Terms-of-Trade Uncertainty and Economic Growth: Are Risk Indicators Significant in Growth Regressions?

Board of Governors of the Federal Reserve System

International Finance Discussion Papers Number 491

December 1994

TERMS-OF-TRADE UNCERTAINTY AND ECONOMIC GROWTH: ARE RISK INDICATORS SIGNIFICANT IN GROWTH REGRESSIONS?

Enrique G. Mendoza

NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors.

ABSTRACT

This paper examines a neoclassical stochastic endogenous growth model in which terms-of-trade uncertainty affects savings and consumption growth. The model explains the positive link between growth and the average rate of change of terms of trade found in recent empirical studies. In addition, terms-of-trade variability, as an indicator of risk, is found to be a key determinant of growth. This implies that welfare costs cf uncertainty are much larger than conventional measures of costs of consumption instability. The

model's key predictions are strongly supported by results of panel regressions.

Terms-of-Trade Uncertainty and Economic Growth: Are Risk Indicators Significant in Growth Regressions?

Enrique G. Mendozal/

1.- Introduction

Many recent empirical studies in growth theory examine the nature of cross-country growth differentials and the economic forces that explain them.2/ These studies produce statistical evidence on the contribution of macroeconomic policies or country characteristics, vis-a-vis exogenous variables, to explain d-fferences in average growth rates across industrial and developing countries. The results of these stucies generally suggest that policy indicators and country characteristics have a variable degree of significance, while the terms of trade are typically a robust determinant of economic growth.

Easterly et. al (1993), for instance, find that economic policies and country characteristics, such as education and political stability, contribute little to explain the observed lack of persistence in growth performance, while terms-of-trade changes are highly correlated with growth, particularly during the 1980s.3/ Barro and Sala-i-Martin (1994) and Fischer (1983) find that country characteristics contribute to explain growth differentials, but the terms of trade continue to play a key role. Barro and Sala-i- Martin's results show that the growth effects of terms of trade compare to those of educational attainment, public spending on education, human capital, and political instability. Fischer shows in addition that the terms of trade are more important at higher frequencies, since they are more significant

in pooled than in between-means regressions.4/

a l/ The author is a Staff Economist, Board of Governors of the Federal Reserve System. Helpful comments and sugges:ions by Michael Kouparitsas, Ravi Kumar, Jong-Wha Lee, Sergio Rebelo, Carmen Reinhart, and Jorge Roldos are gratefully acknowledged. This paper represents the views of the author and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System. 2/ See, for example, Barro (1991), Barro and Lee (1993), Barro and Sala-i-Martin (1994), Razin and Yuen (1994), Easterly and Rebelo (1993), Fischer (1993), and Easterly, et. al (1993). 3/ Poolec! regressions show that if the terms of trade gain as a share of GDP increases by 1 percent per annum, anrual growth rises by 0.42 percent in the 1970s and by 0.85 percent in the 1980s. 4/ This explains why a smoothed measure of terms of trade May not contribute to explain growth in panels

of country means (see De Gregorio (1992)) and suggests that the variance of terms of trade may be relevant for explaining growth.

-2-

While the positive relationship between terms of trade and growth has been clearly identified, most empirical work does not conduct structural tests tied to a theoretical framework in which tke role of terms of trade is explicitly modelled. The explanatory power of country characteristics and policy variables is related to existing growth models, whereas the terms of trade are viewed as an exogenous variable with a somewhat uninteresting role. Moreover, growth effects that could result from uncertainty and risk aversion in the presence of terms-of-trade fluctuations, as well as uncertainty with regard to policy variables and country characteristics, are generally ignored.

This paper attempts to shed some light on these issues by studying a model of savings uncer uncertainty that provides an interpretation for the observed positive relationship between average rates of change of terms of trade and average consumption growth rates. In addition, the model highlights the importance of the variability of terms of trade as a determinant of average growth rates, and thus illustrates potentially important growth effects of uncertainty and risk. These growth effects in turn have important quantitative implications for the welfare costs of uncertainty.

Although the recent empirical growth literature does not discuss much the theoretical basis for the link between terms of trade and growth, there exists a large literature on trade theory dealing with this issue, as reviewed in Findlay (1984). However, the approach followed in this paper differs markedly from the classic models of the trade literature (see Lewis (1954), Prebisch (1950), Singer (1950), and Findlay (1980)) in which the link between terms of trade and growth follows from the existence of laboror goods-markets imperfections. Instead, in this paper the effect of terms of trade on growth reflects the effects of uncertainty on the savings behavior of risk-averse agents in a friction-less neoclassical setting. 1/

The main argument that the paper makes with regard to growth effects of terms-of-trade uncertainty is a feature of growth models of varying complexity, but it remains a feature even of simple

stochastic growth models. Thus, for clarity and technical simplicity, the analysis is based on a model of

1/ The role of market imperfections is likely to be relevant, particularly for explaining growth differentials between industrial and developing countries. However, it is also useful to determine to what extent market forces can contribute to explain the stylized facts.

-3a small open economy that extends the savings-under-uncertainty framework of Phelps (1962) and Levhari and Srinivasan (1969), which is also a one-sector, stochastic version of the endogenous growth model of Rebelo (1991). In the model, the mean and variance of the terms of trade determine the savings rate and consumption growth. Growth is slower in economies in which terms of trade grow at slower rate, on average, because slow terms-of-trade growth reduces the expected real rate of return on savings--in units of imported goods--and this affects the savings rate.

The variability of terms of trade also affects the savings rate and growth, with an effect that is positive or negative depending on the degree of risk aversion. If risk aversion is low, increased variability in the terms of trade, measured as a mean-preserving spread, reduces both growth and social welfare. However, if risk aversion is high, increased terms-of-trade variability produces faster grow at the expense of reduced welfare. Thus, the model predicts that the variance of the terms of trade, as an indicator of risk, contributes to explain growth, and hence suggests that cross-country growth regressions that include only the mean of the rate of change of terms of trade are mispecified.

Given the growth effects of terms-of-trade uncertainty, the welfare gains that result from reducing consumption instability are much larger than suggested in Lucas (1987). Lucas' experiment, as well as sir.ilar exercises in open-economy models (see, for example, Cole and Obstfeld (1991), Mendoza (1991) and Backus et. al (1992)), abstract from the possibility that altering the variance of consumption may not only affect the amplitude of consumption fluctuations around trend, but the level of that trend as well. In the savings-under-uncertainty framework, risk-averse agents adjust their savings rate, and hence the trend level of consumption, in response to changes in the variability of the underlying process driving consumption fluctuations. A similar proposition has been examined in the context of the assessment of the welfare gains of international risk-sharing by Obstfeld (1992). When international diversification affects growth, the gains from risk-sharing are significantly larger than those produced by models of business cycles around exogenous trends (as in Tesar (1994)).

The paper examines the empirical relevance of the model's key predictions using a multi-country

database including 40 industrial and developing countries and using panel estimation methods. The

-4-

model's closed-form solutions establish two hypotheses regarding consumption growth in comretitive equilibrium: (a) a positive time-series linear relationship between the rate of change of terms of trade and consumption growth, and (b) a cross-section relationship such that country average growth rates depend positively on the average gross rate of change of terms of trade and negatively on the variance. These hypotheses are found to be strongly supported by the data. In particular, the variance of the terms of trade includes information relevant for explaining growth not captured in the mean.

The paper is organized as follows. The next section presents the model and conducts m-merical simulations to illustrate the model's welfare implications. Section III examines cross-country empirical regularities of growth and terms of trade, and conducts econometric tests of the model's closed-form solutions. The last section provides some conclusions.

2.- The Terms of Trade and Growth: A Basic Framework

This section examines a basic neoclassical model that provides an interpretation for the relationship between terms of trade and growth under uncertainty. The model extends the savingsunder-uncertainty framework developed by Phelps (1962) and Levhari and Srinivasan (1969) to the case of a small open economy facing terms-of-trade shocks. The model can also be interpreted as a stochastic variant of the one-sector endogenous growth model based on a linear technology developed by Rebelo (1991).1/

The economy is inhabited by households that formulate optimal plans for consumption of an

imported good so as to maximize expected lifetime utility:

c/* 1-y (1)

U(C) - E| YS Bt tO

y>0, 0<B<1

1/ Hopenhayn and Muniagurria (1993)and Obstfeld (1992) examine two-sector applications of the linezr technology setting.

-5where C;, is consumption of the imported good, 8 is the subjective discount factor, and y is the coefficient of relative risk aversion (i.e. 1/y is the intertemporal elasticity of substitution).

The production technology adopts the simple form of a perfectly durable asset, or a linear technology, that yields a stochastic return each period.1/ The return is an exportable commodity that agents exchange for imports of consumer goods in a perfectly competitive world market, where they are a few of a large number of participants. The relative price of imported goods in terms of exports is also subject to random disturbances. Markets of contingent claims are incomplete, and hence households cannot insure away country-specific income shocks resulting from changes in real returns or the terms of

trade. Thus, households maximize utility subject to the following period-by-period resource constraint:

Aw. s R,(A,-p,C,) (2)

Given Ay>0. A, is the stock of wealth in units of exportables, R, is the domestic gross rate of return on savings, and p, is the relative price of imports in terms of exports determined in world markets, or the reciprocal of the terms of trade tot,=p,’. Both R, and p, are non-negative random variables such that the effective rate of return r,=R,p,/P,,, follows a log-normal i.i.d. distribution. Thus, Ln(r) is an i.i.d. probabilistic process with mean p and variance o”, and hence the mean and variance of the r, process are #,=exp(i+ 07/2) and o,’=4,(exp(o7-1)). At each date t, p, is known but R, and p,,, are unknown.

