The Future of Fuzzy Sets in Finance: New Challenges in Machine Learning and Explainable AI

Traditional statistical analysis is oriented towards finding linear relationships between the variables under investigation, often accompanied by strict assump-tions about the problem and data distributions. Moreover, traditional analysis en-dorses data reduction as much as possible before modeling, and, as a result, part of the original information is lost. On the other hand, machine learning does not impose rigid pre-assumptions about the problem and data distributions since the underlying ratio is to “learn from data”, without the need for data reduction or a priori knowledge before the learning. For these reasons machine learning has ex-perienced a rapid dissemination in a large number of sectors including healthcare, finance, transportation, retail and social media services industry. Machine learn-ing is the core technology of the new age of AI applications. Machine learning methods offer tremendous benefits, but are limited by their opaqueness, non-intuitiveness and difficulty to understand. In finance in particular, machine learning methods have played a crucial role in improving the forecasting ability of financial models and trading systems, due to their ability to process a large amount of data and the peculiarity of capturing also non-linear relationships between variables. In recent years, the availability of sample data at very high frequencies (intraday or tick by tick) resulted in a fertile domain for their application, especially in the coding of indicators and patterns of technical analysis. Deep learning systems are the most advanced form of machine learning. They can match humans in recognizing images or driving a car, but why they come up with the solutions remains difficult to tell exactly. Businesses would have used machine learning more widely if they could understand how machines come up with their recommendations on trading, fraud detection, insur-ance and banking. A challenge for AI in finance is the need to analyze and aggregate a large amount of information obtained from different sources. In the financial literature, the use of artificial intelligence (AI) and machine learning techniques is often lim-ited to the coding of technical analysis indicators (such as moving averages or the flag pattern) for trading strategy purposes. As pointed out in [1], most of the con-tributions investigating machine learning methods in financial markets propose trading strategies that rely mainly on technical analysis and focus on a single stock market or market index. Moreover, financial markets are generally treated as compartmentalized and most of the contributions investigate only a single mar-ket or a specific asset [2]. On the other hand, investors and regulators need com-prehensive risk measures able to aggregate and synthesize different types of in-formation in a single indicator. Another challenge is represented by the increasing dominance of computerized trading, which may cause more volatility during market downturns. The rising frequency of 'flash crashes' across many major markets, the increasing incidents of volatility such as the VIX spike on Feb. 5, 2018, the 10-year Treasury bond on Oct. 15, 2014, and the British pound on Oct. 6, 2016, are an important early warning sign that machines have to be closely supervised and understood. New measures and tools to control the volatility of financial markets [3,4,5] should be developed. The semantic properties of linguistic fuzzy sets, their good coverage even in the case of lack of data, their management of the uncertainty, especially in Big Data, make them a very interesting tool for nowadays applications, especially when the practitioner need to understand why a given decision has been made.