Forecasting Economic Growth in Nigeria using Shrinkage and Variable Selection Methods

Economic forecasts are of major importance to set government policies and to make investment decisions. A recent development in economic forecasting is to combine shrinkage and variable selection methods with factorisation. This research assesses the usefulness of this approach in the context of forecasting contemporary Nigerian real GDP growth. A novel type of data set including real GDP growth rates of 52 African countries and a data set containing 35 economic indicators are used to make predictions. Five shrinkage and variable selection methods are used, including least angle regression, ridge regression, elastic net regularisation, bagging, and boosting. A simulation study shows these methods are very effective when the explanatory power of many variables is low. However, their effectiveness is only limited in the application of forecasting Nigeria’s economic growth, due to the low explanatory power of the data. A factor-based approach is generally preferred to using variables directly, although boosting without factors is the optimal method in terms of predictive accuracy. The best shrinkage and variable selection methods that use factors perform more than 20% better than the autoregressive benchmark, indicating these methods can be of added value in practice. Yet, the out-of-sample period of five years does not allow for strong conclusions. Therefore, more extensive data should be used in future research to verify the robustness of the results.