Survey vs ARCH Measures of Inflation Uncertainty.

This paper evaluates estimates of the variance of U.S. inflation based on popular conditional heteroscedasticity (ARCH) models, by comparing them with the variances of subjective probability distributions for inflation provided in the ASA-NBER surveys of U.S. economic forecasters. The results are not encouraging for the ARCH model. Except in the early 1950s, ARCH effects in U.S. inflation are weak; there are no significant correlations between the ARCH and survey estimates of inflation uncertainty; and the ARCH measures give a misleading picture of the causes of inflation uncertainty and its effects on interest rates. Copyright 1993 by Blackwell Publishing Ltd