Akademik Veri Yönetim Sistemi
5th International Conference on Informatics and Software Engineering, IISEC 2026, Ankara, Türkiye, 5 - 06 Şubat 2026, ss.599-604, (Tam Metin Bildiri)
This study provides a comprehensive comparative benchmarking analysis of the financial time series structure of emerging markets (EM), which is characterized by high volatility, low signal-to-noise ratio, and structural heterogeneity. Within this scope, 156 independent test scenarios were created using the stock market indices of 13 emerging countries, and a wide range of models was evaluated, from traditional statistical models to machine learning methods and modern deep learning architectures. In the results obtained, the ARIMA model was the most successful method against the Random Walk benchmark with a success rate of 68.6%. Among deep learning models, Transformer stood out as the most stable modern architecture with a success rate of 44.2%, while more complex structures such as 1D-CNN and xLSTM experienced generalization difficulties in high-noise market conditions. Furthermore, it was determined that prediction success is significantly dependent on country characteristics and the prediction horizon, whereas the impact of data splitting strategies on overall performance is limited. These findings reveal that model complexity alone is insufficient in emerging markets and that prediction strategies must be designed in a manner sensitive to market structure and time scale.