Title: Bankruptcy classification of firms investigated by the US Securities and Exchange Commission: an evolutionary adaptive ensemble model approach

Authors: Sergio Davalos, Fei Leng, Ehsan H. Feroz, Zhiyan Cao

Addresses: Milgard School of Business, University of Washington Tacoma, 1900 Commerce St., Tacoma, WA, 98402, USA. ' Milgard School of Business, University of Washington Tacoma, 1900 Commerce St., Tacoma, WA, 98402, USA. ' Milgard School of Business, University of Washington Tacoma, 1900 Commerce St., Tacoma, WA, 98402, USA. ' Milgard School of Business, University of Washington Tacoma, 1900 Commerce St., Tacoma, WA, 98402, USA

Abstract: This paper develops an adaptive, rule-based model for bankruptcy classification of firms subject to the SEC|s Accounting and Auditing Enforcement Release (AAER). In this paper, we use an evolutionary computing method, genetic algorithm (GA), to generate an optimal set of if-then (comprehensible) rules for bankruptcy classification of AAER firms. In particular, we use bagging to improve the model|s generalisation accuracy; and to develop a doubly controlled fitness function to guide the operations of the (GA) method. Our research contributes to the bankruptcy literature in several ways. First, it fills a gap in bankruptcy classification by developing a domain specific model for AAER firms. Secondly, the derived set of if-then rules used in an expert system adds to the bankruptcy knowledge base. Thirdly, we use bagging to improve generalisation of bankruptcy classification models. Finally, we demonstrate the key role of the fitness function in successful model performance.

Keywords: genetic algorithms; GAs; evolutionary computing; fitness function; concept learning; bagging; cross-validation; accounting and auditing enforcement release; AAER; Securities and Exchange Commission; SEC; bankruptcy classification; ensemble modelling; multi-classifier; expert systems.

DOI: 10.1504/IJADS.2009.031180

International Journal of Applied Decision Sciences, 2009 Vol.2 No.4, pp.360 - 388

Published online: 24 Jan 2010 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article