A meter for mortgage risk.
One can't help but wonder if the United States might have avoided a bubble and subsequent collapse in home prices eight years ago if the National Mortgage Risk Index (NMRI) had been available in 2000. Although looking back a decade or more may seem like idle speculation, we'll know in the years ahead if this index has helped the mortgage finance industry avoid another housing bubble. [paragraph] Time will tell whether a lesson has been learned. However, in the interim, folks in the mortgage and housing industries need to know more about monitoring risk, and the advent of a mortgage risk index is a significant step in the right direction. [paragraph] This article provides an overview of the NMRI, which was launched in December 2013 by the International Center on Housing Risk at the American Enterprise Institute (AEI), Washington, D.C. We'll look at the methodology and structure of the index, what goes into it, how it works and who is behind it.
Background and methodology
Many of the indexes that financial analysts, traders, secondary market executives and economists follow are created from survey results.
Examples abound: the Institute for Supply Management (ISM) Index, the Conference Board's U.S. Consumer Confidence Index, the University of Michigan Consumer Sentiment Index, the National Association of Home Builders (NAHB)/Wells Fargo Housing Market Index, and various regional manufacturing surveys such as the Empire State Manufacturing Index.
Most of them are diffusion indexes based on survey results. A diffusion index shows the number of people saying conditions are better compared with those saying worse.
It does not weigh for size of firm, or the degree of better/worse. As a result, the index may underestimate or overestimate the strength based on input from a small number of respondents. These indexes' methodologies are not based on hard numbers but on respondents' often-subjective assessments.
That's not the way the NMRI works. This index relies on hard data provided monthly (with a one-month lag) by Ginnie Mae for Federal Housing Administration (FHA) and U.S. Department of Agriculture (USDA) Rural Housing Service (RHS) loans; by Inside Mortgage Finance for Fannie Mae and Freddie Mac loans; and by loans guaranteed by the Department of Veterans Affairs (VA), including both purchase loans and refinances.
The information provided for each loan includes the three key risk characteristics: 1) the combined loan-to-value (CLTV), 2) the FICO[R] credit score and 3) the debt-to-income (DTI) ratio--supplemented by other risk characteristics such as loan term (15-year vs. 30-year), type (adjustable rate vs. fixed rate) and purpose (primary residence vs. investor).
Each and every loan received monthly is run through a model that places it into a "periodic table" containing 320 risk baskets based on CLTV, FICO score and DTI, with adjustments made to take the other risk characteristics into account.
The NMRI measures how the loans originated in a given month would perform if subjected to the same stress event as loans were placed under in the 2008 financial crisis.
It is not like the stress tests the Federal Reserve conducts on the largest banks. Rather, this test is similar to those performed to determine an automobile's crashworthiness at 35 mph or a structure's ability to withstand, say, 130-mph hurricane-force winds.
The stressed default rate is calculated in a series of steps. First, a matrix of benchmark default rates is created. It is based on home purchase loans acquired by Freddie Mac in 2007. (Freddie Mac data is used, as it contains greater detail than Fannie Mae's.) All the Freddie loans are 30-year fixed-rate mortgages (FRMs), fully amortizing, fully documented and owner-occupied.
The loans are sorted into 320 risk buckets by combinations of the three risk characteristics already mentioned. Then the share of those loans that had defaulted by Dec. 31, 2012, is calculated. The calculated default rates therefore represent the default experience for owner-occupied, fully documented and fully amortizing 30-year FRMs purchased by Freddie Mac in 2007.
The next step applies the benchmark default rate to each newly originated loan in a given month. Any applicable risk adjustments are then applied to the loan's bucket-level stressed default rate. For example, because a 15-year loan has only 50 percent of the default propensity of a 30-year loan, the bucket-level stressed default rate for a 15-year loan would be reduced by 50 percent. This process is repeated for each of the newly originated loans added each month.
The final step is to calculate the value of the mortgage risk index for a given month as the average of the stressed benchmark default rate across all the loans.
This composite NMRI covers the entire dataset, thus the term "national" in the index name. Subindexes are also published for FHA-, VA- and RHS-guaranteed loans, and for Fannie/Freddie securitized loans for both agencies individually as well as combined.
Figure 1 shows the latest composite index values along with values for Fannie Mae and Freddie Mac combined and for the FHA and VA.
In addition to the comparison with 2007, a comparison is made to 1990 vintage loans. Loans from this period were chosen because 1990 was the last time traditional underwriting standards were broadly prevalent. Estimates for the 1935-1955 FHA vintages are also shown, as these were low-risk loans undertaken during a period of rapid growth in homeownership.
Figure 1 shows the data for September 2014. The latest composite index value is 11.4 percent. This value compares with a 6 percent default rate on the 1990 book and a 19 percent default rate for the 2007 vintage.
