Article Preview
Top1. Introduction
Industry 4.0, or the fourth industrial revolution (4IR), is receiving increasing attention from society, especially its potential impact on human society. It's generally believed that Industry 4.0 will significantly impact the way people live, work, and the way the economy works. As one of the key technologies to unleash more potential of industry 4.0, Artificial Intelligence (AI) has gradually taken out the place of manual processing and traditional optimization. AI technology takes operational data generated by factories to optimize machines and reprogram workflows to improve efficiency and eventual income. However, the potential of AI techniques has not been fully exploited. New AI algorithms can be applied in different aspects of Industry 4.0 to promote efficiency further.
For a long time in the past, qualitative investment has been an essential part of stock investments. By analyzing stock fundamentals and the current stock price, the qualitative investment decides whether to buy or sell according to the investor's personal feelings and experience. However, since the 1970s, the rapid development of computer technology and stock investment began to combine on Wall Street, forming a new concept - quantitative investment. In the age of industry 4.0, artificial intelligence technologies have changed many industries. There is no exception in financial applications as well. More and more fund companies embrace AI technologies to stock selection. It has also been combined with cloud computing and other technologies, like financial modeling and prediction as a service (FMPaaS) (Chang, 2017).
The approach of existing AI quantitative trading strategies generally first filters out some features with predictive value, either manually or automatically (Li, 2019). These features could include fundamental factors, technical factors, macroeconomic data, etc. Then typically, the AI model is trained based on these feature factors to make predictions (Zhang, 2020; Wang, 2019; Xu, 2020; Yang, 2020) and some long-short strategies are made based on the trained model for back-testing.
However, the current AI quantitative trading strategy exhibits many problems and challenges. Firstly, it is hard to extract the most important features. The fluctuation of the stock prices is always affected by various factors such as the economic environment, political policies, market news, etc. Therefore, the stock-related data the researchers can obtain has a low signal-to-noise ratio, which makes the data-based model training prohibitively hard. Secondly, because the style of the stock market is constantly changing, the correlation and dependence in the market are fluctuating tremendously as well. Therefore, even though the existing AI models exhibit high prediction accuracy using the simulated data obtained in the past, they may experience drastic performance degradation if tested in the real market.
At the same time, the high complexity and low interpretability of the machine learning algorithm make the model output, as well as its feature extraction process, inaccessible to humans. In some industries, the “black box” attribute of the model may not be a problem. However, one particular requirement in the asset management industry is that the asset manager must understand and inform the client of the risks of the investment strategy. Therefore, without knowing how it really works and how different features contribute to the final prediction, asset managers and common investors may have little confidence in the AI model, especially when they need to make real investments in a financial market. Thus, the interpretability of the model is particularly critical.