The competitive equilibrium is defined by optimal intertemporal consumption allocations that

maximize (1) subject to (2).2/ The optimality conditions are the constraint and the Euler equation:

R UC) - Ry C1) (3)

fol

1/ Although the paper examines a simple representation of the model, the same basic growth equation results if one allows for investment and depreciation in continuous time (see Rebelo (1991) and Hopenhayn and Muniagurria (1993)). International borrowing and lending in one period bonds, yielding the same return as the domestic linear technology, can also be added without altering the results.

2/ An analysis of a similar model for the case of tariffs and trade reforms of uncertain duration is undertaken in Calvo and Mendoza (1994).

-6- Following Phelps (1962) and Levhari and Srinivasan (1969), closed-form solutions for this model are

obtained using dynamic programming techniques. The solutions are:

C- {4 4)

Ay, = G-ARA, (5) Where: 1 1 = - R,P A = L-oifen) r= (9) Pi And lifetime welfare is: Ve A, ly Vv (A,.P,) a (7) (l-y) \ P,

The constant 4 is the marginal propensity to consume with respect to wealth, and r, is the real interest rate in units of importables, or the consumption-based rate of return. Under the feasibility condition that E[r,'"] < 6’, consumption in each period is a positive fraction of the real value of asset holdings in units of importables, and savings (i.e. assets carried over to the following period) are a positive fraction of the gross return on initial asset holdings. Notice also that, because the terms of trade are known but the return on exportables is unknown when C, is chosen, the actual realization of R, does not affect consumption, while the actual realization of p, does.

It is straightforward to show that, since agents cannot insure themselves against fluctuations in r,, increased risk in consumption-based asset returns (i.e. a mean-preserving increase in o,” due to increased

variability in the rate of return on exportables or the terms of trade)1/ leads to reduced savir.gs and

V/ For o,’ to increase while keeping y, unchanged, it must be the case that o” increases in such a way that pu is adjusted to keep »+07/2 constant.

-7increased consumption if the coefficient of relative risk aversion is lower than 1 (y <1), or the intertemporal elasticity of substitution is greater than 1 (1/y > 1), and that an increase in the mean return has the opposite effects. These results are derived by expressing A as a function of the mean and the

mean-preserving variance of asset returns:

y= pup} onl o¢| 8)

In order to examine the growth implications of terms-of-trade uncertainty, the analysis focuses on the probabilistic process that governs logarithmic first differences of consumption in the competitive equilibrium. This is equivalent to adopting a first-difference filter to separate the trend and cyclical components of consumption. The use of this detrending procedure is appropriate in this case because, given (4), (5) and the statistical properties of r,, consumption growth can be expressed as:

—! . (1-Ay, (9)

t

where r, is log-normal. Denote the log first-difference of consumption as AC,=Ln(C,)-Ln(C,,), and define Ln(r,)=+€,, SO that €, is the period-t deviation of the log of the real interest rate from its mean. Then, it follows from

(9) that AC is: 2 AC, - - xB rtntu | [A-vetS + ey ao)

AC is a proxy of the growth rate of consumption.1/ The first two terms in the right-hand side of (10) define the trend of AC, and the error term ¢ is the cyclical component. The variance of the cyclical component is given by o”. However, because Ln(r,) is white-noise, business cycles in this economy do not display persistence and the correlation between shocks to asset returns, or to the terms of trade, and

OO rrr

1/ The average growth rate reflected in the expected value of C,,,/C, becomes a poor proxy for the growth rate measured by the average of AC,,, as o” rises because, since r, is log-normal, Ln(E[r,])-E[Ln(r,)] =07/2.

-8-

fluctuations in savings or consumption is perfectly positive. These co-movements are in sharp contrast to what is observed in the data at high frequencies.1/ Thus, the strong assumptions needed to generate closed-form solutions produce implications that render this model inappropriate for business cycle analysis.

The model does provide some interesting implications for the low-frequency relationship) between terms of trade and growth. Equation (9) shows that, even in a simple framework of i.i.d. log-normal shocks, the rate of consumption growth is determined by the savings rate and the rate of change of the terms of trade. In addition, equation (10) shows that the relationship between terms of trade and growth can be expressed simply in terms of the parameters governing the savings rate: (a) the rate of time preference 1/f-1, (b) the elasticity of intertemporal substitution 1/y (or the degree of risk aversion y), (c) the average rate of return on savings p,, and (d) the inherent riskiness of domestic asset returns and the terms of trade o’.

It is important to note that although whether the intertemporal elasticity of substitution is greater or less than unitary determines if changes in p, or o” have positive or negative effects on the /evel of consumption (see equations (4) and (8)), an increase in p, always induces an increase in the growth rate, regardless of the size of y (see equation (10)). This result implies that countries in which the terms of trade grow at a faster rate on average also experience faster average consumption growth. Moreover, (10) also shows that a mean-preserving increase in o” (i.e. an increase in risk associated, for instance, to increased variability in the terms of trade) induces a fall (rise) in consumption growth as long as y <2 (y >2). Thus, when the degree of risk aversion is relatively high, an increase in risk increases. the average growth rate. However, since welfare in (7) is declining in o’, it is also true that welfare in such a high-risk, fast-growing economy is lower than in a low-risk, slow-growing economy with the same degree of risk aversion.

The potential empirical implications of this growth model can be quantified by imposing: some

benchmark parameters in the closed-form solution (10) and exploring the results of numerical

1/ For a quantitative analysis of terms of trade and business cycles see Mendoza (1995).

-9simulations. The model is calibrated to create a benchmark economy that conforms roughly to some empirical evidence for developing economies. Parameter values for the discount factor and the mean of asset returns are set to B =0.95 and u,=1.07. B=0.95 is the value that Lucas (1987) used to calculate welfare losses of business fluctuations. The econometric evidence from Ostry and Reinhart (1992) would suggest that this value of B is biased downwards relative to estimates for developing countries, but it is a convenient benchmark for illustrating how the results of Lucas’ welfare analysis are altered by growth effects of terms-of-trade shocks. The real interest rate at 7 percent is consistent with historical evidence documented in Mehra and Prescott (1985) for the mean real interest rate on risky assets in industrial countries, which is a good proxy if developing countries are assumed to be small open economies.

Table 1 lists the model's equilibrium savings rate, 1-4, the growth rate of consumption, gc, and welfare effects of uncertainty, W, for various combinations of o and y, given B =0.95 and y,=1.07. The table includes results for y=2.33, which corresponds to the GMM estimate obtained by Ostry and Reinhart (1992) for Latin America, and o=0.12, which is the standard deviation of the log of the terms of trade for Latin American countries in Mendoza (1995). The various parameterizations produce consumption growth rates in the range between 0.33 and 3.7 percent, which includes the average growth rates of real consumption per-capita for several developing countries over the last two decades. In particular, the Latin American benchmark (o =0.12 and y =2.33) results in an average growth rate of 0.94 percent, in line with observed average per-capita growth rate of consumption in that region. The Savings rates range between 93.8 and 98.1 percent, which imply values for the marginal propensity to consume with respect to wealth ranging between 1.9 and 6.2 percent.

In the two cases that y <2, Table 1 shows that consumption growth falls as terms-of-trade variability rises. If y=1/2, growth falls by more than 1.5 percentage points as o increases from 0 to 0.15, whilz if y=1.5 growth falls by only 1/2 of a percentage point. In the first case the savings rate declines by 1/2 of a percentage point, but in the second case the savings rate in fact rises by 0.5. This is

because, as equations (8) and (10) show, both growth and the propensity to consume fall as risk

-10increases when 1 <y <2, whereas when y <1 growth falls but the propensity to consume rises as risk increases. For y=2.33, both the savings rate and growth increase as o rises, the former rises from 94.1 to 95.5 percent and the latter increases from 0.7 to 1.1 percent. Similar implications follow in the case that y=5, except that the effects are stronger--the savings rate rises from 93.8 to 98.1 percent and the growth rate increases from 0.3 to 3.7 percent. When y >2, growth is faster the more variable are the terms of trade, but since welfare in (7) is decreasing in y, this faster growth is accompanied by a reduced level of welfare as explained below.

The third column in each of the four panels of Table 1 reports the welfare costs resulting from the uncertainty of terms of trade and real asset returns. The social costs of uncertainty are measured as in Lucas (1987) by computing compensating variations in time-invariant consumption paths that render households indifferent between a risk-free consumption path and the consumption path of a risky environment, measuring risk by the size of o. After some manipulation of equation (7), the welfare cost

as a function of o, given y,, is:

a W(o) - -1, (11)

Thus, the welfare costs of uncertainty in this model are a function of how uncertainty affects the propensity to consume relative to the risk-free, Pareto-optimal case.