Also shown are the latest index values for Fannie and Freddie combined, the VA and FHA--6 percent, 11.2 percent and 24 percent, respectively. This compares with the 4 percent and 13 percent default rates in 1990 and 2007, respectively, for the two government-sponsored enterprises (GSEs) and the 15 percent and 33 percent rates in 1990 and 2007, respectively, for the FHA.
The poor performance of FHA loans that we now think of as normal represents a departure from the sound underwriting in the early years of FHA's existence. The default rate for FHA's 1935-1955 book of loans was 3 percent--more than 20 percentage points below the latest risk index reading.
As noted, the risk buckets each represent a combination of FICO score, CLTV and DTI. Five risk buckets are shown in Figure 2, with the associated range for FICO score, CLTV and DTI for each bucket. The buckets range from very low to very high risk, depending on the default rate for the bucket.
For example, the very-low-risk bucket consists of loans with FICOs of 770 and above, CLTVs of 61 percent-70 percent, and total DTIs at or below 33 percent. The last column in the figure shows the default rates for each of these five buckets. The stressed default rate on the very-low-risk bucket just described is 0.8 percent.
The very-high-risk bucket at the other end of the spectrum consists of loans with FICOs of 620-639, CLTVs above 95 percent and total DTIs in excess of 50 percent. Almost half of the loans in this bucket, all of which were originated in 2007, ended up defaulting by the end of 2012.
The extremely wide range of default rates across the five buckets illustrates the value of systematically tracking the mix of loan originations.
In addition to the national data, State Mortgage Risk Indexes (SMRIs) are calculated monthly, and metropolitan area indexes are gradually being added. At this writing, only metro areas in California are available. The exact same stress-test method used to calculate the national index is used for the state and metro indexes.
Figure 3 provides additional detail on DTIs in September. It looks at the percentage of loans with total DTI, by program, that exceeds 38 percent, 43 percent and 50 percent. For example, the composite index indicates that 45 percent of all DTIs in the 4.3 million loan database exceed 38 percent, while 15 percent of FHA loans have DTIs greater than 50 percent. loans for the July-to-September period. A grouping of the five states with the lowest and highest scores are shown with their respective index values.
Figure 5 displays the risk index for six areas in California, along with volatility, fraud and key demographic information for each region. Of the six areas, metropolitan San Francisco has the lowest risk index value (8.8 percent as of September), the highest median income and the second-smallest share of blacks and Hispanics in the area population.
Conversely, the Central Valley area has the highest risk index value (14.8 percent as of September), the lowest median income and the third-largest concentration of blacks and Hispanics.
This comparison highlights that within California, lower-income and heavily minority areas will be disproportionately exposed to default risk should we experience another financial crisis.
As for house-price volatility in the area, the difference in the price change from the best year to the worst over 1979-2014 was 66 percentage points in Riverside-San Bernardino, or nearly double the national average for this measure of volatility. The other California metro areas also had price volatility above the national average. The Volatility Index is from Seattle-based Zillow. Figure 5 also includes a fraud index from Agoura Hills, California-based Interthinx, which shows the incidence of fraud for four of the six regions in California and the nation.
Figure 6 provides additional insights into housing risk in the aforementioned six market areas in California. It displays: alphabet-based cycle risk scores and changes in those scores, from Irvine, California-based John Burns Real Estate Consulting; the Intrinsic Home Value Index (also from John Burns), showing current price versus intrinsic value, assuming 6 percent mortgage; sustainable home-price data from New York-based Fitch Ratings, which shows current prices relative to sustainable values; and "bubble watch" data from San Francisco-based Trulia Inc., showing prices relative to fundamentals. The up/down ratios represent the proportion of houses that experienced a price increase or price decrease, respectively, since the previous month and are provided by Natick, Massachusetts-based Weiss Residential Research. The national averages are weighted for up/down ratios based on 70 metro areas provided by Weiss. It measures the change in house prices relative to fundamentals and up/down ratios for January 2012, June 2013 and June 2014.
The overall message is that risk in California is relatively high and rising. For example, look at the Riverside-San Bernardino region. In December 2012, the market was rated an A+ on housing cycle risk. However, a year and a half later, the risk had risen, lowering the grade to a C.
Concerning the potential for a bubble to develop, house prices in the Riverside area stand as much as 20 percent above fundamental values; other California metro areas are estimated to be overvalued as well, in contrast to the absence of overvaluation for the national average.
The International Center on Housing Risk
The source of the index--AEI's International Center on Housing Risk--seeks to improve the operation of housing markets, mortgage lending practices, borrower decisions and housing policy by providing an unprecedented set of tools for measuring housing and mortgage risk. The center's ultimate objectives are to foster sustainable homeownership and stable housing and mortgage markets.
