The World Trade Model: Merchandise Trade





This paper presents a simple model of world merchandise trade disaggregated by commodity class and by country or country grouping. The model is part of a larger model encompassing all current account transactions. It is designed principally to estimate the responsiveness of merchandise trade to variations in income and activity levels in the industrial countries. The responsiveness of trade flows to variations in prices is also considered, but the empirical results on price sensitivity appear to be less reliable.

The trade flows were disaggregated into four commodity classes: foods, raw materials, fuels, and manufactures. Disaggregation is desirable in that the determinants of import demand vary depending on the type of import. For example, many manufactured imports enter directly into final demand, whereas raw material imports are almost exclusively intermediate inputs in the production process. Similarly, the demand for manufactures is likely to be more price sensitive than is the demand for fuels or raw materials. Further, institutional factors tend to play a larger role in the determination of agricultural flows than in the determination of many other merchandise flows. Thus, it would be unrealistic to specify a single import-demand relationship that necessarily assumed a uniform pattern of demand determination across commodity classes. The benefits arising from disaggregation, however, must be weighed against the loss in manageability that necessarily accompanies this disaggregation. In view of this trade-off, the commodity detail was limited to the four groups mentioned.

The model is a closed model in that all countries are covered either individually, as for the 14 industrial countries, or within one of four country groupings. However, since the principal explanatory variables are activity and income variables in the industrial countries, interest centers on the equations for these countries; the equations for the country groups are included for completeness. Attention is focused principally on trade in manufactures and on imports of raw materials and fuels, since these flows are particularly important for the trade balances of the industrial countries.

It is expected that the model will be used for both simulation and forecasting. The presentation here concentrates on the basic relationships in the model, discussing only briefly supplementary links that are needed to make it fully functional in either of these two modes. Its usefulness in terms of simulation is expected to lie in the calculation of countries’ trade balances under alternative assumptions about economic activity levels in the industrial countries. The model is also expected to be useful in projecting countries’ trade balances for 6 to 18 months ahead, thus providing an input to the semiannual forecasting exercise of the International Monetary Fund. However, when the model is used for this purpose, it will be necessary to make liberal use of forecasts and expert information provided by country and commodity specialists in the Fund, not only for the exogenous variables in the model—for example, spot commodity prices and various components of gross national product (GNP)—but also for a number of variables that are endogenous to the model, especially variables relating to particular commodity flows—for example, agricultural trade—or to country groupings.

This merchandise model contrasts sharply with the trade model of Project LINK in that it is not integrated with explicit national models. 1 This is a disadvantage in that some of the interdependence between the foreign and domestic sectors may not be systematically taken into account; it is an advantage in that it is possible to impose a consistent specification across countries, and in that the maintenance and use of the model are greatly facilitated. 2

The present model is quite similar to the trade model of the Organization for Economic Cooperation and Development (OECD) in basic construction.3 Total domestic demand is the dominant exogenous variable in the OECD model, and final domestic demand for manufactures is the dominant exogenous variable in this model. Both models treat domestic price levels and exchange rates as exogenous, and are specified to be semiannual to conform to the time frequency used in forecasting exercises. However, the model described here differs from the OECD model in a number of respects. The OECD model has no commodity breakdown, whereas this model distinguishes among four commodity classes. The OECD model has individual equations for 18 OECD members, composite equations for “other OECD” members, and closing equations for the rest of the world; this model covers the 14 major industrial countries, and the rest of the world is disaggregated into four regions.

I. General Model Structure

The model is semiannual. This time frequency was selected because it is the one most often used in Fund work and, in particular, in the preparation of papers on the world economic outlook. Further, a semiannual data base is expected to reflect random movements to a lesser extent than a quarterly one and to facilitate the specification of lagged relationships.

The model focuses on the trade of the 18 countries or country groups listed in Table 1, and on transactions for four categories of goods under the Standard International Trade Classification (SITC)—(SITC 0+1), (SITC 2+4), (SITC 3), and (SITC 5–8). (SITC 9 is assumed to be determined exogenously.) This disaggregation reflects the view that the determinants of trade in manufactures are quite distinct from those of trade in raw materials, fuels, or agricultural goods. Thus, to base an analysis on aggregate relationships covering all commodities could produce misleading results, especially as regards the price sensitivity of trade flows. An aggregate analysis would also make it more difficult to integrate exogenous forecasts and information relating to particular commodity flows when the model is used in a forecasting mode. Clearly, manufactures themselves might usefully be disaggregated; for example, it would be useful to distinguish between manufactured imports that enter directly into final demand and those that are used as intermediate inputs. However, difficulties with data increase extremely rapidly as the degree of disaggregation increases. Thus, the disaggregation was limited to the four groups specified here.

Table 1.

Summary of Basic Model Specification

article image

Standard International Trade Classification.

This variable is derived from the identity: output = domestic demand - imports + exports.

The 14 countries and four country groups covered are the 14 industrial countries and the groups of developed primary producing countries, major oil exporting countries, other developing countries, and the rest of the world—consisting essentially of the centrally planned economies. These country groups are consistent with the country groupings used in the International Monetary Fund’s International Financial Statistics, and were chosen both because of their analytical interest and because of the availability of data.

