I. Methodology

The usual method for isolating the influence of competitive factors on trade flows is to estimate simultaneously the combined influence of all factors considered to be significant for explaining those trade flows. The estimated coefficient of the variable chosen as a proxy for “competitiveness” and the changes in that variable then yield a measure of the changes in the trade flows that are due to changes in competitiveness.

The technique used here, on the other hand, attempts to eliminate movements in the trade flows that are due to seasonal, cyclical, and longer-run “income” factors and then to examine the residual. This residual may, under certain conditions, be interpreted as consisting of competitive and random phenomena. The first problem with this technique is the fact that cyclical and income effects are likely to be correlated with competitive developments. For instance, if a country benefits from a marked improvement in competitiveness, this will lead to a greater increase in its exports than there would otherwise have been. However, in an estimated relationship, this increase in exports will be attributed to, say, foreign income growth (through an upward-biased estimate of the rest of the world’s income elasticity of demand for that country’s goods) unless the improvement in competitiveness is accounted for properly. This may be seen in Diagram 1, where the country’s exports are, for simplicity, assumed to be a simple function of time, except for an isolated improvement in competitiveness which leaves the trend rate of increase unchanged and occurs over time periods five through nine. If one estimated this function without taking into account the improvement in competitiveness, one would estimate a regression line such as AB, which embodies an upward-biased estimate of the trend rate of growth and a severely distorted view of the “true” residuals. When defined to include competitive influences, the true residuals are represented by the vertical distance between the plots and the line a₁k. (The random component is zero in this illustration.)

View Full Size

Methodology, Showing Use of Exchange Rate Dummy Variables to Account for Shifts In Functions

Citation: IMF Staff Papers 1974, 003; 10.5089/9781451969344.024.A003

• Download Figure
• Download figure as PowerPoint slide

The method used in this paper to account for the nonseasonal, non-cyclical, and nonincome related shifts in the functions is that of dummy variables which equal zero except over designated intervals where they are equal to one. For the intervals where they are nonzero, these variables operate to shift the intercept of the function by the amount indicated by the coefficient of the dummy variable. For instance, in Diagram 1, dummy variable Dl is defined to be equal to one in time periods five and six and to zero elsewhere—and similarly for D2 and D3. A regression of the form

will then yield the step-like regression line a₁defghcg. Note that during time periods one through four all the dummies are zero, so that the estimated regression line is given by Q = a + b T. During time periods five and six, on the other hand, D₁= 1, so that the computed value of Q is a + c₁D₁ +bT or (a + c₁) + bT, with (a + c₁) being the intercept of the function for time periods five and six; that is, the dummy variable operates to shift the function by the amount c₁. The coefficients of the dummies are chosen so as to minimize the sum of squared residuals around the regression line over the interval where the dummy is nonzero. The important point is that the dummies absorb the shifts in the function that are due to factors other than those otherwise accounted for in the equation.

The procedure in this paper is to estimate equations which account for seasonal, cyclical, and income effects as well as other effects, the latter being estimated by means of dummy variables. The end product is a set of residuals that should include all effects operating on the trade flows except seasonal, cyclical, and income effects. One disadvantage of this technique, however, is that the residuals simply record what has happened; they do not explain it. But this is perhaps not inappropriate in a context such as that in the International Monetary Fund, where knowledge of specific historical developments is fairly extensive. Further, the technique does bypass (in the sense that the answers are largely left to the reader) quite intricate questions of lag structure and interpretation of reduced form coefficients that arise in the usual econometric procedures.

