The concept of “globalization” has recently been the subject of considerable attention in both academic and policy circles. This phenomenon broadly refers to the increasing integration of the world economy through financial and trade flows. As economies become more open to international trade, the transmission and propagation of economic fluctuations through trade links has assumed increased importance. An analysis of the cyclical dynamics of international trade therefore has implications in a number of different dimensions, including macroeconomic forecasting, short-run policymaking, and international policy coordination.

Abstract

The concept of “globalization” has recently been the subject of considerable attention in both academic and policy circles. This phenomenon broadly refers to the increasing integration of the world economy through financial and trade flows. As economies become more open to international trade, the transmission and propagation of economic fluctuations through trade links has assumed increased importance. An analysis of the cyclical dynamics of international trade therefore has implications in a number of different dimensions, including macroeconomic forecasting, short-run policymaking, and international policy coordination.

The concept of “globalization” has recently been the subject of considerable attention in both academic and policy circles. This phenomenon broadly refers to the increasing integration of the world economy through financial and trade flows. As economies become more open to international trade, the transmission and propagation of economic fluctuations through trade links has assumed increased importance. An analysis of the cyclical dynamics of international trade therefore has implications in a number of different dimensions, including macroeconomic forecasting, short-run policymaking, and international policy coordination.

The objective of this paper is to provide a comprehensive set of stylized facts concerning the relationship between external trade and aggregate macroeconomic fluctuations. We assemble a data set that covers the OECD industrial economies and that includes the main price and quantity variables relevant to international trade since 1970. Thus, we reexamine, for a broader set of countries, a few stylized facts reported by other authors, introduce a number of new stylized facts, and attempt to provide a unified analysis of various features of the data based on recent theoretical advances in open economy macroeconomics.

In the first part of the paper, we provide a systematic documentation of the relationship, as measured by unconditional correlations, between trade variables and domestic business cycles. We also examine patterns of volatility of macroeconomic quantities and prices relevant for international trade. These stylized facts serve a number of purposes. First, they help isolate key features of the data and thereby provide building blocks for theoretical models of the dynamics of international trade. Second, they provide a metric against which to measure the performance of calibrated general equilibrium models. Third, they provide an indication of the types of macroeconomic shocks that are important determinants of fluctuations in trade variables.

Unconditional correlations, however, provide information only about the average relationship between these variables and output over different phases of the business cycle. Of more interest, from a forecasting as well as policy perspective, is the role of trade as a catalyst for business cycle recoveries. To examine this issue, we construct a quantitative measure of the importance of international trade in generating business cycle recoveries. Somewhat surprisingly, this measure reveals that the trade balance (i.e., net exports of goods and nonfactor services) has in fact played only a limited role in business cycle recoveries in the OECD economies. The average contribution of the trade balance to output growth from business cycle troughs is quite small for most industrial economies.

An alternative hypothesis is that of “export-led recoveries,” wherein an increase in export demand serves as a catalyst for the recovery in domestic demand. A concomitant increase in the imports of investment goods and intermediate inputs could, in this case, result in a smaller improvement or, possibly, even a deterioration of the trade balance in the short run. We do indeed find that exports appear to play a significant role in business cycle recoveries in industrial economies. We also find that the average magnitude of this contribution is closely related to the degree of openness of these economies to international trade.

The relationship between trade and the business cycle could, however, depend crucially on two factors. The first is the correlation of business cycles across countries. If business cycles across countries were perfectly synchronized, international trade would be less likely to have a significant influence on domestic business cycles. The second consideration is that changes in exchange rates are likely to have a strong influence on the relationship between trade and the business cycle. Different shocks that have similar business cycle effects but that vary in their exchange rate consequences could have very different effects on trade variables. To address these issues, we construct a multivariate econometric model that accounts for changes in domestic and external demand conditions and that captures the effects of exchange rate changes on trade. This part of the paper builds upon the work of Prasad and Kumar (1997), although, unlike that paper, we examine here a much broader set of countries than just the Group of Seven (G-7) industrial countries and, therefore, provide a more comprehensive picture of trade dynamics in industrial countries. In addition, we extend the framework to separately examine fluctuations in imports and exports.

