Medium Term Forecasting Model of Macroeconomic Variables in Russia Using Econometric Methods and Machine Learning

To this day, modeling and forecasting of macroeconomic indicators remains an integral part of economics and the work of many government and financial institutions. Therefore, the purpose of this work was to develop a model using econometrics and machine learning methods that can describe the dynamics of key macro-variables of the Russian economy. The study identified the following tasks: studying successful global and Russian approaches to modeling macroeconomic indicators of countries, collecting, and processing historical data on the main indicators of Russian macroeconomics, building models using the best practices of econometrics and machine learning, evaluating the quality of constructed models, and validating the results, as well as interpreting the results and describing forecasts. Historical data were collected from sources such as the Federal State Statistics Service, the Central Bank of the Russian Federation, and others, and include monthly statistical data on key socio-economic indicators for the period from January 2000 to December 2023. To predict the six main macroeconomic indicators, five types of models were used – ARIMAX, ADL, linear regression, random forest, and gradient boosting. Machine learning models were trained until 2021, and also tested until 2023. It is important to notice that econometric models and machine learning models were compared with each other on an out-of-sample forecast for 2023 using the MAPE metric. As a result, the best model was chosen for each endogenous variable. A feature of this study is also the lack of data on exports and imports of goods for 2023, therefore, a forecast was also built on them. It is important to note that the best quality was obtained specifically for machine learning models, so forecasts for 2024-2025 were also built specifically for gradient boosting and random forest models based on three scenarios – inertial, optimistic, and pessimistic.