II. Estimating Equations

The basic estimating equations used were:

where QM and QX represent the volume of imports and exports; Q and WQ represent domestic and foreign industrial production, where the latter is a trade-weighted average of the indexes of nine industrial partner countries and where the superscript T indicates the linear trend of the variable; T is time; D is a vector of dummies; and ZD and ZF are the usual cyclical variables—that is, the ratios of actual to trend income. These last variables are therefore a crude attempt to capture the cyclical component of the trade flow. The income component is presumed to be accounted for by the Q^T and WQ^T variables, while the seasonal movements are eliminated by using seasonally adjusted data. The antilogarithms of the above functions are of the form

where x, y, and z represent the dependent, trend activity, and cyclical variables, respectively. Note that e^dD = 1 if D = 0, and e^dD = e^d if D = 1. Hence, once they are multiplied by 100, the coefficients of the dummies represent the average percentage deviation of the trade flows from what they would otherwise have been. The equations were estimated by using quarterly data for 1958 through 1972 taken from International Financial Statistics.² The 1973 results for exports shown in Charts 1 and 2 are based on projections made by using the appropriate equations in Tables 2, 3, 4, and 5 (see the Appendix).

III. Interpreting the Residuals as Exchange Rate Effects

The preceding discussion has focused on the problems of computing residuals which exclude all seasonal, cyclical, and income effects and include all other effects without attempting to interpret the resulting set of residuals. One purpose of the exercise, however, is to analyze these residuals in order to show the influence of exchange rate changes.

The perspective in this section is quite different, since one is attempting to interpret the residuals. To do so, it should be recognized first of all that the residuals are simply the net result of all the forces acting on the trade flows not otherwise accounted for in the regression equation. To interpret the residuals as exchange rate effects is thus to assume that all the other forces cancel out. This is particularly inappropriate if, concurrently with the exchange rate action, the country is known to have experienced, say, a wage explosion. Secondly, it must be noted that exchange rate changes result in income effects which must be presumed to be embodied in the activity variables Q and WQ. In the case of WQ, this exchange rate effect is likely to be small. This is not the case for domestic industrial production. A currency devaluation typically boosts exports, leading to an increase in industrial production and thus in imports. The residuals computed here do not include these income-induced exchange rate effects, and therefore they understate the extent of exchange rate effects. Alternatively, one may say that the residuals measure the exchange rate effects that would result if economic policies were designed to offset the income effects of the exchange rate change. As a corollary to this point, it may be noted that the addition of explanatory variables to the estimating equations typically affects the exchange rate effects shown in the residuals, since the added variables are themselves likely to include exchange rate effects. This is particularly true of price variables.³

IV. Specification of the Dummy Variables

The basic set of dummy variables used in the regression equations consists of nonoverlapping dummies which were nonzero during the first four to five years immediately following an exchange rate change and zero elsewhere. The individual dummy variables were normally nonzero over a two-quarter interval although, for miscellaneous programming reasons, several dummies covered three or four quarters. The intervals where the various dummies were nonzero are defined in the charts by the length of the horizontal bars. The coefficients of the dummy variables are shown by the height of these bars relative to the base line. This information is also contained in Tables 1–6. Table 1 defines the various dummy variables, and Tables 2, 3, 4, 5, and 6 give the regression results for the four countries (see tables in Appendix).

The set of dummy variables described above and referred to in the charts as exchange rate dummies was perhaps misleading in that it might accentuate the exchange rate effects. This possibility arises from the fact that exchange rate dummies do not allow for any shifts in the functions during periods other than those of adjustments to exchange rate changes. For instance, if an improvement in competitiveness occurred prior to a devaluation, the resulting upward shift in the demand for exports would tend to be absorbed by the coefficients of the exchange rate dummies, the shift thus being improperly ascribed to the subsequent parity adjustment.