An important feature of the empirical model is that it enables us to disentangle different sources of macroeconomic shocks and to analyze the responses of trade variables to these shocks. Much of the literature on current account dynamics has focused on the role of productivity shocks, both global and country-specific (see, e.g., Backus, Kehoe, and Kydland, 1992; Glick and Rogoff, 1995; and Elliott and Fatás, 1996). Through the effects of exchange rate changes, it is likely that other types of shocks could also significantly influence the dynamics of the trade balance. The framework in this paper enables us to simultaneously characterize the short-run and long-run dynamics of output, the real exchange rate, and the trade balance in response to various types of macroeconomic shocks.

This paper is also related to the large body of recent work that has provided theoretical and empirical analyses of the intertemporal effects of productivity and government spending shocks on consumption, investment, and the current account (see Obstfeld and Rogoff, 1996, and references therein). However, much of this literature has implicitly assumed the absence of quantitatively important nominal rigidities, which could potentially play an important role in short-run fluctuations of real variables. The strong positive correlation between real and nominal exchange rates, for instance, suggests that nominal rigidities influence real relative prices in the short run and could, therefore, have real effects. This paper contributes to this literature by providing an empirical characterization of the relative importance of these effects on the dynamics of the trade balance in the presence of nominal rigidities.

I. Stylized Facts

In this section, we compile a comprehensive set of stylized facts concerning the relationship between international trade and the business cycle. First, we present various measures of openness to international trade, to gauge the importance of trade for industrial countries. Next, we examine the volatility of various trade variables relative to the aggregate business cycle in each country. We then present unconditional correlations between trade variables and the business cycle at various lags and leads, to highlight important features of trade dynamics over the business cycle.

The data sources and variables used in the analysis are described in Appendix I. Some of the results presented in this section build upon the work of Backus and Kehoe (1992), Fiorito and Kollintzas (1994), Baxter (1995), and Zimmermann (1995), although we extend the results of these authors to a broader sample of OECD countries and to a larger set of variables related to international trade. Most of the variables examined here are nonstationary in levels over our sample period. To facilitate comparisons with the existing literature, in this section we focus on the cyclical components of all variables obtained using the Hodrick-Prescott (HP) filter. In later sections of the paper, we allow for more general stochastic trends in these variables.

Much of the open economy macroeconomics literature, including recent work on international consumption smoothing and risk-sharing, has tended to focus on variation in the current account. The current account incorporates trade as well as net interest payments and, from the perspective of saving-investment balances, is the appropriate concept. For investigating the dynamics of trade, however, measures of trade in goods and nonfactor services would appear to be more appropriate than the current account. Variations in net factor incomes are, presumably, driven by a different set of determinants. In any case, as noted by Baxter (1995), short-term variations in net exports and in the current account are highly correlated since the discrepancy between these two measures tends to change very slowly over time. For the purposes of forecasting and short-run policy determination, net exports are indeed the relevant concept from the perspective of the national income accounts. Hence, the analysis in this paper is limited to the national income accounts definition of trade in goods and nonfactor services.

The Extent of International Trade

First, we examine the economic significance of international trade for OECD industrial economies. The ratio of total trade volumes to real GDP is often used as an indicator of an economy’s openness to international trade. The average of this ratio over the period 1970-95, shown in the first column of Table 1, ranges from about 20 percent for Japan and the United States to over 100 percent for Belgium, Ireland, and the Netherlands.

Averages of this ratio over different subsamples show that this ratio increased for almost all countries in the 1980s, reflecting the growing importance of international trade to the world economy. This ratio continued to increase in the 1990s for many countries but declined marginally for a few countries. The second panel of Table 1 shows that exports constitute a significant fraction of total output for a number of countries in the sample. The ratio of exports to GDP also increased in the 1980s for virtually all countries in the sample and continued to increase in the 1990s for most countries. As shown in the bottom row of Table 1, rising trade volumes are also reflected in increases of the ratio of the overall trade of OECD economies to total GDP (constructed using 1990 purchasing power parity weights) in these economies.

Table 1 presents clear evidence of the rapid increase in trade volumes for the main industrial economies since the 1970s. Understanding the role of trade in accentuating or dampening domestic macroeconomic fluctuations is thus of increasing importance as the global economy becomes more integrated through trade flows.