The figures reported in Table 1 show in general that variability in domestic asset returns or in the terms of trade is very costly. If one considers the benchmark welfare costs of 1/10 of a percentage point obtained in Lucas (1987), the results show that the savings-under-uncertainty model produces significantly larger welfare losses, except for cases in which o=0.01.1/ For cases in which 0 =:0.12. the welfare costs range from 10.1 percent when y = 1/2 to 106.1 percent when y=5. This compares to the costs of 12-percent standard deviation of consumption in Lucas (1987) that range from 0.65 to 13.6

eee

1/ If o=0.013, y=1.00001 and y, is adjusted to produce consumption growth of 3 percent, the model produces a welfare cost of uncertainty of 0.0016, compared with 0.00008 under similar assumptions in Lucas (1987).

-ll-

percent as y rises from 1 to 20. Part of the large difference between these results and those of Lucas is accounted for by the smaller standard deviations in some of Lucas’ experiments. However, most of the difference is due to the fact that consumption behavior here is the outcome of an optimization exercise in which uncertainty affects not just fluctuations of consumption around a trend growth rate, but that trend growth rate itself. Lucas’ computations, in contrast, are based on a hypothetical consumption function that abstracts from the distortionary effects of o on the propensity to consume and on consumption growth. L/

The welfare costs in Table 1 for the two cases in which y >2 illustrate the quantitative implications of the model's result that increased risk induces faster growth and lower welfare. When y =2.33, the welfare costs of uncertainty rise from around 2/10s of a percentage point to 62.2 percent as o increases from 0 to 0.15, even though this also implies that the average growth rate of the economy rises by nearly 1/2 of a percentage point from 0.7 to 1.1 percent. 3.- Empirical Analysis

This section documents empirical regularities that characterize terms of trade and real per-capita consumption growth in 40 industrial and developing countries over the last two decades. The data are then used to conduct econometric tests based on the model's closed-form solutions. The emphasis on consumption growth, rather than GDP growth, follows from the structure of the endogenous growth model presented in Section 2. The countries studied include 9 industrial countries (the group of seven largest iridustrialized countries, G-7, plus Australia and Spain) and 31 developing countries from different regions of the world.2/

The classification in terms of developing and industrial countries follows the World Economic Outlook (International Monetary Fund, 1994). The 40 countries are also classified as commodity

exporters or diversified exporters using the classification system of the WEO, with the aim of exploring

oe eeeeeeeEeeeSSSSSSFSSSSSSSSSeeeeSSSSSE 1/ Another important difference with Lucas (1987) is that in the savings-under-uncertainty model welfare

costs of wcertainty are zero if y=1 because in this case A is independent of o. Values of y near 1 produce

small weliare losses, but these reflect the fact that the exponent in equation (1 1), which goes to infinity as y

approaches 1, amplifies even minuscule differences in propensities to consume. 2/ The full list of countries is provided later in Table 3:

-12-

whether the export base is a determinant of growth effects of terms-of-trade uncertainty.1/ The data were obtained from the World Bank's World Tables using the Socio-economic Time-series Access and Retrieval System (STARS). The data are annual time series of private consumption at constant and current prices from national accounts; average U.S. dollar exchange rates; U.S. dollar import and export unit values; and total population. The sample covers the period 1970-1991.

To be consistent with the model's structure, imports are chosen as the 'numeraire,' and thus terms of trade are defined as the ratio of export to import unit values.2/ Consumption is expressed in per-capita terms and deflated using two price indices: import unit values and the consumption deflator. The econometric tests are conducted with both measures. Consumption at import prices is the measure consistent with the model, and consumption at consumer prices is used to examine the influenc2 of valuation effects. Growth rates are approximated using logarithmic first differences.

The empirical analysis begins with a general illustration of the potential connection between terms of trade and growth. For this purpose, Charts 1 and 2 plot actual data and Hodrick-Prescott trends of GDP and terms of trade for two countries chosen randomly (Canada and Kenya). Hodrick-Prescott trends suggest the presence of low-frequency trend shifts in the data. The charts highlight important features that distinguish GDP growth and terms of trade developments in many industrial countries from those observed in most developing economies during the sample period. Canada displayed sus:ained growth, with moderate deviations from trend, and experienced improvements in the long-run rzlative prices of her exports, while in Kenya the opposite phenomena occurred. The increasing (decreasing) trend of the terms of trade in Canada (Kenya) reflects to a large extent the protracted and severe decline

of real commodity prices over the last two decades (see Reinhart and Wickham (1994)).

1/ Diversified exporters include all the industrial countries and developing countries in the sample that WEO classifies as exporters of manufactures or diversified exporters. Commodity exporters include developing countries in the sample that WEO classifies as exporters of fuel, non-fuel primary products, and exporters of services or recipients of transfers. WEO classifications are based on average export shares for 1984-86 (for example, a country is a fuel exporter if fuel exports were more than 5O percent of total exports on averaye in 1984-86).

2/ See Mendoza (1995) for a discussion on alternative measures of the terms of trade.

-13-

Charts 3 and 4 plot scatter diagrams of country averages of consumption growth rates and the means and standard deviations of the rates of change of terms of trade, including simple regression lines. These chart provide visual evidence in favor of the model's key predictions, which are tested more formally below. Chart 3 shows that consumption growth is slightly faster in countries where the rate of change of terms of trade is higher--a well-known fact highlighted in the empirical growth literature. Chart 4 shows the less well-known regularity that growth tends to be slower as the variability of terms of trade increases.

As noted in Section 2, the closed-form solutions (9) and (10) provide a framework for testing two hypotheses: (a) the time-series hypothesis that consumption growth rates and rates of change of terms of trade are positively related over time within each country, and (b) the cross-section hypothesis that the variability of the rate of change of terms of trade, as a measure of risk, provides relevant information for explaining growth not captured by the mean. These hypotheses are tested using the multi-country data base described above and the panel estimation techniques typical of the modern empirical growth literature.

The goal of the econometric analysis is limited to establishing whether there is evidence to support or reject the two hypotheses derived from the model, rather than providing a comprehensive analysis of the determinants of growth. In particular, the tests do not consider any variable other than the terms of trade for explaining growth, so panel regressions are not expected to produce high R? or adjusted-R? statistics. Also, since other determinants of growth are not included, results supporting the model cannot rule out the possibility that the significance of terms-of-trade captures the explanatory power of other "true" driving forces of growth. Note, however, that Easterly et. al (1993), Fischer (1993), and Barro and Salz.-i-Martin (1994) find that terms-of-trade contribute to explain growth even in the presence of variables that measure country characteristics and economic policies. Moreover, when these authors address simultaneity problems and apply instrumental variable methods, the implications of the terms of

trade ar2 unchanged if they are treated as a truly exogenous variable. Thus, there is strong evidence to

-14support the view that the contribution of the terms of trade to explain growth can be examined ina simple bi-variate framework.

The hypotheses are tested jointly with other important assumptions of the model (i.e. that consumption growth rates and rates of change in terms of trade are log-normal stationary processes, that the marginal propensity to save with respect to wealth is a time-invariant positive fraction, and that this marginal propensity to save varies across countries with structural parameters and with the mezn and variance of the terms of trade). The empirical tests also assume that the gross domestic real ra:e of return (i.e. R,) is a stationary process with mean R (where R> 1), and independent of the process governing terms-of-trade shocks.

The validity of the assumptions that consumption growth rates and the rates of change in terms of trade in each of the 40 countries are white-noise processes is assessed using Box-Jenkins methods and augmented Dickey-Fuller tests. Since the data are expressed as logarithmic first differences, they are expected to be stationary and display only weak serial autocorrelation. The tests generally support the view that the data are stationary, but for some countries the Box-Jenkins method could not reject the hypothesis that there are significant, albeit low, first-order autocorrelations. To explore the issue of stationarity further, the panel tests discussed below were also performed adding linear trends. These were not statistically significant and did not affect the results markedly. Thus, the results presented in the tables exclude the linear trend.

3.1 Hypothesis 1: Do Terms of Trade Explain Growth?

Tables 2-9 report the results of several panel regressions that test the hypothesis that the rate of change of terms of trade is a determinant of consumption growth. The tables list results for panel models based on total or pooled, between-means, fixed effects, random effects, and independent regressions with the aim of exploring the validity of the hypothesis for each country's time series as well as across countries. This also provides evidence as to whether the structure of the link between terms of trade and growth is homogeneous across countries, and helps establish a close link with the rece:tt

literature on empirical growth studies. Results are reported for tests that use consumption growth

-15measured at import and consumption prices, and they are also applied to sub-samples arranged according to geographical groups and the structure of exports.

The results show that in general the simple endogenous growth model of Section 2 performs well when confronted with the data. Consider first the panel results in Table 2. Regardless of whether data at import or consumer prices are used, total, fixed, and random effects models show that terms of trade are a Statistically significant explanatory variable of consumption growth, although short of the unitary coefficient that the model predicts. An increase of 1 percentage point in the rate of change of terms of trade increases the growth rate of consumption at import prices by 0.2 percent and that of consumption at domestic prices by about 0.05 percent. The weaker effect on the latter is suggestive of the implications of relative price changes on the consumer goods basket. Adjusted R? statistics are generally low, as expected because the role of growth determinants other than the terms of trade is not considered.

In contrast with the other panel models, the between-means models in Table 2 reject the hypothesis that the country average of rates of change in terms of trade explains country average growth rates. This is consistent with Fischer's (1993) finding that the role of terms of trade is more significant in pooled regressions than in between means regressions. Moreover, as will be shown in 3.2, the poor results of the between-means model reflect the fact that relevant information contained in the variability of the terms of trade is missing. Panel models that include a time-series dimension capture some of this information, and this explains in part their good performance.

