ABSTRACT
Since the initial publication of Thirlwall’s (1979) paper on the role of the balance of payments in constraining long-run economic growth, there have been a number of tests of this hypothesis (see Davidson, 1990-91, for an overall assessment of “Thirlwall’s Law”). While these have been reviewed in McCombie and Thirlwall (1994, 1997a), it is useful to examine some new issues that have arisen and to consider some earlier matters in greater detail. Of particular relevance for the testing of the law are implications arising from recent developments in timeseries analysis with regard to integration and cointegration. These issues have been carefully examined by Hieke (1997) with respect to the United States using quarterly data. I shall consider Hieke’s results and those of other studies and also present some further results for the United States (as well as Japan and the United Kingdom). While my results do not lead to any dramatically different conclusions from much of the earlier work, they tie up some loose ends. The tests of the law have generally involved a consideration of how closely
estimates of balance of payments equilibrium growth rates ( yb) approximate to the observed growth rates of national income, or output ( y). Since the balance of payments constraint is deemed to hold in the long run, these growth rates normally average over a number of years. The balance of payments equilibrium growth rate is calculated as either yb = x/πˆ or yb = εˆz/πˆ where x and z are the growth of exports and world income. εˆ and πˆ are estimates of the world income elasticity of demand for a country’s exports and the domestic income elasticity of demand for its imports (see Thirlwall, 1979 and Chapter 1, for the derivation of yb). The estimates of these elasticities are taken from regression analyses of conventional export-and import demand functions. Although a central tenet of the balance of payments constrained model is that relative prices have a quantitatively small role to play in determining the growth of trade flows, the estimates of the income elasticities should be taken from the demand functions that include the relative price term. There are two reasons for this. First, the approach does not argue that relative prices have no effect on trade flows, only that over the long run their impact is quantitatively small. Second, to exclude the effect of relative prices is to assume
what should be tested. The early tests of the hypothesis (e.g. Thirlwall, 1979) used the estimates from Houthakker and Magee’s (1969) seminal study, which estimated import-and export demand functions using annual time-series data and the logarithms of the levels of the various variables. While their approach has been criticized as being subject to a number of potential problems (Morgan, 1970), the estimates have proved remarkably robust and have been confirmed by other studies (see, e.g. Goldstein andKhan, 1978, for the case of the income elasticity of demand for exports). More recent works (such as Bairam, 1988; Bairam and Dempster, 1991) have estimated the demand functions using proportionate growth rates and annual data. The various approaches to the testing of the hypothesis all share a common
rationale: that disparities in the income elasticities of demand primarily reflect disparities in nonprice competitiveness, which are subject to very slow change. Nonprice competitiveness reflects such supply-side characteristics as quality, aftersales service, the effectiveness of distribution networks, and so on. Consequently, while this approach stresses the importance of the growth of demand for exports in the growth process, this is a function of what may be termed a country’s supply characteristics. There is, however, a marked distinction between this approach and the neoclassical emphasis on the supply side (the rate of technical progress and the growth of factors of production) in economic growth. A close relationship between yb and y suggests that changes in relative prices are unimportant in determining trade flows, and the growth of international capital flows plays only a very small role in allowing the divergence of export and import growth rates (see McCombie and Thirlwall, 1997b, for a demonstration of the latter). It is the differences in the income elasticities of demand for exports and imports that play the crucial role in accounting for disparities in economic growth. Consequently, given that in the long run the current account (or, at least, the basic balance) must be in equilibrium, the fact that yb closely approximates y suggests that it is income adjustments (through the Harrod foreign trade multiplier or, more generally, the Hicks supermultiplier) that ensure this occurs (McCombie and Thirlwall, 1994). Thirlwall (1979) originally used Spearman’s rank correlation coefficient to
test the degree of association between yb and y for the advanced countries over the periods, 1953-76 and 1951-73 (using slightly different data sources). This nonparametric test demonstrated that there was a significant positive relationship between the two growth rates. A more rigorous test, originally suggested by McGregor and Swales (1985), is to regress y on yb or, alternatively, ln y on ln yb using pooled data for a number of countries. The null hypothesis is that the intercept of the regression should not be statistically different from zero and the slope coefficient should not differ from unity. Using Thirlwall’s (1979) data, they claimed that the null hypothesis was in fact rejected, hence casting serious doubts on the validity of the law. However, there are two problems with their procedure. First, the values for yb are stochastic, since they are derived from prior estimated coefficients (namely, the πˆs) which have associated standard errors. Regressing y on yb (or ln y on ln yb) suffers from a misspecification analogous to an “errors in
variables” problem. Thus, although no causality, per se, is implied in the regression, which variable is chosen as the regressand and which as the regressor is not immaterial. yb should be regressed on y, and not vice versa. Second, Japan proved to have an actual growth rate that was much less than its balance of payments equilibrium growth rate. This country proved to be an outlier and it was the reason for the rejection of the null hypothesis for all the countries in the sample. The inference is that, while it is plausible that Japan was not balance of payments constrained (it was accumulating large trade surpluses over the period concerned), its inclusion in the sample led to the erroneous conclusion that no advanced country was balance of payments constrained. It should be noted that all countries in a sample are not normally simultaneously balance of payments constrained (see Chapter 5 for a theoretical model that demonstrates the implications of this proposition). Consequently, the finding that a number of individual countries are not balance of payments constrained does not refute the importance of the balance of payments in constraining the growth rates of a significant number of countries (see McCombie and Thirlwall, 1994, ch. 5). This led McCombie (1989) to suggest an alternative test of the law. Define
