An Exploratory Study on the Complexity and Machine Learning Predictability of Stock Market Data

Area of Science:

• Quantitative Finance
• Computational Economics
• Time Series Analysis

Background:

• Stock market data complexity and predictability are crucial for financial analysis.
• Understanding temporal trends in market behavior is essential for accurate forecasting.
• The influence of monetary supply on market dynamics requires ongoing investigation.

Purpose of the Study:

• To analyze changes in stock market data predictability and complexity over the past 50 years.
• To investigate the impact of M1 money supply on stock market dynamics.
• To explore the relationship between signal complexity and machine learning-based predictability.

Main Methods:

• Utilized three machine learning algorithms: stochastic gradient descent linear regression, lasso regression, and XGBoost tree regression.
• Applied complexity measures, including approximate entropy and sample entropy, to stock market indices (Dow Jones Industrial Average, NASDAQ Composite).
• Correlated machine learning predictability results with signal complexity measures.

Main Results:

• Observed a decrease in stock market predictability and an increase in signal complexity in recent years.
• Identified a significant correlation between approximate entropy, sample entropy, and the predictability of machine learning models.
• Established a novel link between machine learning predictability and entropy measures in financial time series.

Conclusions:

• Stock market predictability has declined while complexity has risen over the last half-century.
• The identified correlation between entropy measures and ML predictability provides a new tool for analyzing complex financial data.
• These findings are critical for improving the analysis and prediction of stock market time series.