AR Model or Machine Learning for Forecasting GDP and Consumer Price for G7 Countries

Abstract

This paper examines the validity of forecasting economic variables by using machine learning. AI (artificial intelligence) has been improved and entering our society rapidly, and the economic forecast is no exception. In the real business world, AI has been used for economic forecasts, but not so many studies focus on machine learning. Machine learning is focused in this paper and a traditional statistical model, the autoregressive (AR) model is also used for comparison. A comparison of using an AR model and machine learning (LSTM) to forecast GDP and consumer price is conducted using recent cases from G7 countries. The empirical results show that the traditional forecasting AR model is a little more appropriate than the machine learning model, however, there is little difference to forecast consumer price between them.

Covid-19 Pandemic and Pre-pandemic Economic Shocks to Brazil, India, and Mexico: A Forecast Comparison Evaluating the Impact and Recovery

“…28 But ANN is found to produce superior estimates over DFM. 34,35 As perstatistical models are inappropriate for prediction when the data are highly nonlinear, uncorrelated, nonstationary, and chaotic. 36 ANN imposes fewer assumptions on the underlying data generation process and thereby making it less susceptible to model misspecification, which remains accurate and robust with non-stationary time series.…”

Economic Nowcasting with Long Short-term Memory Artificial Neural Networks (LSTM)

“…LSTMs are an extension of recurrent neural network (RNN) architecture, which introduces a temporal component to ANNs. LSTMs have been used to nowcast meteorological events (Shi et al, 2015) as well as GDP (Kurihara and Fukushima, 2019).…”