Scaled PCA: A New Approach to Dimension Reduction
21 Pages Posted: 14 May 2019
Date Written: March 23, 2019
We propose a novel modification to the popular principal component analysis (PCA) by scaling each predictor according to its predictive power on the target to be forecasted. Unlike the PCA that maximizes the variations of predictors, our scaled PCA, s-PCA, identifies factors that are particularly useful for forecasting the target. Asymptotically, the s-PCA factors converge to true latent factors that are important for the target. Empirically, we find that the s-PCA outperforms the popular PCA substantially in forecasting market return with a variety of investor sentiment proxies and forecasting inflation with a large panel of macro variables.
Keywords: Forecasting, PCA, Big Data, Machine Learning, Supervised Learning
JEL Classification: C22, C23, C53
Suggested Citation: Suggested Citation