In the case of exports, therefore, the equations were re-estimated, using a second set of dummies referred to as structural dummies, which were each nonzero over a 12-quarter interval and which together covered the whole interval (except 1964–66, where the constant term in the regressions acts as the dummy variable for those years). In general, this approach provides considerably more flexibility and serves to bring out any basic shifts in the functions regardless of their timing. The disadvantage of this specification, however, is that it may result in biased estimates of the slope coefficients—that is, the coefficients of the trend income and cyclical variables—in which case the coefficients of the dummy variables would be biased in the opposite direction. For instance, suppose, as in Diagram 2, that the trade flow consists of a trend and a regular cyclical element. Let β be the trend rate of growth (shown by the slope of the line AB) and let ${\hat{\beta }}_i$ be the estimate of the “trend rate of growth” over the interval defined by the dummy variable i (shown by the slopes of the lines a_ib_i in the diagram). In the regression of the trade flow on time and D1–D4, the estimated coefficient of T, $\hat{\beta }$, will be an average of the ${\hat{\beta }}_i$ which may or may not be equal to β. In the diagram, the intervals chosen coincide with the upswings of the cycle. Hence the ${\hat{\beta }}_i$ (and therefore,$\hat{\beta }$) are biased upward relative to β. In order to offset this upward bias of $\hat{\beta }$, the coefficients of the dummy variables display a marked downward trend. The residuals, inclusive of the dummies, would appear as in the lower panel of the diagram. The arbitrariness of this result should also be noted. If, for instance, the intervals were all lagged by two observations, the ${\hat{\beta }}_i$ would reflect the downswings of the cycle. Therefore, $\hat{\beta }$ would be biased downward and the dummy variables would compensate by shifting the function upward over time. The essential point is that the estimated slope coefficient $\hat{\beta }$ reflects the average of the slope coefficients over each of the separate intervals defined by the dummies, and this average may very well not be the same as the “true” or long-run estimate β. To the extent that this is so—that is, to the extent that $\hat{\beta }$ is smaller or larger than β—the dummy variables compensate by shifting the function upward or downward, respectively. The shifts in the long-run function, which are of interest here, would then appear as either a reversal or an acceleration in the trend of the dummy coefficients.

View Full Size

Methodology, Showing Use of Structural Rate Dummy Variables to Account for Shifts In Functions

Citation: IMF Staff Papers 1974, 003; 10.5089/9781451969344.024.A003

• Download Figure
• Download figure as PowerPoint slide

What are the implications of these specifications for the interpretation of the residuals? In the case of the coefficient of the cyclical variable, the implications do not appear to be too serious. By definition, cyclical influences reverse themselves over time. Hence, if the cyclical elasticity is biased downward, one observes a spurious upward shift in the function during intervals of high activity and a similarly spurious downward shift during intervals where the level of economic activity is depressed. But the shifts are offsetting over time, so that one should still be able to detect the more or less permanent shifts which are the main interest of this paper. The real problem with the structural dummy specification is the possible bias of the long-run income elasticity, because trend income does not reverse itself over time. Hence, a biased estimate of this elasticity must be continually offset (as in Diagram 2) by shifts of the function in the opposite direction. However, the fact that trend income is a linear function of time means that, apart from shifts in the long-run function, the coefficients of the dummy variables are also linear functions of time.⁴ In examining the residuals of the equations which use structural dummies, one must, therefore, first eliminate any long-run trend from these residuals. For example, the income elasticity of U. K. exports using the exchange rate dummies specification is 0.49, while in the structural dummies specification the estimate is 0.85. It is not surprising, therefore, to find that the residuals from the latter specification have a strong downward pattern in the period prior to the devaluation (see Chart 5). In evaluating the residuals for the period subsequent to the exchange rate change (using Chart 5 rather than Chart 1), it is important to relate these residuals to the projection of the earlier trend, since the bias in the slope estimate is of course still causing the function to be shifted downward in this latter period.

View Full Size

Germany and the Netherlands: Percentage of Export Volume Unexplained by Seasonal, Cyclical, and Trend Factors, 1958–72, Calculated by Using the Exchange Rate Dummy Variables In Tables 3 and 4

Citation: IMF Staff Papers 1974, 003; 10.5089/9781451969344.024.A003

• Download Figure
• Download figure as PowerPoint slide

View Full Size

France, the United Kingdom, Germany, and the Netherlands: Percentage of Export Volume Unexplained by Seasonal, Cyclical, and Trend Factors, 1958–72, Calculated by Using the Structural Dummy Variables In Table 6

Citation: IMF Staff Papers 1974, 003; 10.5089/9781451969344.024.A003

• Download Figure
• Download figure as PowerPoint slide

Given these difficulties with the structural dummies, it may well be asked whether they contribute anything to the analysis. The answer depends upon the case. In the case of France and the United Kingdom, it is quite clear from a comparison of the estimated elasticities that the exchange rate dummy specification is preferred. This is not true, however, when structural shifts are apparent in times other than those of adjustment to exchange rate changes. In the case of the Netherlands, for instance, the pattern of the residuals from 1966 to 1971 in Chart 4 suggests that a basic shift in the function occurred in these years. However, the magnitude of the improvement is perhaps better brought out by the structural dummy specification (Chart 5).⁵ Consequently, the preferred equation (see Chart 2) was one combining the features of the exchange rate and structural specifications.

