p.1

Autoregressive and Nonautoregressive Elements in Cross-Section Forecasts of Inflation

Edward J. Kane
Burton G. Malkiel

Abstract

Using cross-section data collected from a panel of institutional investors in 1969, 1970, 1972, this paper focuses on how knowledgeable individuals formulate forecasts of future rates of price inflation. We estimate return-to-normality and error-learning forecasting models and inquire whether such equations can be interpreted simply as reduced forms of an autoregressive forecasting structure. Assuming that respondents' expectations were formed rationally in the sense of Muth, a series of tests lead us to reject the hypothesis of a purely autoregressive forecasting structure. Placed in the context of the return-to-normality model, the decisive evidence consists both of significantly nonzero intercepts and of important variations in survey respondents' anticipated normal rates of inflation that cannot be explained as a reduced-form reflection of past variation in observed inflation rates. These findings indicate that information not collinear with past realizations of price-level change plays an important role in the forecasting process, important enough to allow expected near-term rates of inflation to follow observed inflation rates more closely than autoregressive time series models would suggest.