Novel modelling strategies for high-frequency stock trading data

Area of Science:

• Computational Finance
• Machine Learning Applications
• Financial Time Series Analysis

Background:

• Electronic automation in stock exchanges generates high-frequency intraday data, necessitating advanced price forecasting methods.
• Machine learning is crucial for stock price prediction, but raw data processing significantly impacts model performance.
• Data preprocessing techniques like thinning and feature engineering are critical but underexplored in current research.

Purpose of the Study:

• To propose and evaluate three novel data processing strategies for high-frequency stock market data.
• To investigate the impact of these strategies on the performance of machine learning-based price forecasting models.
• To demonstrate statistically significant improvements in prediction accuracy using real-world stock data.

Main Methods:

• Development of three distinct modelling strategies for processing raw, high-frequency stock data.
• Application of these strategies to Support Vector Machine (SVM) models for mid-price stock prediction.
• Empirical analysis using high-frequency data from Dow Jones 30 component stocks.

Main Results:

• The proposed data processing strategies led to statistically significant improvements in stock price forecasting.
• F1 scores for SVM models were enhanced by 0.056, 0.087, and 0.016 across the three novel strategies.
• The findings highlight the critical role of data preprocessing in achieving accurate financial predictions.

Conclusions:

• Novel data processing strategies can substantially boost the performance of machine learning models in financial forecasting.
• Effective data preparation is essential for leveraging high-frequency trading data for near real-time price predictions.
• The study provides practical insights for researchers and practitioners in algorithmic trading and quantitative finance.