V. Interpretation of the Results

The dummy variable technique and the associated charts allow the reader to draw his own conclusions to a large extent. This is as it should be since, as must be stressed once again, the dummy variables are completely void of content in and of themselves. The dummies acquire significance only because they are temporally related to developments which are otherwise unaccounted for in the regression. Nevertheless, the following discussion of the charts may be helpful (the individual cases being analyzed after some general observations).

In the first place, a comparison of Charts 1 and 3 suggests that fairly substantial exchange rate effects on exports occurred, while import volumes were largely unaffected. This is in line with the empirical evidence on price elasticities, where estimated export price elasticities are usually larger than import price elasticities because of the substitutability among competing suppliers. In addition, it was mentioned earlier that, in the case of imports, domestic industrial production may be significantly affected by an exchange rate change (upward, in the case of a devaluation), so that the dummies are only reflecting that part of the exchange rate effect not absorbed by the change in industrial production induced by the exchange rate. The results shown in Chart 3 therefore underestimate the effect of exchange rate changes on imports. On the export side (excluding the German currency revaluations of 1969 and 1971), Charts 1 and 3 suggest exchange rate elasticities ⁶ of approximately 1½ to 3. Given that the relative price movements induced by the exchange rate change are likely to be smaller than the exchange rate adjustment itself, these exchange rate elasticities imply rather large price elasticities. Moreover, these elasticities appear to be smaller for the recent period than for the earlier one. Secondly, the charts suggest that, so far as exchange rate effects are observable, they occur fairly promptly. Both the British devaluation of 1967 and the French devaluation of 1969 appear to have elicited quite substantial changes in the volume of exports within two to three years. The adjustment to the German and the Netherlands revaluations of 1961 appears to have been even more rapid; most of the eventual adjustment took place within the first two years. On the other hand, the charts do not necessarily suggest that the adjustment of export volume to the French and British devaluations of the late 1960s was complete by mid-1973. Further, proponents of delayed adjustments to exchange rate changes may point to the German revaluations of 1969 and 1971 which, by the first half of 1973, had yet to result in any noticeable effect on the volume of exports—because of a low price elasticity, a very delayed adjustment, or both. This experience, which is in marked contrast to all the other cases considered, is all the more surprising since the revaluations of the deutsche mark do appear to have affected the volume of imports.

results for individual countries

France

In September 1949, the rate for the U. S. dollar was set at 350 old French francs. In August 1957, a 20 per cent surcharge was imposed on most payments abroad, and a 20 per cent premium was paid on foreign currency receipts. This system continued until June 1958, when the premium and surcharge were abolished and the exchange rate was changed from 350 to 420 old French francs per U. S. dollar. Concurrently with the August 1957 adjustment, import surcharges of 15 per cent on about 20 per cent of trade were abolished, as were various tax rebates on exports. Also relevant to an assessment of the possible exchange rate effects resulting from the 1957 devaluation of the French franc is the large differential in price movements between France and other countries. Thus, between the second half of 1957 and the second half of 1958, the French consumer price index increased 14 per cent, while the corresponding figures for Germany, Italy, Belgium, and the Netherlands were 1, 3½, minus 2, and 0 per cent, respectively. The French franc was devalued by a further 14.9 per cent in December 1958, and a par value of 493.7 old French francs per U. S. dollar was established with the Fund. The devaluation was combined with a fairly substantial relaxation of import restrictions. Eleven years later, in August 1969, the par value of the new French franc was changed from 4.937 to 5.554 per U.S. dollar, a devaluation of 11.1 per cent. In November 1968, value-added tax rates were increased by 18 per cent, and payroll taxes were largely abolished. The purpose was, in large part, to offset the loss of competitiveness of the foreign trade sector resulting from the large wage increases granted after the events of May 1968. Finally, the 1971 currency realignment did not significantly affect France’s effective exchange rate.

