Rethinking Measures of Sentiment: A New Approach

Abstract

I argue that valid measures of investor sentiment must satisfy additional conditions beyond conventional return predictability tests to imply sentiment-induced misvaluation. Specifically, both positive and negative sentiments should explain returns and volatility contemporaneously, and forecast returns. I show that several well-known sentiment indexes fail to fully meet these necessary conditions and introduce three new empirical indexes that perform better. These proposed measures demonstrate superior predictive power for returns both in-sample and out-of-sample, particularly over longer horizons; and survive the inclusion of non-sentiment variables known to predict returns. Further evidence shows that their robust forecasting ability extends to broader financial outcomes, including changes in flows to actively-managed equity mutual funds, the VIX, and credit spreads.

Keywords: Investor Sentiment, Sentiment-Induced Misvaluation, Return Predictability, Long-Run Return Predictability, Behavioral Finance, Efficient Market Deviations, Empirical Asset Pricing

JEL Classification: G12, G14, G41, E44, C53