Forecasting Macroeconomic Variables Using Data Dimension Reduction Methods: The Case of Korea
50 Pages Posted: 27 Mar 2015
There are 2 versions of this paper
Forecasting Macroeconomic Variables Using Data Dimension Reduction Methods: The Case of Korea
Date Written: December 3, 2013
Abstract
This paper investigates the usefulness of the factor model, which extracts latent information from a large set of data, in forecasting Korean macroeconomic variables. In addition to the well-known principal component analysis (PCA), we apply sparse principal component analysis (SPCA) to build a parsimonious model, and combine the estimated factors with various shrinkage methods, following Stock and Watson (2012) and Kim and Swanson (2013a), to forecast CPI inflation, GDP growth, exports, consumption and gross capital formation (GCF) of Korea from 2003:01 to 2012:12. Our major findings are that, in predicting growth rates, various hybrid models outperform benchmark models including an autoregressive model, and that this result becomes clearer as the forecast horizons lengthens. Specifically, in forecasting for more volatile periods like the global financial crisis during 2008-09, various hybrid models predict the inflection point better than AR model does. The auxiliary finding is that the main ingredients of Korean macroeconomic variables as indicated by SPCA include interest rates, construction orders received, and employment variables. Surprisingly, the monetary aggregates or price variables are never found to contribute to the principal components in our experiment.
Keywords: Prediction, Sparse Principal Component Analysis, Bagging, Boosting, Bayesian Model Averaging, Ridge Regression, Least Angle Regression, Elastic Net And Non-Negative Garrote
JEL Classification: C32, C53, G17
Suggested Citation: Suggested Citation