When Do Markets Fully Process Public Information? Evidence from Real-Time Prediction Markets

Abstract

How efficiently do markets update beliefs when public information arrives in rapid sequence? We use a real-time prediction market setting that combines binary payoffs, precisely observed public signals, and high-frequency market data, allowing us to compare market price changes with changes in a benchmark probability implied by publicly available information. We first show that prices are informative and become more accurate as resolution approaches. During the event, prices respond rapidly to public signals and move in the expected direction. However, directional responsiveness is not the same as efficient updating. Relative to an out-of-sample benchmark probability model, a one-minute change in the benchmark probability is associated with only about a 0.64-for-one contemporaneous change in market prices. The missing adjustment predicts future price drift over the following several minutes, including drift net of subsequent changes in the benchmark probability. We then study the mechanisms underlying this gradual adjustment. Salient public signals are incorporated relatively quickly in liquid markets, but the same signals generate substantially greater underreaction when liquidity is low. Underreaction gaps associated with salient states also predict stronger subsequent drift. The evidence therefore points to gradual price discovery shaped by the interaction between attention and trading frictions. The results contribute to the literatures on prediction markets, market efficiency, and behavioral finance. More broadly, they show that markets can aggregate public information quickly without necessarily incorporating it fully on impact. Market-implied probabilities are often directionally correct, yet adjustment remains incomplete and predictably depends on liquidity and salience.