Uporaba metod strojnega učenja za napovedovanje delniških donosov

Breskvar, Žan Mark (author), Faganeli Pucer, Jana (mentor)

V diplomski nalogi smo raziskovali učinkovitost metod strojnega učenja, in sicer metodo naključnega gozda, ''gradient boosting'' in Lasso, pri napovedovanju mesečnih donosov indeksa S&P 500. Naša analiza je temeljila na različnih naborih podatkov, ki vključujejo tehnične, temeljne in ekonomske značilke. Ugotovili smo, da so se tehnični podatki izkazali za najkoristnejše, ekonomski pa so pokazali najslabše rezultate. Med uporabljenimi modeli je Lasso dosegel najboljše rezultate, medtem ko so bili rezultati metode naključnega gozda in ''gradient boosting'' primerljivi. Kljub temu, da napovedovanje donosov na podlagi različnih naborov podatkov ni dalo dobrih rezultatov, smo ugotovili, da je možno z uporabo dolgo-kratke strategije pravilno izbirati delnice, katerim bo cena padla in katerim narastla.

• Language: Slovenian
• Keywords: metode strojnega učenja, naključni gozd, ''gradient boosting'', Lasso, mesečni donosi, indeks S&P 500
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science ; FMF - Faculty of Mathematics and Physics
• Year: 2023
• PID: 20.500.12556/RUL-148388
• COBISS.SI-ID: 158274819
• Publication date in RUL: 21.08.2023

BRESKVAR, Žan Mark, 2023, Uporaba metod strojnega učenja za napovedovanje delniških donosov [online]. Bachelor’s thesis. [Accessed 17 June 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=148388

Secondary language

• Language: English
• Title: The use of machine learning methods in predicting stock returns

Abstract

In this thesis, we investigated the effectiveness of machine learning methods, namely Random Forest, Gradient Boosting, and Lasso, in predicting the monthly returns of the S&P 500 index. Our analysis was based on various datasets that include technical, fundamental, and economic features. We found that technical data proved to be the most useful, while economic data showed the worst results. Among the used models, Lasso achieved the best results, while the results of the Random Forest and Gradient Boosting methods were comparable. Despite the fact that the prediction of returns based on different datasets did not yield good results, we found that it is possible to correctly select stocks for which the price will rise and fall using a long-short strategy.

• Keywords: machine learning methods, random forest, gradient boosting, Lasso, monthly returns, S&P 500 index