Monte Carlo Simulations of the Nested Fixed-Point Algorithm
Johnson, Erik P.
MetadataShow full item record
There have been substantial advances in dynamic structural models and in the econometric literature about techniques to estimate those models over the past two decades. One area in which these new developments has lagged is in studying robustness to distributional assumptions and finite sample properties in small samples. This paper extends our understanding of the behavior of these estimation techniques by replicating John Rust’s (1987) influential paper using the nested fixed-point algorithm (NFXP) and then using Monte Carlo techniques to examine the finite sample properties of the estimator. I then examine the consequences of the distributional assumptions needed to estimate the model on the parameter estimates. I find that even in sample sizes of up to 8,000 observations, the NFXP can display finite sample bias and variances substantially larger than the theoretical asymptotic variance. This is also true with departures from distributional assumptions, with the mean square error increasing by a factor of 10 for some distributions of unobserved variables.