G. and Skoufias, E., “Risk, Financial Markets, and Human Capital in a Developing Country,” Review of Economic Studies 64 (July 1997): 311-35.

Jensen, R., “Agricultural Volatility and Investments in Children,” American Economic Review 90 (May 2000): 399-404.

Kose, M. A. and Reizman, R., “Trade Shocks and Macroeconomic Fluctuations in Africa,” Journal of Development Economics 65(2) (1998): 55-80.

Krueger, A. O., “The Political Economy of the Rent-Seeking Society,” American Economic Review 64 (June 1974): 291-323.

Lutz, M., “The Effects of Volatility in the Terms of Trade on Output Growth: New Evidence,” World Development 22(12) (1994): 1959-75.

Maddison, A., Monitoring the World Economy 1820-1992, Paris: OECD Development Centre Studies, 1995.

Mendoza, E., “Terms of Trade Uncertainty and Economic Growth,” Journal of Development Economics 54 (1997): 323-56.

Miguel, E., Satyanath, S. and Sergenti, E., “Economic Shocks and Civil Conflict: An Instrumental Variables Ap-proach,” forthcoming in Journal of Political Economy (2004).

Prebisch, R., The Economic Development of Latin American and Its Principal Problems, Lake Success, NY: United Nations, Department of Economic Affairs, 1950. Reprinted in Economic Bulletin for Latin America 7 (1962): 1-22.

Pritchett, L., “Understanding Patterns of Economic Growth: Searching for Hills among Plateaus, Mountains and Plains,” World Bank Economic Review 14(2) (2000): 221-50.

Ramey, G. and Ramey, V.A., “Cross-country Evidence on the Link Between Volatility and Growth,” NBER Work-ing Paper 4959, National Bureau of Economic Research, Cambridge, Mass. (1995).

Rodrik, D., “Where Did all the Growth Go? External Shocks, Social Conflict and Growth Collapses,” Journal of Economic Growth 4 (December 1999): 385-412.

Rosenweig, M.R. and Wolpin, K.I., “Credit Market Constraints, Consumption Smoothing, and the Accumulation of Durable Production Assets in Low Income Countries: Investments in Bullocks in India,” Journal of Political Economy 101(2) (1993): 223-44.

Sachs, J. and Warner, A., "Economic Reform and the Process of Economic Integration," Brookings Paper on Eco-nomic Activity 1 (1997): 1-53

Sachs, J. and Warner, A., “The Curse of Natural Resources,” European Economic Review 45 (May 2001): 827-38.

Safford, F. and Palacios, M., “Colombia: Fragmented Land, Divided Society” Oxford: Oxford University Press, 2002.

Sala-i-Matin, X., “I Just Ran Two Million Regressions,” American Economic Review 87(2) (1997): 178-83.

33

Shah Mohammed, S. and Williamson, J.G., “Freight Rates and Productivity Gains in British Tramp Shipping 1869-1950,” Explorations in Economic History 41 (April 2004): 172-203.

Singer, H.W., "The Distribution of Gains between Investing and Borrowing Countries," American Economic Review 40 (1997): 473-85.

Spraos, J., "The Statistical Debate on the Net Barter Terms of Trade between Primary Commodities and Manufac-tures," Economic Journal 90 (1980): 107-28.

Stone, I., The Global Export of Capital from Great Britain, 1865-1914, New York: St. Martin’s Press, 1999.

Thomas, D., Beegle, K., Frankenberg, E., Sikoki, B., Strauss, J., and Teruel, G., "Education in a Crisis," Journal of Development Economics 74 (June 2004): 53-85.

Turnovsky, S.J. and Chattopadhyay, P., “Volatility and Growth in Developing Economies: Some Numerical Results and Empirical Evidence,” Journal of International Economics 59 (2003): 267-95.

34

Data Appendix

Most data used in this paper come from a novel database constructed by collaborations between Jeffrey Wil-liamson and Luis Bϑrtola, Chris Blattman, Michael Clemens and Yael Hadass, often from primary sources at the Harvard library. This appendix does not attempt a thorough description of sources and methods used in collecting this evidence, since much of the data herein have appeared in previous publications and more detailed descriptions of sources and methods are available within them. Most data for the period 1870-1913 first appeared in Clemens and Williamson, “Wealth Bias in the First Global Capital Market Boom 1870-1913,” Economic Journal 114 (April 2004): 311-44.. Most data for the period 1914-1940 first appeared in Clemens and Williamson, “Why Did the Tar-iff-Growth Correlation Reverse After 1950?” Journal of Economic Growth (2004 forthcoming). Data for the 8 Latin America countries, however, was updated and expanded by John Coatsworth and Williamson in “The Roots of Latin American Protectionism: Looking Before the Great Depression,” A. Estevadeordal, D. Rodrik, A. Taylor and A. Velasco (eds.), FTAA and Beyond: Prospects for Integration in the Americas (Cambridge, Mass.: Harvard Univer-sity Press, 2004), which has a complete listing of changes and additions. Note that several sources are used fre-quently in all this work. They are: Arthur S. Banks, Cross-National Time Series,1815-1973, [Computer File] ICPSR ed. (Ann Arbor, Michigan: Inter-University Consortium for Political and Social Research, 1976), hereafter Banks (1976); Brian R. Mitchell, International Historical Statistics, Europe, 1750-1988 (New York: Stockton Press, 1992); Brian R. Mitchell, International Historical Statistics: The Americas, 1750-1988 (New York: Stockton Press, 1993); Brian R. Mitchell, International Historical Statistics: Africa, Asia & Oceania, 1750-1993 (New York: Stockton Press, 1998). hereafter Mitchell.

GDP and GDP per capita

The units on this variable are 1990 US dollars per inhabitant of any age. For most countries, gross domestic product per capita (in 1990 US dollars) is taken from Angus Maddison, Monitoring the World Economy 1820-1992 (Paris: Development Center of the Organization for Economic Cooperation and Development, 1995).

GDP per capita estimates for 1870-1950 for Australia, Brazil, Canada, China, Denmark, France, Germany, In-dia, Indonesia, Italy, Japan, Mexico, New Zealand, Norway, Portugal, Russia, Spain, Sweden, Siam, and the United States come from Angus Maddison, 1995, Monitoring the World Economy, 1820-1992, OECD, Paris. For countries not reported in Maddison (1995), GDP per capita is calculated by dividing a country’s income (in 1990 US dollars) by population in every year. Sources of the population data have been described elsewhere in this appendix, and the sources of the income estimates follow.

Data for Argentina after 1890 come from Maddison op. cit. Before this date, GDP per capita is assumed to grow at the same year-on-year rate as the estimates of Argentine real wages found in Jeffrey G. Williamson, 1995, “The Evolution of Global Labor Markets since 1830: Background Evidence and Hypotheses,” Explorations in Eco-nomic History, 32:141-196.

GDP per capita estimates for Austria-Hungary before 1914 come from David F. Good, 1994, “The Economic Lag of Central and Eastern Europe: Income Estimators for the Habsburg Successor States, 1870-1910,” Journal of Economic History, 54(4)(December): 69-891. These are converted from 1980 to 1990 dollars using a GDP deflator obtained from the Bureau of Economic Analysis of the United States Department of Commerce (online at http://www.bea.doc.gov/bea/dn/gdplev.htm).

Data for Burma after 1900 come from Maddison op. cit. Before this date it is assumed that Burmese growth mirrored that of India.

Ceylon presented the most difficult data challenge in this category, as we are not aware of any published figures for GDP in Ceylon during this period. Campbell (Burnham O. Campbell, 1993 “Development Trends: A Compara-tive Analysis of the Asian Experience,” in Naohiro Ogawa et al., eds., Human Resources in Development along the Asia-Pacific Rim, Oxford University Press, New York) has estimated that in 1914, GDP per capita in Ceylon was 1.95 times that of India. The same ratio had declined to 1.52 by 1948 according to United Nations Statistical Year-book 1949-50 (New York: 1950), pp. 21-2 and 406. In the intervening years, 1914-1948, it is assumed that the ratio declined annually at a constant rate. Before 1914, it is assumed that real GDP per capita grew at the same rate as did the ratio of the real value of British colonial revenue from Ceylon to the population of the Island. A full series of annual nominal colonial revenues and population figures come from the 1905 and 1914 editions of the annual Cey-lon Blue Book, a statistical publication of the colonial administration in Colombo. Some of these figures were re-corded in rupees, and are converted to pounds sterling using conversion rates from Bryan Taylor II, 2000, Encyclo-

35

pedia of Global Financial Markets, Global Financial Data, Los Angeles, California (online at http://www.globalfindata.com). The resulting figures are converted to real pounds sterling using the deflator in McCusker op. cit.

Data for Chile after 1900 come from Maddison op. cit. Before this date it is assumed that Chile grew at the same year-on-year rate as did our estimates of Argentine GDP per capita.

Data for Colombia after 1900 come from Maddison op. cit. Before this date, it is assumed that that GDP per capita grew at an unweighted average of the growth rates for Mexico and Brazil between 1850 and 1900 given in Coatsworth op. cit.

Estimates for Cuba for 1850 and 1913 are based on estimates of Cuban GDP per capita relative to that of Mex-ico and Brazil presented in John H. Coatsworth, 1998, “Economic and Institutional Trajectories in Nineteenth-Century Latin America,” in John H. Coatsworth and Alan M. Taylor, eds., Latin America and the World Economy Since 1800, Harvard University Press, Cambridge, Mass. An unweighted average of the figures implied by Coats-worth’s proportion of our estimates for Mexico and Brazil is calculated for both years, and the intervening years estimated by geometric interpolation. For the years 1914-1950, Cuba’s Net National Product in current year pesos comes from Mitchell (1993). These NNP values are converted to 1990 US dollars with the help of the peso-dollar exchange rate given in Taylor (2000) and the American historical consumer price index given in John McCusker, How Much Is That in Real Money (Worcester, Mass.: American Antiquarian Society, 1992), pp. 330-2.

Estimates for Egypt after 1900 come from Maddison. Before this date it is assumed that GDP per capita grew at the same year-on-year rate as did estimates of Egyptian real wages from Jeffrey Williamson, 2000, “Real wages and relative factor prices around the Mediterranean, 1500-1940,” in Şevket Pamuk and Jeffrey G. Williamson, eds. The Mediterranean Response to Globalization Before 1950, Routledge, New York. For the years 1900-1950, a trend for Egyptian GDP per capita is calculated with the help of benchmark values given in Maddison (1995). Annual GDP per capita estimates are then calculated under the assumption that Egypt deviated from the Maddison-estimated Egyptian benchmark trend in the same way (percentage-wise) as Turkey did from her GDP per capita trend (after the civil war).