It is difficult to interpret the results for 1958–60, since the charts do not show the pre-exchange rate adjustment position. However, examination of annual data for earlier years and the views expressed by the French authorities during Fund consultations suggest that little or no exchange rate effect was apparent prior to 1958 because of excessive domestic demand pressures. If so, the exchange rate effects shown in Charts 1 and 3 are rather modest, since the cumulative devaluation of the franc was almost 30 per cent. On the other hand, the net price advantage conferred on French producers by developments between the summer of 1957 and the fall of 1958 was probably considerably less than would otherwise be implied by a 16.7 per cent devaluation. If so, one may assume that the exchange rate effects shown in Charts 1 and 3 for 1959 and 1960 are due to the December 1958 devaluation alone. By 1962 these effects seem to have been reversed, so that the trade balance (see Chart 6 and Section VI of this paper) returned to its pre-1958 position by 1963. Fund staff estimates of the period attribute much of this deterioration in the underlying position of the French trade account to the onset of the European Economic Community (EEC).

View Full Size

France, the United Kingdom, Germany, and the Netherlands: Percentage of Net Value of Exports and Imports in Local Currency Unexplained by Seasonal, Cyclical, and Trend Factors, 1958–72, Calculated by Using the Exchange Rate Dummy Variables in Tables 2–5

Citation: IMF Staff Papers 1974, 003; 10.5089/9781451969344.024.A003

• Download Figure
• Download figure as PowerPoint slide

The interpretation of the results for the 1969 devaluation of the French franc is more straightforward. In the case of imports (Chart 3), the peak in the first half of 1969 probably represents imports induced by the anticipated devaluation. The depressed level of imports after the devaluation may then be interpreted as a drawing down of stocks rather than an exchange rate effect. The subsequent observations may represent a modest exchange rate effect on imports. On the export side, a devaluation effect of 10–20 per cent is indicated. Before accepting these figures, however, one must introduce two considerations. First, there is a suggestion in Chart 1 that French competitiveness gradually improved between 1964 and 1968 (the residuals tending to shift from negative to positive).⁷ If so, one should discount some of the subsequent improvement. (On the other hand, the events of 1968 did lead to a sharp increase in wages, which would tend to reduce the apparent exchange rate effects.) Further, part of the positive residuals shown in Chart 1 for French export volume is due to large increases in agricultural exports. The EEC arrangements in this regard are such that, in principle, the French franc devaluation did not affect the price of French agricultural products in other EEC countries but did result in a substantial rise in the price paid to the French exporter or producer.⁸ Any exchange rate effects on agricultural exports would therefore have to originate on the supply side and not on the demand side.

Germany

The various revaluations of the deutsche mark vis-à-vis gold and a trade-weighted average of the currencies of Germany’s main trading partners, are (in per cent):

Date	Gold	Effective Exchange Rate Change
March 6, 1961	5.0	…
October 26, 1969	9.3	11.5
December 21, 1971	4.6	5.8
February 14, 1973	—	4.2
March 19, 1973	3.0	3.3
June 29, 1973	5.5	7.9

View Table

Date	Gold	Effective Exchange Rate Change
March 6, 1961	5.0	…
October 26, 1969	9.3	11.5
December 21, 1971	4.6	5.8
February 14, 1973	—	4.2
March 19, 1973	3.0	3.3
June 29, 1973	5.5	7.9

The 11½ per cent effective revaluation of 1969 takes into account the French franc devaluation of that year. The effective exchange rate change in 1961 was approximately 4½ per cent.

