Understanding Mortgage Risk in Perilous Times
By Kyle G. Lundstedt Managing Director, LPS Applied Analytics
Introduction
The September 15th headline in the Wall Street Journal read “Lehman Files for Bankruptcy, Merrill Sold, AIG Seeks Cash”. A year ago, it would have been unthinkable that such financial titans might ever founder. However, plummeting values for residential mortgage-backed securities, particularly subprime and Alt-A, have put many of the nation’s largest financial institutions under great pressure. Continued declines in house prices, along with increasing delinquencies and foreclosures, have put many mortgage industry participants - investors, servicers, originators - in difficult straits. Hence, it seems timely to review the tools available for understanding mortgage risk.
Today, many secondary marketing professionals are using more granular data and advanced analytical models to understand mortgage risk. At the same time, these sophisticated users are combining information about borrowers, properties and loans in innovative ways to get a detailed view of their risks. In particular, secondary marketing professionals are looking beyond the solutions they have typically deployed, and are scouring the entire mortgage lending spectrum to find sophisticated data and analytic solutions that can help them more effectively assess risk and combat losses. This article will discuss some of these analytical advances, as well as illustrate their uses via several examples.
A Taxonomy of Mortgage Risk Information
For ease of discussion, we will divide up the tools for managing mortgage risk into three categories: borrower, loan and property information. This information can be simply data, or analytics derived from data. For many years, a primary focus for mortgage risk managers has been borrower information, whether credit data alone or analytics such as credit scores. Credit attributes for individual borrowers, as well as credit scores such as FICO which summarize these attributes, offer critical insight into borrower actions that might affect mortgage risk. For instance, data from a credit report may reveal that a borrower has missed payments on a credit card or other debt obligation, which could be a precursor to defaulting on a mortgage. Similarly, a significant drop in a borrower’s FICO score (an analytic from Fair Isaac, Inc.) may indicate potential difficulties in making future payments, possibly leading to default or limiting the ability to prepay.
In the last decade, increased availability of loan information has offered greater insight into mortgage risk. For example, investors previously had access only to weighted-average characteristics for a limited number of variables like coupon or balance. Today, however, secondary marketing professionals can access loan-level data on a much wider range of loan attributes, such as ARM reset date, level of documentation, and occupancy type. This data is often combined with advanced loan-level behavioral models to support highly accurate analytical approaches to measuring mortgage risk.
In addition, property information, including public records data and the property models based on that data, provides a better understanding of mortgage risk. For instance, automated valuation models (AVM) can provide accurate estimates of the current home value, which translates to an accurate understanding of the borrower’s equity position. Similarly, public data records can reveal the existence of additional liens, such as home equity loans or lines, which further reduce a borrower’s equity but are unobserved by the primary lien holder. Understanding collateral value in relation to loan amount thus leads to a better understanding of mortgage risk.
More Granular Data, More Sophisticated Models
Given this taxonomy of information for better understanding mortgage risk, let’s take a concrete example of how mortgage industry participants put this information to work. To begin, we’ll consider how granular data and sophisticated analytics have impacted the asset-liability management (ALM) function in depository institutions.
In brief, the ALM function requires balancing the effective duration of assets (such as credit cards and mortgages) against the duration of liabilities (such as deposits and CDs). In particular, understanding how interest rate changes impact the expected life of complicated instruments like mortgages is critical for the profitability of lenders. Historically, many ALM managers faced significant data and modeling limitations. They were forced to group large portfolios of individual loans into highly aggregated “cohorts” (i.e., all 30 year fixed-rate mortgages), treating them as pools rather than individual loans. The ALM manager then would project the duration of these assets with a prepayment model utilizing a limited set of weighted characteristics for each cohort. This basic approach has the advantage of running quickly on fewer computing resources, and providing a reasonable approximation for portfolio duration.
However, more sophisticated industry participants now can take advantage of loan-level data with greater granularity, as well as more detailed loan attributes as inputs to leading edge models. For instance, grouping together loans with 5.5 percent coupons and loans with 6 percent coupons into a 6 percent coupon weighted average bucket may not be too egregious an approximation. But, recent events have clearly demonstrated that low and no documentation loans do not behave at all like full documentation loans. Similarly, ARMs with different reset dates pose very different risks. Only by using detailed loan-level data, as well as the associated sophisticated analytics can mortgage market participants get an accurate take on mortgage risk.
Combining Borrower, Property and Loan Information
Increasing data granularity and model sophistication is only one way towards a better understanding of mortgage risk. Another important advance is combining different types of mortgage risk information into a single assessment. Let’s look at this approach in the context of setting loss reserves at a financial institution.
In brief, setting loss reserves requires an understanding of how many loans will enter default within the reporting period, as well as what size the loss will be on those loans. In a basic approach to setting loss reserves, as with the ALM function, loan portfolios are typically grouped into cohorts such as fixed-rate loans and adjustable rate loans. This aggregated picture of the portfolio is often applied an institution’s historical “roll-rate matrix”, which shows the average rates at which loans move from one delinquency status to another. Multiplying the initial portfolio composition by the roll-rate matrix repeatedly leads to an approximation of how many loans will be in default at the end of the reporting period. When used with a simple loss severity assumption, this basic approach provides a reasonable approximation of expected losses.
As in the ALM case, increased data granularity and more sophisticated models lead to improved estimates. Utilizing loan-level data allows the institution to account for the changing composition of their portfolio over time. Moreover, applying dynamic transition models enables loss reserves to be examined under a variety of possible future economic scenarios, rather than simply assuming the future will resemble the past. In particular, the enormous impact of recent local area (MSA)-level house price changes has clearly demonstrated the need for scenario-driven assessment of mortgage risk.
Consider, however, the impact of adding property data and valuation models to this process. First, the institution’s loan data can be matched to public records data to reveal the presence of “silent seconds”, vastly improving the estimated “combined LTV” for each loan. Then, automated valuation models can be applied to transform the combined LTV into a current combined LTV. By adding public records data and valuation information to loan-level performance data, it’s possible to achieve much more accurate loss reserve estimates.
Mortgage Risk Information Across the Enterprise
Utilizing better mortgage risk information is not applicable just to the ALM or loss reserving functions; it can bring efficiencies to a wide range of additional mortgage market processes. For instance, investors benefit from improved portfolio management and risk/return analysis. Mortgage servicers also are using more advanced approaches to help with benchmarking to market and improving loss mitigation efforts. Similarly, originators are pricing more accurately for risk and engaging in more targeted marketing efforts. In summary, applying borrower, loan, and property information in all areas of the mortgage market is leading to improved profitability and reduced risk for many leading participants.