winners and losers in the commodity lotterys3.amazonaws.com/zanran_storage/... · winners and...
TRANSCRIPT
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 [email protected] [email protected] Jeffrey G. Williamson Department of Economics Harvard University Cambridge, MA 02138 [email protected] 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