Mining Big Data Using Parsimonious Factor and Shrinkage Methods

A number of recent studies have focused on the usefulness of factor models in the context of prediction using big data (see e.g., Bai and Ng (2008), Dufour and Stevanovic (2010), Forni et al. (2000, 2005), Kim and Swanson (2014), Stock and Watson (2002b, 2006, 2012), and the references cited therein). We add to this literature by analyzing the predictive bene\u85ts associated with the use of independent component analysis (ICA) and sparse principal component analysis (SPCA), coupled with a variety of other factor estimation and data shrinkage methods, including bagging, boosting, and the elastic net, among others. We carry out a forecasting horse-race, involving the estimation of 28 di¤erent baseline model types, each constructed using a variety of speci\u85cation approaches, estimation approaches, and benchmark econometric models; and all used in the prediction of 11 key macroeconomic variables relevant for monetary policy assessment. In numerous instances we nd that a variety of benchmark autoregressive models and model averaging methods are mean square forecast error (MSFE) dominated by more complicated nonlinear methods. For example, simple averaging methods are MSFE best in only 9 of 33 key cases considered. However, in order to beatmodel averaging methods, we must combine new factor estimation methods with interesting new forms of shrinkage. For example, SPCA yields MSFE-best prediction models in many cases, particularly when coupled with shrinkage. In summary, we present empirical results that provide strong new evidence of the usefulness of sophisticated factor based forecasting methods, and as a corollary, of the use of big datain macroeconometric forecasting