The model can be thought of as consisting of three principal blocks. Of major importance are the price block and the volume block for the industrial countries. Of somewhat lesser importance are price and volume relationships for the four regions. The first block determines, for each commodity class, export and import unit values for the industrial countries, given certain spot commodity prices, domestic costs and prices, and exchange rates; the second block determines the volume of exports and imports by commodity class for the industrial countries (excluding fuel exports for countries other than Canada, the Netherlands, and the United States), given prices and the level of final domestic demand. The exogenous and endogenous variables are listed in Table 1. The relationships briefly noted here are specified more fully in Tables 2 and 3, and are discussed in the subsections, specification of the price block for the industrial countries and specification of the volume block for the industrial countries. The third block relates export receipts of the country groups to activity levels in the industrial countries, and import expenditure to their foreign exchange earnings. These relationships are specified in Tables 4 and 5 and are discussed in the subsection, specification of the volume and price relationship for the four regions. Seasonal and special country dummies and lag structures are not indicated in these tables but are described in the Appendix and/or the tables that report the corresponding parameter estimates.

specification of the price block for the industrial countries

Price relationships for manufactures

In Table 2, equation 1, countries’ export prices for manufactures (unit values) expressed in the domestic currency (XPM) are explained by: (1) the costs of production represented by the cyclical (CYCLOMH) and noncyclical (NULC) components of unit labor costs in manufacturing, and by raw material costs on the domestic market (PRM); and (2) the state of demand represented by competitor prices in foreign markets expressed in the domestic currency (PFX*LCD) and calculated as a double-weighted average of other countries’ export prices for manufactures. The variables are expressed in change in log form.

Table 2.

Industrial Countries: Price Relationships

article image
article image

To analyze the main issues that arise from this specification, it is helpful to begin by considering the price formation process for a profit-maximizing firm subject to: (1) diminishing returns to scale because capital is fixed in all but the long run; (2) perfect competition in the markets for the variable factors of production; and (3) less than perfect competition in the market for the firm’s output, that is, some degree of pricing discretion. In the simplest case, such a model4 yields a reduced form price equation of the following form, where all variables are expressed in terms of rates of change:


VC, CP, and Y represent variable costs, competitor prices, and the income (or scale) variable in the demand function for the firm’s output, respectively; α is the elasticity of the marginal cost function with respect to output (α≥0); and η is the absolute value of the price elasticity of demand in volume terms (η≥1 insofar as the individual firm is concerned).

In this expression, the profit-maximizing price charged by the firm is seen to be an increasing function of both cost (VC) and demand factors (CP and Y). In addition, the coefficients of the VC and CP variables sum to one; this implies that a 10 per cent increase in both of these variables results in a 10 per cent increase in P. Because the coefficients are nonnegative and sum to one, an increase in the relative importance of one of these variables in the price formation process entails a reduction in the role of the other. Further, the relative responsiveness of price to each of these two variables (costs and competitor prices) depends upon the quantity αη. If αη is large—because the firm’s monopoly power is limited (i.e., η is large) and/or because there are strongly increasing marginal costs (i.e., α is large)—then competitor prices tend to dominate the price formation process. In the limiting case of perfect competition, which may not be an inappropriate assumption for many of the smaller countries involved in international trade, η tends to infinity, and the coefficient of CP tends to one while that for VC tends to zero. If, on the other hand, αη is small because the “firm,” as a large supplier, enjoys a considerable amount of monopoly power and/or because production is subject to constant or near-constant returns to scale, then pricing is dominated by supply considerations. This outcome is the more likely, the longer is the time horizon of the pricing decision being considered. Other things being equal, α will tend to fall as the proportion of the factors of production that are variable rather than fixed increases.

The relative importance of cost and competitor price variables in the price equation, therefore, depends upon the quantity αη. This result is of particular interest in the present application, since this quantity is likely to be inversely related to the size of the “firm”—that is, the size of a country. Both α and η are likely to be small for a large country, since exports will represent a relatively small fraction of domestic production but a large fraction of the international market. Thus, prices will depend primarily upon costs, with competitor prices being of secondary importance. The converse would be true for small countries, with competitor prices tending to dominate the export price formation process in the short run.

The specification in Table 2 involves several departures from the theoretical model. First, a distinction is made between the cyclical and noncyclical components of unit labor costs. The latter is defined as actual compensation per man-hour deflated by potential output per man-hour.5 The cyclical component of labor costs is represented by the ratio of actual to “normal” output per man-hour. These specifications of the labor cost variables stem from the view that, other things being equal, firms seek to minimize the fluctuations in the prices that they charge, so that the pass-through of cyclical changes in unit labor costs is expected to be less than that of “normal” changes.6 Second, the “income,” or scale, variable that should logically appear in the equation in Table 2 has been omitted. This omission reflects the consistent lack of significance of this variable in all the estimated equations.

Given the requisite bilateral trade and unit value data, import unit values would be identically equal to some suitably weighted average of the export price data. In Table 2, equation 2, therefore, import unit values for manufactures are explained by a weighted geometric average of the export unit values of other industrial countries7 expressed in the local currency (PFMD*LCD). In addition, since exports are recorded before the corresponding imports, a weighted average of the current and last period changes in partner-country export prices was used. Weights of ⅚ and ⅙, respectively, were selected, since the recording lag is thought to average about one month.8 This weighted export price variable may be thought of as a world index for a particular basket of goods. However, pricing decisions for exports to a particular market may be influenced by characteristics of the particular market, such as the price and availability of domestically produced alternatives. Consequently, experiments with two additional variables—the exchange rate (LCD) and the price of domestically produced manufactures (P)— were conducted to estimate the effects of these country-specific influences that may affect bilateral pricing decisions. However, while a few of the individual country results were suggestive and plausible, the results across countries were too disparate to be retained in the final specification of the equation.

