A1150
Title: Bias approximations for LIML and FLIML in the dynamic simultaneous equation model
Authors: Emma Iglesias - University of A Coruna (SPAIN) (Spain) [presenting]
Garry Phillips - University of Exeter (United Kingdom)
Dandan Wang - Carlos III of Madrid (Spain)
Abstract: Small-sigma approximations for estimator bias in the dynamic simultaneous equation model (DSEM) have previously been presented for OLS and 2SLS in the literature, where both dynamic and simultaneity bias components are present. FLIML has been shown to be useful for removing bias of order $T^{-1}$ in the static SEM (SSEM), but its performance is unknown in the DSEM. A bias approximation for FLIML is provided that shows only a dynamic bias component exists and the simultaneity component disappears. A bias approximation for LIML is also included. It is shown that the FLIML estimator is the only one that removes the simultaneity bias component of common k-class estimators in the DSEM, which, surprisingly, all have the same dynamic component. Theoretical and simulation evidence demonstrates that FLIML performs best overall, especially for the simultaneous effect. An empirical application demonstrates the usefulness of these results.
