The Forecasting Accuracy of Consumer Attitude Data

Abstract

Abstract This note evaluates the forecasting accuracy of the index of consumer sentiment constructed by the Michigan Survey Research Center. For the most part, other analyses have focused on the interpretation which should be given to an attitude variable or broadly described movements of the index of consumer sentiment in the vicinity of cyclical turning points. None of the studies have compared the accuracy of equations in which the attitude index is an independent variable against naive forecasting models. Thus, Eva Mueller [9] examined whether the index of consumer sentiment modified consumer responses to financial variables. Most of her equations explained variations in durable goods expenditures or automobile sales with the index and different forms of income, liquid asset, and unemployment variables. Since Mueller was not primarily interested in constructing a forecasting equation, her analysis of a large number of equations emphasized the interpretation and the significance of the index. F. Gerard Adams [1] examined the principal components of the index and also concluded that the attitude data were useful, although Friend and Adams [5] showed that stock prices and the length of workweek series shared the predictive ability of the attitude data. Suits and Sparks [11], however, elected to use the index in the automobile equation of the Brookings-SSRC Quarterly Econometric Model, and the index also appears in alternative formulations of the consumption functions of the Wharton-EFU model [4]. Subsequently, George Katona [7], reestimated Miss Mueller's preferred equation and discussed the predictions of the equation around cyclical turns.1 This well-known equation explains real household durable consumption purchases in the half year following the survey with the index and deflated household disposable income in the previous half year. More recently, E. Scott Maynes [8] refit this equation with an iterative technique which reduced the autocorrelation of the residuals. Neither Katona nor Maynes, however, investigated the entire lead-lag relationship between the index and consumer purchases or evaluated with strict standards the movements of the index. The following sections analyze the forecasting record of the index of consumer sentiment as an indicator of movements in the durable consumption expenditures series. Subsequently, several of the regressions previously found useful are refitted, and the forecasts of these equations are compared with predictions from naive models during both the period of fit and for nine quarters beyond the period of fit.