Hybrid models based on rough set classifiers for setting credit rating decision rules in the global banking industry
Introduction
The world financial crisis during 2007–2009 was due to reckless and unsustainable lending practices under deregulation and securitization of United States (US) mortgages, which were marketed as investments to global individual investors and financial institutions. The risks of a broad-based credit boom arose from a near-global speculative bubble in real estate and equities, eventually exposing other risky loans and inflated asset values, and instigating a global recession. Clearly, the emergence of subprime loan losses from 2007 to present rapidly reduced economic activity and saw numerous financial institutions and firms facing extreme financial difficulties, distress, or even bankruptcy. Despite a significant easing of national fiscal and monetary policy in US in an effort to stem the global recession, the world has failed to shake the financial crisis. Global investors continue to face difficult challenges, particularly in relation to banks in Europe, particularly in Greece, Italy, Portugal, and Spain, which currently have the same debt problems [29]. Banks typically face numerous risks, including credit, debt, interest rate, currency, liquidity, and systematic risk. Serial bank runs and other blows to financial sector confidence inevitably result from extreme events, such as the current financial crisis, which has seriously jeopardized regional and national economic stability [48]. Banks are clearly important to national, and even global, economic stability. However, relative to other industries, banking stability is very dependent on trust and reputation, which is particularly true for large banks. Global banking investors should therefore protect their profits by identifying high-quality targets. Consequently, an indicator is urgently needed to identify a bank’s financial status and operational competence.
Market investors have used numerous indicators to identify superior investment targets in seeking increased profits. Credit ratings [9], [18] evaluate the attractiveness of banks as investments. When properly assigned by rating agencies, such as Standard and Poor’s (S&P), Moody’s, and Fitch, they are invaluable to financial market participants, providing objective opinions about credit worthiness, investment risk, and default probability. Interested parties include owners, customers, management, personnel, investors, competitors, suppliers, creditors, media, regulatory agencies, researchers, and special-interest groups. Each group uses credit ratings in its own way [47]. For example, credit ratings are extremely important to stock market investors. Although the process of assigning a credit rating requires an enormous amount of time and resources [24], classification models based on financial ratios [13], such as capital adequacy, asset quality, management competence, liquidity risk, sensitivity to market risk (CAMELS) [6] and Earnings Before Interest and Taxes (EBITs), can simplify this process [44]. Financial ratios are typically employed to evaluate bank financial and operational competence, and rate overall management effectiveness based on quarterly and/or annual sales and investment performance. Financial ratios are widely used for modeling by both practitioners and researchers, and have been expressed in various forms [25], [56]. Problems associated with assigning credit ratings resemble those related to forecasting financial crises and bankruptcy [13], [30], [44], [53], which can be developed to construct early warning systems (EWSs) [6] using classification models based on financial ratios.
Since the 1960s, numerous studies have constructed models for predicting financial crises and bankruptcy. These studies have applied both statistical methods and artificial intelligence (AI) techniques, including multiple discriminant analysis (MDA) [40], a logistic model [34], support vector machines (SVMs) [9], [24], [30], and neural networks (NNs) [33]. Although these statistical methods are simple and their outcomes are easy to explain, their explanatory power is inferior to that of AI techniques. This creates decision-making difficulties, as policy-makers cannot fully comprehend and follow the results of the models they use. Moreover, almost all studies comparing the efficiency of these methods found that performance was highly dependent on the application field [8], study goals [17], context and data [27], or user experience [3]. Thus, we recommend employing AI techniques to develop efficient classifiers for forecasting. Artificial intelligence techniques, which have been extensively used when generating credit ratings, have outperformed statistical methods [9], [24]. Particularly, intelligent hybrid systems integrate several models for processing classification problems [2], [50], [55]. In practice, an ensemble classifier outperforms stand-alone models [43], [44]. Given the limitations of statistical methods and AI techniques in stand-alone models, an intelligent hybrid model is needed that maximizes the advantages of statistical methods and AI techniques while minimizing their limitations. Interest in designing and applying various intelligent hybrid models has increased considerably over the last decade [43]. To improve prediction performance and increase investor profits, a reliable forecasting tool based on a hybrid model is required for classifying bank credit ratings.
