Forecasting China's inflation rate: Evidence from machine learning methods

Abstract

We conduct a comprehensive analysis of eight machine learning models (partial least squares, scaled principal components, the least absolute shrinkage and selection operator, ridge regression, random forest, gradient boost decision trees, support vector machines, and neural networks) and the forecast combination method to forecast China's inflation. We use an extensive monthly dataset of 28 predictors with the data period covering January 2000 to December 2022. Our empirical outcomes show that these models beat the autoregressive benchmark regarding out-of-sample R squares. We evaluate the gradient boost decision tree (GBDT) and the forecast combination model as the most effective machine learning tools for forecasting China's inflation rate across various forecasting horizons and evaluation criteria. Moreover, our analysis of variable importance (Gu, Kelly, and Xiu 2020) demonstrates that the retail price index of food and the producer price index of total industry products are the two most dominant predictive signals. These outcomes reflect that structural components and cost-push factors primarily influence China's inflation rate. Our conclusions are robust across various settings.