Data for Greece are estimated by projecting Maddison’s (op. cit.) 1913 figure backwards, assuming the growth rate found in James Foreman-Peck and Pedro Lains, 2000, “European Economic Development: The Core and the Southern Periphery, 1870-1910,” in Şevket Pamuk and Jeffrey G. Williamson, eds., The Mediterranean Response to Globalization Before 1950, Routledge, New York.

Data for Peru after 1900 come from Maddison op. cit. Before this date it is assumed that Peru grew at the same year-on-year rate as did our estimates of Argentine GDP per capita.

Data for the Philippines after 1900 come from Maddison op. cit. Before this date it is assumed that Philippine GDP per capita grew at the same year-on-year rate as our estimates for Siam.

Estimates for Serbia after 1890 come from Foreman-Peck and Lains, op. cit. Before 1890 GDP per capita is as-sumed to grow at the same year-on-year rate as it did between 1890 and 1913.

Estimates for Turkey after 1913 come from Maddison. Before this date it is assumed that GDP per capita grew at the same year-on-year rate as did estimates of Turkish real wages from Jeffrey Williamson, 2000, “Real wages and relative factor prices around the Mediterranean, 1500-1940,” in Şevket Pamuk and Jeffrey G. Williamson, eds. The Mediterranean Response to Globalization Before 1950, Routledge, New York.

Data for Uruguay after 1882 comes from Maddison op. cit. Before this date it is assumed that Uruguay grew at the same year-on-year rate as did our estimates of Argentine GDP per capita. GDP for Uruguay is taken from Mitchell (1993) for the period 1935-1940. Annual GDP per capita estimates 1914-1934 are calculated by assuming that Uruguay deviated from her GDP per capita trend (between the benchmark years of 1914, found in Clemens and Williamson (2000), and 1935, found in Mitchell) in the same way that Argentina did.

Data for a small remaining number of missing years are geometrically interpolated.

Terms of Trade Index (or the Net Barter Terms of Trade - NBTT)

Existing data series were employed for the terms of trade for the US, the UK, France, Germany, Sweden, Italy and Austria. Terms of trade for Austria-Hungary after 1882 are found in Scott M. Eddie, 1977, “The Terms and Patterns of Hungarian Foreign Trade, 1882-1913,” Journal of Economic History, 37(2)(June):329-358. An index for 1876-1882 is constructed from indices of the physical quanta and values of exports and imports given in Statistik des Auswärtigen Handels des Österreichisch-Ungarischen Zollgebiets im Jahre 1891, Statistischen Departement im K. K. Handelsministerium, Vienna, 1893, pp. LXVIII-LXIX. For the period 1865-1875 the same source reports only export and import values, not physical quanta. Since the quanta display extremely stable trends during 1876-

36

1892 (unlike the values, which are subject to the vagaries of prices), the quanta for 1865-1875 are extrapolated as-suming the same, stable growth rate observed on 1876-1892. Combining these estimates with the trade value figures given for 1865-1875 yield a ToT estimate for this period. Terms of Trade for France 1870-1896 come from Charles Kindleberger, 1956, The Terms of Trade: A European Case Study, MIT Technology Press, Cambridge, Table 2-1, pp. 12-13. This is linked to a series from 1896-1913 found in P. Villa, 1993, Une Analyse Macroéconomique de la France au XXeme Siècle, CNRS Editions, Monographies d’Économetrie, Paris, pp. 445-6. German ToT for the entire period come from Walther G. Hoffmann, 1965, Wachstum der Deutschen Wirtschaft seit der Mitte des 19 Jahrhunderts, Springer-Verlag, Berlin, Table 134, col. 1, p. 548. Italy’s terms of trade with Great Britain are taken as a proxy for overall Italian terms of trade. The former are found in I. A. Glazier, V. N. Bandera, and R. B. Berner, 1975, “Terms of Trade between Italy and the United Kingdom 1815-1913,” Journal of European Economic History, 4(1)(Spring): 5-48. Sweden’s terms of trade are taken from Simon Kuznets, 1996 [originally published 1967], “Quantitative Aspects of the Economic Growth of Nations: X. Level and Structure of Foreign Trade: Long-Term Trends,” reprinted in C. Knick Harley, ed. The integration of the world economy, 1850-1914, Volume 1, Elgar Ref-erence Collection: Growth of the World Economy Series, Vol. 3. Cheltenham, U.K, Table 12, p. 150. United States ToT are from Jeffrey G. Williamson, 1964, American Growth and the Balance of Payments 1820-1913, University of North Carolina Press, Chapel Hill, North Carolina, Table B4, p. 262.

For the remaining countries, a NBTT series was calculated from original sources. Note that the NBTT is simply the ratio of export prices to import prices, each weighted appropriately. Mathematically,

∑∑

⋅

⋅= M

iMit

Xij

Xijt

jt wp

wpNBTT

for product i, country j, and period t. Note that in this formulation the numerator, the export price index, is country-specific while the denominator, the import price index, is not. This is a simplification employed in this pa-per due to (i) the limited quality and quantity of data on imports and import prices to countries in the periphery, and (ii) the similarity observed, in what records are available, between the composition of imports to developing coun-tries. While detailed data on exports weights and prices are available for virtually all of the countries and all of the years in our sample, import data are much more limited. These limitations and their consequences are discussed below.

Export Weights. For the purposes of this study, export weights have been calculated by individual country us-ing the current value of major commodity exports and fixed weights. The use of a fixed set of weights is essential for disentangling price from quantity movements. Of course, any such approach is fundamentally flawed, not least because over a long period of time the mix of major commodity exports can shift significantly. A compromise posi-tion was taken by changing the export weights at approximately 20-year sub-periods. These sub-periods are 1870-1890, 1890-1913, 1913-1929, and 1930-1950, and within these the weights are calculated using sample year data. Export values for major commodities for Canada, Brazil, Chile, Colombia, Cuba, Mexico, Paraguay, and Peru are taken from Mitchell, International Historical Statistics The Americas 1750-1993, p.506 ff. Table E3. The same data for Australia, Burma, Ceylon, Egypt, India, Indonesia, Japan, Philippines, Siam, Turkey and New Zealand come from Mitchell, International Historical Statistics Africa, Asia and Oceania 1750-1993, p.637 ff. Table E3. Main commodity exports for Denmark, Greece, Norway, Portugal, Spain and Sweden were calculated from Statistical Abstract for Principal and Other Foreign Countries, London 1876-1912 and Die Wirtschaft des Auslandes, Sta-tistisches Reichsamt Berlin 1928.

Export Prices. Export prices are quoted in foreign markets (wherever possible, in the UK), rather than domes-tic ones. Wholesale Prices for Wheat, Maize, Rice, Beef, Butter, Sugar, Coffee, Tea, Iron, Copper, Tin, Lead, Coal, Cotton, Flax, Hemp, Jute, Wool, Silk, Hides, Nitrate, Palm Oil, Olive Oil, Linseed, Petroleum, Indigo and Timber are taken from Sauerbeck, Prices of Commodities and Precious Metals, Journal of the Statistical Society of London, vol. 49/3 September 1886 Appendix C, for the years 1860-85. Sauerbeck, Prices of Commodities During the Last Seven Years, Journal of the Royal Statistical Society, vol.56/2 June 1893 p.241 ff., for the years 1885-1892, Sauer-beck, Prices of Commodities in 1908, Journal of the Royal Statistical Society 72/1 Mar 1909 for the years 1893-1908. Sauerbeck, Wholesale Prices of Commodities in 1929, Journal of the Royal Statistical Society, vol. 93/2 p. 282 ff, 1930 for the years 1908-1929. Sauerbeck, Wholesale Prices of Commodities in 1916, Journal of the Royal Statis-tical Society, vol. 80/2 p. 289 ff. for the years 1908- 1916. Sauerbeck, Wholesale Prices in 1950, Journal of the Royal Statistical Society, vol. 114/3 1951, p. 417 ff. for the years 1916-50. Prices for Cocoa, Crude Oil, Rubber, To-bacco and Zinc are taken from Dodd, Historical Statistics of the United States from 1790-1970, University of Ala-bama Press. Prices for Fruits and Nuts 1880-1914 are taken from Critz, Olmsted and Rhode, International competi-tion and the development of the dried fruit industry 1880-1930 table 8.2, in Pamuk and Williamson, 2000. Prices for

37

Opium 1860-1906 Ahmad Seyf, Commercialization of Agriculture: Production and Trade of Opium in Persia, 1850-1906 Table 4, International Journal of Middle East Studies, 1984. Prices for Beans & Bean Products were calculated from Liang-Lin, China’s Foreign Trade Statistics 1864-1949, Harvard University Press 1974, p.80 ff.

