یک روش جدید ترکیبی دو مرحله ای ریسک اعتباری در صنعت بانکداری
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|18292||2009||9 صفحه PDF||سفارش دهید||7080 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Expert Systems with Applications, Volume 36, Issue 4, May 2009, Pages 8333–8341
The rationale under the analyses is to propose a new approach by three kinds of two-stage hybrid models of logistic regression-ANN, to explore if the two-stage hybrid model outperformed the traditional ones, and to construct a financial distress warning system for banking industry in Taiwan. The differences from the literatures are that this study adopts the “optimal cutoff point” approach proposed by Hosmer and Lemeshow [Hosmer, D. W., & Lemeshow, S. L. (2000). Applied logistic regression (2nd ed.). New York: A Wiley-Interscience], to determine the cutoff point for financial distress. Additionally, cross-validation [Efron, B., & Tibshirani, R. J. (1993). An introduction to the bootstrap. New York: Chapman and Hall; Stone, M. (1974). Cross-validation choice and assessment of statistical predictions. Journal of Royal Statistical Society. Series B, 36, 111–147] is used to evaluate the prediction power of the constructed models. The results find the factors of observable loans to total loans, allowance for doubtful accounts recovery rate, and interest-sensitive assets to liabilities ratio are significantly related to the financial distress of banks in Taiwan. In the prediction of financially distressed, two-stage hybrid model giving the best performance of 80.0% using cross-validation approach and demonstrates stronger prediction power than conventional logistic regression, logarithm logistic regression, and ANN approaches. It demonstrates that the two-stage hybrid model outperforms the conventional method, providing an alternative in handling credit risk modeling which have assessment implications for analysts, practitioners, and regulators.
The assessment of the financial distress warning system of banks is an important issue for supervising authorities and investors. Financial institutions play an important financial intermediation role in the economy; the development of the financial sector influences the level of the economy’s development. Credit risk1 induces financial distress on banks, and its assessment requires advanced modeling techniques that will link it to the sources of uncertainty generated. Accordingly, credit risk remains one of the major threats that financial institutions face, and it is essential to model the credit risk of financial institutions. The proposes a capital adequacy frame that allows banks to measure capital requirement for their banking books using internal assessments of key risk drivers. Thus, there is necessity for systems to assess credit risk for banks. In the moves to liberalize and internationalize the financial markets, Taiwan deregulated the banking industry in 1991 by allowing the establishment of new-privatized banks and encouraged the establishment of financial holding companies in 2002. In the face of increasingly intense competition, banks opt to shift business focus from business finance to consumer finance. However, aggressive and practically unchecked marketing of consumer lending has led to swelling of non-performing loan problems and has caused financial distress in some banks, which in turn creates economic and social turmoil. In the midst of a rapidly changing financial environment, a few poorly run banks fell into financial distress which created considerable adverse impact on the economy as a whole. The United States secondary mortgage turmoil caused by the global credit crisis is a best example. Therefore, banks face the necessity of managing financial distress, and its management has distinguished qualities due to higher exposure to credit risk by the role of intermediaries they play on the financial markets. The above phenomenon implies an appropriate financial distress warning system established which is most urgent of issues. Previous researches on credit risk modeling refers to traditional multivariate statistical and econometric techniques, for instance multivariate discriminant analysis, logit and probit models are among the most widespread for credit risk modeling (Altman et al., 1981, Altman and Saunders, 1998, Beaver, 1966, Gordy, 2000 and Lin and Chang, 2006), as well methodologies from the field of artificial intelligence and operational research (Dimitras et al., 1998 and Zopounidis and Doumpos, 2002). Over the past three decades, more sophisticated methodologies have been developed to support corporate financial distress assessment which included gambler ruin approach and option pricing model. The theoretical foundation about how a firm gets bankrupted when the liquidation value of its assets falls below its debt obligations. Such model was developed and modified by Merton, 1974 and Scott, 1981 and Wilcox (1972). Recent studies have considered new non-parametric methods such as mathematical programming, classification trees, neural networks (Abdou et al., 2008, Angelini et al., 2008, Yeh and Lien, 2008 and Yu et al., 2008) and support vector machines. As Baesens et al. (2003) concluded that credit risk data are weakly non-linear, thus suggesting that highly complex or non-linear models are not expected to have a considerable predicting ability compared to simpler models. Most of the existing studies on the development of credit risk models involve non-financial firms. However, financial institutions, such as banks, investment firms, security brokers, etc., are also apt to default and major interest due to their crucial role in the economic and business activity in financial markets. Therefore, developing credit-rating models for financial institutions, especially for banks, is of major interest to analysts, practitioners, and supervising authorities. Sahajwala and Van den Bergh (2000) review the current practices on the development and use of credit-rating models for banks. On the quantitative approach of financial distress warning system, Demirgüç-Kunt & Detragiache, 1998 use multivariate logit approach to establish an early warning model for banking crisis by sampling the data of developed and developing countries from 1980 to 1994. Their study finds that banking crisis tends to occur when the macroeconomic environment is weak; especially when the growth of GDP is low, inflation is high, and when it is coupled with increase in banking sector risk. Johnson, Boone, Breach, and Friedman (2000) posit that in the 1997 Asian financial crisis, the measure of corporate governance has better explanatory power than macroeconomic measures and in countries with weak corporate governance, financial distress problems are more likely to emerge. Claessens et al., 1999 and Claessens et al., 1999 and La Porta, Lopez-de-Silanes, and Shleifer (1999) also find that many listed firms use the schemes of pyramid structures and cross-holding to gain corporate control and the financial distress of a firm is to a certain extent related to the cross-holding between parent company and