Volatility of Trade Variables

We now examine the unconditional volatility, as measured by the standard deviation, of various quantities and prices relevant to trade dynamics, relative to the aggregate business cycle. The output and exchange rate variables were first transformed into logarithms and their cyclical components were then derived using the HP filter with a smoothness parameter of 1600. The trade variables are expressed as ratios of aggregate GDP in order to control for scale effects. The trade ratios (not in logarithms) were also detrended using the HP filter.

The first column of Table 2 contains the standard deviation (interpretable as the quarterly percentage standard deviation) of domestic output. These standard deviations generally lie in the range of 1 to 2 percent. The next three columns show the standard deviations of the trade ratios, divided by the standard deviation of output. For most countries, net exports are markedly more volatile than aggregate output.1 The United States is an outlier among the major industrial countries in terms of the low relative volatility of net exports.2 The next column shows that exports are generally less volatile than net exports and often display much lower volatility. In general, imports are about as volatile as exports although, in a few cases, they are more volatile than exports.

Table 1.

Measures of Openness to International Trade

article image
Notes: Exports (imports) refers to the national income accounts definition of exports (imports) of goods and nonfactor services. The ratios shown above are average ratios over the relevant periods. The OECD average is a weighted average constructed using 1990 GDP weights based on purchasing power parities.
Table 2.

Measures of Volatility of Cyclical Components

article image
Notes: The numbers reported above are standard deviations of the respective cyclical components computed using the HP filter. For all variables except domestic output, the standard deviations are expressed as a ratio of the standard deviation of domestic output. The variable definitions are as follows: y: domestic output; netx/y: the ratio of net exports to domestic output; exp/y: the ratio of exports to domestic output; imply: the ratio of imports to domestic output; y*. index of output in trading partner countries; y-y*: relative (domestic minus foreign) output; reer. real effective exchange rate; neer. nominal effective exchange rate. All variables other than the trade ratios were used in logarithmic form.

An important consideration for the dynamics of the trade balance is the commonality of shocks across countries. The pattern of short-run trade dynamics could depend on whether shocks that drive business cycle fluctuations are primarily global or country specific. In addition, the nature of shocks could be important—global shocks are more likely to take the form of productivity or other supply shocks, while demand shocks would tend to be country specific. This issue will be dealt with in more detail in the formal econometric model below. Nevertheless, it is useful at this juncture to examine the relative volatility of global output from the perspective of each country.

Rather than construct one uniform measure of global output, for each country we construct a trade-weighted measure of total output in its trading partner countries. The volatility of this measure of partner country output, which is interpretable as effective foreign demand, is shown in the fifth column of Table 2 (y*). For most countries, partner country output has lower volatility than domestic output. In large part, this is because the measure of partner country output is, by construction, an average measure of output across many countries.

Fluctuations in relative output, defined as domestic minus partner country output, provide a rough indication of how important global shocks are for each country. For instance, a strong positive correlation between domestic and world output fluctuations would imply that relative output fluctuates much less than domestic output. This is indeed the case for most countries in the sample, suggesting that there is a significant common component in international economic fluctuations.3

Finally, we examine the volatility of the relative price that is most closely associated with international trade, that is, the exchange rate. The cyclical component of the real effective exchange rate fluctuates more than output in all countries. Interestingly, for most countries, the volatility of nominal and real effective exchange rates is very similar. The correlations between the real and nominal measures of the exchange rate (not reported here) also turned out to be very strongly positive for all countries in the sample, consistent with a large body of literature that has documented the close relationship between these variables at business cycle frequencies (see, e.g., Mussa, 1986).4

Unconditional Correlations

Next, we turn our attention to correlations between the cyclical components of output and the various trade variables. We examine contemporaneous correlations as well as correlations at various leads and lags in order to explore the dynamic patterns in the data.

The first panel of Table 3 shows the correlations between the cyclical components of output and net exports. The countercyclical behavior of net exports that has been documented by numerous other authors is evident for virtually all the countries in our sample. The second panel of this table documents that this result is driven largely by the strongly procyclical behavior of imports, as shown by the strong positive correlations between the cyclical components of imports and output for all countries except Ireland, New Zealand, and Norway. In these three countries, the correlations between output and leads of imports turn positive, indicating a delayed response of imports to changes in domestic output. In all other countries, the positive correlations between these variables peak contemporaneously or at very short leads, indicating that changes in domestic demand are translated into changes in import demand quite rapidly.