Price relationships for nonmanufactured commodities9

The basic exogenous information used in this block of equations (Table 2) is a set of 35 time series on spot quotations in world markets for precisely defined primary commodities.10 These data are aggregated to obtain spot price indices (e.g., PAWDX and PRWDX) with commodity coverages similar to those of the agricultural or raw material unit value indices that they are intended to explain. In a forecasting context, the projected values of these spot prices are provided by commodity experts.

The export price equations posit that the export unit values for agricultural goods or raw materials depend on spot prices in world markets (weighted on the basis of the 1970 commodity structure of the country’s exports) and labor cost developments in the exporting country. The inclusion of the first variable reflects the view that—apart from the problems discussed later of linking spot prices and unit values—most prices for a specific type of raw material and agricultural good are determined in a single world market. Hence, the export price is simply equal to this world market price, the latter being expressed in terms of domestic currency. Such a view, however, takes no account of the fact that industrial countries’ exports of these commodities often are somewhat more processed than are the commodities to which the spot prices refer. The purpose of the labor cost variable, therefore, is to account for at least part of this extra processing. On the premise that wage developments are fairly uniform across the various sectors of the economy, this allowance for additional domestic value added is based on hourly labor compensation data for manufacturing.11

As was true for the price of manufactured imports, the import unit values are explained primarily in terms of the corresponding partner-country export prices, again lagged to take account of the recording lag between exports and imports. However, in the present instance, allowance must also be made for the nonavailability of unit value data broken down by major commodity groups for the four major geographic regions—regions that happen to be significant suppliers of nonmanufactures to the industrial countries. This gap in the data has been circumvented. Indices of the spot commodity price data, using as weights the 1970 commodity structure of each importing country’s imports, were included in the import price equations to reflect the influence on import prices of commodities imported from the nonindustrial countries.12

The use throughout the commodity block of spot price data to explain unit value indices has several drawbacks. First, the commodity coverage of the spot price data is considerably more restrictive than that of SITC 0+1 and 2+4 (i.e., the coverage of the unit values). Second, the “transactions” coverage of the spot price data (for comparable commodity classes) is (1) considerably less than that of the unit values (i.e., many transactions are carried out under long-term contracts that bear no necessary relationship to current spot quotations), and (2) temporally distinct from that of the unit values (i.e., the latter tend to reflect spot prices of an earlier period). A first consequence of these deficiencies is that, since the discrepancy in commodity coverage is most pronounced with respect to slightly processed commodities, the prices of which are likely to be somewhat less volatile than those of unprocessed commodities, there is an expectation that (other things being equal) the elasticity of the unit value series with respect to the spot price indices will tend to be less than one, particularly in the short run. The dependence of the unit values on transactions that are under long-term contracts would operate in the same direction. Second, the implication of the timing discrepancy between spot and unit value data is that one may reasonably expect the unit values to respond to changes in spot market quotations with a lag.

Finally, two sets of supplementary price equations are used in specific applications of the model. The first set is used in a forecasting mode to link domestic prices of raw materials (PRM) to an index of world spot prices for raw materials in local currency (PRWDM*LCD) and labor cost developments. The second set of equations (which link each of the 35 individual spot commodity price series to general indicators of inflation and underutilization of resources in the industrial countries) is used, for simulation purposes, to estimate the cyclical element in commodity prices.

specification of the volume block for the industrial countries

Volume relationships for manufactures

The relationships in the volume block for manufactures are outlined in Table 3. The value series are adjusted for most countries to exclude transactions in ships and aircraft, which cause considerable variability in the series and are thought to be affected by factors not generally important in determining the demand for other manufactured goods. Trade in automobiles between the United States and Canada has also been omitted from the equations on aggregate manufactures because of its very rapid development from 1965 to 1970 after the conclusion of the U.S.-Canadian Automotive Agreement. The value series for a few countries (Belgium-Luxembourg, Switzerland, and the United Kingdom) are further adjusted to exclude transactions in pearls and precious stones, since these countries tend to import large quantities for re-export. The adjusted value series are then deflated to obtain volume series.

Table 3.

Industrial Countries: Volume Relationships

article image
article image

The import volume equation for the industrial countries posits that the demand for imports of manufactures is the result, in part, of a two-stage decision process. The first stage of that process determines the total final demand in the country for manufactured goods given the country’s income level, the aggregate price for manufactures, and the aggregate prices for a broad spectrum of other commodities.13 This stage in the decision making is outside the purview of the trade model. Given the aggregate final demand for manufactures, the problem then becomes one of deciding how much of the demand will be satisfied from domestic sources, and how much of it will be satisfied from imports. This is hypothesized to depend on relative prices. 14

The import volume figures, however, relate to goods that are used as intermediate inputs as well as those that enter directly into final demand, while the specification suggested earlier focuses on the final demand component. Thus, it was necessary to modify this specification somewhat to take into account determinants that exert particularly strong influence on the demand for intermediate inputs. Consequently, the demand variable was specified as a weighted average of manufacturing output (QM) and final domestic demand for manufactures (DVM), with weights reflecting the share of manufactured imports used as intermediate inputs and the share going directly to final demand, respectively.

It is also important to distinguish between the responsiveness of imports to short-term fluctuations and the responsiveness to longerterm developments. To introduce this distinction into the relationships for manufactured imports, the demand variable presented here was deflated by an index of potential output in manufacturing to provide a measure of fluctuations in demand around its longer-term value; the index of potential output in manufacturing was included separately to capture the responsiveness of imports to the longer-run growth of the economy. The relationships are specified in logarithmic form.

