英文原文： A credit scoring approach for the commercial banking sector Ahmet Burak Emel, Arnold Reisman and Reha Yolalan Yapi Kredi Bank, Levent, 80620, Istanbul, Turkey. The Graduate School of Management, Sabanci University, Istanbul, Turkey Available online 15 March 2007 The economic and, therefore, the social well-being of developing countries with fairly privatized economies is highly dependent on the behavior of a countrys commercial banking sector. Banks provide credit to sustain anufacturing, agricultural, commercial and service enterprises. These, in turn, provide jobs thus enhancing purchasing power, consumption, and savings. Bank failures, especially in such settings, send shockwaves affecting the social fabric of the country as a whole and, as experienced recently, (Latin America and Asia) have the potential of a quick global impact. Thus, it is imperative that lending/credit decisions are made as prudently as possible while keeping the decision making process both efficient and effective. Commercial banks provide financial products and services to clients while managing a set of multi-dimensional risks associated with liquidity, capital adequacy, credit, interest and foreign exchange rates, operating and sovereign risks, etc. In this sense, banks may be considered to be “risk machines”. They take risks, and transform or embed such risks to provide products and services. Banks are also “profit-seeking” organizations basically formed to make money for shareholders. In their typical decision-making processes (i.e. pricing, lending, funding, hedging, etc.), they try to optimize their “risk-return” trade-off. Management of risk and of profitability are very closely related. Risk taking is the basic requirement for future profitability. In other words, todays risks may turn up as tomorrows realities. Therefore, banks may not live without managing these risks. Among the different banking risks, credit risk has a potential “social” impact because of the number and diversity of stakeholders affected. Business failures affect shareholders, managers, lenders (banks), suppliers, clients, the financial community, government, competitors, and regulatory bodies, among others. In the age of telecommunications, the ripple effect of a bank failure is virtually instantaneous and such ripples hold the potential of global impact. In order to effectively manage the credit risk exposure of a modern bank, there is thus a strong need for sophisticated decision support systems backed by analytical tools to measure, monitor, manage, and control, financial and operational risks and inefficiencies. Conscious risk-taking decisions call for quantitative risk-management systems, which, in turn, provide the bank early warnings for predicting potential business failures. Thus, an effective risk-monitoring unit supports managers judgments and, hence, the profitability of the bank. A potential clients credit risk level is often evaluated by the banks internal credit scoring models. Such models offer banks a means for evaluating the risk of their credit portfolio, in a timely manner, by centralizing global-exposures data and by analyzing marginal as well as absolute contributions to risk components. These models can offer useful insight and do provide an important body of information to help a bank formulate its risk management strategies. Models that are conceptually sound, empirically validated, backed by good historical data, understood and implemented by management, augment the business success of credit quality. Over the past decade, several financial crises observed in some emerging markets enjoying a recent financial liberalization experience, showed that debt financing built on capital inflow may result in large and sudden capital outflows, thereby causing a domestic “credit crunch”. Experience with these recent crises forced banking authorities, i.e. the Bank of International Settlements (BIS), the World Bank, the IMF, as well as the Federal Reserve. to draw a number of lessons. Hence, they all encourage commercial banks to develop internal models to better quantify financial risks. The Basel Committee on Banking Supervision, English and Nelson, the Federal Reserve System Task Force on Internal Credit Risk Models.Lopez and Saidenberg and Treacy and Carey represent some recent documents addressing these issues. Credit scoring has both financial and non-financial aspects. The scope of the current paper, however, is limited to the evaluation of a bank clients financial performance. Studies attempting to measure firm performance on the basis of qualitative data are exemplified by Bertels et al. Formal or mathematical modeling of finance theory began in the late 1950s. The work of Markowitz represents a major milestone. The practice reached its “take-off” stage as a sub-discipline of Finance during the early 1960s. Some of the early efforts were directed at evaluating a firm for purposes of mergers and acquisitions; some dealt with using investment portfolios to manage risk; others dealt with improvement/optimization of a firms