The cyclical behavior of exports, shown in the third panel of Table 3, is, on the other hand, very different across countries. The contemporaneous correlations are positive for a few countries, negative for a few, and not statistically significantly different from zero for a majority of the countries. Among the G-7 countries, this correlation is positive for Canada, negative for Germany and Japan, and close to zero for the others. A number of possible reasons could be cited for these mixed results. For instance, variation in exports could be influenced not just by domestic demand conditions but by changes in external demand from trading partner countries. In addition, exchange rate developments over the course of the business cycle could also have a significant influence on these correlations. These factors would not be reflected in unconditional bivariate correlations. Thus, a model that, at a minimum, captured the effects of changes in external demand conditions and exchange rates, would appear to be necessary to model export dynamics in an appropriate manner. Before turning to such a model, however, it is useful to examine the importance of trade in business cycle recoveries. This could potentially yield a very different picture of trade dynamics compared to the correlations discussed above that depict only the average relationship over different phases of the business cycle.

II. International Trade and Business Cycle Recoveries

It is quite typical for international trade to be attributed a critical role in short-term macroeconomic forecasts, especially for smaller industrial economies. Even in large economies that are relatively closed, the external sector is often viewed as being an important catalyst for business cycle recoveries. The evidence on the quantitative importance of international trade in generating economic recoveries is, however, rather limited.

Table 3.

Correlations of Trade Variables and the Business Cycle

article image
Notes: The cross-correlations reported here are for HP-filtered cyclical components of real GDP and the respective trade variables. Lag 8 indicates the correlation of output and the 8th lag of the trade variable. A negative lag denotes a lead. The approximate standard error of these correlation coefficients, under the null hypothesis that the true correlation coefficient is zero, is about 0.10.

To address this issue, we construct a measure of the contribution of the trade balance to output growth from business cycle troughs. For each country in our sample, we identify historical business cycle troughs and then construct a measure of the contribution of the trade balance to total output growth over different time horizons relative to those troughs. Our procedure for identifying business cycle troughs is described in Appendix II. The formula for calculating the growth contribution of the trade balance over different time horizons is as follows:

CB(j)=TB(t+j)TB(t)Y(t),(1)

where CB(j) indicates the contribution to output growth over a. j-period horizon from the cyclical trough, TB(t) indicates the trade balance at time t (the cyclical trough), and Y(t) denotes aggregate GDP at time t.5

The first panel of Table 4 reports the average contributions of the trade balance to output growth over different horizons ranging from 1 to 12 quarters. There are notable differences across countries. For instance, at a horizon of 4 quarters from cyclical troughs, the average contribution of the trade balance to output growth ranges from a high of 2.2 percentage points for Ireland to a low of-1.9 percentage points for Austria. Over horizons of 1 to 12 quarters, the contributions of the trade balance to output growth are consistently negative for a number of countries, including Austria, Denmark, Japan, Spain, and Switzerland. For these countries, there is no evidence in historical data that the trade balance has contributed, in economically significant terms, to output recoveries from recessions. It is quite striking that, for a majority of the OECD economies, the average contributions of the trade balance to output growth at 4-, 8-, and 12-quarter horizons from cyclical troughs are negative or close to 0.

Table 4.

The Rote of International Trade in Business Cycle Recoveries

article image
Notes: The contributions shown above are average percentage point contributions of trade to output growth from the respective cyclical troughs. For a list of business cycle trough dates for each country, see Appendix I

In the context of the national income accounting identity, the net trade balance would appear to be the appropriate variable for examining the contribution of international trade to business cycle recoveries. However, an alternative hypothesis is that export demand, rather than net exports, provides the catalyst for economic recovery. An increase in current and projected exports due to strong external demand or an exchange rate depreciation could trigger a recovery in domestic output. It is therefore possible that, with a concomitant increase in imports of investment goods and intermediate inputs, the trade balance could deteriorate despite the export stimulus to domestic demand.

To examine the alternative hypothesis of export-fed recoveries, we now examine the contribution of exports to output growth using the same measure used above for the trade balance. These results are reported in the second panel of Table 4. The main finding here is that, for virtually all countries, exports do make a significant and sizable contribution to cyclical recoveries. There are, however, marked differences across countries. For Belgium and Ireland, the contribution of exports to output growth from cyclical troughs is quite large and rises to over 13 percent over a 12-quarter horizon. The United States has a zero growth contribution from exports at a 1-quarter horizon and a contribution of only 1.3 percentage points from exports to output growth over 12 quarters, the lowest in the sample.

