PERFORMANCE EVALUATION OF ROBUST ESTIMATORS IN THE PRESENCE OF MULTICOLLINEARITY USING ECONOMETRIC MODELS

Abstract

Multicollinearity remains a persistent challenge in econometric modeling, often
leading to inflated variances and unstable parameter estimates when
conventional estimation techniques are employed. This study evaluates the
performance of selected robust regression estimators under varying degrees of
multicollinearity using a Monte Carlo simulation framework. Five estimators
Ordinary Least Squares (OLS), M-estimator, S-estimator, MM-estimator, and
Least Absolute Deviation (LAD) are compared across different sample sizes
(20, 40, 60, 100, and 200) and collinearity levels (0.7, 0.8, 0.9, 0.99, and 0.999).
Estimator performance is assessed using the Root Mean Square Error (RMSE)
criterion. The findings reveal that estimator accuracy deteriorates as
multicollinearity intensifies, while increases in sample size generally improve
performance with diminishing marginal returns under severe collinearity.
Among the estimators considered, the MM-estimator consistently outperforms
others across all simulation scenarios, followed by the S-estimator. The results
provide empirical guidance for researchers and practitioners in selecting robust
estimation techniques when confronted with multicollinearity.