Use this URL to cite or link to this record in EThOS:
Title: Testing for weak instruments in two-stage least squares estimation of linear instrumental variable models
Author: Sanderson, Eleanor
ISNI:       0000 0004 5915 3482
Awarding Body: University of Bristol
Current Institution: University of Bristol
Date of Award: 2014
Availability of Full Text:
Access from EThOS:
Instrumental Variable (IV) methods are widely used in the analysis of economic data when the explanatory variable of interest is endogenous and so OLS estimation of the model is biased. However, if the instruments used do not strongly predict the endogenous variable being instrumented then the IV estimator will also give biased results. Weak Instrument Asymptotic theory can be used to model the strength of the instruments in Two-Stage Least Squares (2SLS) IV models and critical values have been developed to test for Weak Instruments in models with one time period. In the first part of this thesis I extend Weak Instrument Asymptotics to a model with multiple endogenous variables where the instruments available can strongly predict each of the endogenous variables separately but correlation between the endogenous variables means that they cannot be jointly predicted and so the overall strength of the instruments in the model is weak. I develop a partial F-statistic to test for 'Weak Instruments of this form and show that this test has the correct size using currently existing critical values for testing for Weak Instruments. I then extend the Weak Instrument Asymptotics to Panel Data models with multiple time periods, and one endogenous variable. I show that it is no longer possible to use the F -statistic to test for Weak Instruments but it is possible to use the Effective F-statistic developed by Olea and Pflueger (2013) with appropriately clustered standard errors to test for 'Weak Instruments in Pallel Data models. Finally, I extend this analysis to look a AR(l) panel data models and show that it is possible to control the strcngth of the instruments asymptotically by changing the persistence of the autoregressive process. I also show the Effective F-statistic has the correct size in these AR(l) models.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available