Intelligent forecasting of economic growth for developing economies

It is challenging to accurately forecast economic and financial variables in developing economies mainly because they operate in economic environments that are characterized by sudden stops, external shocks, and chaotic behaviour of input variables. Models based on computational intelligence systems that mimic the biochemical processes of the human brain offer an advantage through their functional flexibility and inherent ability to adapt to changing conditions via training and learning processes. Nevertheless, these class of models have hardly been applied to forecast economic time series in these environments. This study investigates the forecasting performance of artificial neural networks and non-parametric regression models in relation to the more standard Box-Jenkins and structural econometric modelling approaches used in forecasting economic time series in developing economies. The results, using different forecast performance measures, show that artificial neural networks and non-parametric regression models perform better than structural econometric and ARIMA models in forecasting GDP growth in selected African frontier economies, especially when the relevant commodity prices, trade, inflation, and interest rates are used as input variables. The magnitude of the gains in forecast performance per unit of time rises up to 150 basis points in some cases. Thus, there is significant scope for practitioners to improve forecast accuracy in developing economies through the use of artificial neural network and non-parametric regression models.