THE EFFECTIVENESS OF NEURAL NETWORKS IN NOWCASTING GDP

Abstract

My paper adopts current advanced Neural Network methodologies which have been used and proven effective in other sectors and applies them to nowcasting GDP growth. More specifically, Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) are being reviewed. Looking at plots to determine the ability of models to time changes in GDP growth as well as looking at the root mean squared error (RMSE) of the nowcasts, my paper concludes with mixed results. RMSE of Neural Network models are similar to benchmark autoregressive (AR) model but Neural Network models are more volatile and move in tandem to changes in macroeconomic variables, giving more timely nowcasts of GDP. This is shown especially in analyses of two recent recessionary periods - the great recession of 2008-2009 as well as during the COVID-19 pandemic in 2020.