The Next-Day Forecasting Performance of Technical Indicators in Stock Markets: A Comparative Analysis with Tree-Based Machine Learning Models
VII. International Applied Statistics Congress (UYIK-2026), İstanbul, Türkiye, 11 - 13 Mayıs 2026, ss.430-441, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.430-441
- Bilecik Şeyh Edebali Üniversitesi Adresli: Evet
Özet
Forecasting the next-day direction in financial markets is one of the most challenging problems in the literature due to the high noise and nonlinear nature of financial time series. While traditional models that do not sufficiently account for this structural complexity struggle to produce stable and reliable predictions, machine learning models overcome these limitations and deliver more successful results. In this study, a leakage-free and explainable machine learning architecture is constructed using data from the Dow Jones Industrial Average, a widely followed benchmark index of 30 major blue-chip U.S. companies. In the first stage, the Borda count method was used to filter a large number of technical indicators, and those with the highest predictive power were selected. In the second stage, using the TimeSeriesSplit strategy, the directional prediction performance of powerful tree-based algorithms—including Gradient Boosting, LightGBM, AdaBoost, and XGBoost—was analyzed comparatively. The findings reveal that tree-based ensemble learning algorithms yield successful results in technical indicator-based predictions, and specifically, the XGBoost model demonstrates higher performance compared to other models. In conclusion, it has been shown that when technical indicators are used in conjunction with appropriate feature selection methods and powerful machine learning algorithms, they offer an effective and viable framework for predicting the direction of stock markets.