Machine Learning for Efficacy Improvements in Automated Decision-Making in Financial Trading: using SigTech platform
Mehta, Mittal (2022)
2022
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-202205026788
https://urn.fi/URN:NBN:fi:amk-202205026788
Tiivistelmä
Machine Learning (ML) for finance is a fruitful approach to detect patterns in data. However, when it comes to predicting financial markets based on financial data with a low signal-to-noise ratio, it has been shown to be a complex problem to solve. A low signal-to-noise ratio means that there is more irrelevant information in the data compared to actionable events, and if a model relies just on data to determine the underlying drivers, then it will most likely learn to infer noise. Instead of predicting financial markets directly, this work focuses on machine learning as a risk management tool that is taught to identify the price trend.
The paper explores novel machine learning areas applied to finance, including meta-labeling, fractional differentiation, and ensemble learning algorithms predicting five different financial instruments for five different periods. Furthermore, we inspect the impact of the strength of trend and distribution of the return of the financial security (target variable) on the financial return generated by the system developed in this study.
Finally, the paper shows promising results for the developed trading strategy, both in terms of cumulative and risk-adjusted return on the assumption that the predictive instrument has a strong trend for a longer period, return distribution is negatively skewed, and has superior machine learning performance metrics, particularly F1-score. The trading system's ability to identify when good investment decisions occur, generated from signals using a naïve and human-understandable model, further offers a path to explainability for AI in finance.
The paper explores novel machine learning areas applied to finance, including meta-labeling, fractional differentiation, and ensemble learning algorithms predicting five different financial instruments for five different periods. Furthermore, we inspect the impact of the strength of trend and distribution of the return of the financial security (target variable) on the financial return generated by the system developed in this study.
Finally, the paper shows promising results for the developed trading strategy, both in terms of cumulative and risk-adjusted return on the assumption that the predictive instrument has a strong trend for a longer period, return distribution is negatively skewed, and has superior machine learning performance metrics, particularly F1-score. The trading system's ability to identify when good investment decisions occur, generated from signals using a naïve and human-understandable model, further offers a path to explainability for AI in finance.