subsidiaries and poor management, and not entirely attributed to financial factors. Thus, to effectively reduce the likelihood of financial distress, the supervision mechanism of board of directors needs to come into play. The nonmonotonic view of Morck, Shleifer, and Vishny (1988) proposes that based on entrenchment hypothesis and convergence-of-interests hypothesis, low ownership by the board of directors limits the board’s ability and willingness to supervise the management and might be positively related to the likelihood of financial distress. As shown by the discussions above, corporate governance plays a vital role in the financial soundness of a firm. Therefore, this study also includes the variable of ownership by directors and supervisors to discern whether higher shareholding by directors and supervisors would increase or decrease the likelihood of financial distress. Lee, Chiu, and Lu (2002) and Lee and Chen (2005) explore the performance of credit scoring using a two-stage hybrid modeling procedure with artificial neural networks and multivariate adaptive regression splines (MARS). The proposed hybrid approach outperforms the results using discriminant analysis, logistic regression, artificial neural networks and MARS and hence provides an alternative in handling credit scoring tasks. Consequently, this study proposes three two-stage hybrid models of logistic regression-ANN, compare with logistic regression, Altman’s (2006) logarithm logistic regression, artificial neutral network (ANN), aims to explore if the two-stage hybrid model outperformed the conventional ones, to construct a financial distress warning system suitable for Taiwan’s banking industry, and to provide an optimal model of credit risk for supervising authorities, analysts and practitioners in conducting risk assessment and decision making. The differences from the literatures are that this study adopts the “optimal cutoff point” approach proposed by Hosmer and Lemeshow (2000), to determine the cutoff point for financial distress. In addition, cross-validation (Efron and Tibshirani, 1993 and Stone, 1974) is used to evaluate the prediction power of the constructed models. The objective of this study is to construct a financial distress warning model of banks in Taiwan, in distinguishing financially sound and financially distressed banks by the criteria of capital adequacy ratio below 8%, loss of capital by one third or more, current ratio below 7%, non-performing loan ratio over 5%, or delisted declared during the 2002–2004 period. In order to improve investor awareness of credit risk, prior to establishing the financial distress early warning system, detection advance firms may change the deal to reduce the preventable loss for stakeholders. The remainder of this paper is structured as follows. Section 2 discusses the model and research design issues. Section 3 presents and analyzes the data description and results while the final section concludes with a summary of results.
نتیجه گیری انگلیسی
This study proposes three two-stage hybrid models of logistic regression-ANN, compare with logistic regression, Altman’s (2006) logarithm logistic regression, artificial neutral network (ANN), aims to explore if the two-stage hybrid model outperformed the conventional ones, to construct a financial distress warning system suitable for Taiwan’s banking industry. The differences from the literatures are that this study adopts the “optimal cutoff point” approach proposed by Hosmer and Lemeshow (2000), to determine the cutoff point for financial distress. Additionally, cross-validation (Efron and Tibshirani, 1993 and Stone, 1974) is used to evaluate the prediction power of the constructed models. The results of logistic regression show that that factors of observable loans to total loans (X6), allowance for doubtful accounts recovery rate (X7), which are measures of asset quality for the bankruptcy of many banks is materially related to the quality of their loan assets, and interest-sensitive assets to liabilities ratio (X21) are significantly related to the financial distress status of banks in Taiwan. Further logarithm logistic regression identifies allowance for doubtful accounts recovery rate (X7) and interest-sensitive assets to liabilities ratio (X21), which is associated with the earning assets and interest-bearing liabilities of a bank as its income and costs are subject to the fluctuation of interest rates, as significant variables in the financial distress situation of banks. The significant variables identified above warrant the avid attention of banking regulator and may serve as early warning signals of financial distress for banks. The logistic regression model achieves a prediction accuracy rate of 92.4% in the financially sound group, 46.7% in the financially distressed group, and 75.6% for all samples, in contrast to the respective accuracy rate of 84.4%, 36.7%, and 71.2% obtained using the Altman (2006) methodology. It is clear that the Altman (2006) methodology does not necessarily improve the prediction rate. The ANN model achieves a prediction accuracy of 55.58% in financially distressed group, 86.12% in the financially sound group, and 75.34% for all samples. In comparing the performances of one-stage models, it is found ANN is superior to logistic regression in the prediction of financial distress. Our finding that ANN performs better than logistic regression in the prediction of financial distress is consistent with that of previous works (Tam & Kiang, 1992). For the two-stage model, this study used significant variables identified by logistic regression and then used different criteria to determine the accuracy of prediction so as to develop three different models. Results show that model (1) achieves a prediction accuracy of 77.4% in financially the sound group, 78.0% in the financially distressed group, and 77.0% for all samples; model (2) achieves a prediction accuracy of 80.7% in financially sound group, 80% in financially distressed group, and 80.8% for all samples; model (3) achieves a prediction accuracy of 80.7% in financially sound group, 73.3% in financially distressed group, and 79.8% for all samples. It is thus concluded that model (2) and model (3) have the highest accuracy in predicting financial soundness of banks, while model (2) has the highest accuracy in predicting financial distress of banks as well as in overall prediction. Overall, the empirical results show that in the prediction of financially sound banks, two-stage hybrid model demonstrates stronger prediction power than the conventional ones in the prediction of financially distressed banks, with two-stage model (2) giving the best performance of 80%. In the prediction of full samples, two-stage model (2) achieves the highest rate of 80.8%. As the results reveal, the proposed two-stage hybrid model outperforms the results using logistic regression, logarithm logistic regression and artificial neural networks. As a result, it demonstrates that two-stage hybrid model provides an alternative in handling financial distress warning tasks and better prediction power.