Estimating dynamic stochastic decision models: explore the generalized maximum entropy alternative

This paper proposes a generalized maximum entropy (GME) approach to estimate nonlinear dynamic stochastic decision models. For these models, the state variables are latent and a solution process is required to obtain the state space representation. To our knowledge, this method has not been used to estimate dynamic stochastic general equilibrium (DSGE) or DSGE-like models. Based on the Monte Carlo experiments with simulated data, we show that the GME approach yields precise estimation for the unknown structural parameters and the structural shocks. In particular, the preference parameter which captures the risk preference and the intertemporal preference is also relatively precisely estimated. Compare to the more widely used filtering methods, the GME approach provides a similar accuracy level but much higher computational efficiency for nonlinear models. Moreover, the proposed approach shows favorable properties for small sample size data.