Designing a rule-based model that can reasonably and powerfully explain data is a significant trend in knowledge discovery. Notably, AI techniques for classification can automatically extract knowledge-based decision rules from a dataset and construct different model representations to explain that dataset [54]. Market investors are very interested in rule-based models that are based on AI techniques and germane to the global banking industry. Research to improve models that solve credit rating problems is valuable for two reasons. First, although the financial industry focuses on investors seeking personal benefit, AI techniques have rarely been used in credit rating research to generate comprehensive decision rules, particularly when compared with statistical methods; therefore, this work fills this knowledge gap. Second, the authors have extensive experience in the financial industry, about 14 years in total, and thus have relevant knowledge.
Because each interested party has an opinion of how to best apply intelligent hybrid systems to real-world problems, particularly the current financial crisis, a reliable model that predicts credit ratings is welcomed. To objectively address the practical problems of classifying bank credit ratings and generate decision rules in the form of knowledge-based systems, this work applies two hybrid models. This work has the following four objectives: (1) implement the two hybrid models that use rough sets (RSs) to classify credit ratings in the global banking industry, minimize the number of selected attributes and generated rules, and increase prediction accuracy; (2) examine the main determinants influencing credit ratings; (3) assess the performances of the proposed hybrid models; and (4) generate comprehensive decision rules that can be applied to knowledge-based systems using RSs and the LEM2 algorithm, and provide reasonable explanatory power to interested parties.
The remainder of this paper is organized as follows. Section 2 reviews the literature on credit rating classification. Section 3 describes the proposed hybrid models and algorithms. Section 4 gives verification details and compares the models. Finally, Section 5 draws conclusions and provides directions for future research.
Section snippets
Literature review
This section reviews credit rating literature, including that associated with rough set theory (RST), the LEM2 algorithm for rule extraction, rule filter (RF), minimize entropy principle approach (MEPA), and factor analysis (FA).
Proposed hybrid models
This work designs two hybrid models for forecasting bank credit ratings and assesses the quality of RS classification systems. The proposed hybrid models are based on the finance-related experiential knowledge (EK) of the authors, as well as various combinations of four components (treatments): FA, a discretization method (e.g., MEPA), RSs, and the RF. The models are ordered as follows: (1) EK + FA + RS + RF (called FA-RS) and (2) EK + MEPA + RS + RF (called MEPA-RS). The FA-RS model first integrates the
Verification and comparison
A labeled Bank dataset, extracted from the Bankscope database, is used to verify the classification performance of the proposed hybrid models.
Conclusions
For highly educated and skilled investors, bank financial reports are the most important available source of research data. A bank’s financial report helps investors stay up to date on a bank’s outlook, carefully study relevant indicators, and make forecasts regarding specific targets. In this era of global economic uncertainty, investors are building intelligent forecasting models for classifying credit rating problems. Successful investors use knowledge-based systems to differentiate good and
Acknowledgments
The authors would like to thank the Editor-in-Chief, associate editor, and anonymous referees for their useful comments and suggestions, which were very helpful in improving this manuscript. The authors also wish to thank the National Science Council of the Republic of China, Taiwan, for financially supporting this research under Contract No. NSC 98-2410-H-146-002.
References (61)
Multi knowledge based rough approximations and applications
Knowledge-Based Systems
(2012)Do capital adequacy requirements reduce risks in banking?
Journal of Banking & Finance
(1999)- et al.
Prediction of commercial bank failure via multivariate statistical analysis of financial structures: the Turkish case
European Journal of Operational Research
(2005) - et al.
