Effects of Exchange Rate Volatility on Trade: Some Further Evidence

The volatility, frequency, and erratic pattern of exchange rate movements witnessed since the beginning of generalized floating have led to widespread interest in the nature and extent of the impact of such movements on trade. A principal concern is that exchange rate volatility appears to increase the risk and uncertainty in international transactions and may therefore adversely affect trade and investment flows. This concern has strengthened in recent years in response to increasing protectionist trends and slowing growth of world trade, and numerous empirical studies have been written on the subject. The generally inconclusive findings of these studies, however, have failed to provide any empirical basis for the view that exchange rate volatility has discouraged international trade. Indeed, a recent survey of the empirical studies examining the effects of increased exchange rate volatility on international trade concluded that “the large majority of empirical studies … are unable to establish a systematically significant link between measured exchange rate variability and the volume of international trade, whether on an aggregated or on a bilateral basis” (International Monetary Fund (1984, p. 36)). A recent paper by Akhtar and Hilton (1984a) examines afresh the issue of whether exchange rate uncertainty, proxied by observed exchange rate volatility, has had statistically significant adverse effects on international trade. 1

Abstract

The volatility, frequency, and erratic pattern of exchange rate movements witnessed since the beginning of generalized floating have led to widespread interest in the nature and extent of the impact of such movements on trade. A principal concern is that exchange rate volatility appears to increase the risk and uncertainty in international transactions and may therefore adversely affect trade and investment flows. This concern has strengthened in recent years in response to increasing protectionist trends and slowing growth of world trade, and numerous empirical studies have been written on the subject. The generally inconclusive findings of these studies, however, have failed to provide any empirical basis for the view that exchange rate volatility has discouraged international trade. Indeed, a recent survey of the empirical studies examining the effects of increased exchange rate volatility on international trade concluded that “the large majority of empirical studies … are unable to establish a systematically significant link between measured exchange rate variability and the volume of international trade, whether on an aggregated or on a bilateral basis” (International Monetary Fund (1984, p. 36)). A recent paper by Akhtar and Hilton (1984a) examines afresh the issue of whether exchange rate uncertainty, proxied by observed exchange rate volatility, has had statistically significant adverse effects on international trade. 1

The volatility, frequency, and erratic pattern of exchange rate movements witnessed since the beginning of generalized floating have led to widespread interest in the nature and extent of the impact of such movements on trade. A principal concern is that exchange rate volatility appears to increase the risk and uncertainty in international transactions and may therefore adversely affect trade and investment flows. This concern has strengthened in recent years in response to increasing protectionist trends and slowing growth of world trade, and numerous empirical studies have been written on the subject. The generally inconclusive findings of these studies, however, have failed to provide any empirical basis for the view that exchange rate volatility has discouraged international trade. Indeed, a recent survey of the empirical studies examining the effects of increased exchange rate volatility on international trade concluded that “the large majority of empirical studies … are unable to establish a systematically significant link between measured exchange rate variability and the volume of international trade, whether on an aggregated or on a bilateral basis” (International Monetary Fund (1984, p. 36)). A recent paper by Akhtar and Hilton (1984a) examines afresh the issue of whether exchange rate uncertainty, proxied by observed exchange rate volatility, has had statistically significant adverse effects on international trade. 1

The results of Akhtar and Hilton’s study differ from the findings of other researchers. They find that exchange rate volatility, as measured by the standard deviation of indices of nominal effective exchange rates, has had significant adverse effects on the aggregate trade in manufactured goods of the United States and the Federal Republic of Germany. On the basis of regression results for export and import price and volume equations, the authors report a marginally significant adverse effect of exchange rate volatility on U.S. export volumes and U.S. import prices and significant adverse effects on German export and import volumes. Therefore, the authors conclude that nominal exchange rate uncertainty has had a significant negative effect on trade. Although Akhtar and Hilton’s results from a similar exercise based on a measure of real exchange rate volatility are less conclusive, they find the weight of the evidence sufficient to conclude that “from the perspective of international trade, it is desirable to reduce exchange rate uncertainty or variability” (1984a, p. 73). They go on to suggest that this objective may be accomplished through changes in macroeconomic policies, official intervention, or substantial changes in the exchange rate system. The authors do note that, notwithstanding the possible adverse effect of exchange rate uncertainty on trade, other considerations may still support present floating exchange rate arrangements.

The purpose of the present study is to test the robustness of Akhtar and Hilton’s empirical results, with their basic theoretical framework taken as given. The analysis has two parts. The first simply extends Akhtar and Hilton’s analysis, which was limited to the United States and Germany, to include France, Japan, and the United Kingdom. The second examines the robustness of their results with respect to changes in the choice of sample period, volatility measure, and estimation techniques.

The main conclusion of the analysis is that the Akhtar-Hilton methodology fails “to establish a systematically significant link between measured exchange rate variability and the volume of international trade” (International Monetary Fund (1984, p. 36)). The results obtained are not sufficiently robust to indicate the presence of such a link. This is not to say that significant adverse effects cannot be detected in individual cases, but rather that, viewed in the large, the results tend to be insignificant or unstable. Specifically, the results suggest that straightforward application of the Akhtar-Hilton methodology to three additional countries (France, Japan, and the United Kingdom) yields mixed results; that the Akhtar-Hilton methodology seems to be flawed in several respects, and that correction for such flaws has the effect of weakening their conclusions; that the estimates seem to be quite sensitive to fairly minor variations in methodology; and that “revised” estimates for the five countries do not, for the most part, support the hypothesis that exchange rate volatility has had a systematically adverse effect on trade. In sum, the empirical results do not, in the author’s judgment, provide strong grounds for modifying the agnostic conclusions of the Fund survey cited above. Needless to say, and as already noted in that survey, “the failure to establish a statistically significant link … does not, of course, prove that a causal link does not exist” (International Monetary Fund (1984, p. 36)).

The remainder of the paper is organized as follows. Section I outlines the model used by Akhtar and Hilton and discusses its empirical implementation. Section II presents empirical results for the United States, Germany, Japan, France, and the United Kingdom based on the Akhtar-Hilton methodology and the conclusions they suggest. Section III discusses a number of technical problems with Akhtar and Hilton’s estimations and illustrates the empirical significance of these shortcomings by reference to Akhtar and Hilton’s results for the United States and Germany. Section IV outlines preferred methodological procedures and applies them to data for the five countries. Section V presents the conclusions and outlines some avenues for further research. Data definitions and sources, and a full set of regression results for the five countries, are given in the two appendices.

I. The Akhtar-Hilton Model

One of the principal arguments against floating exchange rates has been that they lead to heightened risk and uncertainty in international transactions and thus discourage trade and in-vestment flows. If market participants are risk averse, exchange rate uncertainty and the need to provide against unfavorable changes will lead to supply and demand decisions that will yield higher prices or reduced levels of transactions at any given price. In addition, exchange rate uncertainty and the resultant uncertainty about the price to be paid or received in international trade may lead, other things being equal, to a preference for domestic over foreign markets. This preference in turn may lead to a gradual reduction in the volume of trade through a backward shift in supply and demand schedules. The size of the shift will depend on traders’ perceptions of the risks involved, on the extent of exchange rate uncertainty, and on the elasticities of supply and demand. 2

International traders can, of course, avoid or minimize foreign currency uncertainty in a short-term trading transaction by hedging in the forward market. But forward markets for maturities beyond one year are not well developed, and thus most trading activity, which requires decisions made with respect to a medium-to long-term time horizon, is unprotected by forward cover. Even if forward cover were available for longer maturities, such markets could not eliminate exchange rate uncertainty as long as traders are unable to predict the magnitude and timing of all their foreign exchange payments or earnings (Lanyi (1969)).

Akhtar and Hilton, therefore, disregard the possibilities for forward cover and postulate a standard set of demand and price equations, with each equation augmented to include the exchange rate volatility variable. In the volume equations this variable is expected to reflect the effect on demand of the price uncertainty associated with exchange rate uncertainty when invoices are denominated in a currency other than that of the demander. Similarly, in the price equations the volatility variable is expected to reflect the increase in supply prices induced by increased exchange rate uncertainty when invoices are denominated in a currency other than that of the supplier.

