The U.S. Mortgage Crisis: What the Models Missed
We’re still experiencing the aftershocks of the U.S. mortgage crisis. Conventional wisdom tells us that we know how the crisis happened: A fall in house prices nationwide (step one) triggered a wave of losses in the mortgage industry (step two). The losses shut down the mortgage-backed securities (MBS) market (step three) as traders lost confidence in their ability to value those assets. That led to a liquidity crisis for the banking industry (step four) that triggered a general recession (step five).
That’s the quick and simple explanation. But how exactly does a drop in housing prices initiate such a drastic economic meltdown? The ultimate answer is: It does not. When housing prices flatten or show a modest drop, no increase should occur in the number of defaulted loans. Rather, banks face a severe problem in which foreclosed homes cannot be sold to recover the outstanding loan balance. Only when the drop in home prices becomes severe would we expect to see an increase in the number of defaults, as some consumers choose a “strategic default” or walk away from their property even though they have the ability to continue paying their mortgage.
Not until July 2008 did home prices fall 20 percent below their 2006 peak—the level at which strategic default begins to make financial sense, according to the Case-Shiller Index. That drop was well into the crisis and certainly not an initial cause. Likewise, default severity would have caused problems for some lenders—but not a true crisis—if the number of defaulting loans had remained unchanged. The answer, which has also been pointed out, was that the loans were not normal.
We need to include an initial step (step zero) in the sequence of events. Before a drop in housing prices, we experienced an extended period of time where mortgage lenders were aggressively lowering their lending standards in a “rush to the bottom.” Because the loans could be securitized, less attention was given to underwriting. The focus, instead, was on volume growth. In 2005, the standard planning scenario by mortgage lenders for 2006 was: We want to grow our mortgage originations by 40 percent year-over-year, but not drop our cut-off scores.
That last phrase is critical. Establishing minimum acceptable levels for credit scores and loan-to-value ratios was the standard in risk management for a decade. The clause was meant to soften the aggressive growth targets, but it had an unintended consequence. Many factors beyond credit score and loan-to-value can drive credit risk. Because managers’ bonuses were often tied to growth targets, the instructions could be translated as, “Grow 40 percent year-over-year by defeating our credit scores.”
All of this can be seen in the data. At the onset of the crisis, portfolio managers commonly pointed to flat origination scores as proof that underwriting standards had not been lowered. At the same time, volume soared for sub-prime loans, negative amortization loans, option arms, and adjustable-rate mortgages (ARMs) with unsustainably low teaser rates.
Here is where most discussions of the mortgage crisis end. However, we have additional information that doesn’t fit this simple pattern. Using sophisticated analytical methods, we can normalize the default rate data for changes in the economy to measure the intrinsic credit risk of each month’s originations. That analysis confirms that credit quality was deteriorating, as described in step zero, but it also shows that the problem began earlier.
Credit quality began to deteriorate in late 2003, before the dramatic increases in sub-prime and unconventional mortgages in 2005 and 2006. Furthermore, when we studied auto loans, credit cards, and student loans, all showed deterioration in credit quality with the same timing as observed in the mortgage industry, only with less severe swings.
Further analysis of long historical time series has led us to the conclusion that there was actually another step (minus one), “macroeconomic adverse selection.” Normally, we think of adverse selection as arising from the competition between lenders. If a lender’s pricing is too high or too low relative to the marketplace, the applicants for its loans will not be what the models expect, because “normal” consumers will be shopping the middle market.
Macroeconomic adverse selection applies a similar concept through time. Consumer appetite for new debt will vary over time. When debt is cheaper, lower-risk consumers are drawn into the market. Falling interest rates and low home prices appeal to the value shoppers in the population. Conversely, rising interest rates and rising home prices will interest only those who are forced into the market, are not financially savvy or are betting on prices to continue rising such that they can refinance later. All are risk-taking behaviors that are not conducive to sound lending.
An analysis of data back to 1990 shows three distinct reversals from good-credit-risk mortgage origination to poor-quality origination: 1994, 1999 and 2003. In each of those years, mortgage interest rates switched from falling to rising, and home prices began increasing rapidly.
