چكيده به لاتين
Optimizing investment portfolios in financial markets is an important consideration for various market participants. selecting a collection of assets that maximizes returns while managing risk is crucial for portfolio managers, traders, and all types of investors operating in financial markets. Extensive research has been conducted on factors impacting stock selection and theories underlying investment portfolio construction, sometimes yielding meaningful insights. However, the plethora of approaches for constructing investment portfolios, including modern portfolio theory, fundamental and technical analysis, behavioral finance paradigms, heterogeneous investor behaviors, and other realities necessitate employing novel methodologies and models. The aim of this study is to develop a framework for forecasting stock prices utilizing machine learning and deep learning algorithms and optimizing investment portfolios accounting for predictive errors, thereby better aligning with behavioral models and investor characteristics to facilitate optimal allocation. A comprehensive literature review of the most widely applied machine learning and deep learning techniques was conducted initially. By examining the most cited works in this domain, algorithms such as linear regression, decision trees, gradient boosting, random forests, support vector regression, long short-term memory neural networks, and convolutional neural networks were selected for evaluation. The performance of the chosen algorithms was assessed monthly using chemical industry data from the Tehran Stock Exchange from 2018 to 2023. First, stock price forecasting was performed using machine learning and deep learning. Comparing outcomes based on error metrics revealed convolutional neural networks and support vector regression exhibited superior performance, while the worst was also identified. Subsequently, the mean-variance model incorporating predictive errors was employed to optimize investment portfolios based on Tehran Stock Exchange chemical industry assets using forecasts. To validate the proposed framework, the standard mean-variance model excluding forecast errors was also applied for comparison. Results indicated accounting for algorithms exhibiting less predictive errors yielded better returns and risk-adjusted outcomes for investors.
