Semiparametric estimation of latent variable asset pricing models

Abstract

This paper studies semiparametric identification and estimation of consumption-based asset pricing models with latent state variables. First, we measure how consumption, dividends, and prices depend on Markovian state variables describing aggregate output growth. Subsequently, we identify state-dependent components in the stochastic discount factor using the Euler equation. We develop tractable algorithms for filtering, smoothing, and sieve maximum likelihood estimation, and establish its consistency. Empirically, we find sizable nonlinearities in the impact of expected growth and volatility on the price–dividend ratio and discount factor.