Specifically, Akhtar and Hilton postulate the following four equations. The export demand equation is

XV=f[YF,(PXr/PFf),SXf],(1)

where XV is the quantity index of total manufacturing exports delivered; YF is the real foreign activity level; PX is the price of manufacturing exports, in domestic currency; r is the foreign currency price of domestic currency; PFf is the price of foreign-produced substitutes of exports, in foreign currency; and SXf is the exchange rate risk facing demanders of exported goods.

The import demand equation is

MV=g(YD,PM/PD,SMd),(2)

where MV is the quantity index of total manufacturing imports delivered; YD is the real domestic activity level; PD is the price of domestically produced substitutes for imported manufactures, in domestic currency; PM is the price of foreign-produced manufacturing goods faced by domestic consumers, in domestic currency; and SMd is the exchange rate risk facing demanders of imported goods.

The export supply equation is

PX=F(UCD,SXd),(3)

where UCD represents input costs of domestic manufactured output, in domestic currency, and SXd is the exchange rate risk facing domestic producers of the exported commodity.

Finally, the import supply equation is

PM=G(UCFf/r,SMf),(4)

where UCFf represents input costs of foreign manufactured output, in foreign currency; r is the foreign currency price of domestic currency; and SMf is the exchange rate risk facing foreign producers of the imported commodity.

It is assumed that prices are set on the date a contract is made rather than on the delivery date. The price term in equations (1) and (2) is the relative price that exporters and importers expect to receive or to pay on delivery, which is when payment is assumed to be made. If the contract price is quoted in domestic currency—say, for the importer—the importer faces no price uncertainty. When payment must be made in foreign currency, however, the (domestic) price of imports cannot be known if uncertainty exists about future exchange rates or if the importer does not hedge. This exchange rate risk is denoted by SMd. A similar reasoning applies to the export demand equation, in which the exchange rate risk is denoted by SXd.

Equations (3) and (4) imply that the foreign supply of imported manufactured goods and the domestic supply of manufactured exports are both assumed to be perfectly elastic with respect to the volume of trade. Although not very realistic, this assumption permits one to use ordinary least-squares regression procedures for estimating the structural equations, since it implies that the supply price of traded goods is unaffected by the volume of trade. Akhtar and Hilton note that this assumption permits a distinction between the different effects of volatility on demanders and suppliers in terms of the price and quantity of traded commodities. They point out that the perfectly competitive market structure imposed by this assumption implies that, whereas the uncertainty faced by suppliers cannot directly affect the volume of trade demanded, such uncertainty can indirectly affect trade volumes by raising the price of traded goods.

An important methodological issue concerns specification of the exchange rate volatility measure. A first question is whether to base the measure on the nominal or the real exchange rate, a question that hinges on which rate better captures the risk or uncertainty faced by traders, particularly over the medium-term planning horizon adopted by them. It is frequently argued that, over this time horizon, the real exchange rate is the more relevant measure because the effects of uncertainty on a firm’s revenues and costs that arise from fluctuations in the nominal exchange rate are likely to be offset in large part by movements in costs and prices. Akhtar and Hilton, however, opt for the nominal exchange rate: first, because of the highly unpredictable nature of exchange rate changes and, second, because of the lack of empirical support for purchasing power parity over the medium term. Given the short time horizon over which exchange rate variations are examined in Akhtar and Hilton’s analysis, it is probably correct to suppose that most of the variability in the real exchange rate comes from the variability in the nominal rate.

Given the choice of nominal over real exchange rates, a second question concerns the volatility measure that is appropriate for use in empirical work. The various measures that have been used include the more conventional ones, such as the standard deviation of the levels of exchange rates or of the changes in these rates, and others, such as absolute percentage first differences of exchange rates, nonparametric measures such as Gini’s mean difference, and measures based on the estimated ex ante (rather than ex post) exchange rate. Each of these has advantages and drawbacks. (For a discussion of the merits and limitations, see Brodsky (1984), Kenen (1979), Lanyi and Suss (1982), and Rana (1981).) The measure chosen by Akhtar and Hilton is the standard deviation of the level of the daily effective exchange rate during each quarter. They experimented with other volatility measures, but these results are not reported in their paper. These alternative measures were the natural log of the volatility measure, the standard deviation of the natural log measure, the standard deviation of the daily percentage changes of the multilateral exchange rate indices, the trade-weighted averages of the standard deviations of the daily observations of a country’s bilateral exchange rates, and Gini’s mean difference coefficient. Akhtar and Hilton note that use of alternative volatility measures in general yielded similar results, although all alternative measures were not used in each supply and demand equation for the two countries they studied (the United States and the Federal Republic of Germany). The rationale underlying their use of the standard deviation of the daily effective exchange rate in each quarter is that the average exchange rate for the quarter is the best predictor of the expected rate for each day of the quarter. (For further discussion of the choice of volatility measure, see the ninth paragraph of Section III.)

However volatility is measured, the relation between trade flows and exchange rate uncertainty may not be independent of, and cannot easily be separated from, other uncertainties faced by traders (such as those relating to other features of the economic environment: for instance, the increase in uncertainty in the economic environment in the last decade may also be attributed to the oil price shocks, which overlapped with the advent of floating exchange rates). There is, therefore, need for considerable caution in interpreting empirical results. To the extent that other sources of risk faced by traders are partially offset by exchange rate fluctuations (or vice versa), the empirical results will overstate (or understate) the effects of exchange rate uncertainty on trade.

Akhtar and Hilton modified the basic structural equations, equations (1) through (4) above, for empirical purposes as follows: variables for domestic and foreign capacity utilization were added to each of the equations; domestic and foreign unit cost variables were proxied by series for domestic and foreign prices for manufactures; price equations were extended to include competitor price variables; seasonal dummy variables were added to all equations; and dock-strike dummy variables were included in the volume equations for the United States.

Independent variables other than the price and volatility variables were assumed to enter the equations either contemporaneously or with a one-quarter lag. The relative price terms in the volume equations were permitted a lag of up to eight quarters to account for order-delivery lags. A similar lag structure was imposed on the volatility measure to take into account not only order-delivery lags but also the gradual adjustment of expected volatility to actual volatility. Lags of up to eight quarters were imposed on the exchange rate volatility and relative price variables in the volume equations and on the volatility variable in the price equations. In all these cases, a second-degree polynomial lag structure was imposed with a zero (far) end-point constraint. A one-iteration Cochrane-Orcutt (CO) procedure was employed to correct for first-order serial correlation in all equations. Except for the volatility measure, the natural log of all variables was used, and hence the estimated coefficients represent elasticities. The resultant equations were estimated in Akhtar and Hilton’s study (1984a) using U.S. and German quarterly data for various estimation periods between 1974 and 1982. Akhtar and Hilton focused on the results for the period from the first quarter of 1974 through the fourth quarter of 1981, although they also presented results for the extended period through the last quarter of 1982.

II. Empirical Results Based on the Akhtar-Hilton Methology

Table 1 presents, for five countries, regression estimates for the volatility coefficient that result from the model and estimation procedures just described. The results for the United States and Germany are, of course, those reported in Akhtar and Hilton’s paper. As noted previously, these estimates suggest that nominal exchange rate volatility has had a marginally significant adverse effect on U.S. export volume, and significant negative effects on German export and import volumes. For U.S. imports, volatility was found to reduce import volumes indirectly through a marginally significant positive effect on import prices. In quantitative terms, the results indicate that a 10 percent increase in exchange rate volatility can lead to ½ percent reduction in German exports and a reduction in U.S. exports of about ½ percent. On the import side, a 10 percent increase in volatility can cause a 1 percent decline in German imports and a ½ percent decline in U.S. imports. It is these results that are the basis for the authors’ conclusion that exchange rate uncertainty has had a significant adverse effect on the volume of international trade.

Table 1.

