Subsymbolic and Hybrid Artificial Intelligence Techniques in Financial Engineering
by László Monostori
The major difficulties in financial engineering, the most computational intense subfield of finance, are as follows. The problems are usually multivariate, nonlinear, difficult to model. The information available is imprecise and incomplete, can have both numerical and linguistic forms. Chaotic features and changing environment can be mentioned as further barriers. Obviously, intelligent techniques, more exactly, techniques of artificial intelligence (AI) research, became conspicuous in different fields of finance and business. The range of applications is rapidly increasing and companies use intelligent systems even to automate parts of their core business. The fundamental goal of investigations started some years ago at the SZTAKI was to explore intelligent techniques, including intelligent hybrid systems and multistrategy learning applicable in different fields of finance and business.
The potential users of intelligent techniques in finance and business are corporate finance, financial institutions, and professional investors. The applications can be grouped as follows: analysis of financial condition, business failure prediction, debt risk assessment, security market applications, financial forecasting, portfolio management, fraud detection and insurance, mining of financial and economic databases, other (eg macroeconomics) applications.
Intelligent hybrid systems are a very powerful class of computational methods that can provide solutions to problems that are not solvable by an individual intelligent technique alone. Perhaps, the most important class of hybrid systems is the integration of expert systems and neural networks. Several techniques have emerged over the past few years spreading from stand-alone models, through transformational, loosely and tightly coupled models to fully integrated expert system/neural network models. The integration of neural and fuzzy techniques, which can be considered as a full integration, is an approach of high importance.
Software tools for technical and financial applications have been developed at the Research Group on Intelligent Manufacturing and Business Processes (IMBP) at the SZTAKI. The general-purpose neural network simulator (NEURECA) is to be mentioned first, which provides an integrated framework including feature definition and real-time computation; automatic feature selection; various learning algorithms, classification, estimation of unknown patterns; standardized (DDE) interfaces to other programs, etc. NEURECA was written in C++ using its object-oriented nature enabling to dynamically vary the network structure during learning and to implement different hybrid AI models:
- a hierarchically connected hybrid AI system, HYBEXP where neural networks work on the lower, subsymbolic level coupled with an expert system (ES) on the higher, symbolic level
- the neuro-fuzzy version of NEURECA, which is a symbiotic-type of hybrid AI system, uses a multistrategy learning approach combining self-organized clustering for initializing the membership functions (MBFs), competitive learning for selecting the most important fuzzy rules, and back propagation (BP) learning for fine tuning the MBFs parameters
- genetic algorithms supported rule selection within the neuro-fuzzy version of NEURECA.
The applicability of the developed techniques has been demonstrated through case studies:
- ANN-based forecasting of stock prices
- bankruptcy prediction by ANN and neuro-fuzzy techniques
- stock analysis and trading with neuro-fuzzy techniques.
Promising results have been achieved illustrating the benefits of the hybrid AI approaches (eg comprehensive structures, some sort of explanation facility, higher convergence speed, etc.) over conventional methods or pure ANN or ES solutions.
Within SZTAKI, these and related activities in the field of financial engineering are coordinated by the inter-laboratory Research Group of Financial Mathematics and Management. Further impulses are expected from the ERCIM Working Group on Financial Mathematics. This project is partially supported by the Scientific Research Fund OTKA, Hungary, Grant No. T023650.
László Monostori - SZTAKI
Tel: +36 1 466 5644