Real estate quantified; benchmarking commercial Mortgage PD: an update

RMA Journal, The, Sept, 2002 by George J. Pappadopoulos

A series on real estate investing and risk management premised on the belief that a more analytic approach delivers reduced risk-based capital requirements, improved profitability of real estate investment activities, and growth in the capital allocated to real estate.

The Basel II guidelines provide a carrot to banks with more advanced credit rating systems by reducing regulatory capital requirements. Unfortunately, many banks realize that without substantial work, their credit risk rating systems won't meet these guidelines. A major problem is that a consistent measure of credit risk across the entire portfolio does not exist; although available risk management systems are helpful for a number of asset classes, none provides an adequate solution for real estate credit risk.

The issue lies in the shortfall of appropriate inputs. The credit models are certainly mathematically robust, but the real-estate-specific insights are lacking. Although traditional fixed-income markets benefit from a wealth of high-quality data, information regarding commercial mortgage defaults is, unfortunately, scarce.

This lack of actual data on commercial mortgage defaults begs an alternative approach to understanding expected probability of default. Because individual property market and property type cycles are a major driver of default, our focus is to model credit risk through three key systematic risk factors: 1) market volatility, 2) expected growth of NOI and value, and 3) loan structure protection. The mathematical combination of these three factors allows the development of an informed opinion about the relative riskiness of different loan structures in different markets and property types. (1) Most important, this metric can be calibrated to known historical default outcomes and then used to forecast defaults on a differentiated basis.

The data used for this process comes from a seminal study of defaults originally conducted by Mark Snyderman (2), and updated by Esaki, L'Heureux, and Snyderman. (3) The study tracks average cumulative default rates for a commercial mortgage pool of more than 15,000 loans garnered from American Council of Life Insurers (ACLI) data. The important outcome is a portfolio-level benchmark for the overall cost of default. This study is appropriate for our uses as it tracks each individual loan throughout its life span and currently provides the best benchmark for observing actual cumulative loan default.

By modeling a "market basket" of loans similar to those of the Snyderman study, we were able to fit the outcome of the broad-based, actual mortgage pool extremely well. For the actual loan pool, 18.1% of the loans defaulted, and our benchmarking process exhibited an overall weighted average frequency of default of 18.06%. Further, our modeled expected loss was only 5 basis points below the actual losses of 6.82%.

Obviously, we would expect our aggregate results to be close, but nor perfectly matched. The real test, however, comes from exploring our fit at a disaggregated level. Instead of modeling only the cumulative default rate over the entire life of each origination, we also calibrated the periodic default rates through each quarterly time period. The big benefit of the methodology is that loan seasoning and loan survival factors are more accurately reflected. Loan seasoning represents changes in the periodic default rate as a function of the time lapsed since loan origination. Generally, the default rate will be low during the early years of the loan term, will increase until default rates peak around year 6 or 7, and will then decline in the later years. Obviously, loan terms and economic factors influence the particular pattern. The next factor, loan survival, reflects the pattern of declining default rates as the total percentage of loan pool defaults increases. That is, the default rate for a loan pool that h as already had 40% of its loans go into default will be less than the default rate of a pool in which only 5% of the pool has defaulted. This is because some borrowers have much greater unobservable costs of default (psychological, etc.). Or, stated more simply, a certain percentage of the original loan pool is very unlikely to default, no matter what.

Overall, the process yielded results that were quite striking. Figure 1 outlines cumulative default curves for a range of origination cohorts outlined by the Snyderman study and contrasts each curve with our modeled results. These curves depict the total percentage of loans in that cohort that defaulted at each time period since origination. Once again, the results are not a perfect match, but are, nonetheless, very compelling. Many cohorts exhibit an extremely tight fit, especially during the first four or five years since origination.

The modeled results exhibit the classic "S-shaped" pattern we would expect for timing of defaults. That is, mortgages are less likely to default during the very early life of the loan, and frequencies then increase more rapidly until we reach a peak in the three-to-seven-year range. Then, of course, default speeds tail off during the latter part of the loan term. Interestingly, other than a couple of sizable long-term misses, the model captures many of the subtle differences actually experienced by the sample loan pool in both level and timing (changes in slope).