Import Weights. A single set of import weights is employed for all countries in the sample. Import data, unlike that of exports, is almost uniformly poor, in particular in countries outside the European Core. Traditionally, studies of country terms of trade have compensated for this lack of data through the use of British export data as a proxy for the imports of less developed nations. This approach is undesirable given that the composition of British exports can hardly be considered representative of the imports of developing countries as a whole, and because the use of current-year weights means that movements reflect changes in composition, not just prices. As an alternative, however, we employ a fixed index of non-primary goods from US statistics. This import index, like the British one, is country invariant. In the end, the differences are not material; the two series are almost identical (probably due to the heavy content of metals and textiles in both indices). This US manufactured export statistic is a weighted sum of the prices of textiles (55%), metals (15%), machinery (15%), building materials (7.5%), and chemicals and pharma-ceuticals (7.5%). Obviously a fixed weighting for all developing nations is unrepresentative of their particular im-port mix (but while not representative of the specific import mix of the country, such a metric may be relevant for measuring the changing value of the country's exports versus a fixed package of manufactured products available for import. In this sense our terms of trade represent the purchasing power of local commodities in terms of rich-country goods.) Moreover, a review of each nation's external commerce documents turns up remarkably similar import compositions. For the years 1870-1900, import composition for Australia, Canada, Ceylon, India and New Zealand was examined from Statistical abstract for the several colonies and other possessions of the United King-dom no.1-40, 1863-1902. Data for Burma comes from Terulo Saito and Lee Kin Kiong Statistics on the Burmese Economy, Singapore 1999 pp 177, table VII-4. Import weights for China, Denmark, Egypt, Greece, Japan, Norway, Portugal, and Russia were calculated from Statistical Abstract for Principal and Other Foreign Countries, London 1876-1912 no. 13. Data for the Philippines are taken from Quarterly Summary of Commerce of the Phillipine Is-lands, Washington 1908 p.27 for the year 1893. Import composition for Serbia before 1914 is recorded in Sundhaus-sen, Historische Statistik Serbiens 1834-1914, Munich 1989 pp. 352-355. Main imports for Turkey are calculated from Mulhall, Dictionary of Statistics, London 1892 p. 145, for the year 1888. For the years 1900-1940, import weights for Australia, Canada, Ceylon, India, New Zealand are calculated for several reference years from Statistical abstract for the several British self-governing dominions, colonies, possessions, and protectorates no.41-53, 1903-1915, Statistical abstract for the several British oversea dominions and protectorates no.54-59, 1917-1927, Statisti-cal abstract for the British Empire no.60-68, 1929-1938, Statistical abstract for the British Commonwealth no.69-70, 1945-1947 and Statistical abstract for the Commonwealth (trade statistics) no.71-72, 1948-1951. Composition of main imports for reference years after 1900 for Argentina, Chile, Greece, Indonesia, Japan, Mexico, Norway, Portu-gal, Russia, Serbia, Spain, Siam, Uruguay comes from Die Wirtschaft des Auslandes 1900-1927, Berlin 1928. Data for Burma comes from Terulo Saito and Lee Kin Kiong Statistics on the Burmese Economy, Singapore 1999 pp 177, table VII-4. Data for the Philippines is taken from Foreign Commerce of the Phillipine Islands,Washington 1912-1913 for the reference years 1907, 1908 and 1910. Composition of main imports for Turkey was calculated from Annuaire Statistique , Republique Turque, vol.1 pp. 103, 106 vol. 3 pp. 313, 314 for the years 1923, 1926 and 1929.

Import Prices: US price series for textiles, metals, machinery, building materials, and chemicals and pharma-ceuticals come from Dodd, Historical Statistics of the United States from 1790-1970, University of Alabama Press.

An Additional Note on Import and Export Price Data. UK and US prices are employed in the theory that the prices in these large, integrated and (in the UK, at least) unprotected markets would supply us with a relatively reliable "world" price index for each commodity group. A chief disadvantage of using such world price indices, however, is that home market prices in each country may diverge from the world ones in the short and even long term. This may be because of differences in product features and quality, because of variations in the composition of the products within a category, or because of less-than-perfect market integration combined with local market conditions and shocks. Kindleberger (1958) illustrates the wide divergence in the prices of bulky products such as coal and lumber between two markets as closely integrated as the US and UK. Another disadvantage of not using the home market price is the distortion created by changes in transport costs. One would prefer a terms of trade measure that is independent of transport costs. In a moment we will discuss the adjustments made to our terms of trade figures to account for transport cost changes. Such adjustments as we can make, however, cannot truly repre-sent actual freight-adjusted prices. Overall, though, we feel the advantages of employing world price indices out-weigh these disadvantages. First and foremost, home market prices are not typically on hand for the periods and countries in question. Rather, only the somewhat less desirable unit prices (calculated as the value of imports di-vided by the volume) are available. Second and more important, we believe UK and US market prices to be more reliable, accurate and comparable given the quality of reporting (at the time) and the quality of scholarship on these

38

prices since then. Third, to the extent that commodity markets are well integrated worldwide, the UK and US mar-ket prices should approximate the world price. This is especially true because we are interested in price changes, not levels. To the extent that UK and US prices move in similar directions and similar magnitudes to prices in the rest of the world, these "world" price indices will more or less represent price changes relative to an index year in other nations. We believe this to be a reasonable and necessary assumption. Fourth, these foreign market price indices would have been available to (and probably used) by industrialists and policymakers throughout the period in ques-tion. Accordingly, for questions of policy response (and perhaps price setting) foreign market indices may be a more appropriate data source than home market ones. Fifth, the use of a world price index harmonizes and simplifies con-struction of the indices, enabling us to examine a wider sample of countries at the cost, perhaps, of precision. Fifth, by measuring both the export and import price indices in a common currency, we eliminate any inflationary bias from the figures.

39

ARG

AUS

BRA

BUR

CANCEY

CHL

CHN

COL

CUBEGP

INDIDN

JPN

MEX

NZD

PER

PHL

THA TKY

URU

AHG

DEN

FRA

GER

GRC

IT A

NOR

POR

RUS

SRB

SPA

SWEUK

USA

-1.5

-1-.5

0.5

11.

5M

ean

% T

OT

Gro

wth

0 1000 2000 3000 4000 5000 6000 70001939 GDP per capita

Periphery Core

1939 GDP per capita and Mean Terms of Trade Growth 1870-1939Figure 2:

ARG

AUS

BRA

BURCAN

CEY

CHL

CHN

COL

CUB

EGP

IND

IDN

JPNMEXNZDPER

PHL

THA TKY

URU

AHG

DENFRA

GER

GRC

IT A

NOR

POR

RUS

SRB

SPA

SWEUK

USA

46

810

1214

1618

2022

TOT

Vol

atili

ty

0 1000 2000 3000 4000 5000 6000 70001939 GDP per capita

Periphery Core

1939 GDP per capita and Terms of Trade Volatility 1870-1939Figure 1:

40

C offee

C otton

Iron

R ic e

R ubber

Sugar

Tobac c o

Wh eatWo ol

C offee

C otton

Iron

R ic e

R ubber

SugarTobac c o

Wh eat

Wo ol

C offee

C otton

IronR ic e

R ubberSugar

Tobac c o

Wh eat

Wo ol

05

1015

2025

3035

40P

rice

Vol

atili

ty

-6 -4 -2 0 2 4 6Avg Annual Grow th in Price (%)

1870-1889 1890-1909 1920-1939

* All prices are def lated by a US index of manuf actured goods

Commodity Price Volatility and Growth for Select Commodities *Figure 4:

ARGAUS

BRA

BUR

CAN

CEY

CHL

CHN

COL

CUB EGP

IND

IDN

JPNMEX

NZDPER

PHL

THA

TKYURU

AHG

DENFRA

GER

GRC

IT A

NOR

POR

RUSSRBSPA

SWE

USA

05

1015

2025

TOT

Vol

atili

ty

2 3 4 5 6 7 8 9 10Average Capital Flow s 1870-1913 (in logs)

Periphery Core

1870-1913Figure 3: Capital Flows and Terms of Trade Volatility

41

Figure 5: Long Run Trends in the Terms of Trade

6070

8090

100

110

120

130

140

150

160

Ave

rage

Ter

ms

of T

rade

(19

00=1

00)

1 870 1880 18 90 1900 1910 19 20 1930 1940Year

Industrial LeadersFigure 5a: Trend in the Terms of Trade

6070

8090

100

110

120

130

140

150

160

Ave

rage

Ter

ms

of T

rade

(19

00=1

00)

1 870 1880 18 90 1900 1910 19 20 1930 1940Year

European Industrial LatecomersFigure 5b: Trend in the Terms of Trade

6070

8090

100

110

120

130

140

150

160

Ave

rage

Ter

ms

of T

rade

(19

00=1

00)

1 870 1880 18 90 1900 1910 19 20 1930 1940Year

European PeripheryFigure 5c: Trend in the Terms of Trade

6070

8090

100

110

120

130

140

150

160

Ave

rage

Ter

ms

of T

rade

(19

00=1

00)

1 870 1880 18 90 1900 1910 19 20 1930 1940Year

Latin AmericaFigure 5d: Trend in the Terms of Trade

6070

8090

100

110

120

130

140

150

160

Ave

rage

Ter

ms

of T

rade

(19

00=1

00)

1 870 1880 18 90 1900 1910 19 20 1930 1940Year

Asia & the Middle EastFigure 5e: Trend in the Terms of Trade

6070

8090

100

110

120

130

140

150

160

Ave

rage

Ter

ms

of T

rade

(19

00=1

00)

1 870 1880 18 90 1900 1910 19 20 1930 1940Year

European OffshootsFigure 5f: Trend in the Terms of Trade

42

020

4060

80N

umbe

r of I

ncid

ents

-14 -12 -10 -8 -6 -4 -2 0 2 4 6 8 10 12 14Magnitude of Change in Grow th Rate

Frequency Distribution of Change in Growth Rates, 1870-1939Figure 6:

43

Table 1: Differences in Persistence Between Economic Growth and Typical Explanatory Variables in Growth Regressions

Number of countries

Number of observations

Ratio of within-country variance to total variance

(i) 1960-1999 (quinquennially):Growth of GDP per capita 125 967 0.73Population growth 125 967 0.29Investment rates 125 967 0.28Level of primary education 110 774 0.09Terms of trade growth 106 772 0.90

(ii) 1870-1909 (quinquennially):Growth of GDP per capita 35 280 0.85Population growth 35 280 0.37UK capital inflows 35 272 0.31Level of primary education 35 280 0.05Terms of trade growth 35 280 0.96Primary products in exports 35 280 0.02

(iii) 1870-1949 (decadally):Growth of GDP per capita 35 280 0.84Population growth 35 279 0.60UK capital inflows 35 165 0.40Level of primary education 35 279 0.18Terms of trade growth 35 264 0.90Primary products in exports 35 280 0.10

Variable

Notes: Data for 1960-1999 come from the Penn World Tables v6.1 (for GDP, population, and the investment rate), the Barro-Lee Data Set of International Measures of Schooling Years and Schooling Quality (for education), and the World Bank's 2004 World Development Indicators (for the terms of trade). Countries with fewer than 20 annual observations (or 4 quinquennial ones) were dropped from the analysis. Data for 1870-1939 are discussed in the data appendix.

