Using a Neurofuzzy Expert System to Address Ambiguity Problems in Debt/equity Issues of Closely Held Corporations

Article excerpt

ABSTRACT

With a new round of tax laws being introduced almost every year, complexity has increased for the tax professional. In addition, the tax professional is also facing increased competition, constantly advancing technology and a more litigious environment. Tax planning is becoming more complex and the aid that was promised from the development of expert systems has not materialized. The purpose of this paper is to illustrate how a tax planner might develop decision rules for determining the capital structure of a closely held corporation by the use of fuzzy logic and neural networks. If these decision rules are based upon the judgment of "experts" as they relate to the determination of whether an interest deduction by a closely-held corporation is reasonable in specific situations, less experienced tax planners can benefit from the know-how of the these experts.

INTRODUCTION

Accounting firms have developed several applications of expert systems in accounting and tax. Specific applications include ExperTAX developed by Coopers and Lybrand, FSA developed by Arthur Andersen and ASQ developed by Arthur Young. However expert systems are not being developed and used in accounting as extensively as they were earlier projected. Many problems have slowed their growth. One such problem was the need to use precise variables generally combined with a linear model. It was soon recognized that experts simply don't use precise variables and linear models. Another problem was that for an expert system to be refined with all of its variables and decision rules, the system had to be known and this was not always true. Two current developments are currently being integrated in the area of expert systems that may help solve these problems. Fuzzy logic is replacing Boolean logic in many new systems. And neural networks are being used to help learn complex systems that are not fully understood.

Fuzzy logic was developed to aid computers in solving problems in a manner more characteristic of human expert problem solving. Zadeh (1965) introduced the fuzzy set theory as a branch of classical set theory. It allows for systems with human interaction to tolerate vagueness and ambiguity that is natural in personal judgments. This vagueness and ambiguity should not be confused with the uncertainty that is created by randomness, which is best dealt with using probability. Rather vagueness and ambiguity deal with the imprecise nature of human language.

Neural networks are an attempt to mimic the way the human brain learns. Neural nets use a number of simple computational neurons that attempt to behave as a human brain cell would behave. Each neuron receives inputs and processes outputs to other neurons until an output signal is generated. The computer system can perform numerous iterations at high speed that should allow the learning process to be more efficient and possibly give us insights that may have taken human experts years to discover.

The purpose of this paper is to illustrate how fuzzy logic and neural networks can be used to develop an expert system to aid tax planners in assisting their clients in ambiguous tax planning situations such as determining debt/equity issues of a closely held corporation. If these decision rules are based upon the judgment of "experts" as they relate to the determination of what level of debt is reasonable in specific situations, less experienced tax planners can benefit from the know-how of the these experts. These experts could include the judges of tax decisions if we are trying to predict the outcome of a potential case. Internal Revenue Agents would be used if we were trying to predict the possibility of an audit. Experienced tax planners could be used if we trying to evaluate the best method of structuring a complex tax transaction. The use of fuzzy logic allows for any ambiguities that may occur among the experts in their determinations of the variable that go into the decision and the output. …