Commenting on the NMRI, Mike Fratantoni, chief economist of the Mortgage Bankers Association (MBA), said, "I think it can be valuable to have an objective, consistent measure of credit risk. They have done a good job compiling performance data."
Fratantoni added, "My only critique would be that they tend to be focused on standards from 20 to 30 years ago as the baseline. I do think that the industry learned how to lend to more expansive criteria successfully. The question is whether the right baseline is 1981, 1991 or 2001. No one is arguing that 2005-2007 is the right baseline."
The NMRI and its subindexes for the preceding month are released on the last Monday of each month. The release is accompanied by a conference call held to announce the latest values and discuss the data. In advance of the call, center co-directors Edward J. Pinto and Stephen D. Oliner prepare a detailed slide deck that covers not only all of the key numbers and figures outlined in this article, but also includes a broader discussion about risk that allows participants to raise questions and related issues with the two presenters.
Pinto and Oliner, who developed the NMRI, have complementary skill sets and experience.
Pinto is a resident fellow at AEI. He was an executive vice president and chief credit officer for Fannie Mae until the late 1980s. After leaving Fannie Mae, he worked with the Mortgage Information Corporation (now CoreLogic, Irvine, California) to develop the LoanPerformance database. He has conducted re search on the role of the government in the lead-up to the financial crisis; on the policies, practices and performance of the FHA since 1934; and on the history of mortgage risk and collateral valuation in the United States over the last 110 years.
Oliner is a resident scholar at AEI and a senior fellow at the University of California Los Angeles' (UCLA's) Ziman Center for Real Estate. Before joining AEI, he was an economist at the Federal Reserve Board, where he was closely involved in the Fed's analysis of the U.S. economy and financial markets. Oliner is a student of monetary policy and maintains an active research agenda that focuses on real estate issues and the economy's growth potential.
The mortgage risk index is the first of three indexes the International Center on Housing Risk will create, monitor and publish. On the agenda are a Collateral Risk Index (CRI), a Capital Adequacy Index (CAI) and an assortment of additional metro-level data.
The CRI is designed to measure the growth in collateral risk due to deviations between the market price of a home and its intrinsic value. The index is based on such fundamental factors as land values, rents, construction costs and income growth. A composite CRI will cover 10 major markets across the country. The CRI and its subindexes are slated to be launched in 2015.
The CAI is being designed to measure capital adequacy in the mortgage finance market. It will address capital reserves at major financial institutions, and will also make its debut in 2015.
Pinto and Oliner will be assisted in the creation of the CRI and CAI by Morris Davis, professor and Paul V. Profeta Chair in real estate, and director of the Center for Real Estate Studies at the Rutgers Business School in Newark, New Jersey; and by Michael F. Molesky, an independent economist and expert on mortgage default risk and insurance regulation, based in the Raleigh-Durham, North Carolina, area.
House-price volatility is a fact of life in some markets, but these markets represent only a small number of U.S. cities and metro areas. Most markets experience modest volatility, as indicated in Figure 5. Lenders and investors need to know which is which to minimize the risk of defaults and losses.
Much has been learned from American history about earlier real estate collapses since the 1920s, their causes and effects. As a result of this knowledge, and the International Center on Housing Risk's new measurement instruments created to monitor such risk, it will be interesting to watch and ascertain if this work can reduce, if not preclude, a third major nationwide real estate market collapse.
If that is the outcome over the next decade, all parties will prove winners--including lenders, borrowers, government and taxpayers. Certainly, such an outcome is the center's great hope.
Tom LaMalfa is a 37-year veteran of mortgage market research, whose focus in recent years has been on federal housing policy, and president of TSL Consulting in Cleveland Heights, Ohio. He can be reached at firstname.lastname@example.org.
