Econometrics of Machine Learning Methods in Economic Forecasting

We review the recent methodological advances in machine learning for economic forecasting and nowcasting. We consider the high-dimensional regularized regressions for individual time series and panel data, paying special attention to how time series lags and cross-validation should be used in practice. We also discuss how to do inference and tests such as the Granger causality test with high-dimensional regularized regressions. Lastly, we review the practical implementation of tree-based methods (boosting and random forests) and (deep) neural networks. We refer the reader to the Python and R libraries that can be used to compute the reviewed methods whenever possible.