Modeling uncertainty: predictive accuracy as a proxy for predictive confidence

This paper evaluates current strategies for the empirical modeling of forecast behavior. In particular, we focus on the reliability of using proxies from time series models of heteroskedasticity to describe changes in predictive confidence. We address this issue by examining the relationship between ex post forecast errors and ex ante measures of forecast uncertainty from data on inflation forecasts from the Survey of Professional Forecasters. The results provide little evidence of a strong link between observed heteroskedasticity in the consensus forecast errors and forecast uncertainty. Instead, the findings indicate a significant link between observed heteroskedasticity in the consensus forecast errors and forecast dispersion. We conclude that conventional model-based measures of uncertainty may be capturing not the degree of confidence that individuals attach to their forecasts but rather the degree of disagreement across individuals in their forecasts.