How do experts update beliefs? Lessons from a non-market environment

Abstract

Experts are regularly relied upon to provide their professional assessments in a wide array of markets (e.g., asset pricing, stock and bond ratings, expert witnesses, forecasting), which frequently have characteristics that may generate incentives for experts to provide biased analyses. I ask how experts update beliefs in a relatively simple environment with minimal market incentives. Using data from the Associated Press (AP) Top 25 Poll for college football I find that many standard sets of Bayesian beliefs are rejected by the data, and that experts, while using Bayes’ rule, may still be subject to similar biases as non-experts, including confirmatory bias and lagged signal response, which may be symptomatic of inattention, voter heterogeneity, and signal reassessment. In more complex environments, experts may have strong incentives to substantially deviate from Bayes’ rule, biasing expert predictions in unknown directions.