Can Machine Learning Models Predict Inflation?

Abstract

Abstract This paper studies the performance of Machine Learning models in inflation forecasting. The most popular algorithms have been used, respectively Support Vector Regression, Neural Networks, LSTM, Random Forest, XGBoost and LightGBM in both univariate and multivariate form, in order to predict the inflation in Romania, expressed as CPI, Core-1, Core-2 and Core-3, on multiple time horizons. The results suggest that the heuristic methods are not suited in a data-poor environment, being unable to surpass a simple autoregressive model.