The results in Table 2 also show that F-tests reject the hypothesis that intercept and slope coefficients of the pooled and fixed effects models are equal to those of independent regressions. The failure of these homogeneity tests is not surprising given large cross-country differences between slope and int2rcept estimates in the independent regressions of Tables 3-4. This in turn reflects the model's prediction that the propensity to save varies with the degree of risk aversion, the subjective rate of time

preference, and the mean and variance of the terms of trade, all of which should be expected to differ

-16across countries.1/ In general, however, the negative results of many of the independent regr2ssions cast doubt on the relevance of equation (9) as a model of annual changes in growth rates. Even when these regressions produce significant coefficients for the terms of trade, the coefficients tend to be in excess of 1, which is the value predicted by the model. Thus, as noted earlier, the model of Section 2 seems inadequate for studying high frequency features of the data. It is surprising to note that even at the annual frequency the model performs significantly better for industrial countries than for developing economies, for which terms of trade are widely believed to be a key factor in explaining growt and business cycles. For some G-7 countries, the model explains more than 75 percent of the annual changes in per capita consumption growth at either consumer or import prices. It is also interesting to note that, with the mean rate return on risky assets set at 7 percent, as estimated by Mehra and Prescott (1985), the value of 4 implied by intercept estimates of the independent regressions of industrial countries is A=0.034, which is close to what the calibrated version of the model discussed in Section 2 implies.2/

If the results of independent regressions are compared across geographical regions or be:ween commodity and diversified exporters, a certain degree of homogeneity emerges in coefficient estimates. For instance, many industrial country intercepts in Table 3 are clustered in the range 0.016-0.045, whereas developing countries have smaller intercepts--some even negative values--which are suggestive of lower savings rates. The results for panel models based on data organized by regions (Tables 5 and 6) or export base (Table 7) also show much less heterogeneity on regression coefficients than the general panel regressions. Between-means models continue to perform poorly regardless of the manner in which the data are organized, and of whether consumption is measured at consumer or import prices. Pooled, fixed effects, and random effects models using data at import prices detect strong positive growth effects of terms of trade in the Western Hemisphere but not in Asia, Africa and the Middle Eats, while the data eee

1/ Evidence on differences in the degree of risk aversion and the subjective discount factor across grours of developing countries is reported in Ostry and Reinhart (1992). Evidence on large cross-country differences in

terms-of-trade variability is reported in Mendoza (1995). 2/ A is solved for from the equation 0.033 =(1-a)R, where R=1.07.

-17at consumer prices suggest the opposite. In both cases the results for industrial countries show a significant effect of terms of trade on consumption growth--in Table 6 it is even the case that the hypotheses that the savings rate and the growth effects of terms of trade are similar across industrial countr.es cannot be rejected.

The results for data organized according to export base in Table 7 indicate that regardless of the deflatcr used to measure consumption, the terms of trade are a significant determinant of growth for both commodity-based exporters and non-commodity-based exporters. Thus, the model is more robust to the ordering according to export base than to the ordering according to geography. As before, the model performs generally better when applied to industrial countries, or to diversified exporters, than when applied to developing countries or commodity exporters.

Table 8 provides results of the panel models for two sub-samples, 1971-1980 and 1981-1991, using data organized according to export based and both consumer and import price deflators. The results indicate that the relationship between terms of trade and growth has not been stable over time, and that growth-effects of terms of trade were generally stronger in the 1980s than in the 1970s, as previously argued by Easterly et. al (1993). Surprisingly, when the data are broken down into two subsamples, the models for commodity-based exporters fail to detect a statistically significant link between the terms of trade and growth. The slope coefficients are slightly different from those of the full sample panels in Table 7, so the low t-statistics are mostly due to larger standard errors in the smaller samples used .n Table 8.

3.2 Hypothesis 2: Does Risk Contribute to Explain Growth?

Equation (10) establishes two cross-sectional predictions with regard to growth effects of terms of trade, given preference and technology parameters. First, a higher average gross rate of change of terms of trade increases growth. Second, increased risk, measured as a mean-preserving spread of the stochastic process driving the rate of change of terms of trade, reduces (increases) growth if y <2 (y >2). The assumption that the rate of change of terms of trade follows a log-normal process is critical

for this decomposition of growth determinants in terms of mean returns and risk. The two propositions

- 18 and the log-normality assumption are tested here by running cross-sectional regressions using time-series averages and variances from the multi-country database described earlier.

As in the case of the tests reviewed in 3.1, the tests have implicit the model's assumptions regarding linear technology, isoelastic preferences, and stationary, log-normal random shocks. However, the cross-sectional approach provides an alternative means for examining the robustness of the assumption of log-normal disturbances explicitly. If Ln(r,) is white-noise, with mean p; and variance o; for country i, r, should be white-noise with mean y,,=exp(u;+0,7/2) and variance o,;°=,;(exp(o;-1)). Estimates of 4, can be constructed for each country by computing time-series averages of the ratio P,+,/P,. and the log of these averages can be regressed on (u,+0;/2), where yz, and o, are country i's time-series mean and variances of Ln(p,,,/p,). If the joint hypothesis that the slope coefficient of this regression is equal to | and the intercept is equal to zero cannot be rejected, the log-normality assumption is supported by the data. The results of this regression produce a slope coefficient of 1.13 with a heteroskedastic-consistent t-statistic of 10.5 and an 85 percent adjusted R?. The hypothesis that this coefficient is not different from 1 and the intercept is not different from zero cannot be rejected at the 5 percent level--the Wald F-statistic is 1.538 with a probability level of 0.22.1/ Thus, the data cannot reject the hypothesis that the rate of change of terms of trade follows a log-normal distribution.

The results of the cross-sectional regressions testing equation (10) are reported in Table 9. The regressions use country averages and variances of the log first difference of the terms of trade (u, and 0,7 for country i) and country averages of per capita consumption growth at consumer and import prices, calculated over the full sample 1971-1991, a panel of two sub-samples, 1971-1980 and 1981-1991, and the two sub-samples separately. The dependent variable in each regression is average per capita consumption growth. To be consistent with the mean-preserving spread decomposition expressed in equation (10), the independent variables are the log of the average gross rate of change of terms of trade

Ln(y,,)--proxied as (4,+6;7/2) given the results of the log-normality test--and the variance of the log first-

V/ Excluding Iran and Venezuela, where violation of log-normality is clear from visual inspection of the data, the slope coefficient is 1.01 with a heteroskedastic-consistent t-statistic of 71.4 and an R’* of 0.999.

-19difference of the terms of trade o,7. Thus, the coefficient on o; measures the growth effect of a change in the variance of the rate of change of terms of trade keeping the mean constant.

Table 9 reports, in addition to regression coefficients, the results of Wald F-tests on crosscoefficient restrictions implied by the closed-form solution (10), results of White Heteroskedasticity Tests using cross-products of explanatory variables, and adjusted R? statistics. Results are based on Heteroskedastic-consistent standard errors, although these were not substantially different from ordinary least squares standard errors.

Tie regression results of Table 9 provide strong support for the cross-sectional predictions of the model. The coefficient estimates are generally statistically significant and with the correct sign. The coefficients on the mean gross rate of change of terms of trade, which correspond to the intertemporal elasticity of substitution 1/y, are 0.47 and 0.34 for the full-sample models based on data at import and consumer prices respectively. These coefficients are highly significant, although the tests of crosscoefficie:1t restrictions that link them to the intercepts and the coefficients on the variance of the terms of trade produce poor results. The basic implication that risk, reflecting the variability of the terms of trade, is a key determinant of growth is also strongly supported by the data. According to full sample regressions, a mean-preserving increase of 1 percentage point in the variability of the terms of trade reduces growth by slightly more than 1/2 of a percentage point for both consumption at import prices and at consumer prices.1/ The two full-sample models also explain 10 and 17 percent of the crosscountry clifferences in average growth rates respectively.

The results of the sub-sample models suggest that there have been structural changes in the relationship between terms of trade and growth, as indicated earlier by the results of the panel regressions in 3.1. The growth effects of the mean and variance of terms of trade are significantly smaller during the 1970s than during the 1980s. In the case of consumption at import (consumer) prices,

an increase of 1 percent in the measure of risk associated with terms-of-trade uncertainty reduces growth OO eee 1/ Note, however, that the coefficients on the variance of terms-of-trade growth imply estimates of 1/y

around | end 1.25, which are inconsistent with the values of 0.47 and 0.34 implied by the coefficients on the mean.

-20by more than 1.5 (0.5) percentage points in the 1981-1991 regression, compared to only about 0.25 (0.07) percentage points in the 1971-1980 regression. Compared to the full-sample model with data at import prices, allowing for this structural change by running a cross-sectional regression combining the two sub-samples increases the growth effect of risk from -0.6 to -0.8 and allows the model to account for nearly 1/2 of differences in cross-country growth rates, instead of 10 percent.

