Econometrics of Machine Learning Methods in Economic Forecasting

By Andrii Babii et al

Published on Aug. 23, 2023

Read the original document by opening this link in a new tab.

Table of Contents

1 Introduction
2 High-Dimensional Projections
2.1 Time Series Forecasting
2.2 Panel Data
2.3 Nowcasting, real-time data flow, and textual data
2.4 Granger Causality Tests

Summary

This paper discusses the application of machine learning methods in economic forecasting, focusing on the limitations of traditional maximum likelihood estimation and the benefits of machine learning approaches. It covers topics such as loss functions, optimal decision rules, bias-variance trade-off, nonparametric techniques, and high-dimensional tools. The paper also explores various ML tools like deep learning, random forests, and gradient boosting, tracing back to regression models and regularization techniques. It emphasizes the increasing adoption of ML methods in economics and finance due to the availability of high-dimensional data and computational power. The document reviews recent developments in machine learning literature for economic forecasting, addressing challenges like time series lags, panel and tensor data, nowcasting, Granger causality tests, and classification. It presents methodologies like high-dimensional regularized projections, sparse-group LASSO, and panel data regressions for forecasting economic variables. Furthermore, the paper discusses the importance of nowcasting, real-time data flow, and textual data analysis in improving forecasting accuracy. Lastly, it delves into Granger causality tests using LASSO estimators and HAC estimators to evaluate causal relationships in time series data.