A novel approach for portfolio optimization using machine learning an‎d deep learning algorithms: A two-stage framework for stock preselec‎tion an‎d optimal weight allocation

1/1/1900 12:00:00 AM

This study presents a novel methodology for optimizing portfolios tailored to risk-averse investors, addressing the challenges of return prediction an‎d uncertainty management. The research unfolds through a six-phase process: initial data collection an‎d preparation, development an‎d eva‎luation of advanced predictive models, selec‎tion of the most effective forecasting approach, preselec‎tion of high-potential assets, transformation of returns into triangular fuzzy numbers using expert insights, an‎d optimization via a credibilistic Conditional Value at Risk (CVaR) model. The methodology introduces state-of-the-art neural network models, with prediction accuracy metrics reviewed by a market expert to incorporate factors influencing future stock performance, culminating in a robust portfolio construction framework. Numerical results indicate a base case portfolio with a CVaR of -0.466962535, reflecting positive returns in the worst 5% of scenarios, an‎d an achieved return of 0.46560425, with weights allocating 30% to PLTR an‎d 10% each to seven other stocks, demonstrating resilience an‎d balance. The study’s contributions include pioneering cutting-edge prediction techniques, a novel expert-informed approach to fuzzy return modeling, an‎d an innovative risk optimization strategy, distinguishing it from existing literature. These findings offer practical insights for portfolio managers, including a diversified strategy adaptable to volatile equity markets, an‎d pave the way for future research into dynamic adjustments an‎d broader market applications.

بهينه سازي سبد سهام , يادگيري ماشين , يادگيري عميق , پيش‌بيني قيمت سهام , سنجه‌هاي ريسك

Portfolio optimization , Machine learning , Deep learning , Stock price prediction , Risk measures

Mahdi lotfizadeh ghanjelu

Dr. Emran mohammadi