This thesis is comprised of two parts: Part I It is common in empirical economic applications that use microdata to exhibit a natural ordering into groups. Angrist (1991) use dummy variables based on such grouping to form instruments for consistent estimation. Khatoon et al. (2014) and Andrews et al. (2016) extend the GEL class of estimators to the case where moment conditions are specified on a groupbygroup basis and refer to the resulting estimator as groupGEL. A natural consequence of basing instruments or moment conditions on groups is the degree of overidentification can increase significantly. Following Bekker (1994) it is recognized that inference based on conventional standard errors is incorrect in the presence of many instruments. Furthermore, when using many moment conditions, twostage GMM is biased. Although the bias of Generalized empirical likelihood (GEL) is robust to the number of instruments, Newey and Windmeijer (2009) show that the conventional standard errors are too small. They propose an alternative variance estimator for GEL that is consistent under conventional and manyweak moment asymptotics. In this part of the thesis I demonstrate that for a particular specification of moment conditions, the groupGEL estimator is more efficient than GEL. I also extend the Newey and Windmeijer (2009) manymoment asymptotic framework to groupGEL. Simulation results demonstrate that groupGEL is robust to many moment conditions, and tstatistic rejection frequencies using the alternative variance estimator are much improved compared to using conventional standard errors. Part II Following the seminal paper of Abowd et al. (1999), Linked EmployerEmployee datasets are commonly used in studies decomposing sources of wage variation into unobservable worker and firm effects. If it is assumed that the correlation between these worker and firm effects can be interpreted as a measure of sorting in labour markets, then an efficient matching process between workers and firms would result in a positive correlation. However, empirical evidence has failed to support this ascertain. As a possible answer to this apparent paradox, Andrews et al. (2008) show the estimation of the correlation is biased as a function of the amount of movement of workers between firms, socalled Limited Mobility Bias (LMB); furthermore they provide formula to correct this bias. However, due to computational restrictions, application of these corrections is infeasible, given the size of datasets typically used. In this part of the thesis I introduce an estimation technique to make the biascorrection estimators of Andrews et al. (2008) feasible. Monte Carlo experiments using the biascorrected estimators demonstrate that LMB can be eliminated from datasets of comparable size to real data. Finally, I apply the biascorrection techniques to a linear model based on the Danish IDA, and find that correcting the correlation between the worker and firm effects due to LMB provides insufficient evidence to resolve the above paradox.
Date of Award  1 Aug 2018 

Original language  English 

Awarding Institution   The University of Manchester


Supervisor  Martyn Andrews (Supervisor) & Alastair Hall (Supervisor) 

The Generalized Empirical Likelihood estimator with grouped data ; Bias correction in Linked EmployerEmployee models
Lincoln, J. (Author). 1 Aug 2018
Student thesis: Unknown