Brought to you by EarthWeb
IT Library Logo

Click Here!
Click Here!

Search the site:
 
EXPERT SEARCH -----
Programming Languages
Databases
Security
Web Services
Network Services
Middleware
Components
Operating Systems
User Interfaces
Groupware & Collaboration
Content Management
Productivity Applications
Hardware
Fun & Games

EarthWeb Direct EarthWeb Direct Fatbrain Auctions Support Source Answers

EarthWeb sites
Crossnodes
Datamation
Developer.com
DICE
EarthWeb.com
EarthWeb Direct
ERP Hub
Gamelan
GoCertify.com
HTMLGoodies
Intranet Journal
IT Knowledge
IT Library
JavaGoodies
JARS
JavaScripts.com
open source IT
RoadCoders
Y2K Info

Previous Table of Contents Next


4.2. COGNITIVE ISSUES

One view of developing intelligent systems is to obtain a system that does a good job of solving problems regardless of whether the system mimics human intelligence or not. However, understanding the ability of people to solve problems and perform complex judgement tasks will still be beneficial to designing systems because much can be learned from how people solve problems and perform judgment tasks. An important issue for developing systems is understanding the characteristics of different tasks, characteristics of human expertise, and appropriately matching different types of intelligent systems that have the knowledge acquisition, knowledge representation, and judgment processing characteristics best suited for the task. For example, a general theory for matching the appropriate intelligent system type with characteristics of semistructured judgment tasks, such as financial prediction and classification, can be developed and refined. A related issue concerns how to integrate different types of systems from a system design perspective.

4.3. USER ACCEPTANCE

Development of a dependable and accurate system does not amount to very much if the system is not seen as a useful tool by the intended users. Ease of use and ability of the system to provide explanation of its prediction or "decision" are key ingredients for successful deployment of an intelligent system.

5. FUTURE TRENDS

Financial judgment tasks may exhibit both analytical reasoning and pattern recognition characteristics. Judgment tasks can be viewed as a continuum between pattern recognition tasks and analytical reasoning tasks. Or an overall task may be broken down into pattern recognition and analytical reasoning components. For example, one discrete, definitive rule may be that a buy order is triggered when a certain event in the market occurs. Although such definitive rules are often used, financial judgments require recognition of patterns in the whole set of interdependent financial data. The key to automating analytical reasoning tasks versus pattern recognition tasks is choosing the intelligent system tool with characteristics suited to performing such tasks.

Artificial neural networks (ANNs) are essentially statistical devices with the capability to inductively (from experience with the data) infer complex, nonlinear relationships or patterns within the provided data. This capability provides a strength for ANNs to perform prediction and classification using the complex, nonlinear, interdependent relationships inherent in financial data. Using a neural network for financial prediction and classification still requires substantial expertise in both the ANN development and the financial domain. Choice of variables, data significance, correlation, normalization, etc. are part of preprocessing the data prior to ANN development. During ANN development and validation, experimentation with network parameters is usually extensive and time consuming. Finally, the inductive inference capabilities come with a drawback -- explicit explanation of how the ANN made the judgement is very difficult to obtain from the complex mathematical equations inductively derived from the data (Refenes, 1995). However, ANNs have shown some promise in recognizing trends in the chaotic data of the financial world such that some additional profits may be made due to the use of this technology.

Medsker and Turban (1994) state that integrating expert systems and neural computing has the potential to provide solutions that neither system alone can deliver, or to lead to good solutions with less system complexity. Each technology has deficiencies that can be overcome by the other. For example, an ANN cannot provide explanation, while an expert system can. Expert systems cannot recognize patterns in historical data, whereas ANNs can. The different characteristics of the technologies suggest that they can enhance each other. Typically, a loan assessment system, such as COMPASS, could include an ANN module that would take care of the prediction part and an expert system module that could take care of the rest. By integrating the technologies, we pursue a synergism based on their respective strengths.

ANN can be used in conjunction with a KBS in a hybrid system where, for example, certain data used as input to the KBS are preclassified by a front-end ANN -- for example, for predicting bankruptcies as part of a loan assessment hybrid system.

Brown et al. (1995) state that the financial services industry with its large databases has already fielded several successful neural network applications, and that neural networks based on information about customers or potential customers have proved effective. In fraud detection, they are integrated effectively into expert systems. If large databases exist with which to train a neural network, then use of that technology should be considered. For a neural network, the large database can be used as the equivalent of the human expert. Each type of reasoning can be matched to tasks that require its specific reasoning strengths.

A major barrier to identifying where applications are being used in finance, particularly for capital markets work, is the proprietary nature of the models developed. Due to the potential profitability of any ANN model that does well as a tool for making a profit in the financial markets, many ANN models are kept secret, particularly concerning the details of what variables are used and what parameters are used in the ANN. Occasionally, an article in application journals and newspapers tells of a company using ANNs for financial market transactions, such as Fidelity Investments (see the Wall Street Journal, Oct. 27, 1992) and LBS Capital Management, (see Futures, August 1992, p. 34).

There is still much research to be done into gaining an insight into how to approach different finance problems with intelligent systems. Refenes (1995) and Trippi and Turban (1996) each provide a compilation of papers using neural networks applied to equity applications; including picking stocks, forecasting trading indices, predicting options volatility, pricing, and hedging derivative securities. Debt applications, such as mortgage risk assessment, bond rating, and credit scoring, are also included. Other finance applications presented include commodities trading, exchange rate prediction (see also Episcopos and Davis, 1996), and business failure prediction. These paper collations provide insight into what type of ANN and what variables were used, and what other established types of models the ANN results were compared to, such as linear regression, multiple discriminant, autoregressive (time series) models, etc.

The future will bring more integration of different types of intelligent systems to complement the strengths and mitigate the weaknesses of different intelligent systems. Particularly, the lack of explanation ability of ANNs will be overcome by using other intelligent systems to interpret the ANN. Future research should also include much more empirical documentation concerning the acceptance of intelligent systems as a viable decision aid tool for finance and other professional users.


Previous Table of Contents Next

footer nav
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.