Terms of Trade Volatility and Precautionary Savings in Developing Economies Salim B. Furth∗ Department of Economics, Amherst College

March 31, 2012 Abstract This paper investigates the link between terms of trade volatility and long-term output growth in developing countries. I find that differences in terms of trade volatility account for 20% of the crosscountry variation in growth from 1980 to 2009. The magnitude is arresting: a standard-deviation difference in exposure to terms of trade volatility between two countries is associated in the data with a 16percentage-point difference in overall output growth. A decomposition of output growth distinguishes pure capital accumulation from the dynamic effects of productivity growth. The data show that capital accumulation in the 1970’s and 80’s was highest in countries with high terms of trade volatility, which later shifted their portfolios away from domestic capital and into foreign bonds. The reallocation of precautionary savings from domestic to foreign assets led to falling output in countries with volatile terms of trade. A neoclassical capital accumulation model has significant precautionary savings associated with terms of trade risk. Opening foreign bond markets in the model induces a shift away from capital and a fall in output in price-volatile countries, reproducing my finding from the data. JEL classification: E2, F21, F4, O13. Keywords: terms of trade volatility; precautionary savings; GDP growth; developing countries. The author thanks Mark Aguiar, William Hawkins, Michael Insler, Michal Kuklik, Stanislav Kornienko, two referees, and attendees to conferences and seminars for helpful insights. Research for this paper was performed at University of Rochester, to which the author extends his thanks. All errors are my own. ∗

1

Introduction

Developing countries have experienced widely varying growth trajectories. In 1980, income per adult was almost twice as high in Algeria as in neighboring Morocco. Over the ensuing decades, Morocco grew steadily, although slowly, while Algeria contracted. In Central America, Costa Rica saw income rise 61% while neighboring Nicaragua’s income fell 11%. The same story has played out on every continent: some countries grew, others contracted. Economists spend a great deal of effort trying to explain why growth rates differ across developing countries; this work is a contribution to that effort. This paper shows how a shift in savings patterns can explain a portion of the cross-country differences in growth. Developing countries with high terms of trade volatility are motivated to smooth their consumption by saving, often in domestic capital. As global asset markets became more open during the 1990’s, countries which had saved large quantities of domestic capital began to shift their savings to foreign assets. This asset flow away from certain developing countries leads to lower output growth in those economies than in low-volatility countries. Using cross-sectional analysis I find that the negative correlation between terms of trade volatility and GDP growth accounts for 20% of the variation in growth across developing countries. In order to measure the exposure of an economy to terms of trade shocks, I interact the volatility of terms of trade with exports as a share of GDP. A two-standard-deviation difference in exposures to terms of trade volatility of two countries is associated with a 25 percentage point difference in total GDP growth between 1980 and 2009. Examining robustness regressions, it is clear that the correlation is driven by differences in volatilities of terms of trade, not differences in the magnitudes of export sectors, which have independent positive effects countervailing volatility. I decompose the growth difference associated with exposure to terms of trade volatility and find that a 4 percentage point difference in GDP growth can be attributed to differing rates of capital accumulation. While this is not a very large growth difference, median growth over the entire period was 9% in my sample. In order to provide theoretical foundations for the finding, I posit a dynamic stochastic general equilibrium model of capital accumulation that fits these two features of the data. The modeled country is endowed with an 2

exportable good, which it sells on world markets at a stochastic prevailing world price. The randomness of export prices give countries a precautionary savings motive. Export supply is inelastic, and investment in domestic capital is subject to TFP shocks. When international bond markets are unavailable, precautionary savings are invested in the domestic consumables sector. However, when bond markets are available, savings are optimally invested abroad in riskless markets. The reallocation of savings from domestic capital to foreign bonds leads to a decline in domestic GDP. Since the degree of the precautionary motive differs across countries, depending on their export price volatility, the decline in GDP when bond markets open also differs. This creates a negative relationship between export price volatility and GDP growth when bond markets open. I compute this neoclassical model with preference and technology parameters from the literature held constant across countries, and stationary export price processes which match the data for 50 developing countries. Using simulations, I find that in this model precautionary savings alone can account for 12 percentage points lower capital accumulation in a country which experienced two standard deviations higher export price volatility. This matches the 11 point estimate from the data, and leads to 3 points less GDP growth, supporting the hypothesis that precautionary savings due to terms of trade variability has affected growth rates over the last few decades.

1.1

Outline

Section 2 reviews the relevant literature. Section 3 presents the World Bank’s WDI data, from which I draw quantitative inferences using cross-sectional analysis. I also defend the assumption of terms-of-trade exogeneity. In Section 4 I present a simple model, which I calibrate and report the results of in Sections 5 and 6. Section 7 concludes and discusses ongoing extensions.

2

Literature

The narrative given in this paper complements that of Caballero and Cowan (2007). Comparing emerging market to developed economies, they find that higher terms of trade variance is explained by both higher sectoral concentration and higher export price variance in developing countries. They observe that “prudent” emerging market economies have changed their behavior over 3

time: in response to recent export booms, the prudent countries have used less government borrowing, held larger reserves, invested less domestically, and run smaller current account deficits than they did in response to export booms in the 20th century. They interpret this as a response to increased financial integration. Investigating a larger and poorer sample of countries, and using differing empirical methods, I confirm and extend Caballero and Cowan’s findings and show that these effects are strongest in the highestvolatility countries. A number of previous authors have sought a connection between terms of trade volatility and output growth. Lutz (1994) looks at the connection between TOT volatility and growth in a wide sample of annual data from 1968-1988. He defines the cyclical component of terms of trade as the residual after removing a linear trend, and then computes its variance over a moving three-year window for each country. Regressing annual GDP growth on this measure of volatility, he finds a significant negative growth effect of income terms of trade volatility but a significant positive growth effect of net barter terms of trade volatility. With similar methodology, Bleaney and Greenaway (2001) find weak negative effects of terms of trade volatility and on output growth, but not on investment, in a panel of 14 African countries. Mendoza (1997) finds that terms of trade volatility can explain 17% of consumption growth in a cross-section of 40 countries (28 developing) from 1971 to 1991. Blattman et al. (2007) find that volatility of terms of trade was a key determinant of income growth in commodity-exporting countries from 1870 to 1939, while terms of trade growth was irrelevant. Working with a broad sample of countries from 1960 to 2000, Aghion et al. (2009) find negative growth effects of terms of trade volatility, measured in 5-year periods, under fixed exchange rate regimes. In the ‘resource curse’ literature, volatility is generally considered a potential channel through which abundance of natural resources can be associated with slow growth. This channel is explored by van der Ploeg and Poelhekke (2009), who look at GDP volatility as an intermediary between natural resource abundance and GDP growth. A few equilibrium models have been proposed to explain how terms of trade volatility affects growth. Basu and McLeod (1991) use a model with imported inputs in production to show that increased variance in terms of trade can slow output growth due to the convexity of the production function. The effect, however, is second order and small in expectation. Mendoza (1997) proposes an AK model of endogenous growth, and for values of risk aversion 4

less than 2 is able to generate a negative consumption growth effect of terms of trade volatility. However, the model generates positive GDP growth effects even within the range of parameters where volatility decreases consumption, and for risk aversion greater than 2 generates positive consumption growth. Turnovsky and Chattopadhyay (2003) use a Romer (1986) style endogenous growth model with imperfect capital mobility and three sources of exogenous variation, including terms of trade shocks. Their model implies that terms of trade volatility has a negative effect on growth, as they also find in data from 1975 to 1992. In these endogenous growth models, terms of trade shocks drive both business cycles and growth. Others, like Kose (2002), use price shocks to explain business cycles, but not growth. A broader literature models the negative output growth effects of sources of risk other than terms of trade. Ramey and Ramey (1995) link higher output growth rate volatility to lower average output growth. Burnside and Tabova (2009) show that five global risk factors, including three commodity price indices, and country-specific exposure to each factor can account for 70% of the variation in growth volatility. Likewise, Koren and Tenreyro (2007) decompose the sources of volatility in developing country manufacturing sectors, finding that about 50% can be accounted for by sector-specific volatility. In seeming contradiction to the negative growth effects of terms of trade volatility, the precautionary savings motive of high volatility has been noted as well. Agenor and Aizenman (2004) find a positive precautionary savings response to permanent, favorable TOT shocks in a sample of African countries from 1980-1996. They reproduce their finding in a model of habit formation. Ghosh and Ostry (1994), investigating savings behavior by looking at current account surpluses, find that developing countries with higher TOT volatility ran larger current account surpluses in a large sample from 1965-1991. However, they do not look at domestic savings and capital accumulation. On the other hand, Dawe (1996) shows that volatility in the total value of exports increased capital accumulation but decreased income in a sample of 85 countries from 1970-1985, but does not distinguish between price and quantity movements. I confirm his finding in investigating my initial data: high-volatility countries had higher capital but lower productivity in the early 1980’s. This paper confirms many of the empirical findings above and combines them into a cohesive narrative. Rather than emphasize the experience of volatility over a short window, I take volatility over the whole sample as 5