The failure of the cross-coefficient restrictions should not be viewed as a significant drawback. The failure of the restriction on the intercept, which states that the intercept should be equal to the coefficient on the mean rate of change of terms of trade times the log of the subjective discourit factor (assumed at 0.99), may reflect missing relevant information pertaining to real domestic asset returns (R, in the model of Section 2). Suppose, for instance, that domestic assets yield a risk-free real return of 4 percent in units of exportable goods. The intercept should then be equal to the coefficient on the mean of terms-of-trade growth, times the logarithmic sum of the discount factor and the risk-free domestic rate of return. The Wald test for this restriction using the full sample regression with import price data cannot reject the hypothesis that this restriction holds (the F-statistic is 1.521 with a probability value of 0.225). Thus, the rejection of the intercept restriction is in fact a positive aspect of the results indicating that terms-of-trade risk and return factors are not the only determinants of consumption-based asset returns.

The restriction that links the coefficients on the mean and variance of terms of trade examines whether the value of y implied by the coefficient on the mean is consistent with that implicit ir: the risk effect captured by the variance. The failure of this restriction suggests that, while the simple stochastic endogenous growth model proposed here is a good first approximation, it does not account for all aspects of the link between growth and uncertainty. One important missing element is preference and technology parameter uncertainty. The restriction on the mean and variance coefficients assumes that tastes and technology are identical across countries, but this is unlikely to be true. For instance, the estimates of the intertemporal elasticity of substitution of the full-sample and joint sub-sample models of

Table 9, ranging between 0.23 and 0.59, are very close to the developing country pooled estimates

-21produced by Ostry and Reinhart (1992) at 0.38 or 0.5. However, Ostry and Reinhart also found significant differences in the value of this elasticity in Asia relative to Africa and Latin America. 1/

As a way Of illustrating the potential implications of differences in economic structure, the fullsample models of Table 9 were re-estimated introducing a dummy variable set at 0 for industrial countries, 1 for Asian countries, and 2 for all other countries. The F-statistics for the cross-coefficient restriction tests linking coefficients on the mean and variance of terms of trade are now 4.70 and 2.47 for data al import and consumer prices respectively, compared to 9 and 4.4 in Table 9. For data at import (consumer) prices the probability value associated to the F-statistic is 0.04 (0.22). In both cases the restriction cannot be rejected at the 1 percent confidence level. Thus, there is rough evidence suggesting that cross-country differences on preference and technology parameters may account for the failure of the cross-coefficient restrictions.

Another restrictive aspect of the model that could help explain the failure of the coefficient restriction between p, and o” is the assumption of the isoelastic utility function that forces risk aversion and intertemporal substitution to be governed by the same parameter y. The failure of the restriction could be interpreted as evidence suggesting that this assumption is rejected by the data, thus favoring an alternative specification of preferences such as the non-expected utility function proposed by Epstein and Zin (1991). Obstfeld (1992) shows that, in a linear technology setting with log-linear uncertainty similar to the one examined here, the use of non-expected utility results in similar closed-form solutions, except that the ccefficient on o” in the growth equation is a function of the product of risk aversion and (1 minus) int2rtemporal substitution. Only in the case that the latter is truly the reciprocal of the former the

result of expected utility applies. Moreover, increased risk reduces growth only if intertemporal

Or

l/ These authors also found that the decomposition of consumption between traded and nontraded goods affects estirnates of the intertemporal elasticity of substitution. The model of Section 2 abstracts from this sectoral issue.

-22 -

substitution exceeds unity, regardless of the degree of risk aversion.2/ The larger the intertem oral elasticity of substitution relative to the degree of risk aversion , the stronger the negative effect of risk on growth. Thus, if non-expected utility holds, and if intertemporal substitution exceeds both unity and the reciprocal of the coefficient of risk aversion, one could explain the negative coefficients on o° larger in absolute value than the positive coefficients on p, reported in Table 9.

4.- Concluding Remarks

This paper examines the growth effects of terms-of-trade uncertainty using a one-sector, stochastic model of endogenous growth for a small open economy. The model predicts that (a) within a country, consumption growth is linearly related to the rate of change of terms of trade over time, and (b) across countries, average consumption growth is positively related to the average rate of change of terms of trade, and positively or negatively related to the variance depending on the degree of risk aversion. These predictions follow from the model's closed-form solutions that highlight the role of terras of trade as a determinant of the risk and return properties of domestic assets, and hence as a determinant of savings and growth. Thus, the link between terms of trade and growth is derived here as a feature of a neoclassical savings-under-uncertainty framework, without recourse to the market rigidities emphasized in the classic trade models of terms of trade and growth.

Numerical simulations examine the potential growth and welfare effects of terms of trade uncertainty. If the coefficient of relative risk aversion is smaller (greater) than 2, a mean-preserving increase in terms of trade variability reduces (increases) growth. In both cases increased uncertainty reduces social welfare. Estimates of the welfare costs of uncertainty are much larger than those produced in business cycle models because the benefits of reducing consumption instability affect not

only consumption fluctuations around trend, but the trend level of consumption as well.

1/ In Obstfeld (1992) the condition for higher risk to reduce growth is y > 1, not 2, because he focuszs on the growth rate defined as the log of the expected value of Ct+1/Ct, instead of the expected value of the log of Ct+1/Ct. As noted earlier, under log-normal uncertainty, these two measures of growth are not identical. The emphasis on the second measure in this paper reflects the approach followed in the empirical growth literature, which generally uses log-first differences or net growth rates.

-23 -

Econometric methods provide strong support for the model's two key hypotheses, although strong coefficient restrictions are generally rejected. The terms of trade are found to be significant in panel regressions of per-capita consumption on the rate of change of terms of trade for 40 industrial and developing economies, reflecting the findings of recent empirical growth studies. Between-means models gererally fail, but pooled, fixed effects, and random effects models produce very favorable results. Also, the link between terms of trade and growth is found to be stronger during the 1980s than during the 1970s, for non-commodity-based exporters than for commodity-based exporters, and for industrial countries than for developing countries.

The paper's main empirical finding is the strong evidence in support of large growth effects resulting from the variance of the terms of trade as an indicator of risk. Between-means regressions of country averages of consumption growth rates on averages and variances of the rate of growth of terms of trade, produce statistically significant coefficients and high levels of explanatory power. The variance of terms of trade has a negative effect on growth, and includes information relevant for explaining growth not included in the mean. However, cross-coefficient restrictions implied by the model fail reflecting some of its weaknesses. In particular, the model abstracts from considering determinants of growth other than the terms of trade, and its econometric application assumes identical preference and technology parameters across countries. Tests aimed at controlling for parameter uncertainty using dummy variables produce results where cross-coefficient restrictions hold.

The proposition that indicators of risk are relevant for growth can be extended to other explanatory variables typically emphasized in empirical growth analysis. Thus, further empirical research could elaborate on the role of risk and uncertainty using a more comprehensive stochastic growth model. This research could also explore the implications of allowing for heterogenous preference and technology parameters vis-a-vis alterative specifications of utility and production

functions.

-24- References

Backus, David K., Kehoe, Patrick J. and Kydland, Finn E. (1992), “International Real Business Cycles,” Journal of Political Economy, August, 100, 745-775.

Barro, Robert J. "Economic Growth in a Cross-Section of Countries," Quarterly Journal of Economics, v. 106, 407-443, 1991.

___ and Jong-Wha Lee, “International Comparisons of Educational Attainment," Journal of Monetary Economics, v. 32, no. 3, 363-394, 1993.

___ and Sala-i-Martin, Xavier, Economic Growth, forthcoming 1994.

Calvo, Guillermo A. and Mendoza, Enrique G., "Trade Reforms of Uncertain Duration and Real

Uncertainty: A First Approximation, IMF Staff Papers, International Monetary Fund, Washington D.C.., December 1994.

Cole, Harold L. and Maurice Obstfeld, "Commodity Trade and International Risk-Sharing: How Much Do Financial Markets Matter?" , Journal of Monetary Economics 28, 3-24, 1991.

De Gregorio, Jose, "Economic Growth in Latin America," Journal of Development Economics, v. 39, 59-84, 1992.

Easterly, William, Kremer, Michael, Pritchett, Lant and Summers, Lawrence, "Good Policy or Good

Luck? Country Growth Performance and Temporary Shocks," Journal of Monetary Economics, v. 32, no. 3, 459-484, 1993.

and Rebelo, Sergio, "Fiscal Policy and Economic Growth: An Empirical Investigation," Journal of Monetary Economics, v. 32, no. 3, 417-458, 1993

Epstein, Larry G. and Zin, Stanley, "Substitution, Risk Aversion, and the Temporal Behavior of

Consumption and Asset Returns: An Empirical Analysis," Journal of Political Economy, v. 99, 263-286, 1991.

Findlay, Ronald, "The Terms of Trade and Equilibrium Growth in the World Economy," American Economic Review,v. 70, 291-299, 1980.

, "Growth and Development in Trade Models," in Handbook of International Economics, ed. by R. Jones and P. Kenen, North Holland, Amsterdam, 1984.

Fischer, Stanley, "The Role of Macroeconomic Factors in Growth," Journal of Monetary Economics, v. 32, no. 3, 485-512, 1993

Hopenhayn, Hugo A., and Muniagurria, Maria E., "Policy Variability and Economic Growth," Economics Working Paper 30, Universitat Pompeu Fabra, Barcelona, February 1993.

International Monetary Fund, World Economic Outlook, Washington: International Monetary Fund, May 1994.

Lee, Jong-Wha, "Capital Goods Import and Long-Run Growth," mimeo. Korea University. March 1994.

-25-

Levhari, D. and T.N. Srinivasan,"Optimal Savings Under Uncertainty," Review of Economic Studies, 153-163, 1969.

