Estimates of the Variance of U. S. Inflation Based upon the ARCH Model

Abstract

IT IS GENERALLY AGREED that the unpredictability of future inflation is a major component of the welfare loss associated with inflation. Perfectly predicted inflation might induce some costs through institutional rigidities, governmental interference, and transaction costs, but, in the long run, these should be rather minimal as these institutions adopt various forms of indexing. When inflation is unpredictable, risk averse economic agents will incur a loss, even if prices and quantities are perfectly flexible in all markets. Inflation is a measure of the relative price of goods today and goods tomorrow; thus, uncertainty in tomorrow's price impairs the efficiency of today's allocation decisions. Friedman, in his Nobel lecture [8], argues convincingly that higher variability of inflation would lead to decreased output, ceteris paribus. He then conjectures that higher rates of inflation are generally associated with higher variability of inflation and presumably greater uncertainty about future rates. If this is true, then higher rates of inflation would also be associated with low levels of output, which implies a positively sloped Phillips curve. This paper uses a new statistical technique (ARCH) to estimate the conditional mean and variance of inflation from U.S. time series data. The main finding is that the variance of inflation in the seventies was only slightly greater than in the sixties