Regression Results for Exchange Rate Volatility Variable Based on Akhtar-Hilton (A-H) Methodology

article image
Note: Figures in parentheses are t -statistics; * denotes statistical significance at the 5 percent level; ** denotes statistical significance at the 1 percent level.Sources: For the United States and Germany, A-H (1984a); for France, Japan, and the United Kingdom, own calculations (see the text and Appendices I and II).

The relative price variable in the export volume equation is not statistically significant.

In the calculations all coefficients that were not significant at the 5 percent level were assumed to be equal to zero. For the United States, the coefficients in the export volume and import price equations are significant only for a one-tail test.

Table 1 also presents results of the Akhtar-Hilton methodology for three additional countries—France, Japan, and the United Kingdom. These estimates were obtained by replicating the Akhtar-Hilton methodology as closely as possible. The main difference is that the volatility measures for these three countries were derived as the standard deviations of daily observations within each quarter of the effective exchange rate index for each country, weighted according to the Fund’s multilateral exchange rate model, MERM (see Artus and McGuirk (1981); the computation and sources of this and other data series for these countries are presented in Appendix I).

The results for France, Japan, and the United Kingdom are rather different from those obtained by Akhtar and Hilton for the United States and Germany. In the first place, all the coefficients for the United Kingdom are not only insignificant but are also of the “wrong” sign. Second, whereas the effects in Akhtar and Hilton’s results for the United States and Germany came through primarily on the volume side, the results for France and Japan suggest instead that any adverse effect of volatility on trade is indirect rather than direct. Thus, the volatility variables in the volume equations are either of the “wrong” sign or quite insignificant for France and Japan but are of the “right” sign and significant in the price equations. Given Akhtar and Hilton’s interpretation of the export and import price equations as domestic and foreign supply equations, respectively, the results have the rather paradoxical implication that, whereas French and Japanese exporters bear the exchange rate risk (a not unrealistic result if most export contracts are denominated in foreign currency), French and Japanese importers do not. Rather, it is the foreign suppliers to French and Japanese markets who ostensibly bear the exchange rate risk and who accordingly raise their prices to cover themselves for that risk. This result is not very plausible, especially in light of the magnitudes of the corresponding coefficients in the import price equations for the United States and Germany, which suggest that suppliers to those markets bear a substantially smaller exchange risk. (See Section IV, footnote 7.)

The interpretation of the coefficients in Table 1 is somewhat obscured by the fact—stemming from the particular specification chosen by Akhtar and Hilton—that these coefficients are not elasticities. Nevertheless, because the average value of the volatility variable often tends to be around unity, the coefficients in Table 1 can be roughly interpreted as elasticities. Indeed, regression equations identical to those used to produce the results of Table 1, except for a logarithmic transformation of the volatility variable, yield regression coefficients very similar to those shown in Table 1. The only significant change in the results is that three of the eight significant and “right sign” coefficients lose their statistical significance (the coefficients for the U.S. and for the Japanese import price equations, and the coefficient for the U.S. export volume equation; therefore, the results with the logarithmic transformation of the volatility variable show no effect—direct or indirect—of volatility on U.S. import and export volumes). Be that as it may, the statistically significant parameters in Table 1, together with the price elasticity parameters from the volume equations, may be used to calculate the overall (that is, direct and indirect) elasticity of trade flows with respect to volatility. These implied total elasticities are shown, as a memorandum, in the lower tier of Table 1. Taken at face value, they suggest that German trade and Japanese trade are relatively sensitive to exchange rate volatility, French trade and U.S. trade are relatively insensitive to exchange rate volatility, and British trade is quite insensitive to such volatility.

III. Shortcomings of the Akhtar-Hilton Methodology

The reliability of the results reported in Table 1 is undermined by several technical problems associated with the methodology employed by Akhtar and Hilton. These problems are discussed in turn below, and their quantitative significance is illustrated by reference to modified results for the United States and Germany. The section ends with the presentation of a set of improved procedures, which are then applied in the following section to the data for each of the five countries shown in Table 1. It should be understood that the critique that follows pertains only to the empirical procedures adopted by Akhtar and Hilton. Their basic analytical framework is taken as given.

Although Akhtar and Hilton state that they have tested each equation for first-order serial correlation, it appears that they have instead applied the CO correction for serial correlation to all equations as a routine procedure without a preliminary check for the presence of serial correlation in the ordinary least-squares estimation. This seems likely because all equations reported by them incorporate a CO correction, and when their equations are estimated without correction there is no evidence of serial correlation in several of the equations. The use of the CO procedure in equations where there was no serial correlation implies an incorrect assumption about the structure of the error term in those equations. Moreover, as shown below, the ordinary least-squares and CO results were significantly different for many equations.

A second unusual feature of the Akhtar-Hilton methodology, also related to their procedure for correcting for serial correlation, is that the equations were estimated using a one-iteration CO procedure, which belongs to the class of “two-stage” generalized least-squares correction procedures for serial correlation, rather than the more customary iterative CO procedure. In the absence of serial correlation, and with it assumed that the other conditions of the classical regression model hold, ordinary least-squares estimators have all the desirable small-sample and asymptotic properties. Some of these properties are lost, however, when serial correlation is present. In this case, generalized least squares, iterative CO, and maximum likelihood estimators are all consistent and asymptotically equivalent to best-linear unbiased estimators. But the small-sample properties of these estimators are difficult to derive analytically, and any choice between them that is based on sampling properties must be made on the strength of Monte Carlo evidence (see Judge and others (1980, p. 187), and Kmenta (1971, p. 292)). Such evidence has not led to a clear-cut choice. In applied work, however, researchers usually choose the CO iterative procedure over the two-stage procedures.

For reasons noted above, a preferable set of procedures for correcting for autocorrelation would be first to estimate the equations by ordinary least-squares, check the Durbin-Watson (DW) statistic to test for serial correlation, and, for those equations in which serial correlation is present, estimate the equations using the iterative CO technique. The resultant estimates for the United States and Germany are presented in Table 2. The table shows that the results for Germany turn out to be relatively robust (at least with respect to the present change in procedures), but that results for the United States are not. For that country, the ordinary least-squares estimates of the export volume equation do not indicate the presence of serial correlation. Moreover, the ordinary least-squares estimate of the volatility coefficient is not statistically significant, thus invalidating the evidence for the direct (adverse) effect of exchange rate volatility on U.S. export volumes found by Akhtar and Hilton in using the CO correlation. The new results do, however, continue to indicate an indirect effect on import volumes via import prices, as reported by Akhtar and Hilton. Nevertheless, the adjustment in the procedure to correct for serial correlation has the overall effect of undermining a fourth of the evidence adduced by these authors in favor of their conclusion.

Table 2.

Effect on A-H Volatility Results of Changes in Procedures Used to Correct for Serial Correlation

article image
Note: Figures in parentheses are t-statistics; * denotes statistical significance at the 5 percent level; ** denotes statistical significance at the 1 percent level.Sources: A-H (1984a) and own calculations.

A second problem with Akhtar and Hilton’s procedures is their choice of sample period. The issue is twofold. First, why do they feature results for the 1974-81 period, even though they also report results for the 1974-82 period? Second, why do they include 1974, when that date implies, given the eight-quarter lags used in their analysis, use of data from the period before floating exchange rates? These questions are not trivial. On the first point, it is particularly important that the marginally significant negative effect of volatility on U.S. export volume for the sample period 1974-1981 (which is barely significant at the 95 percent level of confidence by a one-tail test) disappears altogether when the estimation period is extended to 1982 (see Table 3). Although the contradictory results for the extended sample period are reported by Akhtar and Hilton in their longer research paper (1984a), their main article on this research (1984b) focuses only on the results for the shorter sample period and presents only those results. Because it is desirable to examine the effects of volatility on trade flows over the longest available estimation period of the floating exchange rate regime, it is more appropriate to look at the extended estimation period, through 1982, rather than to stop with 1981. In sum, although the results for Germany do not differ significantly over the two sample periods (1974-81 versus 1974-82), in the case of the United States the effect of volatility on trade is reduced, under the extended period, to an indirect effect on import volumes via import prices, without any effect on exports.