Figures 1 and 2 show the growth contributions of the trade balance and exports, respectively, during the most recent cyclical recovery. Note that, since the data sample ends in 1995:4, the available lime horizon relative to the trough is rather limited for certain countries. The top panels of Figures 1 and 2 show that, among the G-7 countries, the recent recoveries in Canada and Italy have relied to a significant extent on increases in external demand. For the United Kingdom, external trade, although not an important factor in the early stages of the most recent recovery, has helped sustain and strengthen the expansion of output. Although France and Germany experienced strong increases in exports, the direct contribution of the trade balance to their respective cyclical recoveries has been rather small.

Figure 1.
Figure 1.

Contribution of the Trade Balance to Output Growth During the Most Recent Cyclical Recovery

Citation: IMF Staff Papers 1998, 002; 10.5089/9781451973600.024.A001

Notes: The contributions to output growth are expressed in percentage points. The cyclical troughs for each country (see Table A1) are denoted by T and a j-quarter horizon after the trough is denoted by T + j. The sample period ends in 1995:4.
Figure 2.
Figure 2.

Contribution of Exports to Output Growth During the Most Recent Cyclical Recovery

Citation: IMF Staff Papers 1998, 002; 10.5089/9781451973600.024.A001

Notes: The contributions to output growth are expressed in percentage points. The cyclical troughs for each country (see Table A1) are denoted by T and aj-quarter horizon after the trough is denoted by T + j. The sample period ends in 1995:4.

Among the smaller industrial economies, the picture is quite disparate. Ireland appears to have received the biggest boost from the external sector during the most recent recovery. Some countries, like Austria, Spain, the Netherlands, and New Zealand, experienced robust increases in exports but these were accompanied by large increases in imports. These countries are prime examples of the notion that exports could serve as a catalyst for cyclical recoveries despite a negligible direct contribution from the trade balance to output growth.

The results in these figures and in Table 4 are consistent with the notion that exports are more important engines of recovery for economies that are relatively more open to international trade. For instance, the contribution measures are among the smallest for Japan and the United States, which have the lowest openness indicators of the countries in our sample (see Table 1). Likewise, the two countries with the largest output contribution measures from exports, Belgium and Ireland, are also the most open to international trade.

The average contributions of trade to output growth during cyclical recoveries, however, mask substantial variation in these contributions across different cyclical episodes for each country. We examined the minimum and maximum contributions of the trade balance to output growth at horizons of one, two, and three years from cyclical troughs.6 The differences across cyclical episodes were substantial for every country in the sample. There were also large differences in the contributions of exports to output growth across different cycles. For instance, for Canada, the trade balance contribution measure over a three-year horizon ranged from a minimum of –4.9 percentage points to a maximum of 2.9 percentage points while the export contributions ranged from 4.2 to 13.5 percentage points.

These results suggest that unconditional measures of the contributions of trade to output recoveries do not capture the effects of other factors, such as exchange rate variation, on the cyclical dynamics of the trade balance. Significant information could also be lost by averaging over business cycle episodes that could be driven by different shocks and that, therefore, display very different exchange rate dynamics. To control for these effects, we now develop a more format time-series model.

III. A Structural Econometric Model

In this section, we construct a multivariate structural time-series model that enables us to identify different types of macroeconomic shocks and to examine the effects of these shocks on trade variables. The empirical framework, similar to that proposed by Prasad and Kumar (1997), builds on earlier work by Blanchard and Quah (1989) and Clarida and Gali (1994). We first discuss some theoretical considerations that motivate the empirical framework, present results from a preliminary analysis of the data, and then briefly describe the implementation of the econometric methodology.

Analytical Considerations

There are two main channels that determine the relationship between external trade and cyclical output fluctuations.7 Given external demand conditions, the first channel is through domestic demand and the second channel is through changes in real exchange rates. The existence of these two distinct but related channels implies that the sources of business cycle fluctuations could influence the cyclical dynamics of external trade.