On the basis of the pure exchange rate dummy specification, it appears that the 1961 revaluation of the deutsche mark led to quite significant declines in export market shares, while the 1969 and subsequent revaluations did not (see Chart 4). However, the structural dummy specification exhibits a U-shaped pattern (Chart 5), suggesting that factors other than exchange rate changes were at work. The two specifications were therefore combined, the results being presented in Chart 2. With the negative residuals prior to the revaluation of 1961 taken into account, this chart suggests a 10 per cent downward adjustment in the volume of exports following the 1961 revaluation, the bulk of the adjustment having taken place within 12 to 18 months. These results are broadly similar to those obtained by Spitaller.⁹ Subsequently, an improvement in competitiveness unrelated to exchange rate changes appears to have occurred.¹⁰

The rather marked shifts in the demand for German exports in the face of relatively small price changes during this period make the ensuing lack of response to considerably larger exchange rate changes all the more surprising. The 1969 revaluation and the Smithsonian Agreement of December 1971 implied a cumulative 18 per cent revaluation for the deutsche mark. Given the apparent responsiveness of German exports to previous price changes, one would expect this to have had a significant impact by the end of 1972 and early 1973. The evidence in Chart 2, however, reveals no marked shifts in any direction.

In the case of import volume, Chart 3 shows virtually no effects resulting from the 1961 revaluation. The shifts are in the right direction, but they are small and not statistically significant, in contrast to the post-1969 adjustments, which indicate a gradual upward shift in the volume demand for imports by as much as 10 per cent. It is interesting to note that if, as suggested earlier, the technique of analysis underestimates exchange rate effects in the case of imports because of changes in industrial production induced by exchange rates and accounted for elsewhere in the regression, the argument would not apply to Germany in the 1969–72 period since the 1969 and 1971 revaluations have not, to date, affected exports. This may in part explain the exchange rate effects on German imports.

The Netherlands

The Netherlands revalued the guilder by 5 per cent in March 1961. In view of the large weight of Germany in trade with the Netherlands, the effective revaluation may have been about 4 per cent. Price and wage developments in the Netherlands were substantially in line with those of neighboring countries through 1963, but strong wage pressures developed in 1964 and 1965.

In the case of imports (Chart 3), it seems reasonable to view the revaluation as having increased the demand for imports by 5–8 per cent. The phasing of the adjustment, however, was probably not as rapid as suggested by the chart. Rather, it appears that most of the positive residuals observed for 1961 were due to cyclical factors which are inadequately accounted for in the regression equation.¹¹ On the export side, it appears that the effects of the revaluation were large. The magnitude of these effects, however, is more uncertain. The straightforward exchange rate dummy variables (see Chart 4) yield exchange rate effects of up to 25 per cent for a revaluation of only 5 per cent. The pattern of the residuals for the 1964—71 period, however, suggests that other shifts occurred in the function. This is borne out by the structural dummies (see Chart 5) which have a pronounced U-shaped pattern, and Chart 5 suggests exchange rate effects of 10–15 per cent. However, the structural dummy specification does not account fully for shifts in the function induced by the exchange rate. The exchange rate and structural dummy specifications were therefore combined to produce the results shown in Chart 2 which shows exchange rate effects of the order of 15 per cent. The negative pattern of the residuals in 1960, however, suggests that other influences may have affected this figure, and it is perhaps appropriate to suppose that the exchange rate effects were of the order of 10–15 per cent. Spitäller did find that the revaluation had significantly affected Netherlands exports of manufactures.¹² His estimates, however, do not reflect as large exchange rate effects as those suggested by Chart 2. The upward shifts in the demand for Netherlands exports in the second half of the 1960s appear to be due both to an underlying improvement in competitiveness and to the very fast growth of natural gas exports and of oil refining activities for re-export.¹³

The United Kingdom

The pound sterling was devalued by 14.3 per cent on November 18, 1967. The effective devaluation of about 14 per cent was maintained through the first half of 1972, and a further effective devaluation of about 10 per cent occurred in the second half of that year.