Table 3.

Effect on A-H Volatility Results of Changes in Sample Period

article image
Note: Figures in parentheses are t-statistics; * denotes statistical significance at the 5 percent level;** denotes statistical significance at the 1 percent level.Source: A-H (1984a) and own calculations.

The significance of the choice of sample period becomes more crucial when the analysis turns to the choice of a starting point. Akhtar and Hilton place considerable emphasis on their results being superior to those of most previous researchers in part because they have used observations only from the period of flexible exchange rates. Yet, although their estimation period does indeed begin in 1974 (the initial observation for the dependent variable is for the first quarter of 1974), their analysis necessarily includes observations from the period of fixed rates, since they use an eight-quarter lag structure for the exchange rate volatility variable and for the relative price variables. In effect, therefore, their estimates in part reflect developments reaching back to early 1972. Thus it is possible that bias in specification may be introduced owing to the change in the exchange rate regime. To avoid this possibility and to exclude observations from the period of fixed rates, the sample period should begin with the first quarter of 1975.

But such a truncation of the sample period has a major impact on the estimates. As can be seen from Table 3, the equations failed to reproduce all four pieces of significant evidence reported by Akhtar and Hilton for the 1974-81 period. For the United States, the results do not support either a direct or indirect effect of exchange rate volatility on export or import volumes. For Germany, the re-estimation fails to corroborate the direct effect of volatility on export and import volumes and, indeed, suggests a “perverse” negative effect on import prices not detected in Akhtar and Hilton’s estimation for the period. Overall, Akhtar and Hilton’s results are evidently quite sensitive to the inclusion of observations for the transition years preceding adoption of floating rates. Not only does inclusion of data from the period of fixed rates introduce specification bias into the estimation results, it also complicates derivation of the standard deviations of exchange rates over this period. One may have to rely on monthly rather than daily observations of the exchange rate to derive the quarterly standard deviations and to compare end-period values with period averages, given the discrete rate changes that occurred during this period. Moreover, the appropriate measure for exchange rate volatility during the period of fixed exchange rates may be different from that for the period of floating rates. These problems are all solved by restricting the estimation period strictly to the period of floating rates.

Another potential problem with the Akhtar and Hilton’s procedures is the arbitrary basis for their specification of the second-degree polynomial lag structure, which they defended on the grounds of simplicity. This specification is questionable chiefly because of the extreme sensitivity, amply documented by other researchers, of regression results based on this technique to changes in the specification of the lag structure. 3 It was thus deemed important to test the robustness of Akhtar and Hilton’s results under alternative specifications. Such testing sometimes yielded significantly different regression results for the United States. For instance, with a third-degree polynomial, the adverse effect of volatility on U.S. export volumes disappears once again, and there is no indication of a direct effect on import volumes (Table 4). Moreover, the effect on import prices found with a second-degree polynomial (by Akhtar and Hilton) loses its significance. The effect on export volumes also disappears when alternative lag structures and end-point constraints are used in the estimation procedure. The results for Germany are more robust when subjected to alternative dynamic specifications of the basic equations. As shown in Table 4, the results for Germany obtained with a third-degree polynomial lag continue to show a direct effect of volatility on trade volumes (given, of course, the 1974-81 sample period featured by Akhtar and Hilton).

Table 4.

Effect on A-H Volatility Results of Change in Polynomial Specification of Volatility Variable

article image
Note: Figures in parentheses are t-statistics; * denotes statistical significance at the 5 percent level;** denotes statistical significance at the 1 percent level.Sources: A-H (1984a) and own calculations (see the text).

A further aspect of Akhtar and Hilton’s procedures that merits attention is their specification of the effective exchange rates used to compute the volatility variables for each country. These variables were derived on the basis of trade-weighted exchange rate indices constructed with respect to major trading partners (9 countries for the United States and 13 for Germany). These measures for the effective exchange rate may be somewhat narrow. The broadest possible measure for the effective exchange rate is more likely to capture accurately the way in which exchange rate factors influence trading uncertainty. Furthermore, the Akhtar-Hilton method of weighting by trading partners allows each country’s effective exchange rate to vary depending on which countries were omitted in that computation. It may, instead, be more appropriate to treat all countries being examined symmetrically by standardizing cross-country comparisons through use of the same set of countries (or bilateral exchange rates) for computing the effective exchange rates. (This point is made by Kenen and Rodrik (1984), although not with regard to Akhtar and Hilton’s research.) Therefore, in computing the effective exchange rate, it is more appropriate to use a range of trading partners that is wider than that used by Akhtar and Hilton.

In light of these difficulties, and with a view to extending the results to France, Japan, and the United Kingdom, an alternative measure for exchange rate volatility was computed. This measure—again the standard deviation of the daily observations within each quarter—was calculated by using a daily version of the Fund’s MERM-weighted effective exchange rate index (see Artus and McGuirk (1981) for a description of the MERM weights). This index uses bilateral exchange rates for 18 industrial countries and, therefore, satisfies both the conditions noted in the previous paragraph. Charts 1 and 2, for the United States and Germany, respectively, compare movements in the volatility variables used by Akhtar and Hilton and in the alternative measures described above. Although the two measures of volatility move together, that used by Akhtar and Hilton in general exhibits greater variance, which might account, in part, for the differences in the estimation results described below.

Chart 1.
Chart 1.

Exchange Rate Volatility, United States, 1973-83

Citation: IMF Staff Papers 1985, 003; 10.5089/9781451972863.024.A004

Sources: A-H (1984a) and own calculations using Fund effective exchange rate indices weighted according to the multiple exchange rate model (MERM; see the text and Artus and McGuirk (1981)).
Chart 2.
Chart 2.

Exchange Rate Volatility, Federal Republic of Germany, 1973-83

Citation: IMF Staff Papers 1985, 003; 10.5089/9781451972863.024.A004

Sources: See Chart 1.

For the United States, use of the alternative volatility measure yielded an insignificant coefficient estimate for this variable in the export volume equation (Table 5). The coefficient of the volatility variable continued to have the perverse, positive sign in the U.S. import volume equation obtained by using Akhtar and Hilton’s volatility measure. The results also continued to show a weakly significant effect of volatility on U.S. import prices over the 1974-81 period. The substitution of the MERM-based volatility measure had a somewhat greater effect on the estimates for Germany. On the one hand, the volatility coefficient in the export volume equation became smaller, and the adverse effect on import volume lost its statistical significance. On the other hand, a possible positive effect of exchange rate volatility on import prices came to the fore.

Table 5.

Effect on A-H Volatility Results of Change in Volatility Measure

article image
Note: Figures in parentheses are t-statistics; * denotes statistical significance at the 5 percent level;** denotes statistical significance at the 1 percent level.Sources: A-H (1984a) and own calculations (see the text).

A final point that requires comment is Akhtar and Hilton’s use of dummy variables to correct for seasonality in the regression equations. Although both seasonally adjusted and unadjusted variables appear in the regression equations, the authors do not indicate which variables were adjusted and which were not. In correcting for seasonality, it is preferable to start with all non-seasonally adjusted series and correct for seasonality using dummy variables or other techniques or, alternatively, to use all series that are seasonally adjusted. In practice, time series are often published only in the adjusted form, thus giving rise to the likelihood of both adjusted and unadjusted series appearing in the estimation equation. Upon examination of Akhtar and Hilton’s data sources, it appears that the U.S. export volume series is the only dependent variable that was seasonally adjusted. Moreover, the seasonal dummy variables were statistically insignificant in this equation. The export volume equation was, therefore, re-estimated without such variables. This re-estimation continued to show a weakly significant adverse effect of volatility on U.S. export volumes. It is noteworthy that the volatility coefficient lost its significance when the dock-strike dummy variable was dropped from the estimation equation. As a separate issue, it would be interesting to examine any possible seasonality patterns in the exchange rate volatility series.