Lewis, W. A., "Economic Development with Unlimited Supplies of Labor," Manchester School, v. 21, 139-191.

Lucas Jr , Robert E., Models of Business Cycles, Basil Blackwell, New York, NY., 1987.

Mehra, Rajnish, and Edward C. Prescott, "The Equity Premium: A puzzle," Journal of Monetary Economics, vol.15, 145-161, 1985.

Mendoza, Enrique G., "Capital Controls and the Gains from Trade in a Business Cycle Model of a Small Open Economy," IMF Staff Papers, v. 38, no. 3, 480-505, 1991.

____., "The Terms of Trade, the Real Exchange Rate and Economic Fluctuations," International Economi: Review, February 1995.

Obstfeld, Maurice, "Risk-Taking, Global Diversification, and Growth," Discussion Paper No. 61, Institute tor Empirical Macroeconomics, Federal Reserve Bank of Minneapolis, March 1992.

Ostry, Jonathan D. and Reinhart, Carmen M., "Private Saving and Terms of Trade Shocks," IMF Staff Papers, September 1992, 39, 495-517.

Phelps, Edmund S., "The Accumulation of Risky Capital: A Sequential Utility Analysis," Econometrica v. 30, No.4., 729-743, 1962.

Prebisch, Raul, The Economic Development of Latin America and Its Principal Problems, United

Nations, New York, 1950.

’

Razin, Assaf and Chi-Wa Yuen, "Convergence in Growth Rates: The Role of Capital Mobility and

Internaticnal Taxation," in Capi i estment and Growth, ed. by L. Leiderman and A. Razin, Cambridge University Press, 1994

Rebelo, Sergio, "Long-Run Policy Analysis and Long-Run Growth," Journal of Political Economy, v. 99, no. 3, 500-521, 1991.

Reinhart, Carmen M. and Peter Wickham, “Commodity Prices: Cyclical Weakness or Cecular Decline," IMF Working Paper, 1994.

Roldos, Jorge E., "A Long-Run Perspective on Trade Policy, Instability, and Growth," in The Effects of Protecrioni : , ed. by M. Connolly and J. De Melo, World Bank, Washington DC, 1994.

Singer, H. W., "The Distribution of Gains Between Borrowing and Investing Countries," American Economic. Review, v. 40, 473-485, 1950.

Tesar, Linda L., "Evaluating the Gains from International Risksharing," mimeo, Department of

Economics, University of California-Santa Barbara, 1994 (forthcoming, Carnegie-Rochester Conference Series on Public Policy).

World Bank, World Tables, Socio-economic Time-series Access and Retrieval System, version 1.0, The World Bank/IFC, Washington D.C., June 1994.

- 26 - Chart 1. Canada: GDP and Terms of Trade

3.10

©.10

Gross Domestic Product

3.00

— Actual -- Trend

2.90

2.70 2.70

2.60 2.60 2.50

2.40

2.30 1955

1958 1961 1964 1967 1970 1973 1976 1979 1982 1985 1988

0.05

Terms of Trade

0.00

— Actual --Trend

-').05

-().10

-0.15

-0.20

-0.25

-0.25 1955 1958 1961 1964 1967 1970 1973 1976 1979 1982 1985 1988

-27-

Chart 2. Kenya: GDP and Terms of Trade 1.50 — 1.50 Gross Domestic Product — Actual 1.40 -- Trend + 1.40 1.20} 4 1.30 1.204 1.20 1.10 1.10 1.00 1.00 90 0.90 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1.20 —— 1.20 1.00} . 1.00 0.80} “ts | 0.80 0.60 sy 0.60 “sah 0.49 Ys 40.40 Terms of Trade *s < 0.2) — Actual oe 0.20 --Trend “s 0.09 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989

‘sapoo UOT BOIIssejo ATJUNOD S.J] aJe sjaqe] WBIP—eIY], “saoiid JouNsUOD Je painseaw UONdwNsuUOo, :910N

dpe] JO suI9] Jo sBueYd Jo a}e1 80°0 900 600 c0'0 0 c00-

real OST AHS! p99 €72

ocpep OP

gst BLS

(166L—-IL6L ‘soseraae AQUNOD) Apel] JO SUA! JO YON PUB YIMOIH UO!duINsUOD ‘E HeYO

b00- 90'0- €0'0-

c00-

10°0-

100

00

£0°0

r0'0

$00

900

L00

qyao33 uondunsuoo eyideo Jad

- 29 -

AU

se0

‘sapoo ATNNUNOD Sy] ale sain3y WZIp—seIy], ‘sao1d JouNsuod Je painseow e1eG :210N dPPI] JO SUI9] JO UONIADpP plepur}s €0 Sal) 70 st ‘0 10 $00 0 €0'0-

c00-

100-

100

700

£00

{wMo13 uonduinsuos eqideo sad

r0°0

$00

Cc. ras <>

(1661 —LZ6I ‘sederoae Aunod) apes] JO SWS] Jo A]IGeUeA oy} pUe YMaID UONdWINSUOD “p WeYD

30 -

‘uged uotjdumsuos ssaT-ysTaA e pue Oo ASTI yITM yqed uotydumsuoo e useMQeq JUeTeFFTPUT p[OYssnoy ayQ seaeeT AeYyQ weerQs uoTqAdwunsuoo AAeuoT{eIS eB uT a8ueyo ey se peinseeu AQVUTeQAV0UN Jo YSOD VsAPFTOM 9YQ ST M pue ‘UOoTQdUMSUOD Jo BsoUaTeFFIP-AsATZ-BoT aun Jo uvauw oul st °8 ‘(sitesse TeTATUT uO UANIeI ssoiZ 9Ua Jo poAeS UOTIDeAI 9uUQ ‘a'T) 93eI SBuTAeS 9UI

ST (Y-T) ‘UOTSZ0Ae YSTA VATIETAET Jo QUueTOTJZJa09O 9yQ st A ‘(A)UT JO UOTIeTASp pAepueAsS aeyQ ST O :a]70N

co9e'€ OLEO'O 8086°0 £129°0 ZOTO'O £€656°0 €9SE°O €S00°0 720S6°0 84ST°'O0 6ST0'0 2096'0 STO €T90°T 6720°0 0S96°0 cOvE'O 7600°0 20S6°0 LOT?'O0 €400°0 28%6°0 TTOT'O 6120°0 ¢296°0 aaa?) €é/S'O0 €810°'0 9956°0 98120 2800°0 %7L76°0 SOvT'O 7800°0 2276 °0 cOL0°O0 2S20'0 €€96°0 O1'O 880€°0 6¢T0'0 46%6'°0 8TET'O T800°0 2@5%76°0 0480°0 €600°O €976°0 0S70°O0 64£20°0 11796'0 80°0 97ST°O 2800°0 %776'°0 6040°0 9400°0 €76'°0 £L70°O0 OOTO'O 456%76°0 €S20°0 O0€0'O 81796'°0 90°0 £€90°0 2S00°0 90%6°0 S0oc0O'O0 €£00°0 ¢2%76°0 8020°0 SOTO'O 2S%6°0 €T10°O STEO'O €596°0 70°0 cSTO'0 6£00°0 78€6°0 $100°0 TZL00°0 *176°0 ¢S00°0 80T0°O0 6%776°0 8200°0 2€0'0 9596°0 c0°0

8€00'0 7€00°O0 82t6°0 6100°0 OZ400°0 2t76'°0 €T00°O0 6010°0 617176°0 £000°0 92£0°O0 4596°0 10°0

-- €£00'0 9L€6'0 -- 0L00°0 ZIv6'0 -- 6010°'0 8%776°0 -- /Z7€0'0 2S96°0 0°0 fay °3 (X-T) A °3 (X-T) iy 93) (XT) fay °3 (X-1) a ae rs cect CT== COs F ° (AT) (III) (II) (1)

(40°'T = 47 ‘¢6'0 = 9)

SUOTIE[NUTS Tepom :Y4QmM0IH pue ‘ores TOM ‘sSutaes uo AQUTeQA90UN JO SQOaTJA “T 9TqGeL

- 31 -

Table 2. Consumption Growth and the Rate of Change of the Terms of Trade: Panel test Results 1/

—______—_—_—_—_———— reese

Intercept Slope F-Test Against Total Independent R2 Data at import prices Total 0.006 0.201 -- 2.844% 0.048 (1.128) (6.498) * Between means 0.006 0.162 -- -- -- (1.143) (0.916) Fixed effects -- 0.202 0.699 4.855* 0.001 (6.404) * Randon effects 0.006 0.202 -- -- 0.001

(0.865) (6.367)*

Data at consumer prices

Total 0.018 0.048 -- 1.810* 0.011 (7.535)* (3.211)* Betwe2n means 0.018 0.076 -- -- -- (5.435)* (0.624) Fixed effects -- 0.047 1.470* 2.073* -- (3.147)* Random effects 0.0178 0.048 -- -- --

(6.098)* (3.133)*

————————— nt eereerrer

l/ Intercept and slope coefficients for regressions of consumption growth on the rate of change of the terms of trade and a constant. Numbers in bracsets are t-statistics. An asterisk denotes statistical significance at the 5 percent level. Results for independent time-series regressions are reported in Tables 3 and 4.

