The Intraday High-Frequency Trading with Different Data Ranges: A Comparative Study with Artificial Neural Network and Vector Autoregressive Models.

With the high-frequency trading process, which is a subclass of algorithmic trading transactions, intraday information has increasing importance. Traditional statistical methods often fall short in capturing the intricate patterns and volatility inherent in such high-frequency data. In contrast, artificial neural network (ANN) models demonstrate remarkable capability in handling these challenges, and vector autoregressive (VAR) models provide insights into short-term relationships among variables. This study highlights the importance of using both ANN and VAR models for processing these short time intervals. BIST100 index, which is the main index of Borsa Istanbul, is predicted with two different models in different data ranges with ANN models and VAR models. Both generated ANN models successfully complete the training stages, with extremely high precision, and exhibit exceptionally low error values in their predictions. Although both models are effective, the evidence favors the model evaluated using 5-min data for both the training and prediction phases of ANN models. However, the relative importance of 15-min data in explaining the variation of BIST100 is higher. Moreover, the VAR model results indicate that the short-term relationship between variables can be influenced by the range of data and the 15-min interval data of the variables play a more significant role in explaining the BIST100 index over the longer term. [ABSTRACT FROM AUTHOR]