The foregoing findings on the implications for Akhtar and Hilton’s results of various changes in empirical methods cast doubts on the robustness of these results and thus raise questions regarding the main policy conclusion that the authors have derived from them. The direct, adverse effect of volatility on U.S. export volumes appears to be highly tentative and unstable and therefore may be disregarded. In addition, the replication results fail to provide a strong, consistent basis for any indirect effect of exchange rate volatility on trade volumes through its effect on prices of traded goods. The results for Germany are somewhat more robust than those for the United States but appear to be quite sensitive to the choice of sample period. Taken together, the apparent lack of stability and uniformity of the empirical results for both the United States and Germany significantly undermines the validity of findings reported in Akhtar and Hilton’s paper.

IV. Revised Results

Given the shortcomings associated with certain aspects of the Akhtar-Hilton methodology, the equations for the United States, Germany, France, Japan, and the United Kingdom reported in Table 1 were re-estimated using revised estimation procedures. The following changes were made to the Akhtar-Hilton methodology.

• An iterative CO correction procedure for serial correlation was used, but only when the ordinary least-squares estimation indicated the presence of serial correlation (determined on the basis of the DW statistic and the pattern of residuals).

• 1975-83 was chosen as the sample period, a period that, after allowance for lags, excludes observations from the period of fixed exchange rates but includes the experience of the entire period of floating rates thus far. 4

• Seasonal dummy variables were included in only those equations in which the dependent variable was not seasonally adjusted (for the United States, Germany, Japan, and the United Kingdom) and were excluded when it was seasonally adjusted (for France); the U.S. export volume series was seasonally adjusted, and the export volume equation for the United States was estimated excluding the dummy variables.

• The volatility measures were derived on the basis of standard deviations of daily observations within each quarter of the nominal MERM-weighted, effective exchange rate index for each country, as described in Section III.

• With some misgivings, a polynomial lag specification was imposed on the volatility and relative price variables. It was considered important to use some empirical criterion to determine the appropriate dynamic specification for the equations; that is, the length of the lag, the degree of the polynomial, and the choice of end-point constraints. The Akhtar-Hilton procedure of arbitrarily picking one specification for estimation was not used because it is unlikely that the identical specification would necessarily be appropriate for all four equations for all the countries. Furthermore, given the sensitivity of the results to the dynamic specification of the equations, it was deemed important to try alternative polynomial specifications. 5 Predictive testing was used to choose among the various specifications. In this procedure, one makes alternative assumptions about the correct lag length and the correct degree of the polynomial and chooses among them on the basis of their predictive ability. For this purpose, the equations were estimated for the 1975-81 and 1975-82 periods and extrapolated through 1983. The polynomial lag specification that yielded the smallest root mean square error was chosen as the preferred specification, to be used in the re-estimation over the full sample period, 1975-83.

The new results are shown in Table 6 (for a complete set of regression results, see Appendix II). It is difficult to interpret them as supportive of the hypothesis that exchange rate volatility has systematically undercut world trade. 6 On examining the new results for the United States and Germany, one notes that, of the four statistically significant results that formed the basis of Akhtar and Hilton’s policy conclusions, three no longer are either of the “correct” sign or statistically significant. The only robust coefficient in this respect is that in the equation for German export volumes. On the one hand, the adverse effect on German import volumes reported by Akhtar and Hilton is now quite insignificant, as are the effects on U.S. export volumes and import prices, which are now of the opposite sign. On the other hand, the new results point to a positive effect on U.S. export prices that was not detected by Akhtar and Hilton, an effect that would indirectly have an adverse effect on U.S. export volumes. The evident instability of these parameters points to the more general conclusion that the testing procedures used here do not appear to lend themselves to any very confident conclusions.

Table 6.

Revised Regression Results for Exchange Rate Volatility Variable, 1975-83

article image
Note: Figures in parentheses are t-statistics; * denotes statistical significance at the 5 percent level; ** denotes statistical significance at the 1 percent level. For the complete estimation results, see Appendix II.Source: Own calculations.

Estimation equation excludes seasonal dummies. Estimation results did not change substantively upon inclusion of seasonal dummies.

Because of unavailability of data, equations were estimated through the second quarter of 1983.

The (relative) price variable was not statistically significant in the corresponding volume equation.

In the calculations all coefficients that were not significant at the 5 percent level were assumed to be equal to zero.

If one broadens the analysis of the results in Table 6 to include France, Japan, and the United Kingdom, a first point to note is the paucity of any direct adverse effect of exchange rate volatility on trade volumes. Of the ten such coefficients in Table 6, only one is statistically significant and of the correct sign—the one for Germany. Moreover, half of the coefficients have the wrong sign. The same is true of the price equations, where again close to half of the coefficients are negative. Of the positive coefficients in the price equations, however, four meet the standard test of statistical significance, suggesting that exchange rate volatility may adversely affect trade volumes by significantly raising international trade prices.

This conclusion, however, must be qualified in several important respects. First, the significance of two of these four coefficients (those for France) is undermined by the nonsignificance of the relative price terms of the volume equation. Second, the pattern of the coefficients is at odds with commonsense notions about the nature of the links between the various countries and the world market. Taken at face value, the export price equations suggest that U.S. and French exporters bear the exchange rate risk on their exports, whereas exporters from Germany, Japan, and the United Kingdom do not. In light of conventional beliefs about the relative market positions of countries in world trade and of information on the currency composition of countries’ trade, it is difficult to provide a clear rationale for this pattern. 7 Rather, one would expect that Japanese and French exporters—and, perhaps to a lesser extent, British exporters—would bear relatively more exchange rate risk. The results on the import side are in this respect scarcely more reasonable. From Akhtar and Hilton’s perspective, the import price equations represent the world’s supply price to the country in question. Hence the positive volatility coefficients for France and Japan suggest that the world bears the exchange risk in exporting to these countries, a risk that it does not incur when exporting to the United States, Germany, or the United Kingdom. Again, these results are not very plausible. Rather, one would expect that Japanese importers—and to a smaller extent French, British, and German importers—would bear the exchange risk, whereas the world would bear the risk in exporting to the United States.

With respect to the results by country, the main points to be noted, subject to the qualifications made above, are as follows. For the United Kingdom, the revised estimates do not show either direct or indirect effects of exchange rate volatility on trade, a result that corroborates the findings of another re-estimation of Akhtar and Hilton’s results for the United Kingdom. 8 For France, the results indicate positive effects on trade prices, but these effects are undermined by the nonsignificance of the relative price terms in the volume equations. For Japan, the results suggest that volatility might have reduced import volumes through its effect on prices of imported goods. Even here, however, there is no indication of a direct impact on trade volumes. In quantitative terms, the results for Japan suggest that a 10 percent increase in volatility could indirectly reduce import volumes by close to 1½ percent. For the United States, the results indicate that volatility might have reduced export volumes through an effect on export prices, with a 1 percent reduction in export volumes likely to result from a 10 percent increase in volatility. Once again, however, there is no evidence of a direct effect on export or import volumes. Only for Germany is there evidence of a direct adverse effect of volatility on export volumes. In quantitative terms, the results suggest that a 10 percent increase in volatility could reduce German export volumes by about 1 percent.

V. Conclusions and Avenues for Further Research

In sum, the empirical results for the five countries do not provide conclusive evidence that exchange rate volatility has had a statistically significant effect on trade flows. The results suggest that, even if there is some residual effect of exchange rate uncertainty on trade, this effect has not operated in a stable and consistent manner. Although the present analysis incorporated some improvements in the statistical methods used by Akhtar and Hilton, it would be advisable to adopt further refinements and stability tests in future work, given that the results are sensitive to alternative specifications of equations and sample periods. (For instance, to deal with the problem of choosing the optimal polynomial specification, a new estimator derived by Kashyap and others (1984) might be used. This is a Bayesian distributed-lag estimator that is consistent with Shiller’s (1973) “smoothness prior” and uses sample data to improve the operating characteristics of both Almon’s and Shiller’s estimators; it has been shown to be superior in forecasting when compared with either Almon’s or Shiller’s estimator.)

