![]() |
|||
![]()
|
![]() |
![]() |
![]() |
3.1.4. System Validation The system was validated by testing the relationship between the decisions developed by the system and the decisions developed by human experts. Recommended prices were compared with the prices posted by the expert specialist on various days when the system was in use at the exchange. The designers logged all user input for a sample of different days over a 1-month period. AESOP presents the recommended quotations of the symbolic model along with the theoretical prices generated by the mathematical model. The user can override any recommendations, ask for an explanation or trace of the symbolic model, and/or change parameters and rerun the entire system. The results indicated that AESOP tended to recommend a call price that was too high and a put price that was too low. However, it was clear that AESOP represented an improvement over the regular Black-Scholes model for this application. 3.1.5. System Performance The user interacts with the system through the menu-driven user interface, which is managed by AESOP's control module and activated by function keys. AESOP provides the following functions:
3.1.6. Summary The AESOP experiment provides evidence that the integration of symbolic and mathematical models offer great promise for matching models to specific problem domains. AESOP was used on an experimental basis successfully for 2 months by a specialist on the exchange market. Reference: Condensed from Clifford, J., Lucas, H.C. Jr., and Srikanth, R.: Integrating mathematical and symbolic models through AESOP: an expert for stock options pricing. Information Systems Research, Vol. 3 No. 4, Dec. 1992, pp. 359-378. 3.2. COMMERCIAL LOAN ANALYSISLoan approval systems have become very popular expert system applications. Typically these are cumulative hurdle decision-making systems. This is where a number of decisions are made linearly, but the problem may be solvable without overcoming every decision hurdle. An expert system can handle the first hurdle -- logical consistency of the loan application, basic credit worthiness, etc. When a reject decision or perhaps an obvious grant decision cannot be made, the application can be passed over to an expert for further review. This separation of routine and more complex decision-making can be very beneficial when the number of routine problems is a large proportion of the volume and the cost of the complex decision making is large (O'Keefe and Preece, 1996). 3.2.1. D&B Expert System The Dun & Bradstreet (D&B) credit clearing house provides risk analysis to manufacturers, wholesalers, jobbers, and marketers in the apparel industry. It maintains and updates a database of credit rating on approximately 200,000 businesses in the U.S., which it then uses in order to recommend specific credit amounts to its clients. Difficulties in the updating of data and inconsistencies in recommendations led to the development of their expert system for credit analysis. The system can handle more than 90% of all requests and it was developed using ART-IM. It consists of a knowledge base with more than 1000 rules and accompanying databases. D&B has benefited from the system being able to reduce the response times from 3 days to less than a minute and a high level of consistency in the recommendations generated. References: This condensed version is from Turban, E., McLean, E., and Wetherbe, J. (1996) It is based on Neuquist, H. P. III, No summer returns, AI Expert, October 1990. 3.2.2. CREDEX CREDEX helps in the assessment of the risk involved in granting a loan. Its purpose is to help the analysts responsible for bank loans to judge the quality of a company seeking funds. It makes a match between the level of risk and the loan request, together with an explanation. It uses economic, financial, and social data on the company and its sector of activity, and also on the bank's lending policy. It provides a diagnosis of each of the firm's functions in terms of weaknesses and strengths, and it states the degree of partial risk inherent in each function. Moreover, it states the degree of overall risk associated with the granting of the loan and provides suggestions regarding loan acceptance. 3.2.2.1. Model The CREDEX system consists of a meta-expert, experiential experts, and judgmental experts. The meta-expert distributes control among the different experts and manages the problem-solving process. It determines which expert should be activated. The experiential experts are activated in order of importance. When one of the experiential experts has finished its task, the meta-model selects one of the judgmental experts to handle the task in question. The experiential experts consist of a commercial expert, a financial expert, a production and operations expert, and a management expert. These experts deduce the importance of their respective subdomain characteristics and make an initial assessment in terms of strong and weak points and elementary associated risks. The commercial expert analyzes the quality of the commercial function. The financial expert assesses financial and accounting data. The production and operations expert handles production, research, and development data, and the management expert assesses the managing team and capital structure. The judgmental experts consists of a compensatory expert, a lexicographic expert, a disjunctive expert, and a conjunctive expert. They contain general cognitive models of information processing and carry out deeper analysis. They examine the elementary risks and their importance, and combine them into a partial risk through a set of general lexiographical, compensatory, or conjunctive decision rules.
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |