Forecasting Senegalese quarterly GDP per capita using recurrent neural network

This article evaluates the predictive efficiency of RNNs comparing two types of architecture on quarterly GDP per capita data from Senegal over the period 1960-2020, namely a recursive neural network with re-estimation and a recursive neural network without re- estimate. The RMSE, MAPE and MAE values of the chosen neural network are respectively 7.41%, 8% and 7.73% lower than those of the RNN model has one hidden layer without re-estimation. Indeed, the architecture with two hidden layers converges less quickly than that with only one hidden layer. Thus, the one hidden layer RNN with re-estimate remains the best forecast of Senegal's quarterly GDP per capita during the test period considered. These results suggest the use of artificial neural networks for forecasting economic variables. than those of the RNN model has one hidden layer without re-estimation. Indeed, the architecture with two hidden layers converges less quickly than that with only one hidden layer. Thus, the one hidden layer RNN with re-estimate remains the best forecast of Senegal's quarterly GDP per capita during the test period considered. These results suggest the use of artificial neural networks for forecasting economic variables.