FIGURE 1 CALIBRATING MORTGAGE SAFETY Latest Latest 1935-1955 NMRI--Purchase Loans Date Value Vintages (est.) Composite Index Sept. 11.4% NA Fannie Mae/Freddie Mac Sept. 6.0% NA FHA Sept. 24.0% 3% VA Sept. 11.2% NA 1990 2007 NMRI--Purchase Loans Vintage (est.) Vintage (est.) Composite Index 6.0% 19.0% Fannie Mae/Freddie Mac 4.0% 13.0% FHA 15.0% 33.0% VA NA 15.0% SOURCE: AEI International Center on Housing Risk NOTE: NMRI = National Mortgage Risk Index FIGURE 2 STRESSED DEFAULT RATES Risk Bucket FICO[R] CLTV Very Low [greater than or equal to] 770 61-70% Low 720-769 76-80% Medium 690-719 81-85% High 660-689 91-95% Very High 620-639 >95% Risk Bucket Total DTI Default Rate Very Low [less than or equal to] 33% 0.8% Low 34-38% 4.2% Medium 39-43% 9.3% High 44-50% 22.7% Very High >50% 45.8% SOURCE: AEI International Center on Housing Risk NOTE: Default rates represent cumulative defaults through year-end 2012 for Freddie Mac's 2007 vintage of acquired loans. The loans included in the calculation are all primary owner-occupied, 30-year fixed-rate, fully amortizing, fully documented, home purchase loans. FIGURE 3 ADDITIONAL DETAIL ON TOTAL DEBT-TO-INCOME (DTIs) September 2014 Total DTI Total DTI Total DTI >38% >43% >50% Composite 45% 23% 5% Fannie/Freddie 35% 13% [approximately equal to] 0% FHA 65% 42% 15% RHS 40% 13% [approximately equal to] 0% VA 55% 35% 12% SOURCE: AEI International Center on Housing Risk FIGURE 4 RANGE OF STATE MORTGAGE RISK INDEXES (SMRIs) FOR HOME PURCHASE LOANS ACROSS STATES July-September 2014 Low Five States (Avg.) High Five States (Avg.) Composite 8.7% 13.3% Fannie/Freddie 5.3% 6.7% FHA/RHS 19.7% 24.0% VA 9.5% 11.9% FIGURE 5 HOME PURCHASE MORTGAGE RISK INDEX (MRI) IN MAJOR CALIFORNIA METRO AREAS Price Fannie/Freddie/FHA MRI Volatility September 2014 Best Year-- November (level and change Worst Year Metro or Region 2012 from prior month) (1979-2014) San Francisco Metro 7.4% 8.8% (+0.2 ppt) 47 ppt San Diego CBSA 8.4% 9.9% (+0.8 ppt) 49 ppt Los Angeles CBSA 9.2% 9.9% (+0.2 ppt) 48 ppt Sacramento CBSA 10.4% 11.5% (unchanged) 44 ppt Riverside-San Bernardino CBSA 13.3% 14.5% (+0.9 ppt) 66 ppt Central Valley Region 14.6% 14.8% (+0.1 ppt) NA California Average 10.3% 11.4% (+0.3 ppt) NA National Average 10.7% 11.1% (+0.3 ppt) 34 ppt Fraud Demographics 2013: Q4 Area Median National: Income/Black or Metro or Region Q1:12 = 100 Hispanic Share San Francisco Metro 145 $101,200/30% San Diego CBSA NA $72,300/37% Los Angeles CBSA 149 $61,900/51% Sacramento CBSA NA $70,900/27% Riverside-San Bernardino CBSA 135 $62,600/56% Central Valley Region 145.5 $53,000 (approx.)/49% California Average 139 $69,600 / 44% National Average 101 $64,400 / 29% SOURCES: AEI International Center on Housing Risk, Zillow, Interthinx, Department of Housing and Urban Development (HUD), U.S. Census Bureau FIGURE 6 HOUSING RISK IN MAJOR CALIFORNIA METRO AREAS Indexes Housing Cycle Risk August 2014 Metro or Region December (change 2012 prior month) San Francisco Metro A + C+ (unch.) San Diego CBSA A C- (C) Los Angeles CBSA A C/C- (unch.) Sacramento CBSA A+ B- (unch.) Riverside-San Bernardino CBSA A+ C (unch.) Central Valley Region A C+ (B-) California Average A+ / A C+ (unch.) National Average A / A- B- (unch.) Indexes Intrinsic Sustainable Bubble Home Value Home Price Watch Metro or Region September 2014 2014:Q1 2014:Q3 San Francisco Metro +10% +10-15% +10% San Diego CBSA +9% +5-10% +3% Los Angeles CBSA +17% +10-15% +15% Sacramento CBSA +6% NA -3% Riverside-San Bernardino CBSA +12% +15-20% +11% Central Valley Region +8% NA N/A California Average NA +10-15% NA National Average +3% Sustainable -3% Indexes Up/Down Ratios Metro or Region January June June 2012 2013 2014 San Francisco Metro 88% / 8% 94% / 4% 76% / 19% San Diego CBSA 96% / 2% 92% / 5% 66% / 28% Los Angeles CBSA 93% / 4% 90% / 6% 68% / 27% Sacramento CBSA 85% / 10% 94% / 2% 69% / 26% Riverside-San Bernardino CBSA 91% / 4% 86% / 10% 73% / 23% Central Valley Region 82% / 11% 89% / 7% 75% / 20% California Average 88% / 7% 91% / 6% 73% / 22% National Average 73% / 18% 74% /17% 66% / 28% SOURCES: AEI International Center on Housing Risk, John Burns Real Estate Consulting, Fitch Ratings, Trulia Inc., Weiss Residential Research