financing mix. They were all directed at enhancing extant finance theory toward the goal of guiding decision-makers. One of the fields in which formal or mathematical modeling of finance theory has found widespread application is risk measurement. A firms financial information plays a vital role in decision making of risk-taking activities by different parties in the economy. An extensive literature dedicated to the prediction of business failure as well as credit scoring concepts has emerged in recent years. Financial ratios are the simplest tools for evaluating and predicting the financial performance of firms. They have been used in the literature for many decades. The benefits and limitations of financial ratio analysis are addressed in a widely used text on managerial finance. Financial statements report both on a firms position at a point in time and on its operations over some past period. However, there are still some limitations in using ratio analysis: (i) many large firms operate in a number of different industries. In such cases it is difficult to develop a meaningful set of industry averages for comparative purposes; (ii) inflation badly distorts a firms balance sheet. Moreover, recorded values are often substantially different from their “true” values; (iii) seasonal factors can distort a ratio analysis; (iv) firms can employ “window dressing techniques” to make their financial statements look stronger; (v) it is difficult to generalize about whether a particular ratio is “good” or “bad”; and (vi) a firm may have some ratios looking “good” and others looking “bad” making it difficult to tell whether the firm is, on balance, strong or weak. Across different countries, sectors and/or periods of time, financial ratios that have been found useful in predicting failure differ from study to study. To deal with the above shortcomings of unidimensional financial ratio analysis, a variety of methods have appeared in the literature for modeling the business failure prediction process. An excellent comprehensive literature survey can be found in Dimitras et al. In the late 1960s, discriminant analysis (DA) was introduced to create a composite empirical indicator of financial ratios. Using financial ratios, Beaver developed an indicator that best differentiated between failed and non-failed firms using univariate analysis techniques. Altman established that ratios found not to be very significant by univariate models, could prove somewhat useful in a discriminant function which considers the relationships among variables. Hence, he considered several variables simultaneously using multiple discriminant analysis (MDA). He argued that MDA had the advantage of considering an entire profile of interrelated characteristics common to the relevant firms. That study also aimed to predict future failure on the basis of financial ratios. He concluded that his bankruptcy prediction model was an accurate forecaster of failure for up to 2 years prior to bankruptcy and that the models accuracy diminishes substantially as the lead-time increases. In spite of widespread use of MDA, Altman, confesses to the following weakness of discriminant analysis: Up to this point the sample firms were chosen either by their bankruptcy status (Group 1) or by their similarity to Group 1 in all aspects except their economic well being. But what of the many firms which suffer temporary profitability difficulties, but in actuality do not become bankrupt. During the years that followed, many researchers attempted to increase the success of MDA in predicting business failure. Among these are Eisenbeis; Peel et al.; and Falbo. Such work also involved Turkish firms. Examples are Unal, and Ganamukkala and Karan. Linear probability and multivariate conditional probability models (Logit and Probit) were introduced to the business failure prediction literature in late 1970s. The contribution of these methods was in estimating the probability of a firms failure. The linear probability model is a special case of ordinary least-squares regression with a dichotomous dependent variable. In the 1980s, studies utilizing the recursive partitioning algorithm (RPA) based on a binary classification tree rationale were applied to this problem by Frydman et al. and Srinivasan and Kim. In the 1980s and 1990s, the use of several mathematical programming techniques enriched the literature. The basic goals of these methods were to escape the assumptions and restrictions of previous techniques and to improve classification accuracy. In the early 1990s, decision support systems (DSS) in conjunction with the paradigm of multi-criteria decision-making (MCDM), were introduced to financial classification problems. Zopounidis, Mareschal and Brans Zopounidis et al. Diakoulaki et al., Siskos et al. and Zopounidis and Doumpos were among the studies that measured firm performance aiming at predicting business failure by making use of DSS and MCDM. The ELECTRE method of Roy and the Rough Sets Method of Dimitras et al. represent studies addressing these issues. Development and