OPPORTUNITIES FOR USING ECONOMETRIC AND MACHINE LEARNING METHODS IN FORECASTING SHARE PRICES ON THE UZBEKISTAN STOCK EXCHANGE

Abstract

This thesis explores the opportunities and priority directions for using econometric (ARIMA, GARCH) and machine learning (Random Forest, XGBoost, LSTM) methods in forecasting share prices on the Uzbekistan stock exchange. Based on an analysis of regulatory documents, analytical data from the Tashkent Republican Stock Exchange for 2025, and scholarly sources, it is shown that although trading infrastructure is digitized, discrete trading and low liquidity lead to high forecast errors in standard linear econometric models, while ML algorithms are not substantively linked to institutional investment decisions. A conditional sequence for model implementation, the "data preprocessing – hybrid ARIMA-GARCH and LSTM/XGBoost ensemble – backtesting validation" principle, and five readiness assessment indicators are substantiated.