the best estimate of the volatility of each country’s underlying price process. This must come at the expense of the time dimension in my analysis, making it similar to the cross-sectional approach of Mendoza, Turnovsky and Chattopadhyay, and Blattman et al. I find an effect of TOT volatility on GDP growth statistically equal to that found by Blattman et al., despite our non-overlapping sample periods. Likewise, my estimates of TOT volatility’s effect on productivity growth are of comparable magnitude to that of Aghion et al., although our volatility metrics differ. Since the most recent of these papers have data through 2000, and most cease before 1996, this paper updates the empirical literature with a decade or more of new data. I uniquely use the long-term measure of volatility to analyze growth spanning the recent era of globalization. Having analyzed more recent data, I come to different conclusions than previous researchers about the best way to model the effects of terms of trade. The most comparable previous papers, those of Mendoza and Turnovsky and Chattopadhyay, rely on externalities to capital accumulation causing endogenous growth. In both their papers, terms of trade volatility leads to GDP growth in some or all cases. By contrast, the model presented below allows diminishing marginal returns to capital, casting excess capital accumulation as an inefficiency rather than a springboard. In addition, Turnovsky and Chattopadhyay’s benchmark calibration results in average per-worker growth of 4% per year, which is higher than the maximum growth rate among the 62 countries in my expanded sample. The model I present here, better matches the experience of developing countries in the last thirty years than the existing work on the topic.

3

Data & Analysis

Throughout the paper, data discussed are from the World Bank World Development Indicators dataset, accessed in 2011 (WDI). Terms of trade (TOT) are defined in the WDI as the price of exports divided by the price of imports. Thus, to speak of ‘improving’ terms of trade denotes an increase in the price ratio. Terms of trade volatility refers here to the standard deviation of detrended net barter terms of trade. I measure long-run growth in terms of trade as its percent change over the sample period, aggregating over the first and last five years of the sample to abstract from short-run movements. 6

A key assumption throughout the analysis is that terms of trade volatility is exogenous to each small economy’s decisions; this assumption is discussed at length in Subsection 3.2. In order to measure each economy’s exposure to terms of trade volatility, I multiply by the mean export share of GDP. I introduce this cross-term in regressions, but find that it is insignificant in the presence of both of its components, and therefore do not use it in my analysis or model. I choose Gross Domestic Product (GDP) as the best measure of income to analyze output growth. From investment and population data, I construct series of GDP and capital stock per working-age adult. In order to identify Total Factor Productivity (TFP) and decompose per-adult growth into capital growth and productivity growth, I employ a riskless neoclassical growth model with Cobb-Douglas production. I proceed to show that peradult growth of GDP, capital stock, and TFP are negatively linked to TOT volatility. The differing magnitudes of these linkages suggest that while TFP changes account for the majority of the fall in capital and GDP associated with exposure to TOT volatility, an additional growth channel exists through differential capital accumulation. Further analysis constrains the search for a mechanism by which higher TOT volatility would cause lower capital accumulation, by ruling out potential channels which are not linked to growth or volatility in the WDI data. Taking these constraints into account, the model presented in Section 4 emphasizes the reallocation of precautionary savings between two steady states as a channel for differential growth as the world transits from one steady state to the other.

3.1

Data description

I analyze the volatility and trends of terms of trade for a sample of developing countries over the period 1980-2009. The data are drawn from the World Bank’s World Development Indicators. The countries I included are those which were not largely industrialized in 1980 and for which the WDI had sufficient data1 over the sample period. I exclude China as a potential nonprice-taker and importer (rather than exporter) of primary commodities. For estimations, this leaves 50 countries, of which 17 are in Sub-Saharan Africa, 1

Countries were excluded which did not have GDP, population, investment, and terms of trade data for most years in the sample.

7

Table 1: Data Summary

Moment n Mean Median Std Dev Minimum Maximum

GDP Capital Growth Growth 50 0.15 0.09 0.37 -0.54 0.99

50 0.13 0.17 0.53 -1.13 1.18

TOT Mean Trade Volatility Share 50 0.099 0.093 0.041 0.038 0.239

50 31% 27% 14% 9% 65%

Primary Exports (80’s) 44 74% 76% 20% 31% 99%

For each country, GDP growth is measured in log differences of GDP per adult in constant 2000 US$ between averages across the period 1980-84 and 2005-09; capital growth likewise. Terms of trade is normalized to mean 1 for each country. TOT volatility is the standard deviation of HP(100)-filtered annual terms of trade. Mean trade share is imports plus exports divided by twice GDP, averaged from 1980-2009. Primary exports are reported as a share of total merchandise exports, averaged for each country over 1980-1989.

17 in Latin America, 4 in Asia, 7 in the Middle East and North Africa, and 5 are islands in the Indian or Pacific Oceans. An expanded sample of 62 countries, for which capital data are not available, is discussed in Appendix B. The researcher is faced with a non-trivial choice of how to measure price volatility. I am aware of no theoretical reason for assuming that price ratios are either autoregressive or contain a long-term trend. Basu and McLeod (1991) are able to reject a unit root process of terms of trade in 11 of 19 developing countries they study. I proceed along the same lines as Mendoza (1995) by Hodrick-Prescott filtering the price data with a coefficient of 100, a common convention for annual data. I find that filtering the data at other frequencies, or not at all, preserves the qualitative results and ranking of countries by volatility while shifting the point estimates. In order to identify income in an environment of shifting prices, I looked for data on production per worker abstracting from price changes. Unlike measures of Gross National Income, which use a common price deflator to deflate both imports and exports - thus internalizing the relative movement of terms of trade - Gross Domestic Product aggregations deflate each item in 8

the national accounts using its own price, washing out direct terms of trade valuation effects from the income measure. Thus, as is noted by Kehoe and Ruhl (2008), terms of trade changes do not enter directly into output like a technological innovation. To control for country size, I divide GDP by the labor force2 . I then aggregate over the first and last five years of the sample period and take logged differences to compute GDP growth. Henceforth, I use the terms ‘output’ and ‘GDP’ to specify the natural logarithm of GDP per labor force member3 in constant US$. I find that my results are qualitatively robust to other measures of income and population. Over the sample period, 31 of the countries experienced positive perworker output growth while 19 shrank. Cote D’Ivoire and Algeria experienced the worst growth record over the period. Thailand and Mauritius grew the fastest. I find only slight evidence of unconditional convergence in my sample: the correlation of mean output from 1975-1979 with subsequent growth is −0.17. Initial income is not significant in my subsequent growth regressions, and has an inconsistent sign. For capital, I construct a per-worker variable analogous to GDP. Taking investment data since 1960 (or its first year of availability), I use the perpetual inventory method with a depreciation rate of 10% per year to estimate total capital stock, which I then report in per-worker terms.

3.2

Exogeneity of Terms of Trade

An underlying assumption throughout much of the literature on developing country terms of trade is that short-term volatility and even long-term growth of terms of trade are exogenous to small economies. Developing countries have historically exported commodities in which they possess a geographically-determined comparative advantage. Modern comparative advantages are generally independent of local history: mineral deposits and oil fields were often unknown until the 20th century, and agricultural options are dictated by climate, soil, and evolving world tastes. In the 1980’s, primary goods made up the majority of exports for 39 of the 44 countries in my sample for which data is available. In the median country, 76% of exports were primary goods, and no country had less than 30% primary-good exports. Furthermore, the productive processes for agricultural and mineral primary 2

For years prior to 1980, labor force data is not available, so I instrument with population aged 15-64. 3 I will henceforward use ‘worker’ as a shorthand for labor force member.

