Bias assessment and reduction for limited information estimation in general dynamic simultaneous equations models

Wang, Dandan 2017. Bias assessment and reduction for limited information estimation in general dynamic simultaneous equations models. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (1MB) | Preview
	PDF - Supplemental Material Restricted to Repository staff only Download (94kB)

Abstract

Most of the literature which has considered the small sample bias of limited information estimators in simultaneous equation models has done so in the context of the static rather than the dynamic simultaneous equations model (DSEM). Therefore, an analysis of the performance of estimators in the general dynamic simultaneous equations case is timely and this is what is provided in this paper. By introducing an asymptotic expansion for the estimation errors of estimators, we are able to obtain bias approximations to order T−1. Following this we constructed bias corrected estimators by using the estimated bias approximation to reduce the bias. As an alternative, the use of the non-parametric bootstrap as a bias correction procedure was also examined. In Chapter 2, we analyse the Two Stage Least Squares ( 2SLS ) Estimator in the general DSEM. Based on the result in Chapter 2, Chapter 3 compared the Fuller modification of the limited information maximum likelihood estimator (FLIML) with the 2SLS estimator. The bias approximation and reduction in the pth-order dynamic reduced form are analysed in Chapter 4. The results indicate that FLIML gives much less biased estimates than the 2SLS estimation in the general DSEM. We have also observed that the bias correction method based on the estimated bias approximation to order T−1 provides almost unbiased estimates and it does not lead to an inflation of the mean squared errors compared with the associated uncorrected estimators. We suggest that the corrected estimators, based upon the O(T−1), should be used to reduce the bias of the original estimators in small samples. Alternatively, the numerical results show that the bootstrap method leads to an effective reduction of the bias and an inflation of MSE, however this reduction is not as effective as the first one.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Business (Including Economics)
Subjects:	H Social Sciences > HB Economic Theory
Date of First Compliant Deposit:	27 February 2017
Last Modified:	20 Apr 2021 09:25
URI:	https://orca.cardiff.ac.uk/id/eprint/98538

Actions (repository staff only)

Edit Item