Time-Varying Market Price of Risk and Investor Sentiment: Evidence from a Multivariate GARCH Model

We test a conditional version of the CAPM in the U.S. equity market using a parsimonious generalized autoregressive conditional heteroskedasticity (GARCH) model in which the risk premia, betas, and correlations vary through time. We introduce U.S. investor sentiment from two surveys as conditional information variables, whereas previous studies generally use economic fundamentals. We assume that investor sentiment is not entirely irrational and decompose it into its irrational and rational components; using the irrational components as information variables. We find that irrational investor sentiment measures are statistically significant, and contain information which may be valuable in asset pricing models.