Effects of multicollinearity in simultaneous equation models and comparisons of alternative estimators

Simultaneous equation models are used quite extensively in economics, econometrics and statistics. Using ordinary least squares (OLS) for the estimation of simultaneous equation model produces biased and inconsistent estimates. Estimation methods overcoming simultaneity between error terms and endogenous variables are used for the estimation of these models. However, in the presence of multicollinearity, variance of these estimators are inflated and these estimators produce unstable estimates. In such cases, estimators producing more stable estimates are used to overcome the effect of multicollinearity. In this study, Klein's simultaneous equation model with multicollinearity problem presented in his study titled "Economic Fluctuations in the United States" in 1950 is used. The model is estimated with traditional estimators two-stage least squares (2SLS), three-stage least squares (3SLS) and biased estimators ridge, generalized maximum entropy (GME). Performances of these estimators are compared according to the mean square error (MSE) criteria obtained with bootstrap. As a result, in the presence of multicollinearity, GME estimator is decided as the most efficient estimator.