44

Table 2: Profile of the Core and Periphery

GDP per capita

Primary Products as

a % of Exports

Top 2 Exports as a % of Top 5

ExportsExports as a % of GDP

GDP per capita

Primary Products as

a % of Exports

Top 2 Exports as a % of Top 5

ExportsExports as a % of GDP

GDP per capita

Primary Products as

% of Exports

Top 2 Exports as a % of Top 5

ExportsExports as a % of GDP

PERIPHERYEuropean "Frontier" Offshoots

Australia 4,442 97% 98% 15% 4,394 97% 84% 20% 5,268 96% 76% 15%Canada 1,822 95% 96% 12% 2,801 91% 68% 15% 3,993 74% 51% 19%New Zealand 3,668 99% 100% 16% 4,295 96% 89% 23% 5,232 99% 69% 25%

3,311 97% 98% 15% 3,830 95% 81% 19% 4,831 89% 65% 20%

Latin AmericaArgentina 1,676 100% 87% 15% 2,823 99% 58% 19% 3,912 99% 47% 14%Brazil 755 100% 86% 17% 749 100% 91% 21% 1,087 100% 92% 9%Chile 1,185 99% 100% 22% 1,923 99% 100% 19% 2,800 100% 100% 12%Colombia 1,113 99% 100% 4% 1,034 99% 100% 6% 1,486 99% 100% 7%Cuba 1,647 80% a 49% 1,825 83% 100% 41% 1,440 96% 100% 41%Mexico 835 100% 99% 4% 1,183 100% 86% 5% 1,463 99% 62% 7%Peru 497 99% 74% 24% 802 99% 55% 10% 1,451 100% 67% 12%Uruguay 1,676 100% 74% 22% 2,823 100% 72% 19% 3,912 100% 85% 18%

1,173 97% 89% 20% 1,645 97% 83% 17% 2,194 99% 82% 15%

Asia & the Middle EastBurma (Myanmar) 628 91% 100% 14% 622 86% 93% 12% 751 98% 83% 45%Ceylon (Sri Lanka) 730 98% 100% 11% 932 98% 100% 13% 1,114 98% 100% 15%China 565 98% 73% 1% 648 93% 66% 1% 769 85% 43% 1%Egypt 369 93% 100% 29% 487 96% 100% 25% 650 97% 100% 18%India 660 98% 55% 4% 723 96% 51% 5% 1,083 97% 59% 6%Indonesia 581 91% . 3% 629 86% 60% 4% 651 75% 54% 4%Japan 800 71% 100% 1% 1,108 70% 99% 4% 1,948 44% 89% 7%Philippines 955 96% 81% 5% 1,086 95% 90% 4% 1,527 91% 81% 5%Siam (Thailand) 751 99% 100% 2% 811 99% 100% 6% 815 98% 87% 7%Turkey 831 99% 50% 6% 941 97% 59% 9% 936 92% 72% 5%

834 92% 83% 4% 987 90% 87% 6% 1,307 81% 82% 6%

COREIndustrial Leaders

France 2,119 43% b 13% 2,739 40% b 15% 4,100 35% b 10%Germany 2,184 38% b 9% 3,007 33% b 9% 4,001 24% b 11%United Kingdom 3,598 12% b 14% 4,419 17% b 14% 5,112 22% b 12%United States 2,952 86% b 6% 4,126 80% b 6% 5,969 57% b 5%

2,713 45% . 10% 3,573 43% . 11% 4,795 35% . 9%

European Industrial LatecomersAustria/Austria-Hungary 1,108 35% b 9% 1,545 41% b 9% 3,211 35% b 10%Denmark 2,105 96% a 14% 2,913 96% 88% 22% 4,791 90% 91% 20%Italy 1,516 87% b 6% 1,784 74% b 7% 2,888 48% b 5%Norway 1,446 90% 100% 13% 1,763 74% 94% 18% 3,126 64% 100% 20%Sweden 1,875 85% b 9% 2,483 75% b 12% 3,722 58% b 15%

1,610 79% . 10% 2,098 72% . 13% 3,548 59% . 14%European Periphery

Greece 1,343 94% a 7% 1,487 90% 73% 7% 2,321 96% 86% 4%Portugal 1,151 96% 75% 6% 1,348 92% 73% 7% 1,562 75% 85% 7%Russia/USSR 976 97% 79% 4% 1,178 96% 73% 4% 1,593 85% 48% 1%Serbia/Yugoslavia 852 96% 73% 6% 944 96% 77% 7% 1,225 92% 64% 5%Spain 1,588 73% 64% 5% 2,009 75% 46% 7% 2,557 80% a 4%

1,182 91% 73% 5% 1,393 90% 69% 6% 1,852 86% 71% 4%

Sources: See data appendix.Note a: No data available for this period.Note b: Existing terms of trade series, for which an export breakdown was not available, were used for the Industrial Leaders and several of the European Latecomers.

1870-1889 1890-1909 1920-1939

45

Tabl

e 3:

GD

P G

row

th S

low

dow

ns a

nd A

ccel

erat

ions

Indu

stria

l Le

ader

s

Euro

pean

In

dust

rial

Late

com

ers

Euro

pean

Pe

riphe

ryEu

rope

an

Offs

hoot

sA

sia

&

Mid

dle

East

Latin

A

mer

ica

# of

cou

ntrie

s4

55

310

835

1875

-187

92

20

20

06

1880

-188

40

00

02

02

1885

-188

92

00

10

03

1890

-189

41

00

13

05

1895

-189

90

00

01

01

1900

-190

43

00

02

05

1905

-190

90

00

01

01

1910

-191

42

10

20

38

1925

-192

91

10

10

47

1930

-193

43

42

25

521

1935

-193

90

01

01

02

Tota

l14

83

915

1261

Indu

stria

l Le

ader

s

Euro

pean

In

dust

rial

Late

com

ers

Euro

pean

Pe

riphe

ryEu

rope

an

Offs

hoot

sA

sia

&

Mid

dle

East

Latin

A

mer

ica

# of

cou

ntrie

s4

55

310

835

1875

-187

90

00

03

03

1880

-188

42

10

21

06

1885

-188

90

00

03

03

1890

-189

40

01

00

01

1895

-189

92

00

21

05

1900

-190

40

00

02

13

1905

-190

91

10

01

03

1910

-191

40

10

02

14

1925

-192

91

20

12

28

1930

-193

40

01

00

12

1935

-193

93

32

35

521

Tota

l9

84

820

1059

A. I

ncid

ence

& M

agni

tude

of G

row

th A

ccel

erat

ions

Not

e: A

gro

wth

slo

wdo

wn

(acc

eler

atio

n) is

def

ined

as

a de

crea

se (i

ncre

ase)

in th

e gr

owth

rate

of p

er c

apita

of 2

per

cent

age

poin

ts o

r mor

e in

co

mpa

rison

to th

e pr

evio

us h

alf-d

ecad

e. T

he p

erio

d 18

70-7

4 is

exc

lude

d du

e to

lack

of d

ata

on th

e th

e pr

evio

us fi

ve y

ears

, and

the

perio

ds

1915

-24

and

1940

-49

are

excl

uded

due

to th

e di

srup

tion

of th

e w

ar y

ears

.

Cor

ePe

riphe

ry

Full

Sam

ple

Perip

hery

Full

Sam

ple

Cor

e

A. I

ncid

ence

& M

agni

tude

of G

row

th S

low

dow

ns

0.1.2.3.4Fraction of All Slowdowns

-15

-14

-13

-12

-11

-10

-9-8

-7-6

-5-4

-3-2

Mag

nitu

de o

f Sl

owdo

wn

(% p

oint

s)

0.1.2.3Fraction of All Accelerations

23

45

67

89

1011

1213

1415

Mag

nitu

de o

f A

ccel

erat

ion

(% p

oint

s)

46

Tabl

e 4:

GD

P G

row

th a

nd th

e Te

rms

of T

rade

, 187

0-19

39D

epen

dent

Var

iabl

e: D

ecad

al a

vera

ge G

DP

per

cap

ita g

row

th

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

TOT

Gro

wth

(Not

e 1)

0.41

10.

422

-0.0

41-0

.044

0.19

40.

157

0.08

00.

084

[0.1

81]*

*[0

.158

]**

[0.1

35]

[0.1

30]

[0.1

27]

[0.1

19]

[0.1

01]

[0.0

95]

TOT

Vola

tility

0.01

50.

005

-0.1

05-0

.106

[0.0

36]

[0.0

42]

[0.0

42]*

*[0

.037

]***

Obs

erva

tions

7979

125

125

7979

125

125

R-s

quar

ed0.

450.

500.

340.

430.

390.

430.

300.

39D

ecad

e D

umm

ies

YY

YY

YY

YY

Cou

ntry

Dum

mie

sY

YY

YY

YY

YC

ontro

ls (N

ote

2)N

YN

YN

YN

Y

Mea

n Va

lues

[Std

Dev

]:G

DP

Gro

wth

1.38

1.38

0.99

0.99

1.38

1.38

0.99

0.99

[1.2

8][1

.28]

[1.7

9][1

.79]

[1.2

8][1

.28]

[1.7

9][1

.79]

TOT

Gro

wth

0.36

0.36

-0.4

9-0

.49

0.30

0.30

-0.6

7-0

.67

[1.0

9][1

.09]

[1.5

3][1

.53]

[1.5

6][1

.56]

[2.3

9][2

.39]

TOT

Vola

tility

7.67

7.67

9.46

9.46

[5.5

6][5

.56]

[5.5

2][5

.52]

Impa

ct o

f a 1

Std

Dev

Incr

ease

TOT

Gro

wth

0.45

0.46

-0.0

6-0

.07

0.30

0.25

0.19

0.20

TOT

Vola

tility

0.08

0.03

-0.5

8-0

.59

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Not

e 2:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

Not

e 1:

Not

dec

ompo

sed,

TO

T G

row

th is

the

deca

dal g

row

th ra

te in

term

s of

trad

e. D

ecom

pose

d, it

is d

ecad

al g

row

th o

f a H

odric

k-P

resc

ott f

ilter

ed tr

end.

No

deco

mpo

sitio

n of

TO

TTO

T de

com

pose

dC

ore

Perip

hC

ore

Perip

h

47

Tabl

e 5:

GD

P G

row

th a

nd T

rend

and

Vol

atilt

y in

the

Term

s of

Tra

de in

Alte

rnat

e Ti

me

Perio

dsD

epen

dent

Var

iabl

e: D

ecad

al a

vera

ge G

DP

per c

apita

gro

wth

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

TOT

Gro

wth

0.42

2-0

.044

0.91

1-0

.294

0.50

90.

130

0.98

4-0

.079

0.80

7-0

.420

1.22

6-0

.031

[0.1

58]*

*[0

.130

][0

.298

]***

[0.1

40]*

*[0

.175

]***

[0.1

47]

[0.2

74]*

**[0

.134

][0

.416

][0

.169

]**

[0.7

26]

[0.2

51]

TOT

Vola

tility

0.00

5-0

.106

0.00

1-0

.101

0.06

2-0

.093

0.06

2-0

.105

0.00

2-0

.026

0.06

4-0

.079

[0.0

42]

[0.0

37]**

*[0

.039

][0

.035

]***

[0.0

35]*

[0.0

44]*

*[0

.036

]*[0

.044

]**

[0.1

87]

[0.0

60]

[0.2

64]

[0.0

63]

(TO

T Tr

end

Gro

wth

) x (E

xpor

ts/G

DP)

-5.0

721.

