A New Approach to Analyzing Default Risks

A global asset manager with a multibillion-dollar MBS portfolio uses IM-Suite to quickly reevaluate its MBS portfolio strategy in light of the challenging dynamics in the secondary mortgage market during 2008. The existing methodologies for measuring mortgage loan default risks based on utilization of CDS spreads are proving to be misleading in the dynamic market environments of the mortgage bubble. The required analysis needs to evaluate default risks at the individual loan level.

Background

The manager needs to access large amounts of data at the individual loan level from multiple loan servicers and issuers. In addition, the manager wants to apply specific market economic data to the analysis of each loan.

With data in hand the manager wants to provide the following analysis:

• Develop complex stress tests for very large mortgage loan pools with hundreds of thousands of mortgages.
• Run simulations of different macroeconomic factors to better understand default rates under differing regional macroeconomic scenarios.

Data Problem

To solve the problem, the client will need to integrate large data sets from a variety of sources, including servicers, government agencies such as Freddie and Fannie, and regional econometric data in both static and real-time data modes. The client’s existing data model and data warehouse are inadequate to handle the complexities and size of the databases required to properly access risks at the individual loan level.

Solution

Data Agent’s security master can manage an extensive set of robust data sources pertinent to the clients’ needs. Our integrated data loaders and adapters enabled the client to quickly build and populate the Data Mart with the required data sources. The data set for this particular project included the following:

• Market data vendors (Bloomberg, Reuters, Compustat, S&P)
• Synthetics (Markit Partners)
• MBS analytical data (Fannie Mae, Freddie Mac, HUD, FHA)
• Economic statistics from MSCI, IMF, and U.S. government data sources (census data, Bureau of Labor Statistics, etc.)

The Need for a New Analytical Approach

The client lacks the internal systems and solutions needed to evaluate the multiple dimensions of risk in its MBS portfolio. In addition, it lacks the analytical structure to evaluate hundreds of thousands of individual mortgage loans at the individual loan level. The client will need to run complex analytics and visualizations on data to better determine the nature and scope of the portfolio risks and to gain a clearer insight into a realistic value of its overall portfolio.

The historical approach for valuing MBS portfolios was to utilize CDS spreads as a proxy for default risk. The credit crisis of 2008 invalidated this model as a prudent method of determining default risks and thus threw into question the valuation of MBS portfolios.

The historical model for MBS valuation based on CDS spreads:

Default probability: MBS risk = ƒ (CDS spread, rating)

The client wanted to evaluate its portfolio on an individual loan characteristic basis. The new model would determine default probability using historical macroeconomic data. It was understood that covariance matrix can be reduced to a few factors and prove effective in analyzing subprime loans and alt-A loans. It was also evident through multiple academic studies that it was possible to predict the default probability of a loan at a specific regional level with 85 percent accuracy.

The client’s model for MBS valuation:

Default probability = ƒ (home ownership rate, LTV90, FHA shares, employment growth rate, price appreciations, unemployment rate, industry exposure, large company exposure)

Problem

The client’s problem would require extensive computing power to determine transition matrix and covariance matrix among multiple economic and consumer measures, especially for +1M loans. Additionally, the client would need to run multiple stress tests or scenarios of different macroeconomic factors to better understand default risks at both the regional and individual loan basis.

Solution

The client deployed Data Agent’s Data Analytics solution built on the Microsoft Business Intelligence platform. The relational data warehouse and Cube technology capabilities of Data Analytics provided support for more than 400 million consumer and mortgage loans and over 100 million consumers including relevant macro and regional economic and demographic factors. Microsoft made it possible to do this using SQL server analytical services.

Additionally, our client needed to efficiently manage the hierarchical and recursive structure of credit risk: issuer, insurers, investment vehicles, consumer credit scores. The built-in business intelligence of Data Analytics is based on Microsoft’s Cube technology and provided the client with the following abilities:

• Support for the hierarchical structure of MBS, tranche, and loan analytics and consumer data
• Support for combining multiple loans belonging to a property
• Flexible classification of stress test results at the consumer or loan level
• Support for both assets and liabilities of consumers
• Support for multiple models of default probability rate and loan loss predictions
• Dynamic IRR calculation for loan portfolios

Additionally, Data Analytics allowed the client to run multidimensional advanced data visualizations that helped pinpoint specific default characteristics across various loan dimensions.

• Loan analytics were decomposed across multiple dimensions, with drill-down capability to the loan level.
• MBS default rates and predictive loan loss were aggregated from the underlying loans.

Benefits

We worked closely with the client to quickly incorporate all its data requirements into Data Mart and assisted with the integration of its MBS analytics. Using IM-Suite, the client was able to better understand the underlying default risks in its multibillion-dollar MBS portfolio and was able to successfully restructure its MBS portfolio to avoid the huge losses incurred by other major MBS holders during 2008.