The Application of Artificial intelligence on Finance and Investing

Since economy and economic situations usually have uncertain behaviors, it is hard to predict its trend with traditional approaches. As the uncertain behaviors change by the time, people need to solve nonlinear and time variant problems. Artificial intelligence was introduced to solve similar problems. According to Wikipedia, artificial intelligence is defined as intelligence displayed by machines. Nowadays, three famous artificial intelligence techniques have been mainly applied on real financial problems to achieve the goal of predicting economic situations, including artificial neural networks, expert systems and hybrid intelligence systems.

In artificial intelligence, an expert system (ES) is a computer system which has the similar decision-making ability of a real person expert. There are two sybsystems in one single expert system, the knowledge base and the inference engine. The knowledge base stores facts and if-then rules, while the inference engine applies those rules to the existing facts to generate the new facts, which is the conclusion.

In the area of investment and tax advice, ES works as an assistant for intensive works to give advice on investment, insurance, tax, assets, and so on. The work usually starts from data entry, followed by input of ideas by the planner (Humpert and Holley 78). Then the programme will run on the knowledge base and inference engine. During the work, the inference engine continuously examines the status of the knowledge base, and determines the order where inferences are made. the Finally, a preferred option will be generated as a report. Different from other mathematical models, ES can be summarized with following characteristics. First, ES is not only able to apply mathematical or analog schemes, but also can handle factual or heuristic knowledge. Second, the knowledge base can be continuous updated based on the prior knowledge and the input as the evidence. Third, the ES can also handle simply qualitative information. Fourth, ES is able to cope with uncertain, unreliable or even mission data (Bahrammirzaee 1165).

With these advantages, ES became popular on finance since the 1970s. in 1987, a review of ES on finance presented a variety of ES application in finance, investment, accounting, taxation, and administration since 1977. In 1988, a similar review was published about the application of ES on finance, including investment and tax advice, financial planning, risk assessment and banking practice. After that, a series of review of ES application on finance have been published (Wang et al. 144–145). In 2010, a recent review summarized the main application of ES in financial domain as follows.

The first application is credit evaluation. As credit is the basis of the conditions and the amount of a loan, a loan officer has to track the customer’s credit history carefully. This task is repetitive and unstructured. The advantage of ES in credit evaluation is its high speed and accuracy. For example, it is well known that American Express credit card application do not need to wait a long time for the results. The reason is that they utilize ES to process the requests. After using ES, their bad guess rate dropped from 15% to 4%, which reflects the high accuracy of ES (Bahrammirzaee 1165). During the 1980s, multiple ES systems were developed to manage banking loans. In 1986, a credit-evaluation ES with MuLISP was developed by the academy of economics in Wroclaw, Poland. In 1989, some French industrial companies started to use a knowledge-based decision support system (KB/DSS), FINISM, to conduct financial analysis and planning. In the 21st century, various ES system are continuously developed to meet different demands of loans. In 2001, ALEES was developed to evaluate agricultural loan incorporating with qualitative and quantitative assessment. In 2003, CEEES began to work on granting credit lines to applicant firms. By the comparison to existing methods in finance, developed ESs exhibited better performance on high efficiency and accuracy.

The second promising application of ES is financial prediction and planning. In this area, ES helps marketing executives to generate attractive finance plans for customers who are “interested in making large scale investments in products, services, and to back it up with convincing arguments that take into account the conflicting interests in business and finance” (Wang et al. 88–89). In this area, FAME system is a famous one for financial marketing. It runs on Lisp and provide financial marketing recommendations for mainframe computer business. FAME was written with several subdomains, including customer information gathering, capacity analysis, cash flow generation, financial analysis and explanation and advice. In this system, the costumer can question any part of the explanation only by pointing to it on the screen. Another developed ES is FINEVA, which is a multicriteria knowledge-based ES to assess firm performance and viability. The inference engine of this system utilizes both backward and forward chaining method. The output of this system can suggest the ranking of the analyzed firm based on class of risk.

In recent decades, artificial neural networks (ANN) emerged and found extensive acceptance in many disciplines for modeling complicated practical problems. ANNs was inspired from biological nervous systems and brain structure. As a computational modeling tools, artificial neural networks have been applied for a large number of economic situations. Currently, ANN represents powerful solutions for subjective information processing, decision-making, forecasting and other related problems. Especially in recent years, ANN become a popular tool for economy prediction.

Stock prices is highly-noisy because stock markets are affected by tons of factors. Predicting stock price with the high-noisy data directly will lead to large errors. In 2011, a Chinese team developed a Wavelet De-noising-based Back Propagation neural network (WDBP) to predict the stock prices. The simulation with Shanghai Composite Index from 1993 to 2009 demonstrated that the application of the WDBP neural network to stock price prediction is effective and accurate (Wang et al. 1016).