9

resources are not conducive to rapid increases in production in response to a favorable price shock. At both the extensive and intensive margins, developing countries are constrained from affecting the terms of trade they face in the world. Using data from various years of the Handbook of International Trade and Development Statistics, Kose (2002) shows that 68% of developing country exports are primary goods. These agricultural and mineral products are endowed to specific regions by geography, and often require long-term investment to produce. Riedel (1984) shows that as late as 1978, the top three exports of an average developing country accounted for 52% of its exports, and primary goods in general accounted for 82%. Bidarkota and Crucini (2000) show that three commodities can account for over 50% of the volatility of terms of trade in the typical developing country. Primary exports are rarely good substitutes for home consumables. Thus, the composition of exports for developing countries is largely exogenous to price movements, especially short-term ones. Previous researchers have likewise assumed that terms of trade movements are exogenous to small developing economies. In their review of the growth-regression literature, Barro and Sala-i-Martin (1995) conclude that terms of trade shocks and growth may be treated as exogenous regressors; that is, they are not highly multicollinear with other common indicators. The assumption of exogeneity is also adopted by the authors whose work was discussed above. In addition, the evidence indicates that developing countries are pricetakers in export markets. Using IMF World Economic Indicators data from 1961-1990, Mendoza (1995) cannot reject the hypothesis that the quantity of exports does not cause the terms of trade for a sample of developing countries. Likewise, he finds that the quantity of imports does not cause the terms of trade. Buttressing this finding, Broda (2004) examines the potential for market power among developing country exports and finds that “only 22 goods from nine countries (out of a sample of 1000 goods and 75 countries)” exceed 15% of total world exports of that good. Of these, ten are manufactured goods from China and only nine are primary products. Of countries in my sample, only Cote d’Ivoirian cocoa and Sri Lankan tea stand out, and even these have less than a quarter of the world market. I follow these authors in proceeding with the assumption that terms of trade shocks are exogenous to developing countries. In my data, I find indications of exogeneity as well. One objection to 10

the assumption of TOT exogeneity over the period is that countries might choose to diversify away from harmful volatility. Practically, that could be accomplished by shifting toward exporting manufactures. However, I find that countries that experienced high volatility of terms of trade in the 1980’s saw the share of primary exports rise relative to those with low volatility. Another objection is that if growth in manufacturing, for instance, occurs for orthogonal reasons, the generally lower TOT volatility of manufactured exports would create a positive, but endogenous, correlation of GDP growth and TOT volatility. To test this, I computed TOT volatility over the first half (1980-1994) and the latter half (1995-2009) of the sample period. Correlating these measures with GDP growth over the opposite period, I found no significant differences in the correlations. The magnitudes of both correlations is about -0.2, and varies depending on sample inclusion criteria.

3.3

Linking Volatility to Growth

To measure the correlations between economic growth and exposure to terms of trade volatility, I use the moments described in Subsection 3.1. These include country-by-country growth of GDP and capital from the first and last five years of the sample period and the country-specific volatilities of detrended terms of trade over the same period. I regress the growths in economic aggregates on terms of trade volatility, which I believe to be a source of exogenous variation. The cross-sectional correlation coefficient of GDP growth and the standard deviation of the detrended terms of trade is −0.50. I specify a crosssectional univariate ordinary least squares regression as:   c GDP2005−09 ln = β0 + β1 σTc OT + ǫc c GDP1980−84 In Table 2 I report the univariate regression of GDP growth on terms of trade volatility (2.1), as well as regressions with potentially relevant controls: trade exposure, other forms of volatility, terms of trade growth, regional effects, landlockedness, and previous GDP and capital. In order to control for the effects of population growth, I add the 1960 to 1961 crude birth rate per 1000 people4 . In Regression 2.2, I exclude previous capital, since my 4

I choose the 1960-1961 birth rate as a proxy for population growth so that the measure does not have a structural contribution to the dependent variable. Children born in 1960

11

hypothesis predicts high initial capital as a response to high expected terms of trade volatility. Since terms of trade volatility is transmitted through the share of imports and exports in the economy, I include trade share and an interaction term in Regression 2.4. The interaction term is insignificant. As reported in Regressions 2.1 through 2.3, the coefficient on TOT volatility is negative and significant at the 1% or 5% level, confirming the surprisingly strong correlation. The adjusted R-squared of each of these regressions exceeds 0.2. That is, exposure to terms of trade volatility explains more than a fifth of the variation in output growth. This relationship can be seen in Figure 1. Regression 2.2 and 2.3 show that this result is robust to the inclusion of likely covariates; other controls and samples are reported in robustness checks in Appendix B. In addition, I find that changing the sample of countries to exclude OPEC members or to include a sample of developed economies does not appreciably change the results. Using the mean absolute value of year-over-year change in terms of trade as the measure of volatility leaves my results essentially unchanged. Likewise, using the volatility of unfiltered terms of trade as the regressor preserves the qualitative results. In that case, the correlation between GDP growth and the standard deviation of terms of trade falls to −0.41, and OLS yields a statistically significant (1%) coefficient on exposure to terms of trade variation, with an adjusted R-squared of 0.15. The implication in the unfiltered case is weaker because short-term volatility and long-term growth are conflated. As seen in the regressions, long-term growth of terms of trade has no predictive power on GDP growth. Panel analysis, reported in Table 3 confirms the importance of underlying volatility in driving growth. Regressing annual growth on the annual innovation in terms of trade as well as a set of country characteristics including terms of trade volatility, I find that long-run volatility and annual innovation are of comparable magnitude and significance in explaining annual growth. For comparison, the coefficient on σTc OT in Table 3 corresponds to a negative effect on GDP three quarters as large as that measured in Table 2. Capital growth is strongly correlated with GDP growth (correlation coefand 1961 are adults of working age throughout the sample period (1980-2009). Children born in 1970, by contrast, would enter the denominator of GDP-per-worker in 2009, but not in 1980. In practice, I find that this and other potential measures of population growth have the greatest impact on the coefficient on terms of trade volatility. I conjecture that the model may suffer from omitted variable bias in the absence of a measure of population growth.

12

Table 2: GDP Growth and Terms of Trade

σT OT T RADE σEXCHAN GE µIN F LAT ION σGDP TOT growth BIRTH60 AFRICA L.AMERICA LANDLOCK GDP75−79 K75−79

(2.1) (2.2) (2.3) -4.5*** -3.6*** -3.4** [1.1] [1.2] [1.3] .002 .003 [.002] [.002] .10 .10 [.13] [.13] -.08 -.07 [.05] [.05] 1.0 1.0 [3.3] [3.3] -.02 -.02 [.13] [.13] -.016* -.016* [.009] [.009] -.32** -.33** [.12] [.12] -.01 -.04 [.15] [.16] .07 .07 [.14] [.15] -.15** -.04 [.07] [.26] -.09 [.20]

T RADE ∗ σT OT ¯2 R Countries

0.23 50

0.32 50

0.30 50

(2.4) -2.4 [3.8] .004 [.005] .11 [.13] -.07 [.05] 0.7 [3.6] -.02 [.13] -.017* [.009] -.31** [.13] -.04 [.16] .06 [.15] -.04 [.26] -.08 [.20] -.01 [.05] 0.29 50

* significant at 10% level ** significant at 5% level *** significant at 1% level Dependent variable: Log GDP per worker growth from 1980-84 to 2005-09. Regressands σT OT and σGDP denote standard deviations of the cyclical component of HP(100)-filtered TOT and GDP, reM+X . TOT spectively. Trade share is defined as T RADE ≡ 2∗GDP Growth is percentage growth between the initial and final periods. σEXCHAN GE is the volatility of a demeaned exchange rate index and µIN F LAT ION is the mean absolute value of the rate of inflation. BIRT H60 denotes the birth 13rate per 1000 people in 1960-61. Capital and GDP for 1975-79 are in log per-adult terms.