More fundamentally, the equations should be re-estimated using a volatility measure that reflects only the unpredictable element of exchange rate movements and, thereby, the short-term volatility or deviations of exchange rates around a long-term trend. The observed variability in flexible exchange rates typically reflects both systematic rate movements, which are largely predictable, and uncertain rate movements, which are largely unpredictable. To the extent that risk from predictable rate changes can be diversified away, it may be argued that it is the unanticipated component of exchange rate movements that is the appropriate proxy for the uncertainty in exchange rate transactions. In calculating exchange rate volatility, it may be important to eliminate the movements along some predictable long-term trend, since these movements are unlikely to reflect the risk associated with exchange rate transactions. But the standard deviation (used as a proxy for exchange rate risk) of a nonstationary process, such as the observed exchange rate series, reflects the short-term volatility in the time series as well as the movement along a long-term trend. The principal volatility measure used by Akhtar and Hilton is derived as the standard deviations of (the levels of) the observed exchange rate series and, therefore, includes both the trend movement and the short-term volatility in the exchange rate. With a view to focusing on the unpredictable and short-term volatility in exchange rates, an alternative volatility measure, derived as the standard deviation of percentage changes in the exchange rate, was calculated in the present analysis. Preliminary testing with this measure led to only one significant change in the results: the direct adverse effect of volatility on German export volumes was no longer evident, thus invalidating the only direct effect of volatility on trade volumes, as reported in Table 6. Akhtar and Hilton also undertook some re-estimation with this measure, but these results were not reported in their paper (see Section II).

It is also not clear that the variation in daily exchange rates is the most appropriate unit of observation. As Solomon (1984, p. 16) Solomon has pointed out, daily fluctuations around a “steady or, at least, predictable” trend in the exchange rate may not dissuade traders from making transactions because traders can choose to postpone or speed up foreign exchange conversions over brief periods. It is also noteworthy that earlier empirical findings suggest that it is the longer-term (rather than weekly or daily) movements in exchange rates that might influence the decisions of traders (see a review of the literature by Farrell, with De Rosa and Crown (1983)). Finally, because Akhtar and Hilton’s theoretical specification excludes the more fundamental economic determinants of the exchange rate (and, thus, the basic cause of its variability) and of trade volumes and prices, the identification of a robust empirical relationship between volatility and trade flows could merely reflect the common effects of such omitted variables on these two factors rather than any causal link between them. Conclusive evidence on the effect of exchange rate volatility on trade may, therefore, hinge on the specification of a markedly more comprehensive model.

APPENDIX I: Data Definitions and Sources

This Appendix presents a description of the construction of, and the data sources for, the variables used in estimations. The data used in the estimations reported for the United States and the Federal Republic of Germany for the period 1974-81 were those provided to the author by Akhtar and Hilton. Because it was difficult to extend these series through 1983, new data series were computed by the author, for the entire 1973-83 period, using the same procedures adopted for computing the data for France, Japan, and the United Kingdom (see Sections II and IV of the text). Although the computation methods are broadly similar to those used by Akhtar and Hilton, the new series for the United States and Germany do not appear to be strictly comparable to those provided by those authors.

In general, all variables were constructed on a basis conceptually similar to that in Akhtar and Hilton (1984a).

Import and Export Volume and Average Value Indices

Import and export volume and average value indices for manufactured goods were used to denote the trade volume and price variables, respectively. Non-seasonally adjusted quantity and average value indices from the Organization for Economic Cooperation and Development, Trade Series A (Paris: OECD, various issues) were used for Germany, Japan, and the United Kingdom. Trade volume and price data for the United States were obtained from U.S. Department of Commerce, International Economic Indicators (Washington: Government Printing Office, various issues) and other Department of Commerce data bases. Of the four series for the United States, only the export volume data are published on a seasonally adjusted basis. For France, seasonally adjusted series of export and import value in constant prices were used because quantity indices could not be obtained. These variables and the export and import price indices were obtained from Institut National de la Statistique et des Etudes Economiques, Les comptes nationaux trimestriels: Series longues, 1963-83 (Paris: INSEE, 1984).

Exchange Rate Volatility

The exchange rate volatility variable was defined as the standard deviation of the daily observations of the Fund’s MERM-weighted effective exchange rate index within each quarter over the period from the second quarter of 1973 through the fourth quarter of 1983. Observations for the exchange rate index for the period from the first quarter of 1972 through the first quarter of 1973 were obtained from the Treasurer’s Department of the Fund. For this period, the standard deviations were based on monthly observations of the effective exchange rate index within each quarter. The MERM weights used in computing the effective exchange rate indices are described and listed in Artus and McGuirk (1981).

Domestic Real Income

The domestic real income variable for each country was denoted by an index of each country’s seasonally adjusted real gross national product (GNP), or gross domestic product (GDP). These indices were obtained from a data base, maintained in the Fund’s Research Department, that comprises data from national sources provided by desk economists throughout the Fund. These data series are either identical or quite similar to those maintained in International Financial Statistics (IFS), International Monetary Fund (Washington).

Domestic Capacity Utilization

A measurement of capacity utilization was obtained for each country by taking the ratio of an index of seasonally adjusted industrial production to an index reflecting trend growth in industrial production. The data for industrial production were taken from the IFS. Measures of trend growth were computed for different sample periods.

Domestic Prices

The domestic price variable was represented by the wholesale price index for manufactured goods as reported in the IFS.

Foreign Real Income

The foreign real income variable was computed as a trade-weighted average of the real GNP (GDP) indices for all industrial country and developing country trading partners. These data were derived from the GEE (Global Economic Environment) data base, maintained in the Fund’s Research Department. The industrial countries included in this data base are the United States, the United Kingdom, Germany, France, Japan, Canada, Italy, Austria, Belgium, Denmark, the Netherlands, Norway, Sweden, Switzerland, Spain, Ireland, Finland, Australia, and New Zealand. Both oil exporting and non-oil developing countries are included in the data base, with individual weights assigned to the larger countries in these two groups.

Foreign Capacity Utilization and Prices

The foreign capacity utilization and foreign price variables were computed as trade-weighted averages of the corresponding indices for all industrial country trading partners. The data were drawn from the GEE data base (for the industrial production data underlying the capacity utilization variables) and from the IFS (for the wholesale price indices).

Base-year (1975) total export values were used in computing the trade weights for all of the three “foreign” variables described above.

Exchange Rates

Period average, bilateral exchange rates drawn from the IFS were used as conversion factors in the computation of the foreign real income and relative price variables.

APPENDIX II: Regression Results by Country

This Appendix provides complete regression results for each country that correspond to those reported in Table 6 of the text. The t-statistics are given in parentheses under the coefficient estimates. All variables, except for the dummy variables and the volatility variable, enter the equation in natural log form. For some equations, more than one polynomial lag specification yielded similar root mean square errors. Only one specification is presented in these instances, because the coefficient estimates did not differ significantly for these alternative estimations.

As can be seen from the estimates, the coefficient of the real income variable is of the right sign and is significant in most equations. The results for the capacity utilization variable and the price variables are somewhat mixed. In a few instances, when a low Durbin-Watson (DW) test statistic was found, re-estimation using the iterative Cochrane-Orcutt (CO) procedure yielded an insignificant estimate of the autocorrelation coefficient. This finding may indicate the possible misspecification of the structural equation rather than the presence of serial correlation.

Variables were defined as follows:

article image

The regression results are arrayed by country in the following subsections. R2 is the adjusted coefficient of determination; e is the error term.