836

-6.8

681.

959

-4.9

01-2

.191

[2.3

05]*

*[0

.830

]**

[2.9

41]*

*[1

.016

]*[3

.446

][0

.985

]**

Exp

orts

/GD

P0.

352

4.69

87.

298

13.0

74-7

.372

-4.2

97[3

.895

][3

.964

][5

.283

][7

.092

]*[9

.639

][4

.332

]

Obs

erva

tions

7912

579

125

5684

5684

2341

2341

R-s

quar

ed0.

500.

430.

570.

460.

660.

460.

710.

550.

760.

910.

810.

93D

ecad

e D

umm

ies

YY

YY

YY

YY

YY

YY

Cou

ntry

Dum

mie

sY

YY

YY

YY

YY

YY

YC

ontro

ls (N

ote

1)Y

YY

YY

YY

YY

YY

Y

Mea

n Va

lues

[Std

Dev

]:G

DP

Gro

wth

1.38

0.99

1.38

0.99

1.13

1.13

1.13

1.13

1.87

0.72

1.87

0.72

[1.2

8][1

.79]

[1.2

8][1

.79]

[0.7

6][1

.70]

[0.7

6][1

.70]

[1.8

5][1

.96]

[1.8

5][1

.96]

TOT

Gro

wth

0.36

-0.4

90.

36-0

.49

0.31

-0.0

10.

31-0

.01

0.50

-1.4

80.

50-1

.48

[1.0

9][1

.53]

[1.0

9][1

.53]

[0.8

6][1

.36]

[0.8

6][1

.36]

[1.5

4][1

.38]

[1.5

4][1

.38]

TOT

Vola

tility

7.67

9.46

7.67

9.46

5.98

8.59

5.98

8.59

11.7

911

.25

11.7

911

.25

[5.5

6][5

.52]

[5.5

6][5

.52]

[3.9

1][5

.63]

[3.9

1][5

.63]

[6.8

0][4

.90]

[6.8

0][4

.90]

(TO

T G

row

th) x

(Exp

orts

/GD

P)

0.03

-0.0

90.

030.

000.

03-0

.27

[0.1

3][0

.35]

[0.0

7][0

.22]

[0.2

0][0

.48]

Exp

orts

/GD

P0.

100.

140.

090.

130.

110.

16[0

.06]

[0.1

2][0

.05]

[0.1

1][0

.09]

[0.1

4]

Mar

gina

l Im

pact

TOT

Gro

wth

0.42

-0.0

40.

42-0

.04

0.51

0.13

0.37

0.18

0.81

-0.4

20.

67-0

.39

TOT

Vola

tility

0.01

-0.1

10.

00-0

.10

0.06

-0.0

90.

06-0

.11

0.00

-0.0

30.

06-0

.08

Impa

ct o

f a 1

Std

Dev

Incr

ease

TOT

Gro

wth

0.46

-0.0

70.

46-0

.05

0.44

0.18

0.32

0.24

1.24

-0.5

81.

04-0

.53

TOT

Vola

tility

0.03

-0.5

90.

01-0

.56

0.24

-0.5

20.

24-0

.59

0.01

-0.1

30.

44-0

.39

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; ** s

igni

fican

t at 5

%; *

** s

igni

fican

t at 1

%

Not

e 1:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

1870

-193

9 (1

910-

1919

exc

lude

d)18

70-1

909

1920

-193

9

48

Tabl

e 6:

Pre

dict

ing

Gro

wth

Slo

wdo

wns

and

Acc

eler

atio

ns, 1

870-

1939

Dep

ende

nt V

aria

ble:

Indi

cato

r for

a s

low

dow

n or

an

acce

lera

tion

(Not

e 1)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

TOT

Gro

wth

(Not

e 2)

-0.0

64-0

.026

-0.0

11-0

.075

-0.0

340.

017

0.08

90.

088

[0.0

22]*

**[0

.011

]**

[0.0

77]

[0.0

32]*

*[0

.022

][0

.008

]**

[0.0

76]

[0.0

34]*

*

Lagg

ed T

OT

Gro

wth

0.03

0.00

6-0

.09

0.05

1-0

.024

-0.0

22-0

.121

-0.1

04[0

.023

][0

.010

][0

.087

][0

.028

]*[0

.019

][0

.009

]**

[0.0

67]*

[0.0

32]*

**

TOT

Vola

tility

-0.0

170.

000

0.03

3-0

.015

[0.0

23]

[0.0

08]

[0.0

17]*

*[0

.009

]*

Lagg

ed T

OT

Vola

tility

-0.0

01-0

.004

-0.0

480.

018

[0.0

32]

[0.0

08]

[0.0

19]*

*[0

.008

]**

Obs

erva

tions

175

6917

469

169

7416

874

Slow

dow

n/Ac

celra

tion

Epis

odes

2536

2536

2138

2138

Hal

f-Dec

ade

Dum

mie

sY

YY

YY

YY

YC

ount

ry D

umm

ies

YY

YY

YY

YY

Con

trols

(Not

e 3)

YY

YY

YY

YY

Mea

n Va

lues

[Std

Dev

]:G

DP

Gro

wth

1.38

0.99

1.38

0.99

1.38

0.99

1.38

0.99

[1.2

8][1

.79]

[1.2

8][1

.79]

[1.2

8][1

.79]

[1.2

8][1

.79]

TOT

Gro

wth

0.30

-0.6

70.

36-0

.49

0.30

-0.6

70.

36-0

.49

[1.5

6][2

.39]

[1.0

9][1

.53]

[1.5

6][2

.39]

[1.0

9][1

.53]

TOT

Vola

tility

7.67

9.46

7.67

9.46

[5.5

6][5

.52]

[5.5

6][5

.52]

Impa

ct o

f a 1

Std

Dev

Incr

ease

TOT

Gro

wth

-0.1

0-0

.06

-0.0

1-0

.11

-0.0

50.

040.

100.

13TO

T Vo

latil

ity-0

.09

0.00

0.18

-0.0

8

Coe

ffici

ents

repo

rted

are

mar

gina

l pro

babi

lity

effe

cts.

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Not

e 3:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

Pr(S

low

dow

n)Pr

(Acc

eler

atio

n)

Not

e 1:

A g

row

th s

low

dow

n (a

ccel

erat

ion)

is d

efin

ed a

s a

decr

ease

(inc

reas

e) in

the

grow

th ra

te o

f per

cap

ita o

f 2 p

erce

ntag

e po

ints

or m

ore

in c

ompa

rison

to

the

prev

ious

hal

f-dec

ade.

The

per

iod

1870

-74

is e

xclu

ded

due

to la

ck o

f dat

a on

the

the

prev

ious

five

yea

rs, a

nd th

e pe

riods

191

5-24

and

194

0-49

are

Not

e 2:

Not

dec

ompo

sed,

TO

T G

row

th is

the

deca

dal g

row

th ra

te in

term

s of

trad

e. D

ecom

pose

d, it

is d

ecad

al g

row

th o

f a H

odric

k-P

resc

ott f

ilter

ed tr

end.

Not

Dec

ompo

sed

Dec

ompo

sed

Not

Dec

ompo

sed

Dec

ompo

sed

49

Tabl

e 7:

Brit

ish

Cap

ital F

low

s an

d th

e Te

rms

of T

rade

, 187

0-19

09D

epen

dent

Var

iabl

e: N

atur

al lo

g of

ave

rage

dec

adal

cap

ital i

nflo

ws

from

Brit

ain

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

(14)

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

TOT

Gro

wth

0.03

60.

031.

212

0.00

4-0

.091

0.08

31.

020.

053

0.02

40.

029

0.72

60.

068

-0.0

340.

07[0

.372

][0

.092

][0

.872

][0

.152

][0

.445

][0

.111

][0

.913

][0

.161

][0

.238

][0

.079

][0

.481

][0

.137

][0

.275

][0

.084

]

TOT

Vola

tility

-0.0

39-0

.044

-0.0

31-0

.045

-0.0

48-0

.035

-0.0

54-0

.037

0.02

5-0

.047

0.03

6-0

.046

0.03

2-0

.044

[0.1

16]

[0.0

24]*

[0.1

24]

[0.0

25]*

[0.1

33]

[0.0

24]

[0.1

40]

[0.0

25]

[0.0

66]

[0.0

22]**

[0.0

67]

[0.0

22]*

*[0

.070

][0

.023

]*

(TO

T G

row

th) x

(Exp

orts

/GD

P)-1

6.97

70.

208

-16.

806

0.25

-11.

671

-0.2

75[1

0.84

7][0

.952

][1

1.33

1][0

.916

][5

.597

]**[0

.739

]

Expo

rts/G

DP

20.2

73-0

.041

23.8

01-0

.342

28.6

421.

672

[22.

264]

[3.8

50]

[23.

094]

[3.9

22]

[11.

382]

**[4

.110

]

Obs

erva

tions

5284

5284

5284

5284

104

168

104

168

104

168

R-s

quar

ed0.

690.

780.

720.

790.

690.

790.

730.

790.

650.

730.

680.

740.

650.

74D

ecad

e D

umm

ies

YY

YY

YY

YY

YY

YY

YY

Cou

ntry

Dum

mie

sY

YY

YY

YY

YY

YY

YY

YC

ontro

ls (N

ote

1)N

NN

NY

YY

YN

NN

NY

Y

Mea

n Va

lues

[Std

Dev

]:ln

(Cap

ital F

low

s)5.

806.

425.

806.

425.

806.

425.

806.

425.

595.

815.

595.

815.

595.

81[2

.40]

[2.2

0][2

.40]

[2.2

0][2

.40]

[2.2

0][2

.40]

[2.2

0][2

.56]

[2.5

3][2

.56]

[2.5

3][2

.56]

[2.5

3]

TOT

Gro

wth

0.32

-0.0

10.

32-0

.01

0.32

-0.0

10.

32-0

.01

0.32

-0.5

80.