Consider the case of a domestic fiscal contraction that leads to a contraction in aggregate domestic demand. In the standard Mundell-Fleming type of framework (with capital mobility), this would be accompanied by a real exchange rate depreciation that, in addition to depressed domestic demand, would tend to induce a negative correlation between cyclical output and the trade balance (or exports). Thus, in this case, both the domestic demand and real exchange rate effects work in the same direction. On the other hand, a monetary contraction (which would also tend to depress domestic demand) would lead to an appreciation of the exchange rate, thereby creating an effect on the trade balance (or exports) opposite to that of the domestic demand effect. The relative importance of these two effects is then an empirical issue and would depend on various trade elasticities.

Another important issue, which is crucial from the perspectives of both theoretical models and empirical work, is the persistence of fluctuations in output and the exchange rate. Highly persistent changes in output or the exchange rate could have very different effects on trade dynamics compared to the effects of transitory changes in these variables (see, e.g., Glick and Rogoff, 1995, and Phillips, 1996). Standard open economy macro models can be used to derive implications concerning the persistence of the output and exchange rate effects of different sources of macroeconomic shocks. We exploit these theoretical considerations to derive a set of identifying restrictions that enable us to identify different types of macroeconomic shocks and to characterize the joint dynamics of output, exchange rates, and trade variables in response to these shocks.

In the empirical work, we separately identify three types of shocks: supply, demand, and nominal shocks. Although we do not directly identify fiscal or monetary shocks, it is reasonable, for heuristic purposes, to think of demand shocks as fiscal shocks and nominal shocks as monetary shocks. The empirical model is identified using three long-run restrictions: that demand and nominal shocks have no long-run effects on the level of output while nominal shocks have no long-run effects on the level of the real exchange rate.

It could also be argued that nominal shocks are unlikely to affect the ratio of the trade balance to output in the long run. However, a large body of literature argues that temporary exchange rate shocks can indeed have persistent effects on external trade through hysteresis or “beach-head” effects (see, e.g., Baldwin, 1988 and 1990). Further, as noted by Lane (1997), temporary exchange rate changes could, through resulting changes in net foreign asset holdings, have persistent effects on the trade balance, although there would be no corresponding long-run effects on the current account. Hence, we do not use the restriction that nominal shocks have only transitory effects on the trade balance for identification of the empirical model. The degree of persistence in trade balance fluctuations induced by nominal shocks is therefore an empirical matter and the model can, in principle, help resolve this issue.

An important virtue of the identification approach is that the short-run dynamics are unconstrained. The identification scheme can therefore be evaluated by examining whether the short-run dynamics implied by the estimates of the empirical model appear reasonable and in accordance with the predictions of the theoretical model. Another feature of this identification approach is that it does not require us to take a stand on the causal ordering of the variables in the VAR. This is useful since there is no clear evidence that any of these variables is predetermined relative to the others in a Granger-causal sense.

This discussion indicates the limitations of models of trade (or current account) dynamics that focus solely on productivity shocks (e.g., Backus, Kehoe, and Kydiand, 1992; and Elliott and Fatas, 1996). Econometric models that distinguish only between real and nominal shocks (e.g., Lastrapes, 1992; and Robertson and Wickens, 1997) would also be inadequate for modeling trade balance dynamics since supply and demand shocks, which could both be viewed as real shocks, have different effects on the real exchange rate.

Preliminary Data Analysis

An important consideration for the empirical work is that the above discussion implicitly assumes that external demand conditions remain constant. Since this is unlikely to be the case, the relevant output variable for the econometric model is relative real output, that is, domestic output relative to external demand. For each country, we constructed an index of external demand by taking a trade-weighted average of real GDP in the remaining OECD countries. The logarithm of this index was then subtracted from the logarithm of the index of domestic output in order to derive relative output.8 Similarly, an index of the real effective exchange rate for each country was constructed by taking a trade-weighted average of bivariate real exchange rates vis-à-vis each of the other OECD economies, using domestic and foreign CPIs as the price deflators. Thus, we derive consistent measures of relative output and the real exchange rate, although it should be noted that the merchandise trade numbers are more comprehensive and not limited to trade within the OECD.

It is necessary to first determine the time-series properties of the variables entering the VAR, The model presented in Prasad and Kumar (1997) implies that relative output, the real effective exchange rate, and the ratio of the trade balance to GDP (and other trade ratios) are all stationary in first differences and that their levels are not cointegrated. To conserve space, here we only briefly summarize the results of formal statistical tests for these empirical features of the data. Tables containing detailed results are available from the authors upon request.

