Macroeconomic forecasting of geopolitical changes in the world using machine learning methods

Abstract

The article explores the possibilities of using machine learning methods for macroeconomic forecasting of geopolitical changes based on panel macroeconomic data. The relevance of the study is due to the increasing level of geopolitical instability in the world economy, the intensification of international conflicts, sanctions restrictions and structural crises, which significantly affect the economic development of countries. In such conditions, the need for the use of modern analytical approaches increases, capable of providing more accurate forecasting of geopolitical risks and identifying complex nonlinear relationships between macroeconomic indicators. The purpose of the study is to develop and implement an approach to forecasting the geopolitical risk index using machine learning methods. The work uses ensemble gradient boosting algorithms XGBoost and LightGBM, which allow for effective work with high-dimensional data and take into account the complex structure of relationships between variables. The MAE, RMSE and MAPE metrics were used to assess the quality of forecasting. The information base of the study was formed in the form of a panel structure "country-year", covering 43 countries for the period 2000-2024 and including 499 macroeconomic indicators that characterize production, consumption, foreign trade, institutional development, demographic and environmental processes. The target variable is the geopolitical risk index. To ensure the correctness of the forecast, the sample was divided into training (2000-2018) and test (2019-2024) periods. The results obtained indicate high forecasting accuracy for both models, the MAPE value does not exceed 1.9%, which confirms the effectiveness of the use of ensemble machine learning algorithms in the tasks of forecasting geopolitical processes. It was established that the used models demonstrate the stability of the results even in the presence of missing values ​​and structural breaks in the time series. Additionally, cluster analysis was applied, which allowed us to identify groups of countries with different levels of geopolitical tension. The study confirmed the feasibility of using machine learning methods to analyze and forecast geopolitical changes. The proposed approach can be used as a decision-making support tool in the field of macroeconomic forecasting and economic policy.