32-0

.58

0.32

-0.5

8[0

.87]

[1.3

6][0

.87]

[1.3

6][0

.87]

[1.3

6][0

.87]

[1.3

6][1

.16]

[1.7

8][1

.16]

[1.7

8][1

.16]

[1.7

8]

TOT

Vola

tility

6.23

8.59

6.23

8.59

6.23

8.59

6.23

8.59

7.19

8.71

7.19

8.71

7.19

8.71

[3.9

3][5

.63]

[3.9

3][5

.63]

[3.9

3][5

.63]

[3.9

3][5

.63]

[6.5

6][5

.36]

[6.5

6][5

.36]

[6.5

6][5

.36]

(TO

T G

row

th) x

(Exp

orts

/GD

P)0.

030.

000.

030.

000.

030.

000.

030.

000.

03-0

.11

0.03

-0.1

10.

03-0

.11

[0.8

7][0

.22]

[0.8

7][0

.22]

[0.8

7][0

.22]

[0.8

7][0

.22]

[0.1

3][0

.36]

[0.1

3][0

.36]

[0.1

3][0

.36]

Expo

rts/G

DP

0.09

0.13

0.09

0.13

0.09

0.13

0.09

0.13

0.10

0.14

0.10

0.14

0.10

0.14

[0.0

4][0

.11]

[0.0

4][0

.11]

[0.0

4][0

.11]

[0.0

4][0

.11]

[0.0

6][0

.12]

[0.0

6][0

.12]

[0.0

6][0

.12]

Mar

gina

l Im

pact

TOT

Gro

wth

0.04

0.03

-0.2

30.

03-0

.09

0.08

-0.4

10.

090.

020.

03-0

.41

0.03

-0.0

30.

07TO

T Vo

latil

ity-0

.04

-0.0

4-0

.03

-0.0

5-0

.05

-0.0

4-0

.05

-0.0

40.

03-0

.05

0.04

-0.0

50.

03-0

.04

Impa

ct o

f a 1

Std

Dev

Incr

ease

TOT

Gro

wth

0.03

0.04

-13.

680.

05-0

.08

0.11

-13.

700.

130.

030.

05-0

.67

0.02

-0.0

40.

12TO

T Vo

latil

ity-0

.15

-0.2

5-0

.12

-0.2

5-0

.19

-0.2

0-0

.21

-0.2

10.

16-0

.25

0.24

-0.2

50.

21-0

.24

Rob

ust s

tand

ard

erro

rs in

bra

cket

s* s

igni

fican

t at 1

0%; *

* sig

nific

ant a

t 5%

; ***

sig

nific

ant a

t 1%

Not

e 1:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P pe

r cap

ita, s

choo

ling,

and

pop

ulat

ion

grow

th

By

Dec

ade

By

Hal

f-Dec

ade

50

Tabl

e A1

: Whi

ch P

erip

hery

Cou

ntrie

s D

omin

ated

Whi

ch E

xpor

t Mar

kets

for W

hat P

rimar

y Pr

oduc

ts in

190

0?

Com

mod

ityC

ount

rySh

are

of w

orld

ex

port

s in

co

mm

odity

(%)

Shar

e of

tota

l do

mes

tic

expo

rts

(%)

Com

mod

ityC

ount

rySh

are

of w

orld

ex

port

s in

co

mm

odity

(%)

Shar

e of

tota

l do

mes

tic

expo

rts

(%)

Egyp

t18

.910

0.0

Indo

nesi

a12

.241

.0

Indi

a9.

414

.8C

uba

7.5

67.4

Bra

zil

75.8

68.6

Bra

zil

3.1

3.7

Indo

nesi

a6.

419

.4P

eru

2.5

32.2

Indi

a41

.014

.2In

dia

14.6

Not

e A

Chi

na26

.122

.7A

rgen

tina

14.0

16.2

Cey

lon

24.2

98.9

Rus

sia

4.8

Not

e A

Chi

na36

.335

.4B

razi

l4.

5N

ote

A

Japa

n29

.849

.9Au

stra

lia71

.0B

razi

l57

.321

.0N

ew Z

eala

nd62

.7

Indo

nesi

a8.

20.

3A

rgen

tina

34.8

Tin

Indo

nesi

a8.

616

.6U

rugu

ayN

ote

B33

.9

Indo

nesi

a20

.117

.7Tu

rkey

18.4

Cub

a15

.132

.6R

ussi

a11

.2

Coc

oaBr

azil

24.1

2.7

Per

u7.

5

Chi

le87

.981

.8R

ussi

a22

.158

.9

Per

u9.

4N

ote

AA

rgen

tina

19.3

37.7

Indi

a46

.619

.4In

dia

5.2

Not

e A

Sia

m11

.610

0.0

Phili

ppin

es36

.176

.9S

pain

27.4

9.6

Rus

sia

13.3

1.6

Chi

le21

.818

.2Ja

pan

5.3

21.1

Mex

ico

13.0

11.3

Chi

na2.

48.

7

Indi

a/B

urm

a 77

.26.

5Ju

te m

anuf

actu

res

Indi

a63

.711

.6

Egy

pt

22.8

Not

e A

Indi

a4.

510

.1

Dye

sIn

dia

16.6

Not

e A

Japa

n2.

228

.9

Bol

d=

Indi

cate

cou

ntry

-com

mod

ity c

ombi

natio

ns w

here

the

pric

e-ta

king

ass

umpt

ion

is h

ighl

y un

likel

y to

app

ly

Not

bol

d=

Indi

cate

cou

ntry

-com

mod

ity c

ombi

natio

ns w

here

the

pric

e-ta

king

ass

umpt

ion

is s

omew

hat u

nlik

ely

to a

pply

Sug

ar

Hid

es a

nd s

kins

Woo

l

Silk

man

ufac

ture

s

Whe

at

Hem

p

Cot

ton

man

ufac

ture

s

Toba

cco

Ferti

lizer

Ric

e

Cop

per

Raw

cot

ton

Cof

fee

Sou

rces

: Ea

ch c

ount

ry's

sha

re in

the

tota

l wor

ld e

xpor

ts o

f eac

h co

mm

odity

are

for t

he y

ear 1

900

and

com

e fro

m J

. R. H

anso

n II,

Tra

de in

Tra

nsiti

on: E

xpor

ts fr

om th

e Th

ird

Wor

ld 1

840-

1900

(NY:

Aca

dem

ic P

ress

198

0), A

ppen

dix

D, p

p. 1

71-8

1. T

he s

hare

of e

ach

com

mod

ity in

eac

h co

untry

's d

omes

tic e

xpor

ts is

in fa

ct th

e pe

rcen

tage

of e

xpor

ts

repr

esen

ted

by th

at c

omm

odity

out

of t

he to

p fiv

e e

xpor

ts in

1900

, not

tota

l exp

orts

. Sou

rces

for t

hese

est

imat

es a

re li

sted

in th

e da

ta a

ppen

dix

unde

r 'ex

port

wei

ghts

' in

the

disc

ussi

on o

f the

term

s of

trad

e in

dex.

Not

e A:

"Sha

re o

f tot

al d

omes

tic e

xpor

ts" i

s no

t the

sha

re o

f tha

t com

mod

ity in

tota

l exp

orts

, but

rath

er th

e sh

are

of th

at c

omm

odity

out

of t

he to

p fiv

e m

ajor

exp

orts

for t

hat

coun

try. T

hus

whe

re a

cou

ntry

pro

duce

s m

ore

than

five

exp

orts

, fig

ures

for t

he e

xpor

t sha

re o

f min

or e

xpor

ts a

re n

ot a

vaila

ble.

By

defin

ition

thes

e ex

ports

will

not

affe

ct o

f m

easu

red

term

s of

trad

e, a

nd s

o ar

e no

t a c

once

rn.

Not

e B:

We

do n

ot h

ave

offic

ial w

orld

exp

ort f

igur

es fo

r woo

l in

1900

. On

our r

evie

w o

f the

his

toric

al li

tera

ture

how

ever

, Aus

tralia

like

ly h

eld

mor

e th

an a

third

of w

orld

exp

orts

in

1900

. Arg

entin

a m

ay h

ave

held

mor

e th

an a

qua

rter b

efor

e 18

98, a

nd s

o w

e ex

clud

e it

at th

at le

vel t

o be

saf

e.N

ote

C: T

he 7

7.2%

figu

re fo

r Ind

ia/B

urm

a w

as re

cord

ed fo

r the

two

coun

tries

join

tly in

our

sou

rces

. The

6.5

% fi

gure

for s

hare

of e

xpor

ts is

for B

urm

a al

one,

how

ever

. It s

eem

s un

likel

y th

en th

at B

urm

a w

ould

hav

e in

fluen

ced

the

inte

rnat

iona

l pric

e, b

ut th

at In

dia

may

hav

e. In

dia,

how

ever

, has

alre

ady

been

flag

ged

as a

non

-pric

e ta

ker e

lsew

here

, ho

wev

er.

Raw

silk

Rub

ber

Oils

eeds

(Not

e C

)

Tea

51

Tabl

e A2

: Rep

licat

ion

of T

able

4 (G

DP

Gro

wth

and

Ter

ms

of T

rade

) with

5-Y

ear P

erio

dsD

epen

dent

Var

iabl

e: Q

uinq

ueni

al (5

-yea

r) a

vera

ge G

DP

per

cap

ita g

row

th

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

TOT

Gro

wth

0.37

8-0

.072

0.78

3-0

.307

0.35

20.

094

0.29

5-0

.038

0.60

5-0

.569

1.30

7-0

.541

[0.1

68]**

[0.1

15]

[0.3

07]**

[0.1

27]**

[0.1

35]**

[0.1

22]

[0.2

14]

[0.1

07]

[0.4

08]

[0.1

99]**

*[0

.537

]**[0

.258

]**

TOT

Vola

tility

0.01

4-0

.061

0.00

3-0

.058

-0.0

15-0

.077

-0.0

16-0

.079

0.15

10.

095

0.12

30.

093

[0.0

32]

[0.0

34]*

[0.0

32]

[0.0

33]*

[0.0

37]

[0.0

37]*

*[0

.037

][0

.035

]**[0

.096

][0

.080

][0

.091

][0

.082

]

(TO

T Tr

end

Gro

wth

) x (E

xpor

ts/G

DP)

-4.3

841.

665

0.71

11.

085

-7.5

91-0

.163

[2.3

97]*

[0.8

75]*

[2.5

12]

[0.6

94]

[3.3

57]**

[1.7

63]

Expo

rts/G

DP

-2.1

872.

333

0.58

18.

228

-13.

854

0.02

5[5

.868

][3

.701

][1

0.37

9][4

.708

]*[9

.766

][8

.272

]

Obs

erva

tions

172

272

172

272

126

189

126

189

4683

4683

R-s

quar

ed0.