If one ignores certain timing problems and difficulties in the proper classification of trade flows, countries’ exports to a particular market at a particular time must equal the level of imports in that market multiplied by the share of the exporting country in that market.15 Thus, to determine a pattern of exports consistent with the predetermined level of imports it is necessary to explain bilateral market shares. However, for an 18-region world, this would require approximately 300 bilateral relationships to be determined for manufactures alone. The data requirements and computational difficulties involved in a straight-forward approach to the estimation of such relationships for a model with commodity disaggregation would be severe.

Under these circumstances, it is useful to consider what the restrictions that underlie the import demand function may imply for export determination, and perhaps what further reasonable simplifying assumptions or approximations might be introduced that would significantly reduce the number of separate parameters that need to be estimated. It has been assumed that the elasticity of substitution between manufactured imports from any two exporting countries is the same for all countries competing in a given market, and that the elasticities are constant over time. If the allocation of shares in a particular market is independent of the size of the market, then it can be shown that the allocation of import demand among exporting countries can be formulated as

XVMij = bijδj(Pij/Pj)-δj. MVMj

where XVMij is the volume of exports of country i to country j; MVMj i is the volume of imports of country j; Pij is the price of exports of country i to country j; Pj is the average price of exports to country j; and δj is the elasticity of substitution in market j.16 This formulation provides the basis for the equations for the export of manufactures in the model presented here. It was modified in several respects, however. To make the data requirements more manageable and to facilitate the use of the model in a forecasting mode, price and demand variables were aggregated across countries. The foreign market variable (that is, the weighted average of imports in partner countries) reflects the influence of changes in demand in foreign markets. In addition, exports may grow or contract, as a result of changes in market shares. This may occur, for example, as a result of movements in relative prices, or increasing productive capacity in the manufacturing sector. To reflect these influences, a relative price variable was included in the equation as well as an index reflecting the growth in potential output of the manufacturing sector in the exporting country relative to that in competitor countries. Thus,

XVMi = α0(FMi)α1(RXPMi)α2(RQMTi)α3(Zi)α4

where XVMi is the volume of manufactured exports from country i; FM i is the foreign market for country i; RXPMi is the export price of manufactures for country i relative to that of competitor countries, calculated as a double-weighted index of export prices; RQMTt is the ratio of potential output in the manufacturing sector in country i relative to that in competitor countries, calculated as a double-weighted index of potential output series; and Zi represents other exogenous variables relating to country i.17

Volume relationships for nonmanufactured commodities

The specification of the manufactures block draws on consumer demand theory, but the specification must be simplified and then modified to reflect the substantial proportion of manufactured imports that are used as intermediate goods. For raw materials and fuels, the specification is based on the assumption that imports of these commodities constitute intermediate inputs in the production process. It is postulated that a given increase in output from the manufacturing sector tends to require a proportional increase in the imports of raw materials and fuels; changes in commodity imports may coincide with changes in manufacturing output, or may lag by one period as a result of transportation and delivery delays. It is also assumed that imported raw materials (fuels) can be stockpiled in anticipation of future need, with the expectation of future need being reflected by the current rate of change in manufacturing output.

Clearly, it is not possible from this specification to identify individually the “stockpiling” effect from the lagged effect corresponding to delivery delays. To reflect their joint influence, however, both the current level of output in manufacturing (QM) and its value lagged by one period are entered as exogenous variables in the import demand function for raw materials and fuels. The coefficient on the lagged value of manufacturing output can be positive or negative.18 The more sensitive that raw material (fuel) imports are to the rate of change in output, however, the greater will be the probability that the coefficient on the lagged output variable will be negative.

It was postulated that imported raw materials and fuels are generally not competitive with home-produced raw materials and fuels, since these imports consist of goods that are not produced domestically, or are produced in inadequate quantity. Thus, there would be little price sensitivity.19 However, to capture to some extent the effects of the unusual developments in 1974, two dummy variables were introduced in the equations for fuel imports. A shift dummy to take into account the extremely large change in the relative price of oil in 1974, and an embargo dummy taking the value one in the first half of 1974 and minus one in the second half were introduced.

The volume of food imports (MVA) is related to real consumption expenditure (CC) on the assumption that these imports either immediately or after some processing are destined for the domestic consumer. The determination of the level of real consumer expenditure falls outside the scope of the model. No attempt is made to measure the sensitivity of agricultural imports to changes in domestic food prices. This is so because, to the extent that domestically produced agricultural goods do compete with imported goods, the relevant relative food price ratios are likely to show little movement; to the extent that there are no domestic substitutes, it is more probable that a variation in the import price will affect a wide range of expenditure decisions rather than just the food budget. Consequently, the relative price variable entering the food-import equation is the ratio of the GNP deflator to the price of food imports. It is introduced in an attempt to measure the overall substitution between imported foods and all goods produced domestically.

Export volumes for raw materials and agricultural goods are estimated to close the model, as noted earlier. Foreign market variables (FA and FR) consisting of a weighted average of imports of raw materials or agricultural goods in the individual industrial countries and country groups from the 14 industrial countries taken as a whole are calculated, and these are the principal explanatory variables. As has been discussed earlier, it is assumed that a particular type of raw material or agricultural good is homogeneous and that there is a single world price for such a good. Thus, no relative price term enters these export equations.

specification of the volume and price relationship for the four regions

To provide complete country coverage is outside the scope of the model and of the available data base. However, to add some degree of realism to the equations, the “rest of the world” was disaggregated into four country groups, with particular characteristics being shared by the countries within each group. Because of the aggregative nature of these group relationships, however, and the heterogeneity across countries within each group despite common characteristics, the ad hoc nature of relationships given in Table 4 must be acknowledged. Further, because of a lack of data, disaggregation of export and import equations by commodity class was not possible.

Table 4.

