Semiparametric Estimation of Latent Variable Asset Pricing Models

This paper studies semiparametric identification and estimation of the stochastic discount factor in consumption-based asset pricing models with latent state variables. We model consumption, dividends, and a multiplicative discount factor component via unknown functions of Markovian states describing aggregate output growth. For the case of affine state dynamics and polynomial approximation of the measurement and pricing equations, we provide rank conditions for identification and tractable algorithms for filtering, smoothing, and likelihood estimation. Empirically, we find sizable nonlinearities and interactions in the impacts of expected growth and volatility on the price-dividend ratio and the discount factor.