United States

Export volume equation, 1975-83:

XVt=7.9(4.72)+1.57(2.56)CUFt1+0.10(0.22)RFYt1+0.14(2.70)S0.82(4.05)RP1+0.25(1.94)et1DW=1.90;R¯2=0.879
article image

Export price equation, 1975-83:

XPt=1.30(2.97)+0.02(0.75)D10.04(1.00)D20.06(1.96)D3+0.01(0.14)CUt1+0.55(4.22)PDt1+0.10(4.16)S+0.17(2.01)RP3t1+0.85(36.1)et1DW=1.91;R¯2=0.993
article image

Import volume equation, 1975-83:

MVt=2.46(1.74)0.01(0.50)D1+0.04(2.14)D20.02(0.12)D3+2.33(15.4)RDYt10.68(2.26)(CUF/CU)t10.02(1.36)S1.87(8.56)RP2DW=1.72;R¯2=0.973
article image

Import price equation, 1975-83:

MPt=0.10(0.32)+0.01(0.91)D1+0.04(0.62)D20.02(0.42)D30.01(0.07)CUFt1+0.41(4.87)PDt10.05(0.23)S+0.57(6.86)RP3t1+0.493.70et1DW=2.04;R¯2=0.994
article image

Federal Republic of Germany

Export volume equation, 1975-83:

XVt=4.11(2.17)0.05(4.05)D10.04(3.17)D20.09(6.80)D3+0.52(1.88)CUFt1+1.30(7.72)RFYt10.12S(2.55)1.14(4.58)RP1DW=1.60;R¯2=0.957

Re-estimation using the CO correction procedure yielded an insignificant autocorrelation coefficient (whose value was 0.10); therefore, the ordinary least-squares procedure was used for estimation.

article image

Export price equation, 1975-83:

XPt=0.89(9.89)+0.01(1.82)D10.001(0.21)D20.0001(0.05)D3+0.02(0.47)CUt1+0.88(10.32)PDt1+0.01S(0.58)0.07(0.92)RP3t1+0.38(2.39)et1DW=1.85;R¯2=0.994
article image

Import volume equation, 1975-83:

MVt=12.1(2.10)0.03(1.95)D10.01(0.71)D20.09(6.08)D3+1.70(4.41)RDYt1+0.22(0.29)(CUF/CU)t10.05(1.07)S3.29(3.76)RP2DW=1.60;R¯2=0.975

Re-estimation using the CO correction procedure yielded an insignificant autocorrelation coefficient (whose value was 0.10); therefore, the ordinary least-squares procedure was used for estimation.

article image

Import price equation, 1975–83:

MPt=0.89(7.50)+0.02(4.41)D1+0.01(2.53)D2+0.01(3.62)D3+0.23(2.72)CUFt1+0.43(3.77)PDt10.22(0.70)S+0.37(3.73)RP3t1+0.35(2.84)et1DW=2.21;R¯2=0.991
article image

France

Export volume equation, 1975-83:

XVt=0.34(0.11)+0.34(1.23)CUFt1+1.52(5.45)RFYt10.01(0.40)S0.44(1.03)RP1+0.39(2.51)et1DW=1.79;R¯2=0.977
article image

Export price equation, 1975–83:

XPt=0.50(11.33)+0.22(2.52)CUt1+0.57(8.57)PDt1+0.04(2.73)S+0.32(5.05)RP3t1+0.19(1.02)et1DW=1.85;R¯2=0.997
article image

Import volume equation, 1975–83:

MVt=2.90(0.48)+2.21(2.95)RDYt1+0.67(1.34)(CUF/CU)t1+0.05(0.61)S0.59(1.07)RP2+0.64(5.38)et1DW=1.82;R¯2=0.975
article image

Import price equation, 1975-83:

MPt=1.31(28.7)+0.41(4.36)CUFt1+0.19(2.49)PDt1+0.03(2.67)S+0.52(7.26)RP3t1DW=1.84;R¯2=0.995
article image

Japan

Export volume equation, 1975-83:

XVt=8.12(3.71)0.09(8.45)D10.04(3.72)D20.03(3.37)D3+0.01(0.11)CUFt1+1.22(5.85)RFYt1+0.03(1.50)S1.96(7.08)RP1DW=1.84;R¯2=0.991
article image

Export price equation, 1975-83:

XPt=2.08(2.44)0.05(0.52)D1+0.01(0.85)D20.002(0.25)D3+0.22(1.02)CUt1+0.21(0.79)PDt1+0.03(1.33)S+0.32(2.71)RP3t1+0.65(5.13)et1DW=1.70;R¯2=0.942
article image

Import volume equation, 1975—83:

MVt=5.57(1.97)0.04(2.03)D1+0.03(1.60)D20.02(1.41)D3+1.11(3.47)RDYt10.80(1.14)(CUF/CU)t1+0.09(2.01)S1.33(2.73)RP2+0.51(3.96)et1DW=1.91;R¯2=0.967
article image

Import price equation, 1975-83:

MPt=2.70(3.43)+0.01(0.48)D1+0.03(0.16)D2+0.01(0.06)D3+0.27(1.09)CUFt10.48(1.59)PDt1+0.06S(2.43)+0.89(5.06)RP3t1+0.30(1.74)et1DW=1.66;R¯2=0.906
article image

United Kingdom

Export volume equation, 1975-83:

XVt=1.74(1.86)0.05(2.36)D1+0.02(1.06)D20.06(2.86)D3+0.22(0.54)CUFt1+1.44(3.52)RFYt1+0.04S(0.98)0.82(3.26)RP1DW=2.32;R¯2=0.654
article image

Export price equation, 1975–83:

XPt=0.36(2.34)+0.003(0.84)D10.003(0.94)D20.005(1.51)D3+0.18(2.04)CUt1+0.76(24.7)PDt10.01S(0.68)+0.17(3.60)RP3t1+0.57(5.17)et1DW=1.39;R¯2=0.997
article image

Import volume equation, 1975-83:

MVt=4.86(2.04)0.001(0.03)D1+0.06(2.57)D20.03(1.26)D3+2.91(8.25)RDYt10.54(CUF/CU)t1(1.25)+0.07(1.31)S0.87(3.87)RP2DW=1.89.39;R¯2=0.938
article image

Import price equation, 1975-83:

MPt=0.49(4.24)0.01(0.73)D10.01(0.81)D20.01(1.29)D3+0.36(3.03)CUFt1+0.34(10.37)PDt10.01(0.82)S+0.57(11.85)RP3t1DW=1.85;R¯2=0.993
article image

REFERENCES

  • Akhtar, M.A., and R. Spence Hilton (1984a), “Exchange Rate Uncertainty and International Trade: Some Conceptual Issues and New Estimates for Germany and the United States,” Research Paper No. 8403 (New York: Federal Reserve Bank of New York, May).

    • Search Google Scholar
    • Export Citation
  • Akhtar, M.A., and R. Spence Hilton (1984b), “Effects of Exchange Rate Uncertainty on German and U.S. Trade,” Federal Reserve Bank of New York Quarterly Review (New York), Vol. 9 (Spring), pp. 7-16.

    • Search Google Scholar
    • Export Citation
  • Artus, Jacques R., and Anne Kenny McGuirk, “A Revised Version of the Multilateral Exchange Rate Model,” Staff Papers, International Monetary Fund (Washington), Vol. 28 (June 1981), pp. 275309.

    • Search Google Scholar
    • Export Citation
  • Bank of England, “The Variability of Exchange Rates: Measurement and Effects,” Bank of England Quarterly Bulletin (London), Vol. 24 (September 1984), pp. 34649.

    • Search Google Scholar
    • Export Citation
  • Brodsky, David A., “Fixed Versus Flexible Exchange Rates and the Measurement of Exchange Rate Instability,” Journal of International Economics (Amsterdam), Vol. 16 (May 1984), pp. 295310.

    • Search Google Scholar
    • Export Citation
  • Farrell, Victoria S., with Dean A. DeRosa and T. Ashby McCown, Effects of Exchange Rate Variability on International Trade and Other Economic Variables: A Review of the Literature, Staff Studies No. 130 (Washington: Board of Governors of the Federal Reserve System, December 1983).

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, Exchange Rate Volatility and World Trade: A Study by the Research Department of the International Monetary Fund, Occasional Paper No. 28 (Washington, July 1984).

    • Search Google Scholar
    • Export Citation
  • Judge, George G., William E. Griffiths, R. Carter Hill, and Tsoung-Chao Lee, The Theory and Practice of Econometrics (New York: Wiley, 1980).

