The Congo Case: Forecasting GDP Growth with Factor Models and Machine Learning Methods

This study uses static factor methods to forecast the annual GDP growth of the Democratic Republic of Congo. In a three step approach, factors are computed from a data set of growth variables for other African countries and international variables via sparse and non-sparse Principal Component Analysis. Thereafter, machine learning techniques in the form of Elastic Net and Bayesian Model Averaging are employed to forecast residuals of an autoregressive model. It is found that the forecasts do not outperform estimations of a simple autoregressive model. Further, it is detected that the Congolese data suffer from structural breaks and that models taking such breaks into account gain improvements in predictive performance over comparable autoregressive models.