Forecasting Accuracy Of Sparse Principal Component Analysis In U.S Inflation
In order to forecast target variables using a dataset with a large number of variables, dimensionality reduction methods are often applied to extract factors from the dataset and those factors are used in forecasting models. The sparse principal component analysis is such a dimensionality reduction technique that has the advantage of interpretability of its components. In this paper, I apply sparse principal component to the Stock and Watson dataset where 132 variables are used to forecast the U.S inflation and compare its predicting performances with the ordinary principal component analysis and the partial least squares. Empirical results show that the sparse principal component analysis brings improvements in forecasting accuracy compared to the ordinary principal component analysis. And it works exceptionally well in the cases where the number of variables is greater than the number of observations.