360.

230.

380.

260.

310.

230.

310.

250.

520.

620.

600.

62D

ecad

e D

umm

ies

YY

YY

YY

YY

YY

YY

Cou

ntry

Dum

mie

sY

YY

YY

YY

YY

YY

YC

ontro

ls (N

ote

1)Y

YY

YY

YY

YY

YY

Y

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Not

e 1:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

1870

-193

9 (1

910-

1919

exc

lude

d)18

70-1

909

1920

-193

9

52

Tabl

e A3

: Rep

licat

ion

of T

able

4 (G

DP

Gro

wth

and

Ter

ms

of T

rade

) Usi

ng A

ltern

ativ

e D

ecom

posi

tion

Met

hods

187

0-19

39D

epen

dent

Var

iabl

e: D

ecad

al a

vera

ge G

DP

per

cap

ita g

row

th

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

TOT

Gro

wth

0.42

2-0

.044

0.91

1-0

.294

0.39

7-0

.019

0.73

3-0

.258

0.38

-0.0

30.

701

-0.2

39[0

.158

]**[0

.130

][0

.298

]***

[0.1

40]*

*[0

.142

]***

[0.1

08]

[0.2

54]*

**[0

.113

]**

[0.1

33]*

**[0

.098

][0

.239

]***

[0.1

05]*

*

TOT

Vol

atilit

y0.

005

-0.1

060.

001

-0.1

010.

007

-0.1

14-0

.001

-0.1

050.

009

-0.1

27-0

.005

-0.1

09[0

.042

][0

.037

]***

[0.0

39]

[0.0

35]*

**[0

.045

][0

.043

]***

[0.0

45]

[0.0

41]*

*[0

.042

][0

.060

]**

[0.0

43]

[0.0

57]*

(TO

T G

row

th) x

(Exp

orts

/GD

P)-5

.072

1.83

6-3

.361

1.70

5-2

.964

1.52

[2.3

05]*

*[0

.830

]**

[1.9

07]*

[0.6

93]*

*[1

.776

][0

.636

]**

Exp

orts

/GD

P0.

352

4.69

80.

917

4.99

81.

303

4.54

9[3

.895

][3

.964

][4

.264

][4

.074

][4

.680

][3

.920

]

Obs

erva

tions

7912

579

125

7912

579

125

7912

579

125

R-s

quar

ed0.

50.

430.

570.

460.

520.

430.

570.

470.

520.

430.

570.

46D

ecad

e D

umm

ies

YY

YY

YY

YY

YY

YY

Cou

ntry

Dum

mie

sY

YY

YY

YY

YY

YY

YC

ontro

ls (N

ote

1)Y

YY

YY

YY

YY

YY

Y

Rob

ust s

tand

ard

erro

rs in

bra

cket

s* s

igni

fican

t at 1

0%; *

* sig

nific

ant a

t 5%

; ***

sig

nific

ant a

t 1%

Not

e 1:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P pe

r cap

ita, s

choo

ling,

and

pop

ulat

ion

grow

th

HP

Dec

ompo

sitio

n fr

om T

able

4 (λ

=300

)H

P D

ecom

posi

tion

with

Slo

wer

-Mov

ing

Tren

d (λ

=100

)M

ovin

g Av

erag

e Fi

lter (

7 pe

riods

)

53

Tabl

e A4

: Rep

licat

ion

of T

able

4 (G

DP

Gro

wth

and

Ter

ms

of T

rade

) Usi

ng A

n Al

tern

ativ

e M

easu

re o

f Vol

atili

tyD

epen

dent

Var

iabl

e: D

ecad

al a

vera

ge G

DP

per

cap

ita g

row

th

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

TOT

Gro

wth

0.13

30.

086

0.16

-0.1

680.

317

0.29

80.

233

0.11

7-0

.044

-0.1

740.

115

0.06

7[0

.120

][0

.094

][0

.192

][0

.095

]*[0

.104

]***

[0.1

19]*

*[0

.148

][0

.087

][0

.334

][0

.081

]**

[0.7

66]

[0.1

38]

Std

Dev

(TO

T G

row

th)

0.04

3-0

.109

0.04

6-0

.068

0.02

4-0

.207

0.01

9-0

.19

0.67

4-0

.053

1.02

1-0

.164

[0.0

40]

[0.0

60]*

[0.0

43]

[0.0

56]

[0.0

38]

[0.0

90]*

*[0

.038

][0

.073

]**

[0.2

35]*

*[0

.072

][0

.783

][0

.082

]*

(TO

T G

row

th) x

(Exp

orts

/GD

P)

-0.2

491.

747

0.98

31.

395

1.09

4-1

.795

[1.1

45]

[0.4

25]**

*[1

.800

][0

.667

]**

[3.0

48]

[1.2

65]

Expo

rts/G

DP

1.89

15.

497

8.92

811

.526

18.6

44-3

.687

[4.9

28]

[3.6

04]

[4.6

85]*

[5.7

87]*

[22.

629]

[8.3

04]

Obs

erva

tions

7912

579

125

5684

5684

2341

2341

R-s

quar

ed0.

440.

420.

450.

480.

660.

550.

690.

630.

580.

740.

610.

82D

ecad

e D

umm

ies

YY

YY

YY

YY

YY

YY

Cou

ntry

Dum

mie

sY

YY

YY

YY

YY

YY

YC

ontro

ls (N

ote

1)Y

YY

YY

YY

YY

YY

Y

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; ** s

igni

fican

t at 5

%; *

** s

igni

fican

t at 1

%

Not

e 1:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

1870

-193

9 (1

910-

1919

exc

lude

d)18

70-1

909

1920

-193

9

54

Tabl

e A5

: Rep

licat

ion

of T

able

4 (G

DP

Gro

wth

and

Ter

ms

of T

rade

) with

the

'Eur

opea

n O

ffsho

ots'

in th

e C

ore

Dep

ende

nt V

aria

ble:

Dec

adal

ave

rage

GD

P p

er c

apita

gro

wth

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

TOT

Gro

wth

0.36

2-0

.058

0.79

5-0

.289

0.32

60.

056

0.49

3-0

.057

0.69

3-0

.455

1.23

6-0

.084

[0.1

30]**

*[0

.153

][0

.255

]***

[0.1

45]*

[0.1

39]*

*[0

.141

][0

.281

]*[0

.139

][0

.441

][0

.286

][0

.503

]**

[0.3

66]

TOT

Vol

atilit

y-0

.004

-0.1

22-0

.008

-0.1

160.

045

-0.1

320.

043

-0.1

14-0

.029

0.03

70.

079

-0.0

07[0

.036

][0

.044

]***

[0.0

35]

[0.0

42]**

*[0

.032

][0

.044

]***

[0.0

32]

[0.0

46]*

*[0

.139

][0

.108

][0

.156

][0

.113

]

(TO

T G

row

th) x

(Exp

orts

/GD

P)

-3.2

181.

742

-1.8

082.

219

-6.5

62-2

.261

[1.6

90]*

[0.7

81]*

*[2

.239

][1

.331

][2

.894

]*[1

.113

]*

Expo

rts/G

DP

2.47

22.

953

14.2

12.2

87-1

1.96

8-0

.53

[3.4

36]

[4.3

10]

[7.0

36]**

[9.5

45]

[6.2

77]*

[6.3

84]

Obs

erva

tions

109

9510

995

7612

676

6433

3133

31R

-squ

ared

0.39

0.30

0.42

0.39

0.55

0.23

0.62

0.43

0.51

0.79

0.66

0.82

Dec

ade

Dum

mie

sY

YY

YY

YY

YY

YY

YC

ount

ry D

umm

ies

YY

YY

YY

YY

YY

YY

Con

trols

(Not

e 1)

YY

YY

YY

YY

YY

YY

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; ** s

igni

fican

t at 5

%; *

** s

igni

fican

t at 1

%

Not

e 1:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

1870

-193

9 (1

910-

1919

exc

lude

d)18

70-1

909

1920

-193

9

55

Tabl

e A6

: Rep

licat

ion

of T

able

4 (G

DP

Gro

wth

and

Ter

ms

of T

rade

) with

the

'Eur

opea

n Pe

riphe

ry' i

n th

e Pe

riphe

ryD

epen

dent

Var

iabl

e: D

ecad

al a

vera

ge G

DP

per

cap

ita g

row

th

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

TOT

Gro

wth

0.22

0.13

10.

606

-0.0

120.

261

0.11

40.

86-0

.021

0.13

-0.0

23-0

.139

0.51

9[0

.177

][0

.122

][0

.386

][0

.156

][0

.200

][0

.115

][0

.549

][0

.120

][0

.967

][0

.323

][0

.000

][0

.441

]

TOT

Vola

tility

0.05

2-0

.048

0.05

9-0

.044

0.02

2-0

.058

0.03

1-0

.091

0.14

3-0

.01

0.00

4-0

.045

[0.0

47]

[0.0

32]

[0.0

49]

[0.0

32]

[0.0

43]

[0.0

32]*

[0.0

39]

[0.0

41]**

[0.4

51]

[0.0

66]

[0.0

00]

[0.0

91]

(TO

T G

row

th) x

(Exp

orts

/GD

P)

-2.8

831.

159

-6.3

581.

787

-2.8

21-4

.065

[2.5

83]

[0.9

43]

[5.0

90]

[1.0

11]*

[0.0

00]

[1.7

64]**

Expo

rts/G

DP

-0.6

012.

666

7.80

713

.313

-22.

773

-9.8

72[4

.297

][3

.976

][7

.206

][6

.964

]*[0

.000

][6

.276

]

Obs

erva

tions

5015

450

154

3620

136

104

1450

1450

R-s

quar

ed0.

550.

370.

580.

380.

540.

290.

580.

540.

810.

781.

000.