Regions: Volume and Price Relationships

article image

The total volume of exports of each of the four regions (XVT) is explained by reference to the level of output in manufacturing in the industrial countries (QMIND). The weights used to construct each of these indices are based on the importance of the various industrial markets for the exports of the country group. The levels of output in manufacturing were also used as relative weights, reflecting the view that a “world” rather than a “bilateral” approach might be more appropriate for these countries. However, they did not perform quite as well as the “bilateral” variables.

Similar specifications are used to determine import demand in each of the four regions: imports are explained on the basis of a foreign exchange constraint.20 This type of specification is clearly not as appropriate for the more developed countries as it is, for example, for the non-oil developing countries. For the centrally planned economies, the specification produces a reasonable fit; but the causality may run in the opposite direction, that is, import needs may well determine foreign exchange receipts rather than the converse.

Imports in volume terms (MVT) are explained by the purchasing power of these country groups (XPP). In principle this purchasing power variable should cover countries’ net foreign exchange holdings as well as reflect their current receipts and their ability to borrow. In the estimated relationships, the purchasing power variable is proxied by the total value of the regions’ exports (including within-region exports), deflated by their aggregate import price index.

The equations determining aggregate export unit values for each of three regions (excluding the centrally planned economies, for which no price data are published) specify that this index is determined by a weighted average of spot commodity prices of agricultural goods and raw materials, with the weights reflecting the importance of the individual commodities in the trade of the region, a unit value index for the world price of manufactures, and a unit value index reflecting the price of petroleum.21 The inclusion of the “world” unit value index for manufactures reflects the view that the manufactured export prices for these countries move in line with the export prices of the industrial countries, that is, that these regions are price takers as far as manufactured exports are concerned. This assumption appears reasonable, and further support for this approach arises from the impracticality of collecting data on domestic cost and price developments for the primary producing countries.

The import unit values for the regions are related to a unit value index for petroleum, a weighted average of the export prices for manufactures of the industrial countries (with the weights reflecting the relative importance of the particular country in the imports of the region), and weighted averages of raw material and agricultural export unit values of the industrial countries. The agricultural and raw material export unit values of the industrial countries were included rather than a weighted average of commodity spot prices in the import unit value equations for several reasons: it was thought that the commodities covered by the spot price indices were not particularly representative of the commodity composition of imports of the regions; a substantial portion of the raw materials and agricultural goods imported by the regions originates in the industrial countries.

II. Model Parameters

The historical data base for the model consists of semiannual observations for the period 1955–60, depending on availability, to the first half of 1976. However, unless otherwise noted in the tables, the estimation period is from the first half of 1964 to the first half of 1976, and the equations were estimated by ordinary least squares. The dummy variables used are described in the footnotes to the tables; a rational is provided either in the footnotes or in the Appendix.

estimation of the price block for the industrial countries

Price parameters for manufactures

Table 5 presents estimates of the export unit value equations for manufactures for the industrial countries. To take account of the simultaneity of this block of equations, that is, the interdependence of competitor prices, a type of two-stage least-squares estimation procedure was used. In the first stage, XPMj for each country j was regressed on the exogenous variables for country j and the exogenous variables for all other countries. The latter, however, were first averaged across countries, using the same weights as those used to construct the competitor price variable. The resulting predicted values of XPMj for all i were then averaged to compute a predicted version of PFXj which was then used in the equations shown in Table 5; these equations were estimated using ordinary least squares.

Table 5.

Fourteen Industrial Countries: Estimates of Export Unit Value Equations for Manufactures, First Half 1963–First Half 19761

article image

The t-statistics are given in parentheses.

Because of collinearity problems, the equation assumes that two thirds of the adjustment to changes in normal unit labor costs or competitor prices takes place within the current half year.

Constrained estimates.

These variables are 0 everywhere except as follows: (a) Belgium-Luxembourg—1965 (first half) = –1, 1965 (second half) = ½, 1970 (first half) = –1, and 1970 (second half) = 1; (b) Italy—1963 (second half) = 1 and 1968 (first half) = –1; (c) Japan—1975 (first half) = 1 and 1975 (second half) = 1; and (d) the Netherlands—1974 (first half) = 1 and 1974 (second half) = 1. The dummies for Japan, the Netherlands, and Italy in 1963 are associated with cyclical developments, the oil price increase, and strikes, respectively. The other dummies pick up seemingly random developments, which, however, affect the structural coefficients of the equation.

Several of the regression equations shown in Table 5 have been constrained to pass through the origin. While this constraint has little effect on the reported results for most countries, it does yield somewhat more plausible coefficients than would otherwise have been true in several instances. Similarly, the coefficients on the variable for raw materials prices tended to be either too high (e.g., in the ⅙ to ⅓ range for the United Kingdom, Belgium-Luxembourg, the Federal Republic of Germany, the Netherlands, Sweden, and Japan) or too low (e.g., 0 or negative for the United States, France, Italy, and Canada), and were consequently constrained to 0.075. This value is consistent with an average of the estimates obtained from input-output matrices. A difficulty associated with making the constraints country specific is that the coverage of the raw material price variables varies considerably across countries and bears an often uncertain relationship to that of the input-output matrices.

