Testing linear restrictions in static and dynamic linear simultaneous equations models
2017-11-30T04:39:44Z (GMT) by
This thesis considers the problem of testing linear restrictions on coefficients of a single equation which is part of a linear simultaneous equation system. In particular the Ordinary Least Squares (OLS), Two Stage Least Squares (2SLS) and Instrumental Variables (IV) based F and t tests are considered for testing linear restrictions on structural coefficients. The problem here is that the OLS estimates of structural coefficients are biased and inconsistent. The 2SLS and IV estimates of structural coefficients are consistent but biased. Therefore, in finite samples, the OLS, 2SLS and IV based F and t tests may perform poorly in this case because the exact distributions of these test statistics under the null hypothesis are not known.In this thesis we use large sample and small disturbance asymptotic approximations to the exact distributions of the F and t statistics under the null to obtain appropriate critical values and compare the accuracy of these approximations in small and medium sized samples using Monte Carlo experiments. Also the power properties of the tests based on small disturbance and large sample critical values are compared. In addition we increased the degree of overidentification of the parameters from the equation of interest to see the effects that has on the asymptotic approximations.