85D

ecad

e D

umm

ies

YY

YY

YY

YY

YY

YY

Cou

ntry

Dum

mie

sY

YY

YY

YY

YY

YY

YC

ontro

ls (N

ote

1)Y

YY

YY

YY

YY

YY

Y

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; ** s

igni

fican

t at 5

%; *

** s

igni

fican

t at 1

%

Not

e 1:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

1870

-193

9 (1

910-

1919

exc

lude

d)18

70-1

909

1920

-193

9

56

Tabl

e A7

: Tes

ting

for E

xoge

neity

of T

erm

s of

Tra

de o

n G

row

th --

Exc

ludi

ng 'L

arge

Cou

ntrie

s' fr

om th

e Pe

riphe

ryD

epen

dent

Var

iabl

e: D

ecad

al a

vera

ge G

DP

per

cap

ita g

row

th

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

TOT

Gro

wth

-0.0

44-0

.294

0.02

9-0

.337

-0.0

07-0

.399

0.03

8-0

.416

[0.1

30]

[0.1

40]*

*[0

.185

][0

.199

]*[0

.233

][0

.245

][0

.245

][0

.267

]

TOT

Vola

tility

-0.1

06-0

.101

-0.1

14-0

.100

-0.1

08-0

.087

-0.0

90-0

.064

[0.0

37]*

**[0

.035

]***

[0.0

42]*

**[0

.040

]**

[0.0

51]*

*[0

.051

]*[0

.055

][0

.053

]

(TO

T G

row

th) x

(Exp

orts

/GD

P)1.

836

2.27

62.

152

2.40

1[0

.830

]**[0

.953

]**[1

.018

]**[1

.088

]**

Expo

rts/G

DP

4.69

85.

738

3.49

23.

413

[3.9

64]

[4.7

90]

[5.1

25]

[5.2

05]

Obs

erva

tions

125

125

8989

7171

6565

R-s

quar

ed0.

430.

460.

420.

470.

430.

470.

440.

49D

ecad

e D

umm

ies

YY

YY

YY

YY

Cou

ntry

Dum

mie

sY

YY

YY

YY

YC

ontro

ls (N

ote

1)Y

YY

YY

YY

Y

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Not

e 1:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

Per

iphe

ry

Orig

inal

Dro

p us

ing

1/3

thre

shol

dD

rop

usin

g 1/

4 th

resh

old

Dro

p us

ing

1/5

thre

shol

d

57

Tabl

e A8

: GD

P G

row

th, t

he T

erm

s of

Tra

de, a

nd In

itial

Inst

itutio

ns, 1

870-

1939

Dep

ende

nt V

aria

ble:

Dec

adal

ave

rage

GD

P p

er c

apita

gro

wth

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Cor

ePe

riph

All

Cor

ePe

riph

All

Cor

ePe

riph

All

TOT

Gro

wth

0.42

2-0

.044

0.15

4-2

.917

-0.0

62-0

.016

0.51

2-0

.006

0.20

3[0

.158

]**[0

.130

][0

.097

][2

.155

][0

.160

][0

.127

][0

.163

]***

[0.1

05]

[0.0

94]**

TOT

Vola

tility

0.00

5-0

.106

-0.0

37-0

.376

-0.1

07-0

.124

0.00

6-0

.12

-0.0

18[0

.042

][0

.037

]***

[0.0

27]

[0.3

70]

[0.0

50]**

[0.0

42]**

*[0

.046

][0

.030

]***

[0.0

20]

(TO

T G

row

th) x

(% P

op'n

Eur

opea

n in

180

0)0.

034

0.00

10.

003

[0.0

22]

[0.0

05]

[0.0

02]*

(TO

T V

olat

ility

x (%

Pop

'n E

urop

ean

in 1

800)

0.00

40.

000

0.00

2[0

.004

][0

.001

][0

.001

]***

% P

op'n

Eur

opea

n in

180

0 (N

ote

1)-0

.226

0.00

3-0

.04

[0.0

82]**

*[0

.018

][0

.012

]***

(TO

T G

row

th) x

(Pol

ity)

-0.0

42-0

.03

-0.0

31[0

.024

]*[0

.015

]**[0

.014

]**

(TO

T Vo

latil

ity x

(Pol

ity)

0.00

3-0

.007

0.00

2[0

.004

][0

.005

][0

.003

]

Polit

y (N

ote

2)0.

045

0.08

00.

029

[0.0

52]

[0.0

91]

[0.0

40]

Obs

erva

tions

7912

520

479

125

204

7479

153

R-s

quar

ed0.

50.

430.

360.

530.

430.

410.

560.

720.

58D

ecad

e D

umm

ies

YY

YY

YY

YY

YC

ount

ry D

umm

ies

YY

YY

YY

YY

YC

ontro

ls (N

ote

3)Y

YY

YY

YY

YY

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Not

e 1:

% o

f pop

ulat

ion

Eur

opea

n or

of E

urop

ean

desc

ent i

s a

time-

inva

riant

var

iabl

e on

a sc

ale

of [0

, 100

].

Not

e 3:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

Not

e 2:

Pol

ity is

a ti

me-

vary

ing

indi

cato

r of d

emoc

racy

/aut

ocra

cy o

n a

scal

e of

[-10

, 10]

and

is a

vaila

ble

in th

e P

OLI

TY IV

dat

abas

e fo

r ind

epen

dent

cou

ntrie

s on

ly. I

nitia

l val

ues

at th

e be

ginn

ing

of th

e de

cade

are

use

d. L

ost o

bser

vatio

ns a

rise

from

cou

ntrie

s in

tran

sitio

n an

d co

untri

es th

at re

mai

n co

loni

es.

58

Tabl

e A9

: Rep

licat

ion

of T

able

7 (C

apita

l Flo

ws

and

Term

s of

Tra

de) w

ith A

ltern

ativ

e Vo

latil

ity M

easu

res

Dep

ende

nt V

aria

ble:

Nat

ural

log

of a

vera

ge d

ecad

al c

apita

l inf

low

s fro

m B

ritai

n

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

(14)

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

TOT

Gro

wth

-0.0

980.

051

1.12

90.

010.

298

0.03

30.

747

-0.0

150.

054

0.01

90.

193

0.04

60.

047

-0.0

19[0

.391

][0

.100

][0

.790

][0

.154

][0

.221

][0

.062

][0

.376

]*[0

.119

][0

.181

][0

.066

][0

.305

][0

.107

][0

.068

][0

.028

]

TOT

Vola

tility

-0.0

67-0

.043

-0.0

78-0

.045

-0.1

14-0

.055

-0.0

97-0

.056

0.04

-0.0

660.

039

-0.0

64-0

.028

-0.0

89[0

.145

][0

.028

][0

.150

][0

.029

][0

.130

][0

.045

][0

.132

][0

.044

][0

.067

][0

.027

]**

[0.0

68]

[0.0

27]*

*[0

.071

][0

.033

]***

(TO

T G

row

th) x

(Exp

orts

/GD

P)-1

7.81

90.

341

-6.9

290.

383

-2.4

06-0

.209

[9.9

27]*

[0.8

49]

[4.8

79]

[0.6

02]

[2.8

38]

[0.4

92]

Expo

rts/G

DP

18.5

36-0

.379

13.0

7-0

.596

20.0

930.

786

[21.

805]

[3.8

96]

[25.

742]

[3.7

48]

[11.

029]

*[3

.602

]

Obs

erva

tions

5284

5284

5284

5284

117

189

117

189

117

189

R-s

quar

ed0.

690.

790.

740.

790.

720.

790.

730.

790.

660.

740.

670.

740.

660.

74D

ecad

e D

umm

ies

YY

YY

YY

YY

YY

YY

YY

Cou

ntry

Dum

mie

sY

YY

YY

YY

YY

YY

YY

YC

ontro

ls (N

ote

1)Y

YY

YY

YY

YY

YY

YY

Y

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Not

e 1:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

By

Dec

ade

By

Hal

f-Dec

ade

HP

Dec

ompo

sitio

n w

ith S

low

er-M

ovin

g Tr

end

(λ=1

00)

HP

Dec

ompo

sitio

n w

ith S

low

er-M

ovin

g Tr

end

(λ=1

00)

Gro

wth

rate

and

Std

Dev

of T

OT

(Not

de

com

pose

d)G

row

th ra

te a

nd S

tde

com

59

Tabl

e A1

0: R

eplic

atio

n of

Tab

le 7

(Cap

ital F

low

s an

d Te

rms

of T

rade

) with

Alte

rnat

ive

Def

initi

ons

of th

e Pe

riphe

ryD

epen

dent

Var

iabl

e: N

atur

al lo

g of

ave

rage

dec

adal

cap

ital i

nflo

ws

from

Brit

ain

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

(14)

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

Cor

ePe

riph

TOT

Gro

wth

0.07

20.

111

0.81

20.

063

-0.8

120.

193

-0.9

60.

208

0.13

40.

073

0.42

70.

129

-0.3

270.

13[0

.320

][0

.120

][0

.646

][0

.176

][0

.738

][0

.122

][1

.278

][0

.171

][0

.203

][0

.084

][0

.364

][0

.143

][0

.295

][0

.083

]

TOT

Vola

tility

-0.0

4-0

.031

-0.0

38-0

.035

0.00

5-0

.072

-0.0

05-0

.07

0.02

5-0

.047

0.03

3-0

.044

0.08

5-0

.059

[0.1

24]

[0.0

26]

[0.1

27]

[0.0

28]

[0.1

10]

[0.0

32]*

*[0

.112

][0

.032

]**

[0.0

56]

[0.0

24]*

[0.0

56]

[0.0

25]*

[0.0

66]

[0.0

26]*

*

(TO

T G

row

th) x

(Exp

orts

/GD

P)-9

.312

0.47

90.

294

-0.1

53-4

.049

-0.3

48[6

.274

][0

.929

][1

1.96

9][0

.925

][3

.026

][0

.639

]

Expo

rts/G

DP

16.8

273.

516

41.8

35-1

.247

17.4

794.

263

[18.

911]

[5.4

00]

[25.

104]

[3.8

51]

[9.9

23]*

[3.6

76]

Obs

erva

tions

5284

5284

5284

5284

117

189

117

189

117

189

R-s

quar

ed0.

690.

790.

740.

790.

720.

790.

730.

790.

660.

740.

670.

740.

660.

74D

ecad

e D

umm

ies

YY

YY

YY

YY

YY

YY

YY

Cou

ntry

Dum

mie

sY

YY

YY

YY

YY

YY

YY

YC

ontro

ls (N

ote

1)Y

YY

YY

YY

YY

YY

YY

Y

Rob

ust s

tand

ard

erro

rs in

bra

cket

s*

sign

ifica

nt a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Not

e 1:

Con

trols

incl

ude

a co

nsta

nt te

rm a

nd in

itial

val

ues

of th

e lo

g of

GD

P p

er c

apita

, sch

oolin

g, a

nd p

opul

atio

n gr

owth

By

Dec

ade

By

Hal

f-Dec

ade

Euro

pean

Offs

hoot

s in

Cor

eEu

rope

an P

erip

hery

in P

erip

hery

Euro

pean

Offs

hoot

s in

Cor

eEu

rope

an P

erip

h