Just as was the case in the French franc devaluation, the effect of the 1967 British devaluation on the volume of exports was large and unmistakable (see Chart 1). Several features of the residuals shown in Chart 1 should be emphasized, however. In the first place, abstracting from the effects of the 1967 dock strike, the residuals tended to be positive prior to the third quarter of 1967. It may, therefore, be more appropriate to evaluate the exchange rate effects on a base line of + 5 per cent instead of zero. Secondly, the step-like pattern of the adjustment is rather pronounced. There was an initial upward shift from the suggested base of about 5 per cent, which prevailed for 12–18 months. There followed a second shift of some 10 per cent, which again lasted 18 months. Finally, in the second half of 1971 and first half of 1972, a further shift of 5–10 per cent is apparent. Although one may question whether this latter shift was due to the 1967 devaluation, it should be noted that the Smithsonian Agreement of December 1971 did not affect the pound’s effective exchange rate and that, since the devaluation, relative price movements were adverse.¹⁴ Cumulatively, therefore, it appears that, in the three to four years after the devaluation, the volume of U. K. exports increased by some 20 per cent more than would otherwise have been likely. These estimates cannot be directly compared with those arrived at in a study by the National Institute of Economic and Social Research (NIESR)¹⁵ since the focus in that study is on the devaluation effects in terms of value rather than volume. For 1970, the study estimated that 15–16 per cent of total exports (in sterling) were “due” to the devaluation. This would be equal to an 11–12 per cent volume effect, if the devaluation induced a 4–5 per cent increase in British export prices in local currency.¹⁶ Given a base line of +5 per cent, Chart 2 suggests that, by 1970, the devaluation effects on export volume also were in the 10–15 per cent range. Chart 1, however, suggests that this may not be the complete effect. Further, U.K. exports in the first half of 1973 appear to be quite strong, but this may be due to the effective depreciation in the second half of 1972.

The striking feature about the effects of the devaluation on U. K. import volume is that they are perverse. Instead of the downward shift in the demand for imports that would be expected subsequent to a devaluation, the shift was upward and statistically significant. Similar results were arrived at in the NIESR study, despite some attempts at examining the devaluation effects at a more disaggregated level.¹⁷ Further, while the devaluation-induced increase in the rate of domestic production (to produce the additional exports) explains some fraction of the positive residuals shown in Chart 3, the NIESR study indicates ¹⁸ that this does not account for all or even most of these effects. One is therefore left with the unenlightening suggestion ¹⁹ that there was an autonomous upward shift in the propensity to import which coincided with the devaluation. In support of this proposition, however, is the fact that the volume of imports did respond to the import surcharge, which accounts for the negative residuals for 1964–66 in Chart 3.

VI. Trade Balance Effects

The preceding sections have focused upon the effects of exchange rate changes on the volume of trade. For policy purposes, however, the primary interest is in the effect of an exchange rate adjustment on the trade balance in value terms. Therefore, the same equations used to evaluate the volume effects were re-estimated, using the corresponding value data (specifically, the product of the price and volume data) denominated in local currency.²⁰. The trade balance effects shown in Chart 6 were then calculated by adding the dummy variable coefficients and the residuals from the export equation and subtracting from this the corresponding sum from the import equation. This calculation assumes that 1 per cent of exports is equal to 1 per cent of imports—that is, that exports equal imports, which is not entirely correct. The percentages shown in Chart 6 are, therefore, best interpreted as the percentages by which the trade balance was increased or decreased, using as a base the average of exports and imports. The exchange rate dummy specification was used for France and the United Kingdom, while the modified version of that specification was used for Germany and the Netherlands. It should be pointed out that it is unsatisfactory to explain the value of trade by the volume of activity. The results should, therefore, be treated with caution. It may also be noted that the rapid acceleration in the rate of increase of foreign trade unit values toward the end of the period is accounted for only to the extent that Chart 6 shows the net value of exports and imports (and hence the net export and import prices).

