This article explores the interaction between a bank’s various models and how they may be built into a comprehensive stress testing framework, contributing to the overall performance of a bank.
The role of stress testing is to reduce the likelihood of a bailout in future crises. The notion that the government will bail out a bank if credit losses spike is no longer considered a valid capital adequacy plan. Stress tests and the resulting restrictions on bank dividend payouts are designed to ensure that, when the next crisis occurs, all banks will be able to weather the storm with sufficient capital already on the books.
From a modeling standpoint, the fact that excess credit losses can cause a depositor retreat belies the notion that a bank is merely the sum of its individual parts. Despite this, the models developed during the short history of stress testing have been highly compartmentalized.
Banks typically have models for mortgages, credit cards, commercial and industrial portfolios, and deposits, but nothing that considers how these individual components of the balance sheet interact with each other. Similarly, banks may have separate models of pre-provision net revenue (PPNR) and credit losses that fail to account for the fact that both sets of cash flows are derived from the same loans overseen by the same group of managers.
This article explores these interactions and how they may be built into a comprehensive stress testing framework. Ultimately, these interactions are profoundly important to the overall performance of a bank.
Suppose that you work in the home equity division of a big bank and are tasked with the job of building a PPNR stress testing model for a portfolio. You pull out all the stops and build a really nice model of future outstanding balances (a crucial component of PPNR) that takes account of house prices, interest rates, unemployment, and household income, as well as a careful representation of the effect of bank management strategy in controlling credit line size and the scale and quality of new originations. The model is parsimonious and watertight, passing validation easily. The CFO is impressed and wants to use the model both as a Comprehensive Capital Analysis and Review (CCAR) tool and to help her formulate alternative strategies under a variety of economic conditions. Unfortunately, the related credit loss models are built by the team in Boise, ID, and the models do not interact. The credit loss models have also easily passed scrutiny and are ready to roll in the CCAR submission.
The CFO wants to consider the strategy where, under the assumption of a strong economy, credit lines are increased and origination standards are lowered. These actions will tend to increase the volume of new account origination, while also boosting the scale of legacy loans. Revenue should be quite a bit higher under this strategy than the baseline benchmark results, where credit standards are maintained at their current levels. Under an adverse economic scenario, revenues will decline as demand for credit will be lower. It is conceivable, however, that stressed loose-policy revenues will remain higher than baseline current-policy revenues.
Does this mean that the strategy is a winner and should be implemented with gusto? Probably not. Increasing line size necessarily causes the potential loss given default (LGD) for the existing legacy book to rise; it may even have a more subtle effect on the underlying probability of default. Further, the losses derived from a higher volume of poorer quality new loans may overwhelm any observed increase in revenues. Compared with the baseline, lax standards lead to higher revenue, higher credit loss, and normally higher profitability. When adding the factor of a stressed economy, the result is an indeterminate revenue, much higher credit loss, and generally drastically reduced profitability.
The benefits of having a holistic home equity model – one that covers credit loss and PPNR in a coherent, inter-related manner – should be obvious. If the stress testing framework is used to inform the operations of the home equity loan business and thus the bank, the ability to run “what if” scenarios is absolutely critical. Of course, these questions do not apply merely to home equity loans; banks should also always build PPNR and credit loss models to be interactive.
Imagine another scenario. You have just been laid off from your job and your prospects are grim. You drive home in a bad mood and find three envelopes in your mailbox. Your mortgage, car loan, and credit card payments are all due and you have insufficient funds to cover all three. You panic. What should you do?
At an individual level, the notion that the probability of default (PD) on any one of these products is independent of the PDs on the other two can be immediately dispelled. If you pay the mortgage, the car loan and credit card will become delinquent. Going delinquent on the mortgage would free up enough funds to remain current on the two other products.
At a macroeconomic level, prior to the Great Recession it seemed that, on average, people in trouble tended to favor mortgages for payment, followed by auto loans and credit cards. Because falling house prices were at the root of the troubles of 2008, this determination to defend the house melted remarkably quickly. To this day, total mortgage delinquency exceeds auto and credit card delinquency by almost two-to-one (see Figure 1).
If the macroeconomic payment hierarchy was an immutable law of nature, the aggregates of auto, credit card, and mortgage loans could be safely modeled separately, despite the continuing dilemma faced by our hypothetical job seeker. As the payment hierarchy flipped as a direct result of stress, however, it seems prudent to question whether modeling consumer loan products in isolation is wise. Not paying the mortgage frees up a lot more auto repayments than not paying the credit card. Could auto and credit card losses have been made lower during the recession as a direct benefit of the many missed mortgage payments? An isolated model of auto loans that ignores the mortgage market would miss this possibly important dynamic feature. A complete model would capture interdependencies between credit loss and PPNR across the consumer credit book. One imagines that similar interdependencies exist on the wholesale side.
The other benefit of modeling across different credit portfolios is that the value of banking relationships can be explored and its effects quantified. Bankers often criticize risk modelers for failing to account for the value of relationships painstakingly built by those on the business side of the bank. Models that span different aspects of the bank could, however, start to bridge this divide.
Suppose a bank has two mortgage clients who live in adjoining identical houses. They have the same income, job tenure, and credit score; they are both paid the same and borrowed the same amount of money at the same interest rate. Now suppose the first person has everything with the bank – checking accounts, CDs, auto loans, credit cards, etc. The other person has all of these things but with different banks. Would the bank prefer the first or the second client? If the second client defaults, one upside would be that the risk is spread across a number of banks and that losses would be limited. In assessing credit risk, it seems logical that modelers should be able to obtain a tighter read on the more loyal client. Put simply, there are so many more signals about what is happening in the life of that person. There is no reason why these additional signals cannot be harnessed and the benefits and costs of loyal customers quantified; helping assess relative risk more accurately. Indeed, those who espouse the wonders of relationship banking should welcome such moves from risk modelers.