In Table 5, the cyclical productivity variables are of secondary importance. As expected, the coefficients tend to be small. Indeed, the variable is excluded from more than two thirds of the equations. The relative lack of success of these variables (as well as measures of demand pressure in foreign markets, which were also experimented with) is thought to be due in part to the cyclical element embedded in the competitor price variable, which plays an important role in many of the equations.22

The dominant variables in the export unit value equations are normal unit labor costs and competitor prices. Indeed, the coefficients for these two variables sum to approximately unity for the majority of countries. The importance of the competitor price variable is particularly striking: it is included in the equations for all countries except the United States. Further, while the classification is not unambiguous, there is a tendency for the magnitude of the coefficient of the competitor price variable relative to that for normal unit labor costs to be inversely related to the country’s size. Thus, competitor prices are relatively more important than labor costs in export price formation for Austria, Denmark, Italy, the Netherlands, Norway, and Sweden, while the reverse is true for the United States. For the other countries—Belgium-Luxembourg, Canada, France, the Federal Republic of Germany, Japan, Switzerland, and the United Kingdom—the competitor price and the unit labor cost coefficients are approximately equal. With perhaps a few exceptions, these results are more or less in accordance with a priori expectations as to the relative market position of the various countries. However, the results pertain only to a short-run analysis. In the longer term, prices cannot diverge markedly from costs. This implies that the importance of costs must increase as the time horizon is extended, and/or that costs are in fact endogenous (rather than exogenous) and depend upon competitor prices. In either case, it is clear that the equations in Table 5 are viable only for the short run and for rates of increase in domestic costs that, after correction for exchange rate changes, are not drastically out of line with cost developments elsewhere.

The import unit value equations for manufactures shown in Table 6 are straightforward. Changes in import prices depend only on changes in partner-country export prices. That is, separate trend influences, as reflected by a constant term, are also excluded from the equations. Constant terms were significant in only two countries—Austria and Switzerland—and in each of these, the corresponding slope coefficient diverged from unity (e.g., to 1.6 for Austria). The resulting import unit value equations specify that import prices follow partner-country export prices with a one-month lag. The coefficients all cluster around unity; the average is 0.95, and only the coefficients for Canada and the Federal Republic of Germany are significantly different from one.

Table 6.

Fourteen Industrial Countries: Estimates of Import Unit Value Equations for Manufactures, First Half 1963–First Half 1976 1

article image

The t-statistics are in parentheses.

Weighted average of current and previous period partner-country export prices using weights of ⅚ and⅙, respectively.

The simplicity of the equations given in Table 6 is due partly to rejections of alternative results that were not, taken at face value, incompatible with the a priori arguments set out earlier. For instance, the addition of the domestic price variable (P), on the assumption that exporters consider price developments in specific markets in the price formation process, results in equations with noticeably lower standard errors of estimate for the United States, Belgium-Luxembourg, and Sweden. These results were rejected, however, because the coefficients of the P variable tended to be too large. For the United States, the coefficient was approximately unity and highly significant, while the coefficient for PFMD was not significant. Such coefficients seem too extreme to be accepted for working purposes. In the other two countries with large P coefficients—Belgium-Luxembourg and Sweden—the importance of the P variable could well be the result of causation running in the opposite direction from that postulated, that is, from import to domestic prices; consequently, this variable was again excluded from the final specification.

Price parameters for nonmanufactured commodities 23

Tables 710 present the parameter estimates for the export and import unit value equations for agricultural products and raw materials. Generally, these equations support the main relationships that are postulated for these prices. Thus, while the export unit values for nonmanufactures depend in large part upon world market prices for similar commodities, there is also a domestic value-added component that responds to changes in domestic wage costs. Similarly, on the import side, it is clear that developments in world commodity prices have an influence on industrial country import unit values quite apart from that exerted via the export unit values for the industrial countries.

Table 7.

Fourteen Industrial Countries: Estimates of Export Unit Value Equations for Agricultural Goods, Ssecond Half 1962–First Half 1976 1

article image

The t-statistics are in parentheses.

For Denmark, this variable is intended to allow for the effect of entry into the European Economic Community. The variable is equal to zero prior to the first half of 1973.

Cochrane-Orcutt autoregressive correction.

Estimate is affected by the origin constraint. SEE is a better measure of the goodness of fit.

Table 8.

Fourteen Industrial Countries: Estimates of Export Unit Value Equations for Raw Materials, Second Half 1962–First Half 19761

article image

The t-statistics are in parentheses.

Price of petroleum for Belgium-Luxembourg and the United States; dummy variable taking on the value 1 from 1974 on for Denmark, to reflect a jump in the price of nonmineral oils.

Cochrane-Orcutt autoregressive correction.

Table 9.

Fourteen Industrial Countries: Estimates of Import Unit Value Equations for Agricultural Goods, First Half 1963–First Half 19761

article image

The t-statistics are in parentheses.

Weighted average of current and previous period partner-country export prices using weights of ⅚ and ⅙, respectively.

Cochrane-Orcutt autoregressive correction.

Table 10.

Fourteen Industrial Countries: Estimates of Import Unit Value Equations for Raw Material Prices, First Half 1963–First Half 19761

article image

The t-statistics are in parentheses.

Weighted average of current and previous period partner-country export prices using weights of ⅚ and ⅙, respectively.

Cochrane-Orcutt autoregressive correction.

Nonetheless, the statistical results, as evidenced by the large standard errors of estimate, are frequently unsatisfactory. The average of these errors across all equations in Tables 710 is 3½ per cent. A third of the standard errors falls in each of the following categories: less than 3 per cent, 3 to 4 per cent, and 4 per cent or more. Several factors account for at least part of these rather large errors. First, the standard errors for many of the equations are greatly affected by several extreme residuals, so that, in many cases, the median error is likely to be smaller than the average error. Second, while these extreme observations often appear to be random (for example, large price changes during the 1960s, when foreign trade prices were quite flat), others can be traced to specific shortcomings of the model. For instance, the omission of wood from the list of spot commodity price indices is a critical shortcoming insofar as Canadian and Scandinavian exports are concerned. In the main, however, the worst equations are those that refer to trade flows of secondary importance both to the country in question and to international trade generally. Third, the residuals from the various equations are often correlated across countries, which suggests that certain common influences may not have been taken into account. Symptomatic of this is the fact that the addition of a dummy variable that takes the value one from 1974 onward to the equations in Tables 8 and 10 would yield statistically significant coefficients in virtually every raw material price equation. This dummy could be interpreted as reflecting the large change in relative fuel prices, but petroleum is known not to be covered in the raw material unit value indices for most of these countries. In short, there was an explosion in foreign trade prices for raw materials in 1974 that is inadequately accounted for by the equations.