For the trade ratios, the results indicated that, in nearly all cases, the null hypothesis of a unit root could not be rejected against the alternative of sta-tionarity around a deterministic trend. To maintain a uniform specification, the trade variables for all countries were included in first-difference form in the VARs. Since relative output and the real exchange rate also appeared to be first-difference stationary for the countries in the sample, their logarithmic first differences, that is, their growth rates, were included in the VARs.9 We then tested for cointegration in each of the trivariate systems using the Stock-Watson (1988) common trends test. With the exceptions of a few borderline rejections of the null hypothesis of no cointegration (for Austria, Belgium, Iceland, and Switzerland), there was little evidence of cointegration for the specifications discussed below. Even for those countries where the null hypothesis of no cointegration could be rejected using the trade balance to GDP ratio, there was no evidence of cointegration when using the imports to GDP ratio (except for Iceland) or the exports to GDP ratio. Hence, to maintain a uniform specification across countries, the reduced-form VARs include first differences of the relevant variables. This specification also has the virtue of facilitating the interpretation of the results.

The Econometric Model

The econometric model builds upon the work of Blanchard and Quah (1989) and Clarida and Gali (1994).10 The methodology involves the estimation of a three-variable VAR comprising the first differences of relative output, the real exchange rate, and the ratio of the trade balance (or exports or imports) to domestic output, with the first two variables used in logarithmic form. Using a set of long-run restrictions, the VAR errors are then transformed into a set of “fundamental” disturbances that have an economic interpretation—supply, demand, and nominal shocks. The short-run dynamics of the model are unconstrained and identification is achieved by imposing constraints on certain long-run multipliers in the system.

As noted earlier, we do not explicitly include monetary or fiscal variables in the estimation. Since we have identified the exchange rate as the mechanism through which different shocks influence the relationship between external trade and the business cycle exchange rates are included directly in the estimation in order to identify these shocks.11 Thus, the econometric approach is structural in that relative output fluctuations, variations in the real exchange rate, and changes in the trade variables are jointly determined in response to different shocks. Also note that, since relative output growth is used in the estimation, the shocks are more appropriately thought of as relative supply shocks, relative demand shocks, and relative nominal shocks. For brevity, this terminology is used sparingly below.

To conserve space, and since the basic econometric model is now widely used, we refer the interested reader to our working paper (Prasad and Gable, 1997) for details on implementation of the econometric methodology.12

IV. Results

Although the discussion thus far has focused on the dynamics of the trade balance, it is straightforward to extend the methodology to separately examine the constituents of net trade—exports and imports. The maintained assumption here is that exports and imports are driven by the same set of determinants as the trade balance.

The empirical model was estimated separately for each country. As will be clear from the results, there is substantial heterogeneity across countries in the dynamics of trade variables, making a panel approach to estimation inadvisable. The sample period is 1975:1-1995:4, obviating possible concerns about parameter instability associated with the breakup of Bretton Woods in 1973 and the first OPEC oil shock in 1974.13 To capture higher-order dynamics in the empirical model, the estimated equations in the reduced-form VARs included a constant and eight lags of each of the three variables.14

We first examine the estimated impulse response functions, which show the dynamic effects of different types of shocks on the trade variables, and then present forecast error variance decompositions. Since the trade balance, exports, and imports are tied together by an identity, we report the impulse responses of only the first two variables. However, we separately examine variance decompositions for all three trade variables since this is of interest from the perspective of reduced-form equations for import and export volumes.

Impulse Responses

Table 5 presents the impulse responses of the trade balance and exports to different types of shocks.15 Note that, although the trade variables are expressed as ratios of output in order to control for scale effects, these variables are substantially more volatile than output (see Table 2), implying that the responses of the trade ratios can be regarded as indicating purely the responses of the trade variables rather than reflecting changes in output.16

A striking feature of the impulse response functions for the trade balance is that, in every country, nominal shocks have a positive effect on the trade balance. Since relative nominal shocks lead to increases in relative output accompanied by exchange rate depreciations, this suggests that the exchange rate effects on the trade balance tend to dominate the output effects of these shocks. This is consistent with the findings of other authors that nominal shocks result in rapid and sharp exchange rate responses (see, e.g., Eichenbaum and Evans, 1995) but have only very small and transitory effects on output. These patterns were also evident in our estimated impulse responses for output and the exchange rate (not shown here). Interestingly, positive effects of nominal shocks on the U.S. trade balance have also been uncovered recently by Lane (1997) using VARs with identification schemes based on more traditional short-run restrictions and using direct measures of innovations in monetary policy (also see Betts and Devereux, 1997).