Of course, such a step is not possible unless encompassing models of the entire retail (or wholesale) book are first considered.
A valid question from a modeling perspective, where credit losses and deposit balances are going to be jointly considered, is whether such forces also act at the margin. Would a small credit loss shock to a bank cause a small decline in the size of the deposit book or a small increase in the cost of obtaining funds in financial markets?
We think that it would. In a highly competitive investment and banking landscape, where investors can move their funds around with a mouse click, even small shifts in relative prices could have a meaningful impact on the demand for banking services. If we further assert that a bank’s pricing power is inherently tied up with its reputation for successful money management and that this reputation would be harmed if its credit losses rose relative to the industry, it becomes easy to sketch out ways that a bank’s funding costs could be affected by the scale of its credit losses.
Two additional points should be noted here. First, we are discussing relative prices – if a bank’s direct competitors suffered similar detrimental credit loss shocks, we would expect no ramifications for the deposit book of the bank in question. The observation made here concerns only the situation where one bank’s credit losses rise while others’ do not. The final point is that we are not talking about systemic risk. If system-wide credit losses rise as they did during the Great Recession, we would expect the overall scale of deposits to contract as confidence in the continuity of the institution of bank deposit-taking takes a hit. These systemic problems should be accounted for in modeling deposits under stress, though there are no obvious ways that an individual bank manager might respond in a bid to mitigate their effect.
While the existence of the FDIC should calm the nerves of rational investors, people often fail to behave rationally. When choosing between two equally priced, equally convenient deposit products at two different banks, perceived fiscal soundness may be a valid way to break the tie. For those whose funds are not insured, or for those who would be willing to line up to withdraw their insured deposits from a failing bank, questions of bank prudence may be a more important determinant of who wins their deposit business.
The potential for credit losses should therefore be factored into liability-side stress testing models.
In banks, head office managers do more than simply aggregate performance outcomes from individual business lines. Complex banking corporations exist to take advantage of synergies, scale economies, pricing power, and risk mitigation techniques that derive from having many such business lines under a single umbrella. If the total were less than the sum of the parts, financial markets would demand that banks be dispersed to create greater shareholder value. If a stressful situation occurs, presumably the advantages of corporation and coordination do not suddenly disappear. Banks should be capable of mitigating stress risks, and these efforts should be assisted by the ability of senior managers to coordinate across individual lines of business.
When reviewing the models used to address stress testing challenges, one would think that banks were completely uncoordinated collections of unrelated businesses. Each line item in the CCAR submission is estimated by a model that typically bears no relationship to the behavior of any other line item. Even within business units, there are cases where the revenue and credit loss models used for stress testing are completely unrelated to each other. Few people have proposed models for “household” or “business” credit risk while many have used models specific to auto loans or commercial real estate to address the stress testing imperative.
If CCAR results are to be woven into the fabric of the bank, and referenced at all levels of management, this lack of coordination in stress testing models must be addressed. For this to happen, model infrastructure needs to mirror and mimic the way banks actually operate. The models need to account explicitly for the actions managers take in controlling portfolio outcomes, both within and across business lines, between revenues and expenses, and between assets and liabilities. If stress testing models remain siloed, informative intelligence may be difficult to gain in the best management responses to an adverse economic environment.
Dr. Juan M. Licari is a managing director at Moody's Analytics. Juan and team-members are responsible for the research and analytics that enable our quantitative solutions. The team helps our customers solve complex business problems; adding value through data and analytics.
Explores how North American financial institutions can leverage stress testing regulations to add value to their business, for compliance and beyond.
We look at climate risk and consider how a heating planet might impact a bank's performance
Expanding Roles of Artificial Intelligence and Machine Learning in Lending and Credit Risk Management
With ever-expanding and improving AI and Machine Learning available, we explore how a lending officer can make good decisions faster and cheaper through AI. Will AI/ML refine existing processes? Or lead to completely new approaches? Or Both? What is the promise? And what is the risk?
When banks manage risk, conservatism is a virtue. We, as citizens, want banks to hold slightly more capital than strictly necessary and to make, at the margin, more provisions for potential loan losses. Moreover, we want them to be generally cautious in their underwriting. But what is the best way to arrive at these conservative calculations?
The traditional build-and-validate modeling approach is expensive and taxing. A more positive and productive validation experience entails competing models developed by independent teams.
The industry is currently a hive of CECL-related activity. Many banks are busily testing their systems or finalizing their preparations for the go-live date, which is either in January 2020 or somewhat later, depending on the organization. Some are still making plans for implementation, and the rest are worried that they should be.
The theory that banks are now safer because of CCAR, though, has not yet been tested.
Loan-loss provisioning models must take a variety of economic and client factors into account, but, with the right approach, banks can develop sensible loss forecasts that are more accurate and less susceptible to volatility.
As evidence of climate change builds and threats materialize,data will be invaluable in creating a framework for making future credit decisions.
In recent years, attention has increasingly turned to the promise of artificial intelligence (AI) to further increase credit availability and to improve the profitability of banks and other lenders. But what is AI?
Good-quality CECL projections can be developed using high-quality data that is available free of charge.