Table 3: Panel Analysis

σTc OT

-0.071* [0.040] 0.034*** [0.009] 0.019* [0.010] 0.036*** [0.010] 0.242*** [0.068] 4.46e-5 [4.51e-5] -9.96e-5 [8.99e-5] 0.22 52

∆T OT c,t ∆T OT c,t−1 ∆XRAT E c,t−1 c σGDP

T RADE C X1c Xc

¯2 R Countries

* significant at 10% level *** significant at 1% level Dependent variable: Annual log capital per adult growth. Unreported regressands: lags of GDP (t-1,t-2,t-3), time dummies, AFRICA and L.AMERICA dummies, and a constant. ∆T OT c,t is the difference in normalized, unfiltered TOT between years t and t− 1 in country c. X1 is primary commodity exports.

14

Table 4: Capital Growth and Terms of Trade

σT OT T RADE σEXCHAN GE µIN F LAT ION σGDP TOT growth BIRTH60 AFRICA L.AMERICA LANDLOCK GDP75−79 K75−79

(4.1) (4.2) (4.3) -5.8*** -4.7*** -3.7** [1.6] [1.7] [1.6] .002 .005* [.003] [.003] .16 .21 [.18] [.17] -.09 -.06 [.07] [.06] 1.0 1.3 [4.5] [4.2] .00 -.00 [.18] [.17] -.016 -.018 [.012] [.010] -.42** -.42** [.16] [.15] .14 -.06 [.20] [.20] .00 -.00 [.20] [.18] -.36*** .45 [.09] [.32] -.64** [.25]

T RADE ∗ σT OT ¯2 R Countries

0.19 50

0.36 50

0.44 50

(4.4) -4.2 [4.8] .004 [.006] .21 [.17] -.07 [.06] 1.5 [4.5] -.00 [.17] -.018 [.012] -.43** [.17] -.05 [.20] .00 [.19] .45 [.33] -.65** [.25] .01 [.07] 0.43 50

* significant at 10% level ** significant at 5% level *** significant at 1% level Dependent variable: Log capital per worker growth from 1980-84 to 2005-09. See Table 2 for further notes.

15

ficient is 0.88) and is negatively correlated with 1975-79 capital level (−0.45) and initial capital-output ratio (−0.47). This is consistent with my story, though it may reflect measurement error or artifacts of the perpetual inventory method. Capital growth, like GDP growth, is negatively related to terms of trade volatility (−0.46). Figure 2 exhibits the relationship between terms of trade volatility and capital growth, and regressions in Table 4, confirm that TOT volatility has explanatory power, including in the presence of other significant regressors. The inclusion of K75−79 in Regression 4.3 diminishes the effect of TOT volatility on capital to nearly the level of its effect on GDP; this is consistent with my hypothesis of 20th century precautionary home investment. Identifying Productivity. In order to compute a measure of productivity from the data, I assume that the economy takes inputs of capital and labor and produces output according to Cobb-Douglas aggregate technology with total factor productivity z. That is, Y = zK α L1−α . Taking α = 0.35, I apply the production function to the output and capital per adult series and compute TFP for each country, as usual averaging over the first and last five years in the sample and computing the growth between the two. Total factor productivity growth is negatively correlated with terms of trade volatility (correlation coefficient is −0.44), and Regressions 5.1 through 5.4 quantify and confirm the statistical significance of the relationship. The relationship between terms of trade volatility and productivity is illustrated in Figure 3. Interpreting the Regressions. Consider two countries with terms of trade volatility a standard deviation below and above the mean, respectively. Regression 2.2 associates that difference with a 25.4 percentage point difference in GDP growth over the 30-year sample period. Regression 5.2 implies that TFP in the more volatile country will shrink by 14.6% relative to the other. Meanwhile, Regression 4.2 associates the terms of trade volatility difference with a 32.1 percentage point difference in capital growth. The TFP difference, however, has dynamic effects on capital and output. Taking the regression coefficients back to the Cobb-Douglas production function allows us to evaluate the dynamic effects. In order to evaluate the growth effects of a change in TFP, I hold the capital-output ratio constant. Solving the system Y = zK α L1−α K =c Y 16

Table 5: Productivity Growth and Terms of Trade

σT OT T RADE σEXCHAN GE µIN F LAT ION σGDP TOT growth BIRTH60 AFRICA L.AMERICA LANDLOCK GDP75−79 K75−79

(5.1) (5.2) (5.3) -2.4*** -1.9** -2.1** [0.7] [0.8] [0.8] .001 .001 [.001] [.001] .04 .03 [.08] [.09] -.04 -.05 [.03] [.03] 0.7 0.6 [2.1] [2.1] -.02 -.02 [.09] [.09] -.011* -.010* [.006] [.006] -.18** -.18** [.08] [.08] -.06 -.02 [.10] [.10] .07 .07 [.10] [.10] -.03 -.19 [.04] [.17] .13 [.13]

T RADE ∗ σT OT ¯2 R Countries

0.17 50

0.22 50

0.22 50

(5.4) -0.9 [2.5] .002 [.003] .03 [.09] -.05 [.03] 0.1 [2.3] -.02 [.09] -.011* [.006] -.16* [.08] -.02 [.10] .05 [.10] -.20 [.17] .14 [.13] .001 [.032] 0.20 50

* significant at 10% level ** significant at 5% level *** significant at 1% level Dependent variable: TFP growth from 1980-84 to 2005-09. See Table 2 for further notes.

17

18

19

for any constants c and z, α = 0.35, and L = 1 yields specific values of K and Y . When TFP falls by 14.6%, capital and output fall by 21.6%. Likewise, the capital decrease implied by Regression 4.1 will also decrease output dynamically. A residual 10.5 percentage point difference in capital accumulation remains after the balance is explained by the dynamic TFP effect. Using the same Cobb-Douglas production function, the 10.5 point difference in capital growth leads to a 3.8 point difference in output growth. Taken together, the capital and TFP differences would lead to a 25.4 percentage point difference in output growth, equal to the estimate implied by Regression 2.25 . The decomposition here implies that five sixths of the GDP growth difference and two thirds of the capital growth difference are caused by TFP growth differences. The remainder comes through the channel of capital accumulation. Although that portion is small, it is a persistent feature of the data, and it is economically significant in that it can explain a 4-percentage point difference in growth between two developing countries over a period when median growth was only 9%. This paper shows that precautionary savings can account for that difference, and for the observed differences in capital accumulation. Instead of using the *.2 regressions, I could have used the *.1 regressions, without controls, and found largely the same results. When capital stock from 1975-1979 is included as a control (as in the *.3 regressions), there is only a 3% difference in capital growth between a low- and a high-volatility country unaccounted for by TFP. That is consistent with my story that there was an initial buildup of capital in high-volatility countries which has since been dispersed abroad. Thus, when capital just previous to the sample period is included as a control, most of the initial capital buildup is accounted for by the new control. Evidence from initial conditions suggests that capital had been overaccumulated in countries that subsequently experienced high TOT volatility. Capital-output ratio in the early 1980’s is positively correlated (0.24) with overall terms of trade volatility; the same correlation in the late 2000’s is 0.00. The initial condition illuminates Dawe’s 1996 finding that export volatility was linked to rising capital but falling output from 1970 to 1985. Likewise, WDI data show significantly lower interest rates in countries with higher 5 I do not impose a constraint that the productivity-growth effect plus the capitalaccumulation effect must equal the estimated output effect implied from Regression 2.2. Their equality merely indicates that the output, capital, and TFP regressions are internally consistent.

20

terms-of-trade volatility. The model presented in Section 4 will be consistent with over-accumulation of capital during the 20th century, followed by decumulation in the 1990’s and 2000’s as better savings instruments became available to risk-averse developing countries.

