محمدرضا دهقاني

Portfolio optimization based on asset return forecasting using machine learning models is regarded as an innovative approach in modern investment management. This study was designed in two main stages with the aim of enhancing the accuracy of the capital allocation process. In the first stage, stock market data, including historical prices, technical indicators, an‎d essential influencing factors, were collected an‎d preprocessed appropriately for modeling purposes. To predict future asset returns, a variety of machine learning algorithms—including Elastic Net, Ran‎dom Forest, an‎d XGBoost—were employed, an‎d the performance of each model was eva‎luated using accuracy metrics such as Mean Absolute Error (MAE) an‎d R-squared (R²). In the second stage, the forecasted returns generated by these models served as input data for an advanced portfolio optimization model using the semi-variance criterion. The semi-variance model, with an emphasis on downside risk an‎d consideration of the asymmetry in return distributions, demonstrated greater effectiveness compared to the traditional mean-variance approach, particularly in turbulent market environments. The optimization problem was formulated as a multi-objective model aiming to maximize the predicted portfolio return while minimizing downside risk, leading to the determination of the optimal asset weights. The desirability of the optimized portfolios was assessed based on metrics such as annual cumulative return, Sharpe ratio, an‎d maximum drawdown. The findings revealed that the use of machine learning models for return forecasting, combined with semi-variance-based optimization, resulted in significant differences in portfolio performance depending on the model chosen. The comparative analysis of the three selec‎ted models indicated that the optimized portfolio based on the outputs from the Elasticnet model outperformed the others in terms of cumulative return, Sharpe ratio, an‎d downside risk control. This practical comparison can assist investors in selec‎ting appropriate data-driven approaches for portfolio optimization an‎d provides valuable insights for eva‎luating the effectiveness of various machine learning techniques in financial decision-making.

بهينه سازي سيد سهام , يادگيري ماشين , نيم واريانس , بهينه سازي چند هدفه , مدل سازي ريسك , سري زماني

Portfolio Optimization , Machine Learning , Semi-variance , Risk Modeling , Multi-objective Optimization , Time Series

mohammadreza dehghani

rouzbeh ghousi