Model Averaging Estimators For Handling Persistent Predictors

This book addresses the challenge of endogeneity induced by persistent predictors in linear regression models (LRM), an overlooked factor that increases model uncertainty. It aimed to develop model averaging estimators capable of accounting for the persistence property in predictors. The LRM was modified with persistence-adjustment terms, designed to correct for endogeneity by incorporating first-differences of the predictors. Two estimators, Modified Bayesian Model Averaging (MBMA) and Modified Weighted-Average Least Squares (MWALS), are proposed; and compared with existing variants - Bayesian Model Averaging (BMA), Weighted-Average Least Squares (WALS), and Ordinary Least Squares (OLS) - through Monte Carlo simulations under various persistence scenarios (none, low-transient, high-transient, and permanent). Performance was evaluated using metrics like Absolute Bias, Relative Efficiency, Posterior Inclusion Probability, and Relative Root Mean Square Error. Results highlight the importance of persistence-adjusted model averaging for improving estimation accuracy and forecast reliability in models with persistent predictors; and their characteristic sample sample properties.