The results suggest that the French franc devaluation of 1969 led to a quite substantial improvement in the trade balance. This estimate should, however, be treated with particular caution because of: (1) the suggestion in Chart 6 that the underlying pre-1968 trade balance position was positive rather than zero; and (2) the “terms of trade effect” (Table 2) implied by the volume and value regression equations.²¹ While this terms of trade effect is initially negative, as would be expected, it becomes positive within a year. This is in contrast to the U. K. case, for instance, where the devaluation-induced terms of trade effect (Table 5) is adverse by about 4 per cent. While this asymmetry may be due to a faulty specification of the French value equations, it may also be rooted in the fact that France’s competitive position in 1969 was considerably stronger than that of the United Kingdom in 1967, so that French exporters lowered their prices relatively less in foreign currency and increased them relatively more in local currency. This would also partly explain the much more modest benefits (shown by Chart 6) resulting from the British devaluation of 1967, although another major factor is, of course, the perverse results obtained for import volumes. On the other hand, Chart 6 implies somewhat larger exchange rate effects for the United Kingdom than suggested by the NIESR study, which concluded that the 1967 devaluation resulted in a 1970 trade balance £130 million higher than it would otherwise have been.²² For comparison, the value equations in Table 5 imply the following “devaluation effects” on the trade balance (customs basis, in millions of pounds):²³

	1968	1969	1970	1971	1972
Devaluation effect	–163	357	379	1,265	642

View Table

	1968	1969	1970	1971	1972
Devaluation effect	–163	357	379	1,265	642

The 1970 estimate is about three times that of the NIESR study. Except for 1971, the pattern does not look unreasonable, and while the effect may seem modest (the 1972 estimate is equivalent to only 3 per cent of the sum of 1972 imports and exports), it is not negligible in terms of the trade balance.

The results in Chart 6 with regard to the 1961 revaluations highlight the importance of terms of trade effects. The volume effects shown in Tables 3 and 4 are substantially similar for Germany and the Netherlands, except perhaps for a somewhat larger effect on import volumes for the latter. The trade balance effects, on the other hand, tend to be quite different. The German trade balance registered only a modest deterioration (particularly in the light of the improvement between 1965 and 1968), whereas the Netherlands trade balance showed a marked downward adjustment. While Germany benefited from a 5 per cent improvement in its terms of trade after the revaluation (Table 3), the improvement in the case of the Netherlands was only about 1 per cent (Table 4). This difference does not appear to be due to import prices, which seem to have fallen by about 5 per cent in both cases, but rather to export prices, which were largely unaffected in the case of Germany but declined ²⁴ by about 3 per cent in the Netherlands case. Finally, Chart 6 suggests that the German revaluations of 1969 and 1971 had virtually no net effect upon the German trade balance. While this result is in agreement with the results from direct observation, it should be treated with some caution in view of the absence of favorable terms of trade effects following the 1969—but not the 1971—revaluation.

VII. Suggestions for Further Research

The evaluation of exchange rate effects is a delicate matter requiring the specification of general equilibrium models that pay particular attention to linkages which are often neglected. In this paper, these complexities have been by-passed by estimating a special type of reduced form. While this technique has led to some rather suggestive results, the methodology used implies that the results are only suggestive. Conclusive results await the more detailed research effort referred to above. Nevertheless, the technique outlined here may have some further uses of which two may be mentioned.

The methodology described in this paper has been applied to aggregate trade data for which unit values are available. Its results can therefore be compared with those of the more traditional procedures based on computed price elasticities. In most cases the present study found that volume changes in response to exchange rate adjustments were somewhat larger than those that could be expected on the basis of the price elasticities estimated by standard procedures. One reason for the difference may be the difficulties that typically arise with the traditional methods in the process of estimating distributed lags for the price variable.²⁵ This suggests that the dummy variable technique used in the present study could be useful in providing clues to a more appropriate lag structure which, in turn, would allow a more accurate quantification of the coefficients to be estimated.

The dummy variable technique could, however, be used most profitably in cases where the relevant price data are not available—for example, in a disaggregated analysis of foreign trade by commodity and/or geographic markets. An important advantage of such a disaggregated analysis would be the multiplication of the number of observations, since exchange rate effects could be sought for each market and commodity subgroup. The results would therefore be less subject to sampling error than those of the more aggregative studies.