Macroeconomic adverse selection contributed significantly to the U.S. mortgage crisis, because the aggressive growth goals of 2005 and 2006 were based on the low industry delinquency rates observed up to that point. Even high-risk mortgages do not default immediately. The peak delinquency usually occurs two to four years after origination. Thus, in 2005, the weakening originations from 2004 were not yet apparent in simple reports, but the conservative, low-credit-risk consumers were rapidly pulling out of the market, leaving only risky consumers to meet the aggressive growth targets of lenders.
The data makes it clear that mortgage delinquency shocks are cyclical: 1991, 1996, 2002, and 2006-2010. In an attempt to prevent future crises, underwriting standards, leverage ratios and securitization are experiencing many modifications. However, none of the changes address cyclicality in consumer appetite for credit. All of the mortgage lender models and policies assume that the same prospective borrowers will be available over time. In fact, the pool of available borrowers changes dramatically through time, and lenders and their models must adapt if we are to soften future cycles.
In the modern age of retail lending, good decision-making starts with sound models and informative reports. Poor models and inadequate reporting can assume significant blame for the crisis. The workhorse models in retail lending have always been credit scores and roll rates. For trading mortgage-backed (MBS) and asset-backed securities (ABS), the focus has been on valuation models that take those model outputs as inputs or pure trading models that try to predict trends in market prices. Unfortunately, neither method can predict turning points in consumer appetite, underwriting standards or macroeconomic conditions that were key to the last crisis.
The industry response to the problems has largely been to focus on loan-level models and include more factors. Certainly, refreshed house prices, combined loan-to-value ratio (CLTV), and refreshed credit scores can improve the performance of failed models. Using loan-level models is beneficial if the goal is to trade individual loans. Even traders have begun to realize that market-based pricing models are only as smart as the market, and we have seen that the MBS/ABS market did not have access to or ignored many critical factors for efficient pricing.
Even with these changes, the models will only be as good as the factors they include. At present, none of those factors can capture macroeconomic adverse selection or creative changes to underwriting standards. Nevertheless, properly structured non-linear decomposition models, such as dual-time dynamics and survival models, can analyze the early performance of loans to detect changes to credit quality. Dual-time dynamics successfully predicted the mortgage crisis as early as December of 2005.
But even if we improve the models, we must ultimately fix the reporting and decision-making process. We often hear that the outputs of a model must be reduced to a simple roll-rate or score distribution report “for the executives” because executives are accustomed to those technologies. Unfortunately, roll-rate and score distribution reports hide important information that explain why a crisis is imminent and can inform management on how to respond. As long as we keep producing the same outdated reports, we will continue making the same flawed decisions that led to the current crisis—and may create the next.
As for where we go from here, we have seen that macroeconomic adverse selection leads every mortgage delinquency cycle. Although we’re only starting to quantify and predict this effect, we can be certain that when interest rates begin to rise and home prices increase again, we must look critically at the quality of the loans being booked. Who wants a loan when both the price of the home and the mortgage are expensive?
Joseph Breeden is chief executive officer of Strategic Analytics, a provider of credit risk and capital management solutions to consumer and mortgage lenders. As part of the Interthinx business unit of Verisk Analytics, Strategic Analytics provides advanced solutions and professional services critical to loss forecasting and the stability of the U.S. residential mortgage market. For more information, call (505) 995-4755 or visit www.strategicanalytics.com.
FMJ Job Listings
- Real Estate Disclosure Coordinator - Provident Credit Union - Redwood City, CA
- Premier Mortgage Consultant - HSBC - Brooklyn, NY
- President/Chief Executive Officer - County Educators Federal Credit Union - Roselle Park, NJ
- Systems QA Analyst 4 - Wells Fargo - Saint Louis Park, MN
- Systems QA Analyst 4 - Wells Fargo - West Des Moines, IA
- Senior Technology Manager Commercial Mortgage Servicing (CMS) (Technology Manager 5) - Wells Fargo - Charlotte, NC