John J. Cheh

A roadmap for future neural networks research in auditing and risk assessment

Abstract This paper focuses on the use of neural networks (NNs) as an enabler of the new business risk auditing framework and provides insight into future research opportunities. In general, NNs, which are classifiers by nature, offer the capacity to simultaneously consider multiple types of evidence and can assist auditors in assessing risks and making judgments [Int. J. Intell. Syst. Account. Finance Manag. 2 (1993) 19; Decis. Sci. 26 (1995) 209; Eur. J. Oper. Res. 103 (1997) 350; Audit. J. Pract. Theory (1997) 14; Int. J. Intell. Syst. Account. Finance Manag. 7 (1998) 21; The use of neural networks as an audit tool in fraud risk assessment. Proceedings of American Accounting Association Northeast Regional Meeting, Rochester (NY), 1999; Int. J. Intell. Syst. Account. Finance Manag. 8 (1999) 159]. Furthermore, NNs may be superior to other approaches in circumstances where data are available in relatively large samples, the range of values to be analyzed for each case is large, and the underlying associations among the data are fuzzy and ill-defined [Decis. Support Syst. 11 (1994) 497.]. The paper reviews several published studies, which are grouped into six categories—preliminary information risk assessment (1 study), control risk assessment (2 studies), errors and fraud (6 studies), going-concern audit opinion (3 studies), financial distress (3 studies), and bankruptcy (12 studies). The paper includes a brief introduction to NNs, followed by a description and analysis of the methods employed by and findings of researchers who used NNs as a tool for research in the auditing and risk assessment domain. The literature review leads to discussion of several broad foci areas that need further exploration in order to gain a better understanding of the efficacy of NNs as an enabler of business risk-based auditing. A discussion of the general limitations of NNs as an auditing and risk assessment tool and an outline of implications for future research opportunities conclude the paper.

Modeling an intelligent continuous authentication system to protect financial information resources

Abstract This paper examines fundamental principles of continuous authentication (CA) and proposes a four-tier CA architecture to secure financial information systems. We define CA as a process that verifies the identity of an information systems user continuously for the entire duration of an authorized session. While organizations can, in theory, strengthen the security of their financial information systems through CA, several challenges need to be addressed in designing a CA architecture. A primary challenge involves the constantly changing user profiles in globally networked business environments. Profile content may include user knowledge and characteristics, access location, job characteristics, and transaction attributes. We propose swarm intelligence, which has the capacity to handle complex profile changes, as a technology for implementing CA in a dynamic, distributed network environment where user profiles are constantly changing. The paper explores model implementation challenges and discusses opportunities for future research.

Investing in Growth Stocks vs. Value Stocks: Does Trading Frequency Matter?

This article examines whether the performance in the investment in value stocks vs. growth stocks is different in terms of the holding period and stock market conditions. The study analyzes the performance of high- and low-P/E portfolios for the quarterly and monthly rebalancing as well as the annual rebalancing, using a sample period of April 1986 to March 2003, divided into a general bull market period of April 1986 to March 2000 and a recent bear market period of April 2000 to March 2003. The findings show that regardless of the frequency of portfolio rebalancing, high-P/E portfolios outperform low-P/E portfolios especially for the bull market period. However, when adjusted for the risk inherent in different P/E portfolios, low-P/E portfolios outperform high-P/E portfolios for all rebalancing frequencies. Finally, the evidence shows that more frequent rebalancing tends to improve the performance of low-P/E portfolios but reduces the performance of high-P/E portfolios.

Multiple criteria linear programming data mining approach: an application for bankruptcy prediction

Data mining is widely used in today’s dynamic business environment as a manager’s decision making tool, however, not many applications have been used in accounting areas where accountants deal with large amounts of operational as well as financial data. The purpose of this research is to propose a multiple criteria linear programming (MCLP) approach to data mining for bankruptcy prediction. A multiple criteria linear programming data mining approach has recently been applied to credit card portfolio management. This approach has proven to be robust and powerful even for a large sample size using a huge financial database. The results of the MCLP approach in a bankruptcy prediction study are promising as this approach performs better than traditional multiple discriminant analysis or logit analysis using financial data. Similar approaches can be applied to other accounting areas such as fraud detection, detection of tax evasion, and an audit-planning tool for financially distressed firms.

A heuristic approach to efficient production of detector sets for an artificial immune algorithm-based bankruptcy prediction system

Bankruptcy prediction has been extensively studied. These studies provide a rich library of important variables to be considered in predicting whether a particular company faces bankruptcy. Furthermore, systems designers can utilize the findings of these studies as a reservoir of knowledge that complements the knowledge accumulated from the advancement of computer immunology in designing and developing a bankruptcy prediction system. In this paper, the author proposes a heuristic approach to efficient production of detector sets for an artificial immune algorithm (ARIA) that takes advantages of the knowledge derived from bankruptcy prediction literature, and explores the issues related to time and space complexities of different artificial immune algorithms. Furthermore, he provides a preliminary evidence on the time complexity associated with the new approach to detector set production and designing an ARIA-based bankruptcy prediction system.

An Application of Self-Organizing Maps to Financial Structure Analysis of Keiretsu versus Non-Keiretsu Firms in Japan

It has been argued that keiretsu in Japan allows its member firms to maintain a financial structure different from that of non-keiretsu member firms. In this paper, we use two different types of financial statement ratio analysis techniques to discover whether Kohonens self-organizing map (SOM) is able to uncover the differences in financial structures between keiretsu and non-keiretsu firms: ad hoc financial ratios and valuation-based financial ratios. We have found some evidence that SOM enables both financial analysis techniques to recognize different financial structures between the two groups of the firms. Implications of this finding for investment decisions have been discussed.