-~ 32 -

Table 3. Real Per-Capita Consumption Growth at Import Prices and Rate of Change of the Terms of Trade: Country Time Series Regressions

(Data for 1971-1991)

ss

Country Intercept Slope F-statistic R2 D.W Industrial countries United States 0.016 1.183* 51.010* 0.714 1.770 United Kingdom 0.033* 1.549* 54.234* 0.727 1.929 Japan 0.075* 1.429* 334.535* 0.943 1.156 France 0.045* 1.361* _ 82.947* 0.804 1.729 Italy 0.039* 1.241* 130.020* 0.866 2.013 Canada 0.019 0.421* 4.722* 0.166 1.192 Germany 0.032* 1.189* 163.584* 0.890 1.798 Australia 0.020 0.506* 10.794* 0.329 2.274 Spain 0.052* 1.204* 74.048* 0.785 1.089 Developing countries Asia Hong Kong 0.055* 0.216 0.892 — 2.188 India ~0.017 1.289* 25.541* 0.551 2,331 Indonesia 0.017 0.073 0.254 -—- 1.688 Korea 0.086* 1.481* 59.421* 0.745 1.338 Malaysia 0.006 0.374* 4.096* 0.134 1.073 Pakistan -0.058 -1.005* 8.912* 0.284 1.238 Philippines -0.018 -0.096 0.387 -- 1.395 Thailand 0.015 0.255 1.233 0.012 1.540 Africa Algeria 0.001 0.037 0.121 — 0.866 Cameroon 0.001 -0.158 1.125 0.006 1.743 céte d'Ivoire -0.004 0.159 0.551 -— 1.895 Gabon 0.042 -0.154 0.629 -—— 2.508 Ghana -0.048 -0.172 0.210 -— 1.200 Kenya -0.035 0.156 0.683 -- 2.233 Nigeria -0.072 0.331* 5.484* 0.183 1.684 South Africa -0.005 0.696 3.123 0.096 1.234 Tunisia 0.014 -0.349 2.697 0.078 2.278 Morocco -0.004 0.168 0.266 -0.038 2.046 Middle East Egypt -0.023 0.091 0.108 -—- 1.459 Iran 0.058 0.006 0.002 “= 1.241 Israel 0.028 0.714 1.542 0.026 2.088 Kuwait -0.010 0.102 1.299 0.016 1.721 Western Hemisphere Brazil 0.034 0.845* 15.452* 0.419 2,223 Colombia 0.003 0.152 2.247 0.059 0.701 Ecuador -0.013 0.175* 5.714* 0.191 1.642 El Salvador 0.007 0.135 1.572 0.028 1.650 Guatemala -0.009 0.239 1.357 0.018 1.878 Mexico 0.006 0.808* 9.248* 0.292 2.135 Peru -0.003 -0.038 0.014 — 2.039 Venezuela 0.002 -0.046 0.178 ~0.043 1.969 chile -0.019 0.293 0.55 —— 1.144 Notes: An asterisk denotes that intercept, slope, or Fr-statistic estimates are significantly from zero at the 5 percent confidence level. Countries classified as commodity exporters,

the text, are in bold typeface.

- 33 -

Table 4. Real Per-Capita Consumption Growth at Consumer Prices and the Rate of Change of the Terms of Trade: Country Time Series Regressions

(Data for 1971-1991)

Country Intercept Slope P-statistic R2 D.W.

Industrial countries

United 3tates 0.019* 0.114* 5.570* 0.186 0.986 United X<ingdom 0.024* 0.017 0.024 _ 1.131 Japan 0.033* 0.031 1.115 0.001 1.898 France 0.023* 0.050 3.268 0.102 0.931 Italy 0.028* -0.059 2.439 0.067 1.490 Canada 0.025* -0.008 0.010 _ 1.077 Germany 0.026* 0.079* 4.724* 0.157 0.806 Australia 0.020* 0.143* 26 .000* 0.556 1.982 Spain 0.024* -0.001 _ _ 0.602 Developing countries Asia Hong Kong 0.061* -0.070 0.269 - 1.841 India 0.019 0.150 1.154 0.008 2.806 Indonesia 0.037* 0.095* 6.803* 0.225 1.569 Korea 0.063* 0.193 3.034 0.092 2.095 Malaysia 0.045* 0.287* 7.310* 0.240 1.122 Pakistan 0.003 -0.123 1.162 0.008 1.632 Philippines 0.008 0.028 0.231 -- 1.133 Thailand 0.039* 0.245* 5.427* 0.181 1.927 Africa Algeria 0.020 0.125* 10.217* 0.315 2.139 Camercon 0.005 0.023 0.056 _ 1.404 céte d'ivoire -0.016 0.216 3.715* 0.120 1.917 Gabon 0.005 -0.293* 7.164* 0.236 2.171 Ghana —0.007 -0.081 0.334 _ 2.101 Kenya 0.025 -0.019 0.011 —_ 1.546 Nigeria 0.001 0.216* 6.057* 0.202 2.396 South Africa 0.005 0.263 1.478 0.023 2.582 Tunisia 0.037* 0.019 0.043 — 1.655 Morocco 0.020* 0.020 0.048 - 2.712 Middle East Egypt 0.030* 0.104 1.389 0.023 1.292 Iran -0.022 -0.118 0.740 _ 2.005 Israel 0.021 -0.508 2.858 0.085 2.586 Kuwait —0.002 0.110 1.646 0.037 1.785 Western Hemisphere Brazil 0.025* 0.020 0.054 _ 1.564 Colozbia 0.019* -0.037 1.173 0.010 1.225 Ecuador 0.020* 0.019 0.753 _ 0.533 El Salvador -0.002 0.105 3.156 0.097 0.933 Guatemala 0.001 0.032 0.309 -_ 0.917 Mexico 0.019 0.123 2.605 0.074 1.251 Peru -0.017 -0.038 0.176 -- 1.304 Venezuela 0.013 -0.082 2.941 0.108 1.065 Chile 0.017 0.262 0.951 —_ 2.236

cen

- 34 -

Table 5. Consumption Growth and the Rate of Change of the Terms of Trade: Results for Regional Panels 1/

(Data_at import prices for 1971-91)

Intercept Slope F-Test Against Total Independent R?

Industrial countries

Total 0.036* 1.142* -- 4.543* 0.722

Between means 0.011 -0.360 -- -- --

Fixed effects -- 1.147* 1.122 7.631* 0.721

Random effects 0.036* 1.145* -- -- 0.716 Asia

Total 0.008 0.131 -- 5.898* 0.009

Between means 0.013 0.569 -- -- --

Fixed effects -- 0.118 2.009 9.0725* --

Random effects 0.007 0.126 -- -- -- Africa

Total -~0.011 0.070 -- 0.923 0.001

Between means -0.011 0.191 -- -- --

Fixed effects -- 0.067 0.550 1.289 --

Random effects -0.011 0.069 -- -- -- Middle East

Total 0.024* 0.088* -- 0.783 0.013

Between means 0.026 -0.328 -- -- --

Fixed effects -- 0.099 0.874 0.704 --

Random effects 0.024 0.093 -- -- -- Western Hemisphere

Total -0.001 0.200* -- 1.003 0.046

Between means 0.005 0.445 -- -- 0.260

Fixed effects -- 0.197* 0.119 1.882 0.001

Random effects -0.001 0.199* -- -- 0.002

1/ Intercept and slope coefficients for regressions of consumption growth on the rate of change of the terms of trade and a constant. Wumbers in brackets are t-statistics. An asterisk denotes statistical significance at the 5 percent level. Results for independent time-series regressions are reported in Tables 3 and 4.

- 35 -

Table 6. Consumption Growth and the Rate of Change of the Terms of Trade: Results for Regional Panels 1/

(Data at consumer prices for 1971-91)

Intercept Slope F-Test Against Total Independent R2

Industrial countries

Total 0.025* 0.036* -- 1.308 0.019

Between means 0.023* -0.067 -- -- --

Fixed effects -- 0.037* 1.094 1.499 --

Random effects 0.025* 0.037* -- -- -- Asia

Total 0.034* 0.101* -- 3.579% 0.048

Between means 0.038* 0.385 -- -- --

Fixed effects -- 0.092* 4.739% 2.175 0.005

Random effects 0.034* 0.094* -- -- 0.005 Africa

Total 0.009 0.061* -- 1.990%* 0.011

Between means 0.001 0.171 -- -- --

Fixed effects -- 0.058* 0.726 3.178* --

Random effects 0.009 0.059* -- -- -- Middle East

Total 0.011 0.036 -- 1.349 --

Between means 0.011 -0.197 -- -- --

Fixed effects -- 0.042 1.084 1.585 --

Random effects 0.011 0.038 -- -- -- Western Hemisphere

Total 0.001 0.026 -- 0.934 --

Between means 0.013 0.184 -- -- --

Fixed effects -- 0.024 0.714 1.150 --

Random effects 0.009 0.025 -- -- --

l/ Intercept and slope coefficients for regressions of consumption growth on the rate of change of the terms of trade and a constant. Numbers

in brackets are t-statistics. at the 5 percent level.

reported in Tables 3 and 4.