    • Search Google Scholar
    • Export Citation
  • Justice, G., The Impact of Exchange Rate Variability on International Trade Flows, Discussion Paper (Technical Series) No. 4 (London: Bank of England, December 1983).

    • Search Google Scholar
    • Export Citation
  • Kashyap, A.K., P.A.V.B. Swamy, J.S. Mehta, and R.D. Porter, Estimating Distributed Lag Relationships Using Near-Minimax Procedures, Special Studies Paper No. 187 (Washington: Board of Governors of the Federal Reserve System, Division of Research and Statistics, September 1984).

    • Search Google Scholar
    • Export Citation
  • Kenen, Peter B., “Exchange Rate Instability: Measurement and Implications,” International Finance Section Research Memorandum (Princeton, New Jersey: Princeton University, June 1979).

    • Search Google Scholar
    • Export Citation
  • Kenen, Peter B., and Dani Rodrik, “Measuring and Analyzing the Effects of Short-Term Volatility in Real Exchange Rates,” International Finance Section Working Paper in International Economics No. G84-01 (Princeton, New Jersey: Princeton University, March 1984).

    • Search Google Scholar
    • Export Citation
  • Kmenta, Jan, Elements of Econometrics (New York: Macmillan, 1971).

  • Lanyi, Anthony, The Case for Floating Exchange Rates Reconsidered, Essays in International Finance No. 72 (Princeton, New Jersey: Princeton University, February 1969).

    • Search Google Scholar
    • Export Citation
  • Lanyi, Anthony, and Esther C. Suss, “Exchange Rate Variability: Alternative Measures and Interpretation,” Staff Papers, International Monetary Fund (Washington), Vol. 29 (December 1982), pp. 52760.

    • Search Google Scholar
    • Export Citation
  • Magee, Stephen P., “U.S. Import Prices in the Currency-Contract Period,” Brookings Papers on Economic Activity: 1 (1974), The Brookings Institution (Washington), pp. 11764.

    • Search Google Scholar
    • Export Citation
  • Page, S.A.B., “The Choice of Invoicing Currency in Merchandise Trade,” National Institute Economic Review, National Institute of Economic and Social Research (London), No. 98 (November 1981), pp. 6072.

    • Search Google Scholar
    • Export Citation
  • Rana, Pradumna B., “Exchange Rate Risk Under Generalized Floating: Eight Asian Countries,” Journal of International Economics (Amsterdam), Vol. 11 (November 1981), pp. 45966.

    • Search Google Scholar
    • Export Citation
  • Schmidt, Peter, and Roger N. Waud, “The Almon Lag Technique and the Monetary Versus Fiscal Policy Debate,” American Statistical Association Journal, Vol. 68 (March 1973), pp. 1119.

    • Search Google Scholar
    • Export Citation
  • Shiller, Robert J., “A Distributed Lag Estimator Derived from Smoothness Priors,” Econometrica (Evanston, Illinois), Vol. 41 (July 1973), pp. 77588.

    • Search Google Scholar
    • Export Citation
  • Solomon, Robert, The Consequences of Exchange-Rate Variability, Brookings Discussion Papers in International Economics No. 24 (Washington: The Brookings Institution, December 1984).

    • Search Google Scholar
    • Export Citation
*

Ms. Gotur, an economist in the Research Department, is a graduate of George Washington University, where she was an assistant professor of economics.

The author is grateful to M. A. Akhtar and R. Spence Hilton for kindly providing the data used in their study, and she acknowledges the beneficial conversations held with her colleagues in the Research Department and with Bonnie Loopesko and P.A.V.B. Swamy.

1

Unless otherwise noted, all subsequent references to Akhtar and Hilton’s work are to their more comprehensive paper (1984a); their second paper (1984b) is a condensation of this first paper and reports only selective results.

2

Besides increasing costs through uncertainties, exchange rate fluctuations may result in costly shifts of resources between economic activities in response to changing price incentives or to greater riskiness perceived for the traded-goods sector. Large and persistent changes in real exchange rates can involve serious adjustment costs that affect investment decisions and trade patterns. Such resource shifts and the related economic costs are not, however, directly associated with the more short-term volatility in nominal exchange rates being examined in the present analysis and, therefore, are not of particular relevance to it. (For a review of the mechanisms by which exchange rate volatility could affect trade flows, see International Monetary Fund (1984).)

3

Misspecification of the lag renders the coefficient estimates biased and inconsistent, and the standard tests of hypotheses unreliable. See Judge and others (1980).

4

The data used in the estimations reported for the United States and Germany for the period 1974-81 were those provided to the author by Akhtar and Hilton. Because it was difficult to extend these series through 1983, new data series were computed by the author, for the entire 1973-83 period, using the same procedures adopted for computing the data for France, Japan, and the United Kingdom. Although the computation methods are broadly similar to those used by Akhtar and Hilton, the new series for the United States and Germany do not appear to be strictly comparable to those provided by Akhtar and Hilton.

5

The problems and difficulties that arise when using the Almon lag specification are well known. (See Judge and others (1980) and Schmidt and Waud (1973) for a review of the issues involved.) Thus it is sometimes preferable to opt for alternative lag specifications. In the present analysis, this option could not be used because the objective was to determine whether the results of Akhtar and Hilton’s analysis could be extended to other industrial countries, and therefore their broad empirical specification had to be adopted. Some researchers have handled the problem of determining the correct dynamic specification of an Almon lag estimation by using a procedure that searches over several possible values for the degree of the polynomial and the length of the lag and then by choosing the combination that minimizes the residual variances (maximizes the corrected coefficient of determination, R2). The changes in R2 for alternative estimations, however, are frequently too small to permit meaningful selection of one estimation over another.

6

Other studies do not help resolve the difficulty. Two recent studies in this area by Justice (1983) and Kenen and Rodrik (1984), published after the Fund survey (International Monetary Fund (1984)), have reported evidence that may be regarded as being more suggestive than conclusive. Despite testing with alternative measures for exchange rate volatility, Justice did not find a significant effect of such volatility on the volume of U.K. exports of manufactures in the period of floating rates. He did, however, find some tentative evidence suggesting that volatility had influenced export pricing behavior, but this result was heavily dependent on the particular volatility measure used in the estimation. Kenen and Rodrik have presented new data on the short-term volatility of real exchange rates for the members of the Group of Ten plus Switzerland and have analyzed the effect of such volatility on trade volumes. Their results are also mixed. For only three of the seven countries examined did they find a significant negative effect of volatility on export volumes; for three others, volatility was actually found to stimulate exports, whereas for the remaining five countries the coefficient of the volatility variable was not significantly different from zero. With respect to import volumes, they found a significant negative effect of volatility for four countries, a significant positive effect for two others, and a coefficient for the volatility variable not significantly different from zero for the remaining five countries.

7

Although information on currency composition has not been fully compiled, findings by Magee (1974) and Page (1981) clarify the issue. Citing estimates for 1979 and 1980, Page shows that over 95 percent of U.S. exports and 85 percent of U.S. imports are invoiced in U.S. dollars; 80 percent of German exports and 40 percent of German imports are invoiced in deutsche mark; 75 percent of U.K. exports and 40 percent of U.K. imports are invoiced in pound sterling; 60 percent of Japanese exports and 90 percent of Japanese imports are invoiced in U.S. dollars (with about 30 percent of exports invoiced in yen); 60 percent of French exports and 35 percent of French imports are invoiced in francs; and, finally, 30 percent of German, U.K., and French imports, respectively, are invoiced in U.S. dollars.

8

Applied to the equations for U.K. manufacturing trade volumes and for unit values in the Bank of England’s short-term model, the Akhtar-Hilton methodology yielded no significant effect of short-term volatility on trade for the United Kingdom (Bank of England (1984)). No significant direct effect was found on trade volumes; the relevant coefficients in the import price equation tended to be negative rather than positive, and those in the export price equation tended to be of the expected sign but statistically insignificant.

IMF Staff papers: Volume 32 No. 3
Author: International Monetary Fund. Research Dept.