Bank performance prediction during the 'great recession' of 2008-'09: a pattern-recognition approach.

Author:Moore, James


The 2008-'09 banking crisis had been developing for several decades, as banks have not exhibited an ability to diversify their business model as their industry evolved (Waldeck, 2009). Since the 1980's, the banking industry has encountered growing competition, from both external and internal sources. Commercial banks witnessed their core business of lending being encroached upon by retailers and specialty finance companies. The 1999 repeal of the Glass-Steagall Act brought open competition between commercial and investment banks in the underwriting and trading of financial instruments. The intensity of the competition resulted in more aggressive portfolio management, lax liquidity risk management, and the emergence of new and creative financial instruments. The trouble with existing bank asset portfolios lies in their recent reliance upon increasingly complex securities packages that were designed to spread risk across participants. Many of these packages are backed by large diverse pools of currently underperforming mortgages, which makes the assets difficult to value and dispose of. A cash liquidity crisis contributed to the collapse of Bear Stearns, Lehman Brothers Holdings, and Washington Mutual among others in 2008. Traditional credit and portfolio risk management policies were relaxed as banks extended credit to too many marginal credit applicants. Evidently, the historical lessons of unchecked optimism of the1920's were simply set aside.

During 2009, 140 U.S. banks and thrifts failed and were closed by regulators; that followed 25 such institutions failing in 2008 (FDIC, 2010). Many of the larger failing banks were troubled by the commercial real estate market, while many smaller banks fell victim to losses in their commercial and consumer loan portfolios, as well as their mortgage lending. These closings have drained billions of dollars from the FDIC Deposit Insurance Fund, reducing it to its lowest level since 1992, at the peak of the savings-and-loan crisis. The 2008 and 2009 failures drained the FDIC's fund below its mandated level, causing the FDIC to require insured banks to prepay the next three years of their annual assessments. The end of 2009 saw the number of banks on the FDIC's 'problem list' grow to 702 from 552 in the previous quarter and from 252 at year-end 2008. While the FDIC creates reports on problem or troubled banks in the aggregate, for obvious reasons, it does not make details of the 'problem list' public nor comment on open financial institutions. Institutions on the 'problem list' have issues that could lead to their eventual failure. Persisting loan losses and credit delinquencies will continue to 'extract their pound of flesh' on institutional performance and place additional banks at risk.

Thus, there exists a need for predictive techniques to provide an early warning system to regulatory agencies and other stakeholders, regarding distressed or failing institutions. Once distressed banks are identified, regulator intervention might be able to prevent ultimate failure, or at least minimize the impact of such failure. Predicting the financial health of individual banking institutions is the intent of this study. As in biology, sociology, and other similar disciplines, both environmental (external) and genetic (internal) factors have a role in developing the individual. This study will examine selected factors internal to the individual bank and their role in the financial soundness of the specific institution, while controlling for the macroeconomic external factors.

The gradually recovering ability of the largest banks to sell their long-term debt in the private markets without government backing and their ability to raise equity funding may insulate them from the urgency to dispose of their toxic assets at distress prices. Moreover, they have sufficient 'critical mass' to enable them to extend the time horizon in which they will dispose of their troubled assets. Thus, the investment banks (e.g. Goldman Sachs, Morgan Stanley), the financial service giants (e.g. Bank of America, J.P. Morgan Chase, Citigroup), and those that focus on wealth management (e.g. Bank of New York Mellon) may use the breadth and depth of their capitalization to survive, and even prosper, in spite of the continued presence of the troubled mortgage-backed assets on their balance sheets. That could mean that the troubled regional and small banks become the at-risk stratum of institutions on the FDIC 'problem banks' list. By early 2010, in an effort to stimulate more lending to small businesses, the Treasury directed $1 billion towards small banks, thrifts, and credit unions that were certified as Community Development Financial Institutions, meaning they target more than 60 percent of their small-business lending to lower-income areas. Given this potential vulnerability of regional institutions, this study will focus on a data set dominated by regional and small banks in one major metropolitan area, in its effort to identify the key internal factors influencing financial health. The use of a common geographic region attempts to control for the influence of regional economic conditions on bank performance.


