Yazar "Erik, Fatih" seçeneğine göre listele

Listeleniyor 1 - 1 / 1
  • Küçük Resim
    Öğe
    Comparison of the performances of BIST 30 portfolios by using machine learning algorithms
    (İstanbul Bilgi Üniversitesi, 2021) Erik, Fatih; Öztürk, Serda Selin
    ABSTRACT: In this study, the prices of BIST 30 stocks are estimated using machine learning algorithms such as linear regression, decision tree, support vector machines, long- short term memory and XGBoost, and based on these predicted prices, various portfolios have been created. Portfolios are generally managed in two ways, active and passive. The most important factor while choosing one is their return. Current studies show that the returns of funds traded on Borsa Istanbul Stock Exchange performed below the market, in other words, passive management beat active management. On the other hand, there are also studies showing that portfolios to be created with certain strategies yielded higher returns than the market. In this study, first of all, the prices of BIST 30 stocks for 1 month, 2 months, 3 months, 4 months, 6 months and 12 months are predicted by using machine learning algorithms and according to these revision periods, 6 strategies were created. In the second stage, these strategies were diversified by changing the portfolio sizes (5-10-15 shares). As a result, a total of 90 strategies were formed with 3 different sizes, 6 different timeframes and 5 different algorithms. As a result of the study, it was determined that the returns of portfolios created by using machine learning algorithms are generally above the return of the BIST 30 index. Also, the LSTM algorithm generally makes more successful predictions, 12-month strategies yield higher returns than other strategies, and portfolios including 5 shares are more successful than other portfolios. The result table shows that the most successful portfolio is the portfolio with 5 shares, revised every 12 month and created by using the XGBoost algorithm.