MACHINE LEARNING APPLICATIONS IN PORTFOLIO OPTIMIZATION

Date

2024-9-5

Author

Uykun, Firdevs Nur

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

119
views

86
downloads

Cite This

This study examines the evaluation of S&P 500 trend movements and their impact on portfolio optimization methodologies. Through the meticulous construction of risk aversion-adjusted portfolios applicable to both single and multiple period analyses, the research employs variance and Conditional Value at Risk (CVaR) for single periods, while using Mean Absolute Deviation (MAD) for multiple periods. The optimization process benefits from the synergistic use of Python for computational modeling and AMPL for executing complex mathematical formulations. To accurately predict risk aversion, the study utilizes six classification models integrated with 29 indicators derived from technical analysis: Logistic Regression (LR), K-Nearest Neighbors (KNN), Support Vector Classifier (SVC), Decision Trees (DT), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). Simultaneously, return forecasting leverages the predictive capabilities of four regression frameworks—Linear Regression, LSTM, XGBoost, and LightGBM—based on various technical indicators. Explainable AI (XAI) techniques, particularly LIME and SHAP, facilitate a deeper understanding of feature importance in the decision-making processes of machine learning algorithms. The findings of this thesis demonstrate that the methods used perform differently across various optimization problems. While DT achieves the highest Sharpe ratio for Mean-Variance portfolios, LR performs best in the Mean-CVaR portfolios, and SVC excels for the Mean-MAD portfolios. Additionally, the KNN, SVC, and LR have the lowest Sharpe ratios, respectively. In the process of predicting returns, Linear Regression produces the best outcomes based on the applied comparison metric. Lastly, the XAI methods highlight the importance of incorporating the Average Directional Index (ADX) with a ten-period setting in the feature design of the KNN and LR models.

Subject Keywords

Portföy Optimizasyonu, Teknik Analiz, Makine Öğrenmesi, Açıklanabilir Yapay Zeka

URI

https://hdl.handle.net/11511/111402

Collections

Graduate School of Applied Mathematics, Thesis