Bagging weak predictors

Often, relations between economic variables cannot be exploited for forecasting, suggesting that predictors are weak in the sense estimation uncertainty is larger than bias from ignoring relation. In this paper, we propose a novel bagging estimator designed such predictors. Based on test finite-sample predictive ability, our shrinks ordinary least squares estimate—not to zero, but towards null of equates squared with variance. We apply reduce variance further. derive asymptotic distribution and show substantially lowers mean-squared error compared standard t-test bagging. An shrinkage representation simplifies computation provided. Monte Carlo simulations showed predictor works well small samples. Empirically, found proposed worked inflation forecasting when using unemployment or industrial production as an application predicting equity premiums, combination positive constraint forecasts delivered statistically significant gains relative historical average wide range