The standard errors tend to be larger in the equations for raw materials, given in Tables 9 and 10, than for agricultural products, given in Tables 7 and 8. One possible explanation is that long-term contracts are more pervasive in trade in raw materials than in agricultural trade. Some support for this thesis is provided by the somewhat longer lags (particularly on the export side) between spot prices and unit values for raw materials than for agricultural products. A second distinguishing characteristic of the tables is that the results on the import side tend to be somewhat better than those on the export side. The explanation here must stem from the near-identity character of the import equation, since it includes a weighted average of the export price indices of industrial trading partners.

It is difficult to evaluate the magnitudes of the parameters in Tables 710, since it is unclear what “reasonable” estimates for these parameters are. On the export side, the spot price data serve both an input cost and a competitor price function, but the importance of these functions is not clear. On the import side, the influence of the spot price data is partly embedded in the partner-country export price variable. With these considerations in mind, note that the cumulated coefficients for the spot price data tend to be rather small, mostly between ¼ and ½, in all the tables. The exceptions—the parameters for Canadian agricultural export prices and Japanese raw material import prices—generally seem to be of the right magnitude.

There is a wide diversity of lags between changes in spot prices and unit values, as evidenced in Tables 710. While this diversity makes generalizations somewhat hazardous, the evidence suggests that the lags on the export side are somewhat longer than those on the import side. This is consistent with the notion that the spot price data in the export equations partly serve an “input cost” function, so that a pass-through lag that does not apply to imports is added to the spot-price unit value lag for exports. Finally, as already mentioned, the average lags seem to be somewhat longer for raw materials than for agricultural products.

estimation of the volume block for the industrial countries

Volume parameters for manufactures

The parameter estimates for the import and export volume equations for manufactures are given in Tables 1112. The price terms used here are based on unit value indices and, for imports, the domestic wholesale price for manufactures. Partly to avoid errors in measurement bias arising from the use of unit value indices as both deflators and price measures, and to economize on the number of variables that need to be introduced in the equations, a variable reflecting short-term developments in relative prices (an average of the relative price terms over the two preceding periods) and one representing longer-term developments (an average over the period from one and one-half years to four years, or three to seven periods, earlier) were introduced. The joint impact of these price terms for both imports and exports of manufactures are summarized in Table 13. Given the rather long lags in the response of flows to price movements that have been found in empirical studies, the omission of the current level of relative prices was not thought to have serious consequences.

Table 11.

Fourteen Industrial Countries: Estimates of Volume Equations for Manufactured Imports1

article image

The t-statistics are in parentheses.

This dummy takes the value 0 for the first half year and 1 for the second half year.

SDYR (e.g., SD71) indicates that dummy represents a change in seasonal pattern starting in YR (e.g., 1971). DYR12 (e.g., D196612) indicates a dummy that takes the value 1 for the first half year (1966:1), and -1 for the second half year (1966:2). DYR (1) YR (2) (e.g., D196768) indicates a dummy that takes the value 1 for the second half of YR (1) (1967:2) and -1 for the first half of year YR (2) (1968:1). DYRPR (e.g., D197201) is a dummy that takes on the value of 1 for the year and period indicated (e.g., 1972:1). CDYRPR (e.g., CD196801) is a constant shift dummy taking the value of 1 from YR, PR (e.g., 1968:1). For Austria, dummy I takes the value 0.5 for the second half of 1972, and 1 thereafter; dummy II = D197601. For Belgium-Luxembourg, dummy I is CD197401; dummy II = D196612. For Denmark, dummy I = D197601. For France, dummy I = CD196801. For the Federal Republic of Germany, dummy I = D197302 + D197401. For Italy, dummy I reflects the effects of strikes on trade in manufactures. For Japan, dummy I = 1, 2, 3 for the second half of 1972 through the second half of 1973, and remains at 3 thereafter; dummy II = 1 for 1965. For the Netherlands, dummy I = CD197401; dummy II = D197301. For Norway, dummy I = CD197301. For the United Kingdom, dummy I = 1 for the first half of 1973, and 2 thereafter. For the United States, dummy I takes the value of 1, 2, …, 7 from the first half of 1968 through the second half of 1970, and remains at 7 thereafter; dummy II = 1 for the second half of 1973, 0 for the first half of 1974, and -1 for the second half of 1974.

The sample period is indicated only when it is different from the first half of 1964 through the first half of 1976 (i.e., 1964:1–1976:1).

Table 12.

Fourteen Industrial Countries: Estimates of Volume Equations for Manufactured Exports1

article image

The t-statistics are in parentheses.

This dummy takes the value 0 for the first half year and 1 for the second half year.

For Austria, dummy I = 2 for the second half of 1972 and -1 for 1973. For Belgium-Luxembourg, dummy I = SD6973. For the Federal Republic of Germany, dummy I = D196802; dummy II = 1 for 1970. For Japan, dummy I = 1,2,…, 7 from the first half of 1968 through the first half of 1971 (i.e., 1968:1–1971:1), and 7 thereafter; dummy II is the level of capacity utilization in Japanese manufacturing. For the Netherlands, dummy I = SD71. For Norway, dummy I = CD197301. For the United Kingdom, dummy I = CD197301. (See Table 11, footnote 3, for an explanation of these symbols.)

