Forecasting African GDP Growths Using Factor Models and Machine Learning Methods

This paper analyses a hybrid forecast method to estimate forecasts for Ghana's Gross Domestic Product (GDP) growth. We use data of 51 African countries of the time period 1963-2016. Kim & Swanson (2018) [8] explored different methods to estimate forecasts using big datasets. In their research they use hybrid methods, which are mixtures of well known factor models such as PCA, ICA and SPCA, and data shrinkage methods, such as Boosting. We use a hybrid model, by applying Boosting followed by ICA, from Kim & Swanson, to predict forecasts and answer our research question, whether we can accurately forecast Ghana's GDP growth using lagged GDP growths from other African countries. First we will use simulations to obtain the accuracy of our hybrid method. Next we compare our forecast results with those of a benchmark model, using the Mean Square Forecast Error (MSFE), and we can conclude that our hybrid model makes better predictions for Ghana's GDP growth, than the benchmark model. We also conclude that using lags of different countries as predictors can help create a good forecasting model.