3.4

Disciplining Mechanism Choice

In choosing how to model an economy in which terms of trade volatility leads to lower growth, I am constrained not only to match the key correlations that I observe in the data, but to avoid mechanisms that contradict the data. Significant in their absence from my data are strong relationships (a) between terms of trade shocks and output business cycles, (b) between the long-term growths of terms of trade and output, and (c) between terms of trade and export quantities. Thus, I will describe a mechanism by which terms of trade volatility hurts growth but is constrained from using business cycles, terms of trade growth, or export volume as a principal channel through which volatility impacts growth. To test the hypothesis that annual terms of trade shocks cause business cycles, I computed country-by-country time series correlations between the annual cyclical components of GDP and terms of trade. These countryspecific correlation coefficients are widely distributed between −0.47 and 0.75, with a mean and median near 0.12. Their standard deviation is 0.29. Similar results obtain when using unfiltered price data. These results largely confirm those found by Mendoza (1995) over the period 1965-1990 (Mendoza, Table 3). He finds correlation coefficients for 23 developing countries varying from −0.46 to 0.89, with a mean of 0.26, using IMF WEO data and the same HP(100) filter. Kose (2002) reports first moments very similar to mine using earlier World Bank data for 28 non-oil exporting countries (Kose, Table 3). In addition, GDP growth is unrelated to GDP volatility in my sample (see Table 2), further evidence that business cycles are not driving growth. On the other hand, when the entire panel of countries is estimated, a significant link appears between terms of trade shocks (current and lagged) and immediate GDP growth (see Table 3). Thus, I cannot conclude decisively on the business-cycle impact of terms of trade shocks. Using a mechanism where terms of trade volatility drives growth without the intermediation of business cycles distinguishes my model from those of Mendoza, Kose, and Turnovsky and Chattopadhyay. The second notably absent relationship in the WDI data is one between 21

the growth of terms of trade and GDP growth in the long run. This is distinct from the Singer-Prebisch Thesis6 , since in this case the GDP deflator abstracts from relative price movements. The cross-country correlation between the two growth rates is −0.07 and in regressions the relationship is never significant. This is surprising, since terms of trade volatility does have a significant negative impact on GDP growth. Intuitively, one would expect that the realization of terms of trade deterioration would be worse than a mere possibility thereof (that is, expected volatility). This does not appear to be the case. Lastly, I find no statistically significant correlation between terms of trade volatility and primary export quantities. While quantities of both imports and exports of manufactured goods are related to TOT volatility in the same way as domestic output, primary commodity export quantities move more independently, and vary more widely.7 Thus, explanations of the link between TOT volatility and GDP growth should not be based in the expansion or contraction of the primary export sector. The model I will posit in Section 4 is consistent with these limitations: growth is not driven by business cycles, long run TOT appreciation has little effect, and export quantities are constant.

3.5

Globalization

The reallocation of savings from productive domestic capital to foreign assets is central to the model I develop below, and is a notable fact in my data. That reallocation causes GDP to fall most in countries which had the highest levels of precautionary capital savings. Changes in the global financial climate over the past thirty years suggest a world economy that has become more open to cross-border savings and debt. Measurement by Chinn and Ito (2008), Quinn (2003), Miniane (2004), and Lane and Milesi-Ferretti (2008), among others, has established that financial openness increased vastly between the 1980’s and the 2000’s. This 6 The hypothesis, due to Prebisch (1950) and Singer (1950) states that terms of trade between primary and manufactured goods will move systematically in the latter’s favor as technology advances. 7 Regressing primary export quantity growth on TOT volatility yields a coefficient similar in magnitude to those for output growth, but not statistically significant, and with an R-squared of just 0.02. The same regression for manufactured exports yields a 1%significant coefficient with an R-squared of 0.21.

22

included freer movement of financial capital across countries, better insurance markets, more transparent accounting, and fewer capital controls. For the question this paper addresses, the openness of each country to foreign capital is not important. Rather, the openness of developed-world financial markets and availability of desirable assets is important. The spectacular rise in debt held by developing countries has been widely noted, as has its heavy bias toward bonds.8 Less well-known is that total domestic capital in developing countries has fallen sharply over the last ten years, despite their increasing wealth. In my sample of developing countries, I find that high-volatility countries had significantly lower net foreign assets and larger current account surpluses. However, high-volatility countries also show significant growth in net foreign assets, and show large gains in current account relative to low-volatility country. Among the limitations of this paper is its inability to account for the large net debt held by many developing countries.

4

Model

This section describes a model which takes into account the features suggested by the data analysis above. I model a small economy open to trade in goods and populated by a continuum of identical agents who produce two goods. The first good, which I refer to as ‘manufactured’, they produce using inputs of capital and labor; it may be exported, invested as physical capital, or consumed. The second good, termed ‘primary’, is endowed in fix quantity every period, and is in demand on international markets but has no value to domestic residents. Agents can save physical capital in non-negative quantities and, subject to constraints, can borrow and save a riskless internationally traded bond. The small economy is subject to two aggregate shocks: a productivity shock that affects the sector employing capital and labor and a world relative price shock that affects the endowment sector. Preferences. The small open economy is populated by identical, infinitelylived households of measure one who consume a single internationally-traded manufactured good. They value consumption over discrete periods of time according to a period felicity function u(c) that is increasing, concave and 8

For instance, Curcuru et al. (2010) find that 92% of US securities held by 19 emerging economies are bonds of maturity greater than one year.

23

differentiable. Expected lifetime utility is given by "∞ # X U(c) = E β t u(ct ) , u(ct ) =

t=0 1−σ ct

1−σ σ > 1, β < 1.

,

Households do not have preferences over time use, and supply labor inelastically at any wage greater than zero. They discount the future by the subjective discount factor β and their risk aversion, σ, is the inverse of their intertemporal elasticity of substitution of consumption, 1/σ. Production, Endowment and Goods Markets. A large number of identical, risk-neutral, profit-maximizing firms use capital and labor to produce the manufactured good according to the Cobb-Douglas production function Y = zK α L1−α , α ∈ (0, 1), z > 0. The manufactured good may be allocated to immediate consumption, traded, or irreversibly invested as capital. It may be freely imported and exported in unbounded quantities. Without loss of generality, I normalize the world price of the final good to unity in every period. Capital is subject to physical depreciation at a constant rate δ, and may be owned only by domestic investors. Capital is irreversible, and thus may not be consumed in future periods. Total factor productivity, z, is constant for now. The representative household is endowed with a non-transferable stream of a tradable primary good, {x}. The endowment quantity x is constant. The primary good has no value within the small open economy, but faces a horizontal demand curve at an exogenous relative price pxt on world markets. Financial Markets. Households face limited insurance opportunities. They may purchase capital goods for domestic investment or foreign non-contingent bonds subject to accumulation and borrowing constraints. Foreign bonds are risk-free and have constant gross return Rb . Negative foreign bond holdings are interpreted as debt and are subject to a No Ponzi Game condition, 24

limN →∞ (Rb )−N Et [bt+N ] = 0. The gross return on domestic capital, Rk , is net of depreciation and determined in equilibrium. Stochastic Processes. Agents are subject to two shocks. The first is the world price of the primary export, px . It evolves according to a finite Markov process. The second is TFP shocks, z, which evolve according to an independent finite Markov process. These shocks fully characterize the world environment faced by the small open economy; thus a state of the world at time t can be characterized by the tuple (pxt , zt ). Planner’s Problem. A beneficent social planner chooses asset and consumption allocations to maximize expected lifetime utility. The solution to this problem can be decentralized as solutions to the problems of the representative firm and representative agent. The recursive formulation of the planner’s problem when national capital assets equal Kt and bond holdings Bt in state of the world (pxt , zt ), is V (Kt , Bt , pxt , zt ) = ct + Kt+1 + Bt+1 ct Kt+1 Bt+1

max

ct ,Kt+1 ,Bt+1

u(ct ) + βE[V (Kt+1 , Bt+1 , pxt+1 , zt+1 )]

≤ Bt Rb + pxt x + zt Ktα + (1 − δ)Kt , ≥ 0, ≥ (1 − δ)Kt , ∈ [bmin , bmax ]

(1) (2) (3) (4) (5)

where (bmin , bmax ) are the constraints on bond holdings. Let λt denote the Lagrangian multiplier on the capital irreversibility constraint, Equation (4). Equilibrium. An equilibrium in this model is a set of policy functions for capital and bond holdings and prices of capital and labor such that the planner solves Equation (1) subject to (2) through (5). Such a solution can be decentralized as an equilibrium in a competitive economy. Note that the small open economy assumption implies that final and primary goods and foreign bond markets clear trivially at world prices. We look for two specific equilibria by considering solutions to the household’s problem under extreme assumptions on bmin and bmax . First, assume bmin = 0 = bmax , and assume current bondholdings, b, are also zero. Then, after substituting under the Envelope Theorem, the Euler Equation for capital becomes α−1 βu′(ct ) = βE[u′(ct+1 )[1 − δ + αzKt+1 ] − (1 − δ)λt+1 ]] + λt .

