Direct tests of the hypothesis that expectations are rational

Abstract

This paper uses the quarterly nominal sales forecasts from the “Manufacturers' Inventory and Sales Expectations Survey” conducted by the U.S. Commerce Department from 1961 to 1976 and tests whether they are consistent with the rational expectations hypothesis. Two sample periods were used, 1961–1972 and 1961–1976, because with the onset of rapid inflation in 1973, there was a significant decrease in forecasting accuracy. Over both samples the sales forecasts test to be unbiased. Over the 1961–1972 sample, the forecast errors made by durable manufacturers test to be uncorrelated, “weakly rational” in the sense that they appear to be orthogonal to past actual sales, and “stronger-form rational” in that they appear to be uncorrelated with past publicly available macroecenomic data. Similarly, the forecasts made by nondurable manufacturers tested to be orthogonal to past forecast errors and past actual sales. However, they did not efficiently utilize the information in past money supply growth or in seasonality. Thus over the first 12 years of the survey, the sales forecasts were found to be generally consistent with the hypothesis of rational expectations. However, when 1973–1975 data were added to the sample, the forecast errors were found to be autocorrelated and to be correlated with publicly available macroeconomic data. Hence, the conclusion one draws depends on one's interpretation of the events of the mid-1970s.