Latent Factor Prediction Pursuit for Rank Deficient Regressors

Abstract

In simulation studies Latent Factor Prediction Pursuit outperformed classical reduced rank regression methods. The algorithm described so far for Latent Factor Prediction Pursuit had two shortcomings. It was only implemented for situations where the explanatory variables were of full colum rank. Also instead of the projection matrix only the regression matrix was calculated. These problems are addressed by a new algorithm which finds the prediction optimal projection.