Secular and Cyclical Responses of U.S. Trade to Income: An Evaluation of Traditional Models
使用谱分析方法，将美国进出口与收入的关系分解为长期趋势和商业周期成分，发现传统回归估计的收入弹性可能存在偏差，因为贸易对长期和周期收入变化的反应不同。
IN the three decades since Guy Orcutt's (1950) classic paper evaluating empirical research on international trade flows, most trade research has focused on the influence of prices. In the decade since the benchmark study by Houthakker and Magee (1969), examination of the role of prices has accelerated. Although income is consistently the most significant explanatory variable in these studies, it is typically introduced only as a background control variable rather than as a variable of central interest. The few explicit studies of the link from income to trade either use proxies such as delivery lags and inventory ratios to investigate the dynamics of the link (e.g., Gregory, 1971 and Artus, 1973), or decompose income into trend, deviation-from-trend, and (in one instance) change in deviation-from-trend in an attempt to separate the secular and business cycle components (e.g., Ball, Eaton, and Steuer, 1966; Khan and Ross, 1975; and Dunlevy, 1980). Stephen Magee (1975, pp. 188-193, 211-214) provides a persuasive rationale for at least the spirit of the second line of research, arguing that trade flows respond differently to secular and business cycle income changes and that regression estimates with income in level form mix these effects. Time domain decomposition is now an established practice, with the trend response identified with secular forces like comparative advantage (hence could be positive or negative in either demand or supply), the deviation-from-trend response with a business cycle Keynesian mechanism (hence is likely to be positive in demand and negative in supply), and the change in deviation-from-trend response with a sneed of adjustment constraint (hence is likely to be negative). The most recent evidence bearing on these issues is provided by Dunlevy (1980), who finds that the effect of the business cycle income proxy (deviation-fromtrend income) on the supply of exports for the United States and the United Kingdom is positive rather than negative, evidence contradicting both theory and evidence in earlier capacity pressure studies. In this paper, we first evaluate traditional regression estimates of income elasticities of demand and supply for U.S. imports and exports from 1955 to 1979, then compare these estimates to cross spectral estimates of the secular and business cycle responses of trade to income. The spectral technique is used because, by definition, the approach estimates separately the secular and business cycle links between trade and income.' If the response of trade to movements in income does differ over cycles, regression estimates of income elasticities may reflect bias and instability due to cycle-varying parameters. Regression elasticity estimates with income in trend and deviation-from-trend forms do provide separate estimates, but the decomposition is useful only to the degree that the two variables are accurate proxies for secular and business cycle movements in income. It should be emphasized that we are not using spectral analysis to detect lead-lag properties, since these are more directly obtained in the time domain and are relatively trivial for our sample. The unique advantage of spectral analysis is that it isolates secular and business cycle components directly, even when the time domain link is only contemporaneous. The paper is organized as follows. Section II presents and evaluates time domain estimates of standard demand and supply equations for U.S. imports and exports and presents caveats regarding interpretation of the trend and deviationfrom-trend income variables. Section III proReceived for publication March 4, 1981. Revision accepted for publication May 12, 1982. University of Oregon. An earlier draft of this paper was presented at the Fall 1980 North American Meeting of the Econometric Society. We acknowledge helpful comments from S. Y. Kwack, John Mutti, and two anonymous referees. We also thank John Pippenger for supplying the spectral program SPECX68, and the University of Oregon for a Faculty Research grant. Remaining errors are solely ours. I Application of spectral methods to investigate trade-income links was first suggested by Ball, Eaton, and Steuer (1966, p. 503). They were unable to use the technique, however, because the data series available at that time were too short.