Quantifying noise in survey expectations

Abstract

Expectations affect economic decisions, and inaccurate expectations are costly. Expectations can be wrong due to either bias (systematic mistakes) or noise (unsystematic mistakes). We develop a framework for quantifying the level of noise in survey expectations. The method is based on the insight that theoretical models of expectation formation predict a factor structure for individual expectations. Using data from professional forecasters, we find that the magnitude of noise is large (10%–30% of forecast MSE) and comparable to bias. We illustrate how our estimates can be applied to calibrate models with incomplete information and bound the effects of measurement error.