1. Introduction.

2. Computer-based Decision Support Systems for Investment.

3. Are Expert Systems an Appropriate Technology?

4. Current Use of Expert Systems.

5. Examples of Successful Applications.

5.1 Stock Trading: RIEM.

5.2 Real Estate: RESRA.

5.3 Personal Financial Investing.

6. Quo Vadis?

7. References.

This paper looks at the application of Expert Systems (ES) in financial investment decision making. Firstly ES are positioned within the other computer-based decision-aiding technologies. Then a closer investigation is made as to whether ES are in fact a suitable and appropriate technology in the area of investment. Thirdly, a survey is made to what extent ES are already implemented; although this is restricted to US financial institutions. Some successful applications are looked at in more detail and the paper closes with some thoughts on expected future developments.

Financial institutions and practitioners have traditionally been at the forefront of implementing computer-based Decision Support Systems (DSS) for various reasons. These include the predominant use of quantitative data (often already in electronic format), strategic issues of competitive advantage, the relative sophistication and computer-literacy of users, the perception that the decision processes are very analytical and the fact that even marginal enhancements of the decision making can yield substantial monetary rewards.

As a result, there have been an abundance of system development in this area and it would be beyond the scope of this paper to even start enumerating the different types of applications that have been developed. For this purpose it will be sufficient to position ES in terms of the state-of-the art Artificial Intelligence (AI) techniques. The many conventional on-line time-sharing services and the plethora of off-the-shelf software packages, often with esoteric names [Trippi, 1990] that are now available usually employ "traditional linear programming techniques", although some systems rate amongst the most complex on-line real-time multi-tasking systems in use in a commercial environment. An illustrative example can be found in Leinweber [1988] who gives a comprehensive account of the historical evolution of securities trading systems from non-integrated "data pictures" via video and digital composites to high-level integrated and synthesized knowledge-based applications with real-time decision support.

At the other end of the spectrum, cutting edge investment DSS would utilize neural networks [Smith, 1990], employ natural language interfaces and the like. From this perspective, it would seem that ES for investment applications are currently beyond the experimental stage and slowly finding acceptance as a "tried and tested" technology; although this is by no means a majority view as evidenced by the survey undertaken by Connell and Powell [1990].

An immediate consequence will probably be that ES will slowly cease to be regarded as AI but may become another standard, "conventional" systems development technique. The ready availability of ES shells and development tools for both mainframe and personal computers may be a sign that this trend is already taking place.

Not all decision making processes can be readily complemented, improved or replaced by expert systems. The potential of initial ES applications was typically assessed on a subjective basis, where the main criterion tended to be the symbolic, complex and systematic nature of the task environment. The decision making process could then more or less easily be extracted by knowledge engineers as a basis for the ES; a good example is given by Bouwman [1987]. Trial and error would typically tell whether or not the problem would yield to more traditional computing methods but often the initial promises made by the advocates of ES were not vindicated by their final results.

It is now clear that, although the nature of the task is still the most critical, it is by no means the only consideration that must be used when assessing the value of applying a ES approach. A very useful set of criteria has been developed by Beckman [1991], who has actually developed a commercial AI software for stock market technical analysis and residential real estate appraisal.

Beckman suggests a modular point-system (100 points) based on 64 characteristics (with unequal weightings) grouped into six major areas. Evaluation processes developed by other authors have been integrated and supplemented, based on academic research and practical experience. It was found that other evaluation methods and checklists focus mainly on the technical aspects but are quite deficient in practical implementation aspects which often determine the success of a project: management support, benefit/cost ratios, user concerns and system designer competence.

The two crucial areas for success remain the task characteristics (sub-divided into 19 specific criteria) and the potential pay-off (13 criteria); which are given total weights of 25 and 20 percent respectively. However, other categories that must be assessed for potential difficulties include customer-management characteristics (11 criteria; 20% weight), system-designer (9; 15%), domain-expert (7; 10%) and user characteristics (5; 10%). The dangers associated with a blind adoption of the checklist evaluation approach are also pointed out by Beckman.

Apart from the above system characteristics, it must be realized that the adoption of any new technology may also bring with it a set of other unexpected potential problems which might be best characterized as "soft" problems. One relevant concern to investment-related ES would be the question of legal liability for recommendations made by the system, especially if systems are developed for more than one user [Mykytyn, 1990]. Another important issue will be the confidentiality and protection of the intellectual assets incorporated in the ES in the form of the knowledge database and rules. Finally, many of the ethical problems associated with information systems in general will become even more pronounced with ES [Mitroff, 1988].

Initially, most investment-related ES seem to concentrate on one specific type of investment only. The first ES were developed to perform technical share analysis because of the ideal task environment. Subsequently, other ES have focused on other types of investments: bonds, derivative financial instruments (options and futures) and real estate. More recently, a number of "personal investment" ES were developed which advise on which type of investment to choose (i.e. high growth shares, residential property or near-cash fixed interest securities). Examples of these will be given later.

As to the actual use of ES, several surveys have found that ES applications in investments are already quite widespread in U.S. financial institutions. A survey by Coopers & Lybrand in 1987 reports that 50% of the largest security firms in the U.S. were researching, developing or using expert systems; and "one third of the firms that [had] not yet started to develop an expert system, expected to do so by 1990" [O'Leary, 1988]. Most of these ES were specifically related to stock market investment i.e. technical share analysis.

Likewise, the annotated bibliography of the use of ES in accounting and related fields [Brown, 1989] suggests that a substantial number of investment ES are already in actual use.

What has been described as "the ultimate expert system in finance" may well be the one in use by the Rosenberg Institutional Equity Management company in California (U.S.A.): RIEM. The input to the real-time trading system is formed by three "live feeds", which include electronic on-line services providing price, earnings and dividends data. All investment decisions are taken by the system and, according to the C.E.O., never overridden by humans.