application of artificial intelligence resulted in the use of expert systems. Neural Network methods were applied to the bankruptcy problem as well. In the late 1990s, data envelopment analysis (DEA) was introduced to the analysis of credit scoring as in Troutt et al., Simak, and Cielen and Vanhoof. As opposed to the broadly known MDA approach for business failure prediction (which requires extra a priori information, i.e. good/bad classification), DEA requires solely ex-post information, i.e. the observed set of inputs and outputs, to calculate the credit scores. Thus, it opened new horizons for credit scoring. DEA, widely known as a non-parametric approach, is basically a mathematical programming technique developed by Charnes, Cooper and Rhodes (CCR) to evaluate the relative efficiency of “decision making units” (DMUs). DEA converts a multiplicity of input and output measures into a unit-free single performance index formed as a ratio of aggregated output to aggregated input. A productivity maximization rationale is elegantly embedded in its original fractional formulation. The capability of dealing with multi-input/multi-output settings provides DEA an edge over other analytical tools. Conceptually, DEA compares the DMUs observed outputs and inputs in order to identify the relative “best practices” for a chosen observation set. Based on these best observations, an efficient frontier is established and the degree of efficiency of other units with respect to the efficient frontier is then measured. Based on its input-oriented DEA formulation, the resulting performance index value (the credibility score, in our context) provides a numerical value E. E lies between zero and one. If E is less than one, the DMU is considered “inefficient” as compared to the efficient frontier derived from best practices. If E is equal to one, the DMU is located on the efficient frontier. Therefore, it can be said that E measures the relative credit riskiness of firms within the bank portfolio. A number of studies have attempted to use statistical methods (such as discriminant, Logit and Probit analyses) with financial ratios to generate early warning signals for distressed banking institutions The idea is to develop meaningful “peer group analysis”, that is, to develop specific financial characteristics that distinguish between two or more groups, for example, failed and non-failed banks, or problem and non-problem banks, with relatively “good” or “bad” financial conditions. However, except when a priori groups are available to provide certain financial profiles for comparison, identifying appropriate peer group analysis is a difficult task. Data envelopment analysis (DEA), which computes a firms efficiency by transforming inputs into outputs relative to its peers, may provide a fine mechanism for deriving appropriate categories for this purpose. An advantage of DEA is that, it uses actual sample data to derive the efficiency frontier against which each unit in the sample is evaluated with no a priori information regarding which inputs and outputs are most important in the evaluation procedure. Instead, the efficient frontier is generated, when a mathematical algorithm is used to calculate the DEA efficiency score for each unit. Although DEA was introduced in the early 1980s, its applications are acquiring more widespread recognition in the financial literature as time passes. 中文翻譯： 商業銀行的信用評分步驟 在經濟相當被私有化的發展中國家，經濟福利和社會福利與國家的商業銀行業的行為有相當高的依賴性。銀行給制造業 、 農業 、 商業服務和服務企業提供信貸。這些能提供工作、提高購買力、影響消費和儲蓄。特別是在此背景下 ， 銀行倒閉其沖擊波會影響到該國的整個社會結構。因此 ， 這是當務之急 ， 貸款 /信貸決定都是一樣謹慎 ， 盡量保持決策過程的效率性和有效性。 商業銀行對客戶提供金融產品和服務的同時，還要管理一套聯系了流動資產、資本充足、信用、利率及匯率方面、操作和主權風險等多維風險，從這個意義上講 ， 銀行可能會被認為是 “ 風險機器 ” 。他們在提供產品和服務時，必須承擔風險 ， 嵌入或改造這種風險。 銀行也是 “ 追求利潤 ” 組織 ， 其股東基本是以賺錢為主要目的。在典型的決策過程（即價格 ， 貸款 ， 資金 ， 套期保值等） ， 他們試圖優化其 “ 風險 -收益 ”權衡。 風險管理和贏利的關系非常密切。風險追求是未來盈利能力的基本要求，換句話說，今天的風險也許作為明天的現實出現。所以，商業銀行部管理好風 險就無法生存。 在不同的銀行業務風險之中 ， 由于賭金保管人數量和變化影響，信用危險有潛在的 “ 社會 ” 沖擊。在電訊日趨成熟的現代社會 ， 銀行倒閉的波動行為幾乎是在瞬間產生全球性沖擊。為了有效管理現代銀行的信用風險，輔助決策支持系統就需要精密的分析工具來衡量，監測和管理，和控制財務與業務風險和低效率。 意識到冒險的決定 ， 呼吁定量風險管理系統提供銀行預警來預測潛在的企業倒閉。因此 ， 盈利的銀行必須使有效的風險監控單位支持經理人的判斷。一個潛在客戶的信用風險水平常常用來評價銀行的內部信用評分模型。這些目標 ， 以確定申請人 是否有能力償還的評估信用風險貸款，這通常是利用歷史數據和統計方法。這些模型能給銀行提供一種手段，以及時評估它們的風險信用組合，集中了全球風險數據并對此進行了邊際分析。這些模型還可以提供有用的見解 ， 并提供了一個重要的信息 ， 以幫助銀行制定風險管理戰略。實驗驗證，數學模型是在概念上健全，輔以良好的歷史數據，并且對此執行管理和理解，以充實業務成功的授信品質。 過去十年，對幾個金融危機的觀測說明，在一些新興市場金融自由化的經驗，表明債務融資興建的資本流入可能導致大資金突然外流，從而造成國內的 “ 信貸緊縮 ” 。縱觀這些金融 危機的起因表明 ， 信貸擴張的資金主要來自資本流入導致投資過 高 ， 使得銀行和公司部門易受沖擊。最近這些危機迫使銀行業監管當局 ，即國際清算銀行、世界銀行、國際貨幣基金組織以及美國聯邦儲備委員會 ， 吸取一些教訓。因此，他們鼓勵各商業銀行發展的內部模式，以更好地量化金融風險。巴塞爾銀行監督委員會， English 和 Nelson、聯邦儲備系統專責小組內部信用風險模型， Lopez、 Saidenberg、 Treacy 與 Carey 用最近的一些觀點和文獻來解決這些問題。 信用計分有財政和非財務兩個方面。然而，當前文件的范圍被限制對銀行客戶的財政表現的評估， Bertels 試圖研究以衡量公司業績的基礎上的定性數據。 數學建模金融理論始于 50 年代末， Markowitz 的工作是一個重大的里程碑。財政部在 60 年代初，將其作為一個分學科，使其從實踐中達到了 “ 起飛 ” 階段。早期一些嘗試 ， 是針對評價一個公司用于兼并和收購；一些處理利用投資組合風險管理；一些人處理改進 /優化企業的融資結構。他們都是針對增強現有金融理論的指導決策者。 其中 1948 年的數學建模的金融理論已廣泛應用 ， 稱作為風險度量 。在決策中形成不同黨派經濟活動的風險 ， 一家公司的財務信息方面發揮了重要作用。廣泛的文獻致力企業倒閉的預言，并且近年來涌現了信用計分的概念。財務比率是為評估和預言企業財政表現的最簡單的工具，財政比率分析的好處和局限演講廣泛應用在管理財務的文獻。財政決算報告堅定了公司的立場和關于過去某一期間的業務。但是，仍然有