A1433
Title: Asymptotic expansions and approximate moments for non-linear panel data models with separable errors
Authors: Paul Rilstone - York (Canada)
Gubhinder Kundhi - Memorial University of Newfoundland (Canada) [presenting]
Abstract: Third-order stochastic expansions are derived for the fixed and random effects versions of non-linear panel data models with separable errors. These are used to derive the approximate bias, mean-squared error, skewness and kurtosis of the estimators. In turn, these moments are used to derive Edgeworth, saddlepoint and related expansions for these estimators. In a Monte Carlo experiment, the performance of these expansions for these models is compared to the first order approximation and other techniques commonly used in finite samples. The simulation results indicate that the confidence intervals based on these expansions have substantially better coverage properties than those based on standard asymptotic results and somewhat better than those provided by the bootstrap.