A study of Taiwan’s issuer credit rating systems using support vector machines
Expert Systems with Applications
(2006) - et al.
Entropy-based and trapezoid fuzzification-based fuzzy time series approaches for forecasting IT project cost
Technological Forecasting & Social Change
(2006) - et al.
Bankruptcy prediction models based on multinorm analysis: an alternative to accounting ratios
Knowledge-Based Systems
(2012) Modelling of insurers’ rating determinants: an application of machine learning techniques and statistical models
European Journal of Operational Research
(2007)- et al.
Credit rating analysis with support vector machines and neural networks: a market comparative study
Decision Support Systems
(2004) - et al.
A simulation of factors affecting machine learning techniques: an examination of partitioning and class proportions
Omega
(2000) - et al.
The use of hybrid manifold learning and support vector machines in the prediction of business failure
Knowledge-Based Systems
(2011)
Bankruptcy prediction in banks and firms via statistical and intelligent techniques – a review
European Journal of Operational Research
Rough sets and the role of the monetary policy in financial stability (macroeconomic problem) and the prediction of insolvency in insurance sector (microeconomic problem)
European Journal of Operational Research
A case-based approach using inductive indexing for corporate bond rating
Decision Support Systems
Jmax-pruning: a facility for the information theoretic pruning of modular classification rules
Knowledge-Based Systems
Economic and financial prediction using rough sets model
European Journal of Operational Research
Dimensionality reduction based on rough set theory: a review
Applied Soft Computing
Simple instance selection for bankruptcy prediction
Knowledge-Based Systems
Bipolar fuzzy rough set model on two different universes and its application
Knowledge-Based Systems
A hybrid KMV model, random forests and rough set theory approach for credit rating
Knowledge Based Systems
On classification and segmentation of massive audio data streams
Knowledge and Information Systems
Rough set algorithms in classification problem
Ridge estimators in logistic regression
Applied Statistics
A model for customer-focused objective-based performance evaluation of logistics service providers
Asia Pacific Journal of Marketing and Logistics
Forecasting IPO returns using feature selection and entropy-based rough sets
International Journal of Innovative Computing, Information and Control
Entropy Minimax Sourcebook
Diversification, size, and risk at bank holding companies
Journal of Money, Credit, and Banking
Bond rating: a nonconservative application of neural networks
Proceedings of IEEE International Conference on Neural Networks USA
When is a correlation matrix appropriate for factor analysis? Some decision rules
Psychological Bulletin
Comparative study of generalized born models: protein dynamics
Chemical Theory and Computation Special Feature
Cited by (69)
A qualitatively analyzable two-stage ensemble model based on machine learning for credit risk early warning: Evidence from Chinese manufacturing companies
2023, Information Processing and ManagementA robust TOPSIS method for decision making problems with hierarchical and non-monotonic criteria
2023, Expert Systems with ApplicationsCitation Excerpt :The list of criteria, their description and sources are displayed in Table 2, and descriptive statistics are given in Table 3. Evaluation according to financial characteristics is undoubtedly important as financial institutions entail the element of systemic risk (Chen & Cheng, 2013), the close monitoring of which has been signified after the global financial crisis. However, an evaluation could well be augmented to include other, non-financial attributes that portray also the sustainable aspect of entities.
A comparative study of forecasting corporate credit ratings using neural networks, support vector machines, and decision trees
2020, North American Journal of Economics and FinanceCitation Excerpt :Pai et al. (Pai et al., 2015) used Decision Tree Support Vector Machine (DTSVM) integrated to multiple feature selection strategies and rough set theory (RST) to predict credit ratings. Description for other hybrid techniques are available in (Chandra, Ravi, & Bose, 2009; Akkoç, 2012; Chen & Cheng, 2013). Evaluation and Comparison.
Approaches for credit scorecard calibration: An empirical analysis
2017, Knowledge-Based SystemsCorporate Credit Rating: A Survey
2023, arXiv