The recursivity of λ implies that capital stock is not only a function of the history of states of the economy but also affected by all future states in 25

which capital decumulation might be constrained by irreversibility. In a k competitive decentralization, the gross return to capital is therefore Rt+1 = α−1 [1 − δ + αzKt+1 ] − (1 − δ)λt+1 ]] + λt In contrast, if bond holdings are constrained only by the No Ponzi Game condition, such that bmin and bmax are ‘loose enough’ to be slack for every realization assigned positive probability, the Euler Equation for bonds fixes the ratio of expected utilities: u′ (ct ) = βRb E[u′ (ct+1 )]. In this case, the decreasing returns to investment available in the capital sector are avoided by borrowing or accumulation of bonds. If z is constant, the return to capital is nailed down in the stochastic k steady state, Rt+1 = Rb , since optimal capital level will never change and thus the irreversibility constraint will never bind. In turn, this determines the level of capital in the constant-z stochastic steady state, regardless of shocks to px .

5

Computational Strategy

In order to compute the model, I choose parameters and functional forms in keeping with the literature and to match what I observe in the data, fixing all parameters except the standard deviation of export price shocks. This I vary across countries, matching it to the standard deviation of terms of trade observed in the data. I then employ value function iteration to compute optimal policy functions for each country under financial autarky and under free bond trade. Parameter Choices. The parameters used to compute the model are summarized in Table 6. I follow conventions and previous literature in setting technology and preference parameters. Preferences are characterized by constant relative risk aversion, with the period felicity function given by 1−σ u(c) = c1−σ . I evaluate the model with the risk aversion parameter σ set to 2, a common value used in the literature9 . The annual discount rate β equals 0.96. The return on internationally traded bonds is set at Rb = 1/(0.96 − ǫ) with ǫ greater than zero to support a non-divergent steady state but small 9

See, for example, Lucas (1990), who uses 2.0 as his baseline value, while testing many other values for robustness and discussing the drawbacks of any CRRA coefficient choice.

26

enough to avoid agents short-selling bonds to buy capital in most states of the world. In practice, the range of possible ǫ is very small; I find no difference in computation between ǫ = 0.001 and ǫ = 0.0001. Technology follows the same production function used in the riskless model of Section 3.3. Output is produced following a Cobb-Douglas production with the quantity of labor normalized to unity; hence Y = zK α . The share of income paid to capital is 0.35, which is the developing country average estimated by Bernanke and Gurkaynak (2002). Capital suffers physical depreciation at the rate of 10% per year and is irreversible. Table 6: Parameters

Moment

Value

Source

α

0.35

Bernanke and Gurkaynak (2002)

β

0.96

Convention

δ

0.1

Convention

2

Convention

1/0.959

Calibrated

0.33

WDI Trade as % of GDP

0.83

WDI Net Barter TOT

0.064

Author’s calculations

0.94

Author’s calculations

σ crra R

b

X+M µ( 2∗GDP ) ρ(pxt , pxt+1 ) σ(tf p) µ(tf p)

ρ(tf pt , tf pt+1 ) bmin

−175% of GDP

bmax

550% of GDP

The WDI data provide the remaining parameters. The size of the endowment stream x is calibrated to match the average mean trade share of the sample. I estimate the cyclical component of estimated TFP as an AR(1) process common to all countries, which I convert to a two-state Markov chain following the method due to Tauchen (1986). Likewise, for export price shocks all countries in the simulation share the same persistence and five-state Markov transition matrix. However, the magnitude of the export price shock realizations is taken from the unfiltered WDI data. These shocks are also estimated as AR(1) processes, and converted to 27

five-state Markov processes following Tauchen (1986). For computation, I must pick constraints on bond accumulation even in the open market case. These are set wide enough that very few of the simulated countries hit the bounds. The tightness of the bounds does not appear to affect computational results.

6

Simulation and Results

Using the country-specific policy functions computed per Section 5, I simulate a sequence of relative price shocks and calculate the optimal response of the country to that shock sequence. The years 1965-1994 are simulated (with random TOT and TFP draws) under closed asset markets while 1995-2009 are simulated under open markets.10 After performing this simulation 3000 times for each country, I measure log GDP growth between the periods 198084 and 2005-09 and HP(100)-filtered export price volatility from 1980-2009, thus reproducing the data moments used in Regression 2.1 by employing the same transformations on the simulated data as on the real data. Likewise, I compute capital growth, and reproduce Regression 4.1. Since the model rules out export share influencing GDP through other channels, I can compare these results to the coefficients estimated in Regressions 2.1 and 4.1. As a further exercise, I simulate the model again, this time drawing the terms of trade shocks from the data for the years 1980-2009. The TFP shocks remain random. While this method has its drawbacks, such as the lack of TOT data before 1980, its results are very much in line with the fully random model. In Table 7, I report both the ‘Random’ and the ‘Actual’ simulations. The precautionary motive leads to higher capital savings in higher-volatility countries during the period with closed financial markets. After markets open, all countries revert to the same levels of capital and output, and additional savings and borrowing take the form of bonds. The only crosscountry differences are cyclical and depend on recent TFP realizations. Due to volatile TFP, a risk premium spread of less than one percentage point opens between the return to bonds and the expected return to capital. Regressing simulated GDP and capital growth on simulated terms of trade volatility yields estimates that indicate that a country with two-standarddeviation higher TOT volatility will decrease capital by 10%, thus decreasing 10

In robustness checks, I find that moving the shift to globalized markets a few years in either direction from 1995 did not affect results.

28

Table 7: Growth effects of two-standard deviation difference in TOT volatility

Model TOT Shocks

Data

Random Actual

K growth via accumulation

-10pp

-12pp

-10.5pp

GDP growth via accumulation

-3pp

-3pp

-3.8pp

K/Y ratio growth

-16pp

-18pp

-11pp

3%

4%

2.8% to 5.4%

CA surplus (annual % of GDP) TFP growth

-18%

K growth via TFP

-21.6pp

GDP growth via TFP

-21.6pp

Notes: ‘pp’ denotes ‘percentage point’. Current account surpluses are measured during the last ten years of the sample. Table 8: Average Capital-Output Ratios

Model TOT Shocks

Data

Random Actual

K/Y 1980-84

1.74

1.78

1.75

K/Y 2005-09

1.65

1.65

1.62

output by 3%. Both results are nearly equal in magnitude to the portion of the effects in my data analysis not attributable to TFP growth. Capitaloutput ratios fall in modeled TOT-volatile countries by a larger magnitude than implied by regressions on the data. The estimates from the model and the data are presented in Table 7. In addition, the model closely follows the average capital-output ratios observed in the data, which fell from 1.75 in the early 1980’s to 1.62 in the 2000’s. This is shown in Table 8. Welfare. Recall that the distinction between the two environments is the tightness of a constraint. A social planner, therefore, can always find a weakly better allocation in the less constrained environment. Because the solution to the problems of the representative agent and representative firm are equivalent to a social planner’s, all countries are weakly better off under 29