A comprehensive, pre-crisis review of the U.S. banking industry since the seminal work of Benston et al (1986) is provided by DeYoung (2007). Ironically, he argues that the banking industry is 'almost certainly' safer and sounder at the time of his writing (early 2007) than twenty years prior. Nevertheless, his profile of the banks' changing operating environment provides excellent context for the financial crisis that began later in 2007. Bullard et al (2009) examine the connections among mortgage market problems, the failure of financial institutions, and the impairment to the broader economy in the context of systemic risk. They argue that such systemic triggers are more dangerous with the failure of financial firms than with the failures of non-financial firms. This concern for systemic risk prompted the Federal Reserve and the U.S. Treasury to act to prevent the failure of several large financial institutions in 2008. The specific impact of the regulatory environment on bank operations and their financial health is reviewed by Boerner (2008). He poses thought provoking questions about regulatory reform. This study will focus on the manifestations of the legal environment, namely the bank balance sheet as a predictor of an institution's future financial health.

The history of formal bank failure prediction models goes back over 30 years and includes a host of analytical techniques including multivariate statistical analysis/discriminant analysis (Altman et al, 1977; Sinkey, 1975), Logit / Probit analysis (Avery and Hanweck, 1984; Barth et al, 1985; Estrella et al, 2000; Kolari et al, 2002), survival analysis (Cole and Gunther, 1998; Molina, 2002), neural networks (Tam and Kiang, 1992; Bell, 1997; Alam et al, 2000), Data Envelopment Analysis (Barr and Siems, 1996), and simulation (Lam and Moy, 2002). More recently, Kim and Miner (2007) examine failures and near-failures of banks; they find that the local market has more influence than does the non-local (national) market. Consequently, this study will attempt to control for the 'local' market as it focuses on individual bank's internal portfolio practices. A forensic study of failed banks using Probit analysis was conducted by Dandapani and Lawrence (2008). They distinguished 'brick and mortar' banks from 'virtual' banks in their search for the underlying contributors to bank failures. They found that 'brick and mortar' banks typically failed due to poor asset quality, while 'virtual' banks failed primarily due to high non-interest expense. They conclude that different business models are warranted for these distinct banking channels. In a comparative-methodology study, Kosmidou and Zopounidis (2008) developed a bank failure prediction model using financial ratios. Their multi-criteria decision model outperformed Multiple Discriminant Analysis in predicting failing banks during 1993-2003. Arena (2008) used cross-sectional multivariate Logit analysis on bank-level data associated with the CAMEL rating system to conclude that such fundamentals did significantly influence the likelihood of collapse for banks in East Asia and Latin America during the 1990's.

Using a parallel methodology to that of this study, Ozkan-Gunay and Ozkan (2007) use data mining (via an Artificial Neural Network) to search for predictive structures in financial data that would explain previous bank failures in Turkey. While their validation / hold-out sample was very small (n=29), their correct overall prediction rate, (nearly 85%), is quite strong and thus encouraging for further use of pattern recognition technologies as early-warning mechanisms. Quek et al (2009) use a neural network with 3636 U.S. banks, with data reconstruction of missing financial information, spanning 1979 to 1999. While neural network techniques have resulted in good predictive accuracy with respect to bank failures, they lack the ability to explain/reveal the major contributors, and their relative influence, on such performance. The current study will employ Rule Induction, as an alternative approach to pattern recognition within the financial data of individual U.S. banks, in an effort to build an early warning system for the identification of distressed institutions. Unlike neural networks, Rule Induction can reveal the relative roles of those factors contributing to the classification outcome.


Little has been done to find application for the decision support technologies of pattern recognition / data mining to the issue of institutional performance; this paper seeks to re-open that area of inquiry. Bose and Mahapatara (2001) surveyed data mining procedures in business applications, and found rule...

To continue reading