Table 5.

Impulse Responses

article image
article image
article image
article image
Notes: The impulse response functions of the trade ratios show the dynamic responses of the trade variables to unit (one standard deviation) supply (SS), demand (DD), and nominal (NM) shocks. The impulse responses of changes in the trade ratios were cumulated in order to derive the responses in terms of levels. Bold entries indicate statistical significance at the 5 percent level. Standard errors for the impulse responses were computed using Monte Carlo simulations with 500 replications.

As would be expected, relative demand shocks, which lead to increases in relative output and concomitant exchange rate appreciations, result in declines in the trade balance in almost all cases. The effects of supply shocks, on the other hand, are mixed. Supply shocks typically have large and permanent effects on output. Their effects on real exchange rates, however, differ across countries. Supply shocks generally tend to result in exchange rate depreciations or have small and statistically insignificant exchange rate effects. In some cases, however, the effects of supply shocks on the exchange rate are positive. For this group of countries—Finland, Iceland, the Netherlands, and Spain—it is likely that these effects reflect positive terms-of-trade shocks that result in permanent increases in the levels of both output and real exchange rates. It is also interesting to note that, for some small economies that are highly open to international trade (Iceland, Ireland, Norway, and Sweden), country-specific supply shocks result in significant but temporary improvements in the trade balance.

Intertemporal models of the trade balance (see Obstfeld and Rogoff, 1996) imply that temporary increases in domestic output would tend to increase domestic saving since optimal consumption, which is determined by permanent rather than current income in these models, would increase by less than the temporary increase in output. Hence, transitory output fluctuations would tend to be accompanied by increases in exports and in the trade balance. The estimated trade balance responses to nominal shocks support this implication of this class of models.

The impulse responses for exports also portray a similar picture of nominal shocks, which result in temporary increases in relative output and simultaneous exchange rate depreciations, leading to increases in exports, although these effects are often attenuated at longer horizons. Demand shocks typically lead to a fall in exports, reflecting the exchange rate appreciation that accompanies these shocks. The effects of supply shocks on exports, however, differ markedly across countries and it is difficult to discern a clear pattern.

An interesting feature of the estimated impulse response functions is that nominal shocks appear to have persistent effects on the trade balance. Even at long horizons, the impulse responses of the trade balance in response to these shocks are significantly different from zero for most countries. Since, in this framework, nominal shocks have only transitory effects on both relative output and the real exchange rate, this result suggests that hysteresis and “beach-head” effects are quantitatively important for the medium-term dynamics of international trade. As Baldwin (1988 and 1990) and Baldwin and Krugman (1990) have argued, these effects can translate transitory (but sufficiently large) exchange rate changes into persistent effects on trade prices and volumes. We also find these effects to be important for the dynamics of both export and import volumes.

Variance Decompositions

Next, we examine the forecast error variance decompositions for the trade variables. These decompositions indicate the proportion of the variance in the forecast error of the trade variables that can be attributed to each of the three types of shocks. By providing a quantitative measure of the relative importance of different types of shocks that drive fluctuations in the trade variables, these decompositions complement the information obtained from the impulse responses.

Table 6 shows that, for a majority of the countries, nominal shocks account for the largest fraction of the forecast error variance of changes in the trade balance, at both short and long forecast horizons. The relative importance of nominal shocks, however, tends to decline over longer forecast horizons. For a number of countries, the contribution of demand shocks is quite important, particularly over longer horizons. Supply shocks are quantitatively significant in these variance decompositions for only a handful of countries.

Table 6.

Variance Decompositions

article image
article image
article image
article image
Notes: The forecast error variance decompositions are for the changes in the trade variables. These decompositions indicate the proportion of the variance of the k-period ahead forecast error that is attributable to different types of shocks, i.e., supply (SS), demand (DD), and nominal (NM) shocks. Bold entries indicate statistical significance at the 5 percent level. Standard errors for the variance decompositions were computed using Monte Carlo simulations with 500 replications.