Using Machine Learning Models to Predict Economic Recession Caused by COVID-19

Abstract

Beginning in early 2020, there was a severe global epidemic, which was named COVID-19. Its outbreak has severely disrupted the global economy. This paper attempts to predict the economic fluctuations caused by COVID- 19. I collect data about unemployment rate, inflation rate, Producer Price Indices, house price, population, stock market value, treasury bill rate, corporate and government bond yields, net export, savings rate, average price-earnings ratio of the stock market, total credit, loan rate, personal and government consumption expenditures and investment, and so forth from the United States spanning from 1983:1 to 2023:2 and make prediction of the Gross Domestic Product (GDP) using a VAR model, five machine learning models, including gradient boosted regression model, random forest regression model, K-nearest neighbors regression model, linear regression model and support vector machine regression model, along with a deep learning model which is Long Short Term Memory model. The predictive effectiveness of those models is measured by mean absolute error and the results shows that the gradient boosted regression model after hyperparameter optimization, whose error is minimized to about 75, is the best at predicting the US economy. This model also exhibit superior performance in predicting the Italian economy, which to some extent shows its widespread usage.