The sample period is indicated only when it is different from 1964:1–1976:1.

Table 13.

Fourteen Industrial Countries: Relative Price Elasticities for Manufactures1

article image

These elasticities are based on Tables 11 and 12. The short-term response elasticity is the coefficient on the average of the relative price term over the two preceding periods; the longer-term response is the coefficient on the average for the three to seven preceding periods. The cumulative response coefficient is the sum of these two; t-statistics are given in parentheses.

It was hoped that a disaggregation of trade flows into manufactured and nonmanufactured commodities would result in reasonable and significant coefficients on the relative price terms for manufactures. From Table 13, this will be seen to be true for export and import volume equations for the majority of the industrial countries. The principal exception is Italy. For most of the European countries for which the price coefficients were significant, imports responded in large part to changes in relative prices within three periods; for Denmark, Norway, and the United States, the import response to price movements tended to be slower. A similar though less pronounced pattern of price coefficients was found for exports, with the response time for the Federal Republic of Germany, the Netherlands, and the United States being substantially slower than for the other countries. The price coefficients by themselves suggest that a sustained fall in domestic prices relative to those abroad would tend to contribute to an improvement in the balance on manufactures for Japan, Norway, Sweden, and the United States.24 For a number of the other countries, the price effects on the value of exports and imports resulting from a change in relative prices would tend to be offsetting, so that the outturn for the trade balance for manufactures is much less certain.

These measures of price sensitivity must be viewed with caution. There are a number of reasons why these price measures based on unit values and domestic wholesale prices may perform rather poorly. These reasons include differences in coverage, the influence of import prices on the domestic price level, the effects of competitor prices on export price formation, and simple errors in measurement.

In the equations for imports, the coefficients on fluctuations in the demand for manufactures variable range from about 1.3 to 2. The demand coefficients for Japan, Denmark, and the Federal Republic of Germany tend to be larger (though not significantly larger in a statistical sense) than those of other countries. As a result of substantial collinearity with the trend term, the coefficients on potential output tended to have large standard errors and did not generally appear to be reliable considered one at a time. Consequently, the average of the coefficients for the 14 countries was calculated and used to constrain the coefficients for which satisfactory estimates had not been obtained.25

The principal explanatory variables for the equations for manufactured exports are the weighted averages of manufactured imports in the 14 industrial countries and four regions. The coefficients on the foreign market variable ranged from about 0.6 to 1.5. In Table 12, Japanese exports show the greatest sensitivity to short-term variations in foreign demand followed at some distance by those of the United States, Sweden, Italy, and the Federal Republic of Germany; the Dutch, Canadian, French, and Norwegian coefficients show much lower levels of responsiveness.26 At the same time, the large negative trend coefficient for Japan suggests that the responsiveness of exports to short-term fluctuations in foreign demand may exceed their responsiveness to longer-term developments.27 The trend variable for the other industrial countries in Table 12 does not tend to play an important role.

There was little fluctuation in the rate of growth of potential output over the period of estimation, with the result that the potential output variables and time trends were highly collinear. Consequently, the estimated coefficients had large standard errors and frequently displayed the wrong sign. Only the coefficient for the Federal Republic of Germany appeared satisfactory. When the estimation period is extended, it is expected that more reasonable coefficient estimates will be obtained. For the time being, however, it is necessary for purposes of medium-term projection to be able to take into account the implication for countries’ export performances of the deceleration in the growth rates of potential output from the mid-1970s. To this end, the coefficients on the potential output variable were constrained to one. The constraining of these parameters had only a small effect on the estimated coefficients for the other exogenous variables excepting the trend and constant terms.

A number of dummy variables were included in the individual country equations for manufactures and other commodities and are described briefly in the footnotes to the tables. Many of these are designed to pick up changes in seasonal patterns and institutional changes that have had a significant impact on trade flows. A rationale for the inclusion of many of these dummies is immediately forthcoming, for example, the effects of the oil embargo on fuel imports, or the effects of strikes on trade flows, and further information on these dummies is provided in the Appendix.

Volume parameters for nonmanufactured commodities

The equations for the import and export volumes for raw materials are given in Tables 1415. They do not tend to be very satisfactory. However, for forecasting purposes, exogenous information provided by country and commodity specialists on raw material and agricultural flows is introduced.

Table 14.

Fourteen Industrial Countries: Estimates of Volume Equations for Imports of Raw Materials1

article image

The t-statistics are in parentheses.

This dummy takes the value 0 for the first half year and 1 for the second half year.

For Austria, dummy I = D196869; dummy II = CD197102. For Canada, dummy I = 19, 20, …, 29 from the first half of 1964 through the first half of 1969 (i.e., 1964:1–1969:1), and 30 thereafter. For Denmark, dummy I = CD197401; dummy II = D197601. For Japan, dummy I = 19, 20, …, 32 from 1964:1–1971:2, and 32 thereafter; dummy II = D196901. For Sweden, dummy I = 19, 20, …, 34 from 1964:1–1972:2, and 34 thereafter. For the United Kingdom, dummy I = SD6671. For the United States, dummy I = 19, 20, …, 28 from 1964:1–1968:2, and 29 thereafter; dummy II = SD6870. (See Table 11, footnote 3, for an explanation of these symbols.)

The sample period is indicated only when it is different from 1964:1–1976:1.