the less constrained environment. Even though output falls, on average, in moving to the less constrained environment, welfare improves. For 37 countries with consumption data available, consumption-output ratios in the early 1980’s were negatively associated with terms of trade volatility (ρ = −0.38). By the end of the sample, the correlation disappears (ρ = −0.01), and average consumption-output ratio rises. Note that the relative welfare gain of higher C/Y to high-volatility countries are swamped by much larger losses associated with falling TFP; in the model’s context of focusing solely on the precautionary savings portion, a small welfare gain is realized by the more volatile countries. Current Accounts. An implication of the model is that high-volatility countries will run large, persistent current account surpluses after financial integration. Regressions indicate that a country with price volatility one standard deviation above the mean will run an annual current account surplus 3% of GDP larger than a country with price volatility one standard deviation below the mean during the last simulated years. Returning to the data, I observe an effect of similar magnitude. Identifying the same effect is ambiguous: in the model, current accounts during the 1980’s are all zero; in the data, they are generally negative, and vary widely. Thus, I compute both the average current account surplus and the change in average current account surplus. With each method, I average current account per GDP over 5, 10, and 15 year periods ending in 2009 and, for the latter method, I average current account per GDP over a period of equal duration beginning in 1980 or 1975. Then I regress the current account measure on TOT volatility and a basket of controls. The t-statistics on TOT volatility are in the [1,2] range, and the coefficients are economically significant. I associate a two-standard-deviation difference in TOT volatility with 2.8 to 3.3 percentage points higher annual current account, and with 2.8 to 5.4 percentage points larger change in annual current account, depending on the period used. Of the 46 countries in my sample for which I have data, 29 ran current account deficits and 17 surpluses in the 2000’s. By contrast, just 3 averaged a surplus from 1975-1989. Some of the largest relative current account surpluses in the 2000’s belong to hydrocarbon exporters Gabon (16% of GDP), Malaysia (14%), and Venezuela (10%); the largest deficits were those of Nicaragua (17%), Zambia (12%), and Ghana (12%).

30

7

Conclusion

Cross-country differences in terms of trade volatility appear to be a key exogenous factor in explaining the differences in recent growth experiences across countries. This paper showed that a fraction of the output growth effect of terms of trade volatility differences, 4 percentage points, works through the channel of precautionary savings, as well as one third of the capital growth effect, or 10.5 percentage points. Countries with high levels of initial precautionary savings contract relative to those with little precautionary savings as they shift assets from productive domestic capital to international bonds with better yields. A neoclassical representative agent model is able to reproduce the portion of the growth effect attributable to precautionary savings. With parameters drawn from the literature, the precautionary savings shift in the model produces a 10% fall in capital and a 3% fall in output in price-volatile countries relative to less volatile ones. The contraction of high-volatility countries due to portfolio shifts from domestic capital to newly available international bond markets is welfareimproving in a model with no externalities to capital. This should serve as a cautionary note to those interpreting growth regressions: a negative output growth effect may in fact be an improvement when it occurs through optimizing behavior. Though the shift in precautionary savings seen in the data appears to exacerbate the losses due to lower TFP growth associated with TOT volatility, it in fact ameliorates it. The rapid shift in international capital account positions contributes to our understanding of the large current account deficits run by the U.S. in recent years. The model predicts smaller current account deficits, and in some cases current account surpluses, in developing countries during the transition from closed to open international bond markets. In ongoing work, I am investigating channels through which TOT volatility can affect productivity through technology adoption and resource allocation.

31

References Agenor, P.-R. and J. Aizenman (2004). Savings and the terms of trade under borrowing constraints. Journal of International Economics 63 (2), 321 – 340. Aghion, P., P. Bacchetta, R. Rancire, and K. Rogoff (2009). Exchange rate volatility and productivity growth: The role of financial development. Journal of Monetary Economics 56 (4), 494 – 513. Barro, R. J. and X. S. i Martin (1995). Economic Growth, 1st Edition, Volume 1 of MIT Press Books. The MIT Press. Basu, P. and D. McLeod (1991). Terms of trade fluctuations and economic growth in developing economies. Journal of Development Economics 37 (12), 89 – 110. Bernanke, B. S. and R. S. Gurkaynak (2002, March). Is growth exogenous? Taking Mankiw, Romer, and Weil seriously. In NBER Macroeconomics Annual 2001, Volume 16, NBER Chapters, pp. 11–72. National Bureau of Economic Research, Inc. Bidarkota, P. and M. J. Crucini (2000). Commodity prices and the terms of trade. Review of International Economics 8 (4), 647–666. Blattman, C., J. Hwang, and J. G. Williamson (2007, January). Winners and losers in the commodity lottery: The impact of terms of trade growth and volatility in the Periphery 1870-1939. Journal of Development Economics 82 (1), 156–179. Bleaney, M. and D. Greenaway (2001, August). The impact of terms of trade and real exchange rate volatility on investment and growth in sub-saharan africa. Journal of Development Economics 65 (2), 491–500. Broda, C. (2004, May). Terms of trade and exchange rate regimes in developing countries. Journal of International Economics 63 (1), 31–58. Burnside, C. and A. Tabova (2009, August). Risk, volatility, and the global cross-section of growth rates. NBER Working Papers 15225, National Bureau of Economic Research, Inc.

32

Caballero, R. J. and K. N. Cowan (2007). Financial integration without the volatility. Working paper series, Massachusetts Institute of Technology. Chinn, M. and H. Ito (2008, September). A new measure of financial openness. Journal of Comparative Policy Analysis 10 (3), 307–320. Curcuru, S. E., T. Dvorak, and F. E. Warnock (2010). Decomposing the U.S. external returns differential. Journal of International Economics 80 (1), 22 – 32. Special Issue: JIE Special Issue on International Macro-Finance. Dawe, D. (1996). A new look at the effects of export instability on investment and growth. World Development 24 (12), 1905 – 1914. Ghosh, A. R. and J. D. Ostry (1994). Export instability and the external balance in developing countries. Staff Papers - International Monetary Fund 41 (2), 214–235. Kehoe, T. J. and K. J. Ruhl (2008, October). Are shocks to the terms of trade shocks to productivity? Review of Economic Dynamics 11 (4), 804–819. Koren, M. and S. Tenreyro (2007, 02). Volatility and development. The Quarterly Journal of Economics 122 (1), 243–287. Kose, M. A. (2002, March). Explaining business cycles in small open economies: ‘How much do world prices matter?’. Journal of International Economics 56 (2), 299–327. Lane, P. R. and G. M. Milesi-Ferretti (2008, May). The drivers of financial globalization. American Economic Review 98 (2), 327–32. Lucas, Robert E., J. (1990). Supply-side economics: An analytical review. Oxford Economic Papers 42 (2), pp. 293–316. Lutz, M. (1994). The effects of volatility in the terms of trade on output growth: New evidence. World Development 22 (12), 1959 – 1975. Mendoza, E. G. (1995, February). The terms of trade, the real exchange rate, and economic fluctuations. International Economic Review 36 (1), 101–37. Mendoza, E. G. (1997, December). Terms-of-trade uncertainty and economic growth. Journal of Development Economics 54 (2), 323–356. 33

Miniane, J. (2004, August). A new set of measures on capital account restrictions. IMF Staff Papers 51/2, International Monetary Fund. Prebisch, R. (1950). The Economic Development of Latin America and Its Principal Problems, Volume 1. United Nations. Quinn, D. P. (2003). Capital account liberalization and financial globalization, 1890-1999: a synoptic view. International Journal of Finance and Economics 8 (3), 189–204. Ramey, G. and V. A. Ramey (1995, December). Cross-country evidence on the link between volatility and growth. American Economic Review 85 (5), 1138–51. Riedel, J. (1984). Trade as the engine of growth in developing countries, revisited. The Economic Journal 94 (373), 56–73. Romer, P. M. (1986, October). Increasing returns and long-run growth. Journal of Political Economy 94 (5), 1002–37. Singer, H. (1950). The distribution of gains between investing and borrowing countries. American Economic Review 40, 473–85. Tauchen, G. (1986, March). Finite state Markov chain approximations to univariate and vector autoregressions. Economic Letters 20 (2), 177–181. Turnovsky, S. J. and P. Chattopadhyay (2003, March). Volatility and growth in developing economies: some numerical results and empirical evidence. Journal of International Economics 59 (2), 267–295. van der Ploeg, F. and S. Poelhekke (2009, October). Volatility and the natural resource curse. Oxford Economic Papers 61 (4), 727–760. World Bank (2011, May). World Development Indicators. World Bank.