the hypothetical income elasticity of demand that exactly equates the actual and the balance of payments growth rates as π ′ ≡ x/y. Then, if π ′ and πˆ (the leastsquares estimator) are not statistically significantly different, the hypothesis that the country is balance of payments constrained has not been refuted. This has the great advantage that the test can be applied to each country separately. By this test, a significant number of advanced countries were found to be constrained by their balance of payments over the postwar period (McCombie and Thirlwall, 1994). Two further tests of the law that have been proposed but these are not without
their limitations. The first is to use, for a particular country, the current account equilibrium condition PdX = PfEM , where Pd,Pf and E are the domestic price of exports (X ), the foreign price of imports (M ), and the exchange rate. The exportand import demand functions are substituted into this equation. This gives, after expressing the relationship in terms of growth rates and some rearranging,
yb = εz/π + (1+ ψ + η)( pd − pf − e)/π , (9.1)
where ψ(< 0) and η(< 0) are the price elasticities of demand for imports and exports and e is the rate of change of the exchange rate. If the law is to hold, we should expect the estimate of ε/π to be statistically significant. While the estimate of (1 + ψ + η) may also be statistically significant and negative, it should be of a size that gives only limited explanatory power to the rate of change of relative prices. The advantage of this specification is that it enables us to test the law for indi-
vidual countries using time-series data. The disadvantage is that it is essentially testing the proposition that countries are balance of payments constrained in the short run – in other words, on a yearly, or even quarterly, basis. However, it is highly probable that the growth of exports and imports may diverge substantially over such short periods and, in these circumstances, the growth of capital flows
is likely to be important in accommodating the difference. A rapid upswing in the growth of domestic demand is likely to lead to an immediate worsening of the balance of payments and an increase in import growth over exports that will in the short run be financed by capital inflows. The notion of the balance of payments constraint implies that this cannot persist for very long. Eventually the rate of growth of economic activity will fall sufficiently to generate the balance of payments surplus necessary eventually to reduce the net stock of foreign debt to a level more acceptable to the international financial markets. Consequently, the failure to find that equation (9.1) gives a good statistical fit should not necessarily be taken as a refutation of the law. Atesoglu (1993-94), however, uses a moving average to eliminate short-term
fluctuations using Canadian data and finds that export growth is a significant determinant of output growth. He also explicitly includes the growth of capital flows in equation (9.1), but finds that this term is not statistically significant. The second testwas first proposed byAtesoglu (1993), who smoothed the annual
growth rates of exports and income for the United States over the period 1955-90 by calculating a 15-year moving average. Using the (single) value of the income elasticity of demand for imports estimated over the full period, he calculated 21 overlapping balance of payments equilibrium growth rates. To provide a formal test of the law, he regressed the actual growth rates on the calculated values for yb and tested the null hypothesis that the intercept and the slope did not differ significantly from zero and unity, respectively. This is an interesting test, but a couple of observations are in order. First, while an inspection of the actual and predicted growth rates shows a remarkably close correspondence (see Atesoglu, 1993, p. 512, table 1), the data display very little variation. This is inevitable given that they are generated from samples that significantly overlap. Thus, the slope coefficient is likely to be poorly determined simply because of the lack of variation in the data. TheR2 maywell be very low. Hence, itwould not be surprising if the law were refuted even though y and yb were very similar in magnitude. Consequently, the failure to find a significant relationship should be interpretedwith great caution. Alternatively, it could be argued on a priori grounds that the intercept should be constrained to pass through the origin (see Bairam, 1988). This is likely to give a very close fit since we are almost estimating the regression through two points.1 I provide later an alternative test for the United States based on the use of a rolling regression, which has certain similarities to Atesoglu’s procedure. Finally, other indirect evidence is used in the interpretation of the law. The alter-
native hypothesis, in contrast to the law, is that the growth of exports is endogenous, determined by changes in relative prices expressed in a common currency. Hence, if there is an exogenous increase in the growth of imports, relative prices should adjust to increase the growth of exports, bringing the balance of payments back into equilibrium without requiring any income adjustment. Thus, any observed current account deficits are optimal, representing, for example, intertemporal optimization of consumption on the part of the country. A corollary of this approach is that output growth is determined by the exogenously given growth of technical change and the labor force, and the economy is always on its production possibility curve. Of
relevance to this argument is Cornwall’s (1977) careful study that demonstrates how, even in the Golden Age of economic growth 1950-73, the growth of the labor supply in many advanced countries was essentially endogenous. In much of continental Europe, there were either substantial reserves of labor in the agricultural sector or “guest workers” providing an additional source of labor when demand factors warranted. The United Kingdom drew heavily on immigration from the new Commonwealth countries to supplement its labor supply, especially for unskilled or semi-skilled jobs.
