Investor Sentiment and Momentum Across Market States

Abstract

Previous studies link investor sentiment to momentum, yet the forecasting content of that link may vary with the sentiment measure and the market environment. We compare the same US momentum outcomes across two published sentiment indices, their available calendar windows, and the winner and loser legs, and then evaluate pseudo-out-of-sample forecasts. The Baker-Wurgler (BW) series covers 701 months, whereas the relative-sentiment (RIS) series covers 323 months; a common-sample exercise aligns the outcome dates. We estimate retrospective regressions with three-factor controls and HAC inference. Across the 12 primary slope and interaction tests, none remains significant after Holm adjustment at 5%. RIS has a UP-state momentum slope of 2.486 percentage points per sentiment standard deviation (adjusted p = 0.2826). Expanding-window forecasts use training-only scaling and five model specifications. The state-interaction model produces out-of-sample coefficients of determination of 0.745% for BW and -3.615% for RIS, lower than the best simple comparator in each full sample. Secondary associations and changes over time rule out a universal null reading. The evidence is therefore measurement-sensitive and retrospective; it does not establish broad forecast superiority, a causal investor-psychology channel, or real-time trading profitability.