Regularization methods in economic forecasting

Abstract

Modern regularization techniques are able to select the relevant variables and features in prediction problems where much more predictors than observations are available. We investigate how regularization methods can be used to select the influential predictors of an autoregressive model with a very large number of potentially informative predictors. The methods are used to forecast the quarterly gross added value in the manufacturing sector by use of the business survey data collected by the ifo Institute. Also ensemble methods, which combine several forecasting methods are exemplarily evaluated.