An asterisk denotes statistical significance

Results for independent time-series regressions are

- 36 -

Table 7. Consumption Growth and the Rate of Change of the Terms of Trade:

Results for Export Category Panels 1/

(Data for the period 1971-91)

nn

ee

Commodity-based exporters

Total -0.005 0.093%* -- 1.023 Between means -0.003 0.205 -- -- Fixed effects -- 0.090%* 0.516 1.518 Random effects -0.005 0.092* -- -- Non-commodity-based exporters Total 0.025* 0.765%* -- 5.279% Between means 0.022* 0.638 -- -- Fixed effects -- 0.766* 1.153 8.967% Random effects 0.025* 0.765* -- -- (Data at consumer prices) Commodity-based exporters Total 0.007* 0.046% -- 1.432 Between means 0.008* 0.126 -- -- Fixed effects -- 0.044* 0.651 2.175* Random effects 0.007* 0.045* -- -- Non-commodity-based exporters Total 0.028%* 0.067* -- 2.431% Between means 0.037* 0.616 -- -- Fixed effects -- 0.063* 3.145* 1.623 Random effects 0.028* 0.064* -- -- 1/ Intercept and slope coefficients for regressions of consumption

Intercept

Slope

F-Test Against Total Independent

(Data at import prices)

growth on the rate of change of the terms of trade and a constant.

in brackets are t-statistics.

at the 5 percent level. reported in Tables 3 and

oo0o°0o

O11

.339 .007 .319 . 309

Numbers An asterisk denotes statistical significance

Results for independent time-series regressions are

4.

- 37 -

Table 8. Consumption Growth and the Rate of Change of the Terms of Trade: Panels for Decade Subsamples According to Export Category

(Data at import prices for 1971-91)

1971-80 1981-91 Intercept Slope Intercept Slope

(Data _ at import prices)

Commodizy-based exporters

Total 0.005 0.069 -0.012 0.113 Between means -0.001 0.322* 0.014 0.667 Fixed effects -- 0.043 -- 0.102 Random effects 0.006 0.059 -0.012 0.109

Non-commodity-based

exporters Total 0.001 0.659% 0.043* 0.909%* Between means -0.008 0.381* 0.042* 1.137* Fixed effects -- 0.676%* -- 0.892%* Random effects 0.002 0.665* 0.043%* 0.900* (Data at consumer prices)

Commodity-based exporters Total 0.023%* 0.033 -0.006 0.032 Between means 0.020* 0.069 0.008 0.338* Fixed effects -- 0.026 -- 0.027 Random effects 0.022 0.029 -0.006 0.030

Non-comnodity-based

exporters Total 0.034* 0.059%* 0.022* 0.134* Between means 0.039* 0.198 0.022* 0.192 Fixed effects -- 0.051* -- 0.130%* 0.055* 0.022* 0.132*

Random effects 0.034*

- 38

“L6-I86T UT UQMoTs aaTQeBau aB1eT ATaA pue OB-TZ6T UT yymors uctydunsuoo s¥eraae aatzeseu a3reT Aq@aA OF ANp URITT Futon IXg “suzeq ssozrd yqtmM s4seq AZTOTIsepaysoreiay saTyM ‘TaAeT queoded GT ey. Je asouRoTZTUSTS saqouep ++ pue TsA2T quaored oT ayq qe aouRrotytTuadts saqo 4 ‘TeaaT quaored ¢ ayy Je aoUROTJTUBTS TeOTYSTIeIS Sajouep YsTieqyse Uy “SzeyoeTq UT uMOYS aTe SOTISTIEIS-7 BUSEISTSUOD-OTYSepeysorszaH /

NEO

ap

are

I I

x(€66°4-) x(T€9'Z) 2( LTB’ 9) (T661T-T86T)

9290 TLO'E »9S°0 x€LZT'€ #20 £0 L£@S0- £820 6T0'0 6€ /€ eTdues-qns puosss (Pp (SZS°0) 4 (L476 °T) aCLTL°9) (086T-TL6T)

990 °0 €ot 9 "eT 4479'S 4672? TZ£0°0- 60T 0 Teo’ 0 6€ /€ ajtdues-qns ysitq (9

(T66T-T86T

446865 1) x (ZE0' 4%) 2(G¥Z°9) ‘086T-TZL6T)

96T 0 4959 °0T 48L2°2 4€Z OT #100" 222 O- Tez 0 "70°00 BL /€ seytdues-qns om] (q x(€90°Z-) 4(@L2 1) 2(L429°> 4) (T66T-TLET?

TLT 0 OT 9 862 T a9T 4 ¥€Z°O€ €TS O- 44E "0 TE0'O 0” atdwes TInNgZ (e

Taunsuods 4e eyep)

2(ZS0°%-) x(628° 2) a(TSZ°9) (T66T-T86T)

T79°0 TL6°S eé6t Tt x00 TE «1T9°0S BST T- TTZ'0 0”0°0 04 aqTdues-qns puooses (Pp 440969 T-) «(226° 7) (esp '0-) (O86T-TL6T)

€ze 0 96S°E 22490 xOT TT 900°0 £9@'0- 98€°0 €00°0- Ov ajdues-qns qsitq (9

(T66T-T86T

x(8T €-) x(TB€ 8) (06S °€) “O86T-TL6T)

82h 0 “997° HT atT@€ 288° 4T xITT'8t 2408 °0- €6S 0 €z0'0 08 seqtdwes-qns om] (q «(£78 °2) #(690°€) x(Z90°€) On (T66T-TLET)

€0T "0 447888 44226 1 «STO 6 #9 °0OT »e9°0- "L470 0z0°0 ajdues TInNgZ (e

(Seotad yaodut qe e7eq)

a a,

(2) (Te) (On) zu x d _T-z_(192) =20 (gjuqto—00 0 hea qdaore4Uuy SUOTIRA atdues A aasy iT Sys0]-d TO} JUSotzpooo UOTSSoTdsy -BAIeSqo SUOTYOTIYSAY jo zequny

quatotyye0g

i Ii — III omar

/{ apery jo sural ayy Jo aoueTIeA pur uray eyy uo YIMOIH uoTIduNnsuOD JO suOTSSETBey TeUOTIIeS-SSOIQ “6 STAPL

IFDP Number

49]

490

489

488

487

486

485

484

483

482

481

480

479

-39-

International Finance Discussion Papers

Titles

1994

Terms-of-Trade Uncertainty and Economic Growth:

Are Risk Indicators Significant in Growth Regressions

Politics, Economics, and Investment: Explaining Plant and Equipment Spending by U.S. Direct Investors in Argentina, Brazil, and Mexico

On The Dynamic Properties of Asymmetric Models of Real GNP

A distributed block approach to solving near-block-diagonal systems with an application to a large macroeconometric model

Conditional and Structural Error Correction Models

Bank Positions and Forecasts of Exchange Rate Movements

Technological Progress and Endogenous Capital Depreciation: Evidence from the U.S. and Japan

Are Banks Market Timers or Market Makers? Explaining Foreign Exchange Trading Profits

Constant Returns and Small Markups in U.S. Manufacturing

The Real Exchange Rate and Fiscal Policy During the Gold Standard Period: Evidence from the United States and Great Britain

The Debt Crisis: Lessons of the 1980s for the 1990s

Who Will Join EMU? Impact of the Maastricht Convergence Criteria on Economic Policy Choice and Performance

Determinants of the 1991-93 Japanese Recession: Evidence from a Structural Model of the Japanese Economy

Author(s)

Enrique G. Mendoza

Guy V.G. Stevens

Allan D. Brunner

Jon Faust Ralph Tryon Neil R. Ericsson

Michael P. Leahy

Robert Dekle

John Ammer Allan D. Brunner

Susanto Basu John G. Fernald

Graciela L. Kaminsky Michael Klein

Graciela L. Kaminsky Alfredo Pereira

R. Sean Craig

Allan D. Brunner Steven B. Kamin

Please address requests for copies to International Finance Discussion Papers, Division of International Finance, Stop 24, Board of Governors of the Federal Reserve System, Washington, D.C. 20551.

IFDP

Number

478

477

476

475

474

473

472 47]

470

469

468

467

466

- 40 -

International Finance Discussion Papers

On Risk, Rational Expectations, and Efficient Asset Markets

Finance and Growth: A Synthesis and Interpretation of the Evidence

Trade Barriers and Trade Flows Across Countries and Industries

The Constancy of Illusions or the Illusion of Constancies: Income and Price Elasticities for U.S. Imports, 1890-1992

The Dollar as an Official Reserve Currency under EMU

Inflation Targeting in the 1990s: The Experiences of New Zealand, Canada, and the United Kingdom

International Capital Mobility in the 1990s

The Effect of Changes in Reserve Requirements on Investment and GNP

International Economic Implications of the End of the Soviet Union

International Dimension of European Monetary Union:

Implications For The Dollar

European Monetary Arrangements: Implications for the Dollar, Exchange Rate Variability and Credibility

Fiscal Policy Coordination and Flexibility Under European Monetary Union: Implications for Macroeconomic Stabilization

The Federal Funds Rate and the Implementation of Monetary Policy: Estimating the Federal Reserve’s Reaction Function

Author(s)

Guy V.G. Stevens Dara Akbarian

Alexander Galetovic Jong-Wha Lee Phillip Swagel

Jaime Marquez

Michael P. Leahy John Ammer Richard T Freeman Maurice C’bstfeld

Prakash Loungani Mark Rush

William L. Helkie David H. Howard Jaime Marquez

Karen H. Johnson

Hali J. Edison Linda S. Kole

Jay H. Bryson

Allan D. Brunner