Can Machine Learning improve inflation forecasting?

Abstract: This paper aims to compare and evaluate the performance of inflation forecasting performance for benchmark time series models and machine learning models. The process is performed for both a developed economy, the US, and an emerging economy, Mexico. The study examines how forecast performance compares between benchmark time series models and machine learning models, as well as how forecast performance overall compares between an emerging economy and a developed economy. To conduct the study, two separate datasets for the US and Mexico are gathered and to perform the forecasts, a rolling window with a fixed window size is used across a total of 12 horizons. The study finds that the benchmark time series models outperform the machine learning models in forecasting inflation for the US dataset, but not for the Mexican dataset. Additionally, the study finds that the inflation forecast performance shows smaller forecast errors across all horizons for the developed economy compared to the emerging economy. These results are mainly attributed to the characteristics of each specific dataset. The US dataset with its higher dimension and more potential predictors is concluded to better suit machine learning models and perform better inflation forecasts. In conclusion, the study provides evidence that machine learning can be a useful tool for macroeconomic policymakers when sufficient data is available.