Title: Forecast combinations for predictive regressions via the Lasso
Authors: Bonsoo Koo - Monash University (Australia) [presenting]
Hong Wang - Monash University (Australia)
Abstract: When a number of specifications are suggested, forecast combination reduces the information in a vector of forecasts to a single summary measure using a set of combination weights. While the reasons are poorly understood, simple equal weighted (EW) forecast combination scheme often outperforms more sophisticated combination schemes in empirical studies. We propose a Least Absolute Shrinkage and Selection Operator (LASSO) estimator of the optimal combination weights of which are estimated from potentially highly correlated covariates (individual forecasts). Motivated by the properties of LASSO, we demonstrate two applications of the proposed LASSO approach in time series setting. The proposed LASSO approach is applied to forecasting stock returns with comparison to the simple equal weighted (EW) combination scheme, which in turn outperforms the best individual predictions.