Model averaging with ridge regularization

This paper studies the properties of a model averaging estimator with ridge regularization. I propose the ridge-regularized modifications of Mallows model averaging (Hansen in Econometrica 75(4):1175–1189, 2007) and heteroskedasticity-robust Mallows model averaging (Liu and Okui in Economet J 16(3):463–472, 2013) to leverage the capabilities of averaging and ridge regularization simultaneously. Via a simulation study, I demonstrate the finite-sample improvements gained by ridge regularization, which, in some cases, amount to a reduction of over 50% in the mean squared error. Ridge-based model averaging allows one to accommodate sets of many correlated predictors without blowing up estimation variance. I also show the superiority of the ridge-regularized modifications via empirical examples focused on wages and economic growth.