34

A

Bootstrapped Estimates

A source of potential inefficiency in estimates of standard errors in the paper’s regressions is introduced by the use of estimated independent regressors, particularly σT OT . In order to check the robustness of the estimates, I bootstrap the results from the detrended terms of trade data, sampling with replacement 500 times for each regression. The results of Regressions (1b), (5b), and (7b) - analogous to (1), (5), and (7) - are presented in Table 9. I find that point estimates of the coefficients fall slightly, but standard errors fall even more. Table 9: Bootstrapped Estimates

Dependent variable σT OT Countries

(2.1b) GDP growth -3.8*** [0.7] 54

(4.1b) K growth -5.8*** [0.7] 54

(5.1b) TFP growth -2.3*** [0.3] 54

*** significant at 1% level

B

Robustness

I check the robustness of my estimates and choice of regressands in Table 10. Regression 10.1 duplicates Regression 2.1 with 9 additional developing countries for which capital data was limited. The new countries are Burkina Faso, Burundi, Cape Verde, Mozambique, Niger, Nigeria, Panama, Togo and Uganda. Using this 59-country sample, I include the volatilities of export and import volume as potential correlates; they have little explanatory power. Regression 10.3 includes the volatility of the component parts of terms of trade: the unit values of exports and imports. These offer little explanatory power, perhaps because import and export price indices are highly correlated: ρ(PM , PX ) = 0.54 in a panel of 110 countries. For Regression 10.4, I add three industrial Asian countries (China, Korea, and Singapore) and six resource-rich Western countries (Australia, Canada, Netherlands, New Zealand, Norway, and the United States). Their inclusion confirms the original estimates of the impact of terms of trade volatility, and markedly increases the explanatory power of terms of trade volatility. 35

Table 10: Robustness Regressions

σT OT

(10.1) -3.1*** [1.0]

(10.2) -2.7** [1.1]

σPX σPM σXvolume

(10.3) -3.0** [1.3] -0.5 [1.6] 1.3 [1.6]

1.0 [0.7] -.59 [.89] 1.7 [3.1] -.07 [.05] -.34*** [.13] -.17 [.13]

σM volume σGDP µIN F LAT ION AFRICA L.AMERICA WESTERN LANDLOCK ¯2 R Countries

0.13 59

.13 [.13] 0.18 59

0.11 59

(10.4) -3.3*** [1.2]

1.5* [0.7] -.83 [.98] 2.0 [3.4] -.07 [.06] -.47*** [.14] -.26* [.14] -.21 [.19] .14 [.14] 0.28 68

* significant at 10% level ** significant at 5% level *** significant at 1% level Dependent variable: Log GDP per worker growth from 1980-84 to 2005-09. σXunitvalue is the volatility of an HP(100)-filtered index of export prices; σXvolume is the volatility of an HP(100)-filtered index of export quantities. Import measures are analogous. See Table 2 for further notes.

36

In further regressions on a smaller sample (unreported), I included primary commodity share of exports (1980-89), and found that it does not significantly affect my results.

37

Terms of Trade Volatility and Precautionary Savings in ...

Mar 31, 2012 - shocks to explain business cycles, but not growth. A broader .... I proceed along the same lines as Mendoza ...... Bt in state of the world (px.

627KB Sizes 1 Downloads 240 Views

Recommend Documents

Terms of Trade Volatility and Precautionary Savings in ...
Oct 21, 2010 - countries, reproducing my finding from the data. ... All errors are my own. ... This creates a negative relationship between export price volatility ...

Terms of Trade Volatility and Precautionary Savings ...
with high terms of trade volatility, which later shifted their portfolios away from domestic capital and into foreign bonds. The reallocation of precautionary savings ...

Trade, the Precautionary Principle, and Post-Modern Regulatory ...
TTIP has been welcomed by the business communi- ties on both ... the one hand, the European Parliament has called for the TTIP not .... Even if not all negotiation sessions ..... Analysis”, Office of Management and Budget, 17 September. 2003 ...

terms of trade pdf
Page 1. Whoops! There was a problem loading more pages. Retrying... terms of trade pdf. terms of trade pdf. Open. Extract. Open with. Sign In. Main menu.

terms of trade pdf
There was a problem previewing this document. Retrying... Download. Connect more apps... Try one of the apps below to open or edit this item. terms of trade ...

Trade, the Precautionary Principle, and Post-Modern Regulatory ...
Jul 18, 2013 - I. Introduction ..... 30 Karel De Gucht, “Speech – Transatlantic Trade and Investment .... could also take hold in the US and start to influence.

Terms of Trade Uncertainty and Business Cycle ...
The right row of Figure 1 displays the monthly growth rates of copper and .... where the mean and the variance of the terms of trade determine the savings rate.

Trade, the Precautionary Principle, and Post-Modern Regulatory ...
allowing the marketing in the EU of products com- plying with US regulations would .... ubiquitous internet and Wikileaks?23 Perhaps, the ne- gotiating parties' ...

The Consumption Terms of Trade and Commodity Prices
trade shares helps us isolate the source of a nationps terms of trade varia' tion in the ..... estimates are inflation rates, in U.S. dollars, of a particular good, i, Api,t,.

Precautionary Bidding in Auctions
IN MANY REAL WORLD AUCTIONS the value of the goods for sale is subject to ex post ... Econometric evidence based on data from timber auctions is provided ..... is for example the case for competing internet auction websites), then it may be ..... Har

Determinants of Consumption and Savings Behavior in ...
relationship between the real interest rate and consumption. The evidence for the Hall ... Lakshmi Raut is an assistant professor of economics at the University of California,. San Diego. ..... cannot be accepted in our tests. This rejection may be .

Formalization and applications of the Precautionary ...
renewable energy sources is an act which does not correspond to an .... (2). We immediately see that µ∗. F is a non additive probability on P(Ω) satisfying. µ∗.

Precautionary Demand and Liquidity in Payment Systems
Aug 1, 2010 - In large-value real-time gross settlement payment systems, banks rely heav- ily on incoming ... a high degree of coordination and synchronization. We construct a ... McAndrews and Potter (2002) give a detailed account .... satisfied wit

Precautionary Demand for Education, Inequality, and Technological ...
This paper offers an explanation for the evolution of wage inequality within and between industries and education groups over the past several decades. The model is based on the disproportionate depreciation of technology- specific skills versus gene

Precautionary Demand and Liquidity in Payment Systems
... those of the. Federal Reserve Bank of New York or the Federal Reserve System. ... Every member maintains an account which contains: b ..... us to analyze:.

Precautionary Demand and Liquidity in Payment Systems
Aug 1, 2010 - Association for Public Economic Theory, IESE Business School, Bank ..... their daylight overdraft capacity, a small number of institutions found their net ...... 800. 900. Eastern Time. Queued payments. Bank A. Bank B. Bank C.

The Volatility of the Extensive Margin of Trade under ...
using disaggregated data on bilateral intraiEMU exports. The covered .... using STAN (the OECD trade and industry database), we collect the total bilateral.

The Volatility of the Extensive Margin of Trade under ...
analyzing the business cycle properties of the extensive margin of intraiEMU .... Flam and Nordstrom (2006) show that the creation of the Euro has led to an increase .... (1 ") 9t (Pc,t#% ... Similar expressions are used for the foreign country. ....

Income Uncertainty and Household Savings in China
household-level data to identify the effect of employment displacement on the ...... rate for a male retiring at age 60 to decline to about 60, 55 and 50 percent of the ..... viewed as an illustration of the predictive content of a stylized model tha

Trends in Health Savings Account Balances, Contributions ...
Jul 11, 2017 - 2. • Annual 2016 contributions are higher the longer an account owner had ..... /national-survey-of-employer-sponsored-health-plans-2016.html.

Trade Integration and the Trade Balance in China
changes in technology, trade costs, and preferences accounting for the dynamics of China's gross and net trade ... Keywords: Trade Integration, Trade Balance, Real Exchange Rate, International Business. Cycles, Net ... models have been shown to best

Precautionary price stickiness - CiteSeerX
Nov 22, 2010 - Matejka, Mirko Wiederholt, and seminar participants at the Bank of Spain, .... Most alternative frameworks, including the Calvo and the menu cost model, ..... holdings with interest rate Rt −1; Tt represents lump sum transfers .....