Possibility of predicting inflation: Using machine learning model

Tegshjargal Sodnomdavaa,  Tsolmon Sodnomdavaa ,  Namuun Amgalanbat

Corresponding Author: Tsolmon Sodnomdavaa

¹School of Management, Mongolian University of Science and Technology, Mongolia, tegshjargal@gmail.com

^2*School of Engineering and Economics, Mandakh University, Mongolia, tsolmon@mandakh.edu.mn

³Business Development Department, Max Group, Mongolia, amgalanbatbadmaa@gmail.com

Digital Object Identifier: https://doi.org/10.53468/mifyr.2025.05.02.49

Abstract: In recent years, global inflationary pressures have intensified due to multiple overlapping risk factors, including the COVID-19 pandemic, supply-chain disruptions, geopolitical conflicts, and escalating trade wars. These shocks have heightened macroeconomic uncertainty and increased the demand for accurate, timely, and robust inflation forecasting, particularly in emerging and small open economies such as Mongolia. This study examines the applicability and performance of machine learning techniques in forecasting inflation in Mongolia and compares them with conventional econometric approaches. The analysis employs monthly data from January 2004 to January 2025, comprising 6,325 observations of inflation and 25 macroeconomic indicators, sourced from official national and international sources. Five forecasting models are evaluated: XGBoost, Random Forest, Quantile Regression, SARIMA, and GARCH. The dataset is split into training (70%) and testing (30%) sets, and model performance is evaluated using the coefficient of determination (R²), root mean squared error (RMSE), and mean absolute error (MAE). To ensure robustness and mitigate overfitting, five-fold cross-validation is applied. The empirical results indicate that machine learning models significantly outperform traditional time-series models in forecasting accuracy. Among all approaches, XGBoost achieves the strongest performance, with an R² of 0.88, an RMSE of 0.12, and an MAE of 0.11, reflecting its superior ability to capture nonlinear relationships and complex macroeconomic dynamics. In contrast, SARIMA and GARCH models demonstrate limited effectiveness, particularly for medium- and long-term inflation forecasting. These findings suggest that conventional linear and volatility-based models may be insufficient in environments characterized by frequent structural changes and external shocks. Moreover, the results highlight the importance of incorporating high-dimensional information and nonlinear learning mechanisms when modeling inflation in volatile emerging economies. Overall, the study underscores the substantial potential of machine learning-based models, especially XGBoost, to enhance inflation forecasting and support data-driven monetary policy formulation, inflation targeting, and macroeconomic planning in Mongolia.

Keywords–Inflation forecasting; Machine learning; XGBoost; Economic shocks; Time series

Article History: Received 20 May 2025, Received in revised form 22 May 2025, Accepted 9 June 2025

Download file: Possibility of predicting inflation Using machine learning model