Rules are based on common sense, modern portfolio theory and most likely "proprietary theories" on risk, reward and portfolio optimization. The equity portfolio amounted to US$ 3 billion and almost doubled the performance of the S&P 500 (although no indication was given as to the relative risk profile of the portfolio). [Marion, 1988].

RESRA is a stand-alone ES developed by Security Pacific and implemented using the PROLOG programming language on HP Vectra workstations. RESRA is structured around a standard property appraisal form (URAR) which consists of nine key evaluation areas (such as site, neighbourhood or improvements).

The system interface makes use of the touch-screen capabilities of the HP Vectra workstation and performs the tedious cross-checking and reconciliation of the different sections of the report. The expert rules are derived from fairly standard heuristic procedures, complemented by various the in-house criteria and methodologies.

RESRA allowed Security Pacific to embark on a high-growth expansion path based on growth by acquisition. The ES ensured that procedures throughout its subsidiaries remained standard and could easily be updated centrally. [Eliot, 1988]

The "Personal Financial Investment Decision Support System" is an advanced personal investment advisor programmed in PROLOG for the IBM compatible AT designed for usage by fairly sophisticated end-users. Its complex design features two expert systems and employs gaming techniques to measure the degree of risk aversion of the investor.

Input includes long and short term forecasts for 13 general economic indicators, individual goals and personal attributes and circumstances (such as cash flow situation, personal net worth or tax bracket situation). These are used as input by the first ES's engine to generate rules for the second knowledge base which selects suitable investment categories.

The final output includes recommend portfolio mixes which best match the personal goals of the investor. It must be noted that only 6 broad types of investments are recognized by the ES, namely Treasury Bills, Money Market Accounts, AA Bonds, leveraged real estate, income and growth stocks. No mention was made of user or performance feedback. [Shane, 1987]

It is clear that certain standards are already developing with regards to ES nomenclature and methodology. The technology is to a certain extent maturing (as evidenced in the number of development tools and off-the-shelf system shells that are already available) and therefore becoming less and less associated with AI. ES therefore tend to become more transparent and demystified tools and less of a "sui generis".

Secondly, and partly as a result of the above trend, ES will become less of a "stand-alone" stand-alone technology but more and more integrated with different technologies. Prime examples are direct communications support using public access databases; natural language interfaces; and more hybrid systems incorporating perhaps advanced statistical components, additional heuristics, fuzzy logics and neural network technology. Investment analysis is also a prime candidate for implementation of parallel computing and voice I/O. The competitiveness of the financial markets encourage experimentation and will assure at least the development of fairly sophisticated prototypes. This necessitates a well-organized and more integrated approach from the perspective of the main-stream IS management.

Another area of development will be the "soft problems" of ES mentioned higher. Specifically the questions of legal liability, proprietary intellectual property (coded in the rules and knowledge base) and ethics will have to be resolved before they become stumbling blocks for future developments.

Finally, investment ES may move out of the back room or trading offices onto the streets as a result of "off-the-shelf" packages which will run on tomorrow's personal computers and the marketing or strategic positioning efforts of various financial institutions. Initially these will be watered-down versions of the personal investment ES, but gradually more powerful ES will become available. One major macro-economic benefit may well be the increase in the investment market efficiency, particularly for the property and real asset markets.

Jean-Paul Van Belle,

Fish Hoek, 25 March 1991.

Beckman, T.J. (1991) "Selecting Expert System Applications". AI Expert, Vol.6 No.2, pp.42-48.

Bouwman, M.J.; Frishkoff, P.A. and Frishkoff, P. (1987) "How Do Financial Analysts Make Decisions? A Process Model of the Investment Screening Decision". Accounting, Organizations and Society, Vol.12 No.1, pp.1-29.

Brown, C.E. (1989) "Accounting Expert Systems: An Annotated Bibliography". Expert Systems Review, Spring-Summer 1989, pp. 25-129.

Connell, N.A.D. and Powell, P.L. (1990) "A Comparison of Potential Applications of Expert Systems and Decision Support Systems". Journal of the Operational Research Society, Vol.41 No.5, pp.431-439.

Eliot, L.B. (1988) " Security Pacific's RESRA: A Case Study". AI Expert, Vol.3 No.4, pp.48-59.

Leinweber, D. (1988) "Knowledge-Based Systems for Financial Applications". IEEE Expert, Fall 1988, pp.18-30.

Marion, L. (1988) "A Tale of Two Systems: An Example of a Portfolio Management System". Institutional Investor, July 1988, pp. 84-86.

Mitroff, I.I. and Mason, R.O. (1988) "Deep Ethical and Epistemological Issues in the Design of Information Systems". Expert Systems Review, Vol.I No.3, pp.21-25.

Mykytyn, K.; Mykytyn, P.P. and Slinkman, C.W. (1990) "Expert Systems: A Question of Liability?". MIS Quarterly, March 1990, pp.27-36.

O'Leary, D.E. and Watkins, P.R. (1988) "Prior Surveys on Expert Systems in Accounting, Auditing and Related Areas". Expert Systems Review, Vol.I No.4, pp.11-12.

Shane, B.; Fry, M. and Toro R. (1987) "Design of an Investment Portfolio Selection System Using Two Expert Systems and a Consulting System". Proceedings of the Twentieth Annual Hawaii International Conference on Systems Sciences.

Smith, J.C. (1990) "A Neural Network - Could It Work For You?". The Financial Executive, May/June 1990, pp.26-30.

Trippi, R.R. (1990) "Decision Support and Expert Systems for Real Estate Investment Decisions: A Review". Interfaces, 20:5, pp.50-60.

 © Jean-Paul Van Belle University of Cape Town Date: 25 March, 1991.