For some 15 years, from the mid-1950s to the late 1960s, the record of prices and unemployment in the United States supports the notion of a trade-off between unemployment and inflation, implying that unemployment can as a rule be reduced only at the expense of rising inflation unless structural conditions in labor and product markets are changed or direct controls are brought to bear on wages and prices. In recent years, however, as the United States has been confronted with high unemployment and inflation simultaneously, this trade-off seemed not only to have worsened but, at times when both unemployment and inflation were rising, seemed no longer to exist. While there is evidence that the response of prices to unemployment is delayed, so that relatively high rates of unemployment and inflation may temporarily coexist, the lag involved has been no longer than six months on average and it is implausible that it would have lengthened sufficiently to account for the persistence of high unemployment and inflation in the years 1970 and 1971. Rather, the explanation appears to lie elsewhere. The literature has referred to changes in the demographic composition of unemployment—raising the unemployment rate that corresponds to a given degree of demand pressure—and to influences not necessarily associated with unemployment though affecting its trade-off against inflation. Among these influences are differential reactions of wages to increases of different magnitude in the cost of living and the role of expectations with respect to inflation.1
These and other influences may alter any observed dependence of current price movements on past ones. This paper shows that an increase in that dependence has occurred, thus explaining the recent concurrence of high unemployment and inflation. The paper first presents a model of price behavior, based in part on an expectations hypothesis allowing for an inflationary process that is set in motion as price increases reach and exceed a certain level and which thereby raises the dependence of current price movements on past ones. The model is then empirically tested, and the implications of the results for the trade-off between unemployment and inflation are assessed in the short run as well as in the steady state.
Mr. Spitäller, economist in the Special Studies Division of the Research Department, is a graduate of the University of Graz, Austria, and of the School of Advanced International Studies of the Johns Hopkins University, Washington, D. C.
See Morris Goldstein, “The Trade-Off Between Inflation and Unemployment: A Survey of the Econometric Evidence for Selected Countries,” Staff Papers, Vol. 19 (November 1972), pp. 647–95, for a review of the literature. See also Otto Eckstein, editor, The Econometrics of Price Determination, a volume of papers from the Conference on the Econometrics of Price Determination, Board of Governors, Federal Reserve System (Washington, June 1972).
See James Tobin, “The Wage-Price Mechanism: Overview of the Conference,” in The Econometrics of Price Determination, edited by Otto Eckstein, Board of Governors, Federal Reserve System (Washington, June 1972), pp. 5–15; and Morris Goldstein, op.cit.
Extrapolative expectations may take the form of
A distributed lag pattern provides another formulation of extrapolative expectations; it may be written as
where λ is a weight, Σkλk = 1, and k indicates the length of the lag.
Adaptive expectations may be represented by
See Stephen J. Turnovsky and Michael L. Wachter, “A Test of the ‘Expectations Hypothesis’ Using Directly Observed Wage and Price Expectations,” Review of Economics and Statistics, Vol. 54 (February 1972), pp. 47–54.
Price equations of this form were estimated for the United States and other industrial countries, in Erich Spitäller, “Prices and Unemployment in Selected Industrial Countries,” Staff Papers, Vol. 18 (November 1971), pp. 528–67.
However, even if the coefficients on past price movements summed to unity, a trade-off between unemployment and inflation exists in the short run. It can be seen from equation (7) that the rate of inflation then equals the past rate plus a constant plus a factor that depends on the rate of unemployment (as well as on its rate of change):
For a description of the controls during Phases I and II, see “The Annual Report of the Council of Economic Advisers: Chapter 2. Inflation Control Under the Economic Stabilization Act,” in the Economic Report of the President (U.S. Government Printing Office, Washington, January 1973), pp 51–70.
Because of the inclusion of the lagged dependent variable in equation (11), the Durbin-Watson statistic is biased and does not therefore test reliably for absence of autocorrelation in the residuals. An alternative test developed by Durbin, the h-test, provides better information. However, this test applies to large samples and can only be indicative when applied to equation (11). The value of h for this equation is 1.65, indicating that autocorrelation is absent since the hypothesis of zero autocorrelation is rejected at the 5 per cent level if h > 1.645. See J. Durbin, “Testing for Serial Correlation in Least-Squares Regression When Some of the Regressors Are Lagged Dependent Variables,” Econometrica, Vol. 38 (May 1970), pp. 410–21.
The dependence of
This finding is consistent with the conclusions of a study that analyzes the effectiveness of the controls in detail. See Barry Bosworth, “Phase II: The U.S. Experiment with an Incomes Policy,” Brookings Papers on Economic Activity, No. 2 (1972), pp. 343–83.
The acceleration of inflation refers here to the rise in the elasticity of inflationary expectations with respect to past rates of price change as a result of the threshold effect and not to the acceleration that is merely attributable to higher rates of past inflation.
The hypothesis that the sum of the coefficients on past price behavior, over the range where the threshold effect applies, differs significantly from unity at the 5 per cent level is confirmed if
where SE is the standard error and 0.53 and 0.17 are the relevant coefficients. The denominator of this expression is computed as the square root of the sum of the variances of the coefficients plus twice their covariance. This yields