Valuing MBS with AI and Prepayment Hazard Models

This article provides a comprehensive overview of how mortgage servicers and housing economists accurately value mortgage-backed securities (MBS) during volatile interest-rate shocks. By integrating prepayment-speed hazard models with regional home-price artificial intelligence (AI), financial professionals can predict borrower behavior, estimate home equity shifts, and calculate precise cash flows. Together, these advanced analytical tools mitigate duration risk and protect institutional portfolios from sudden market fluctuations.

The Challenge of Interest-Rate Shocks

Mortgage-backed securities are uniquely difficult to value because their underlying cash flows are uncertain. Unlike standard bonds, homeowners have the option to prepay their mortgages at any time, usually through refinancing or selling their homes.

When interest-rate shocks occur—meaning rates suddenly plummet or spike—borrower behavior changes drastically:

  • Falling Rates (Contraction Risk): Homeowners rush to refinance into lower rates. MBS investors receive their principal back much faster than expected, forcing them to reinvest that capital in a lower-yield environment.
  • Rising Rates (Extension Risk): Refinancing grinds to a halt. Homeowners hold onto their existing low-rate mortgages longer than anticipated, locking MBS investors into below-market yields for an extended period.

To price MBS accurately during these shocks, economists and servicers must predict exactly when and where these prepayments will happen.

Prepayment-Speed Hazard Models Explained

Mortgage servicers use prepayment-speed hazard models—originally adapted from biological survival analysis—to calculate the probability that a specific loan will terminate (prepay) at any given month, assuming it has survived up to that point.

Instead of treating a pool of mortgages as a single homogenous block, hazard models evaluate individual loan characteristics. Key baseline variables include:

  • Refinancing Incentive: The spread between the borrower's current mortgage rate and the prevailing market rate.
  • Loan Age (Seasoning): New loans rarely prepay immediately, while older loans follow predictable turnover curves.
  • Burnout: If rates have dropped previously and a borrower did not refinance, they are statistically less likely to do so during the next rate drop.

During an interest-rate shock, the hazard model dynamically adjusts the conditional prepayment rate (CPR) by calculating how the sudden shift in the rate spread interacts with these baseline loan characteristics.

Regional Home-Price AI

A borrower’s desire to refinance is meaningless if they lack the home equity to qualify for a new loan. This is where housing economists deploy regional home-price artificial intelligence.

Traditional valuation models rely on lagging, broad-stroke national or state-level home price indices. Modern AI models, however, use machine learning to process vast amounts of granular, real-time data at the ZIP-code or neighborhood level. These AI systems ingest:

  • Local employment trends and migration patterns.
  • Real-time property listings and days-on-market metrics.
  • Climate risk data and local zoning changes.
  • Satellite imagery and tax assessment records.

By analyzing these localized factors, the AI forecasts precise regional home-price appreciation or depreciation. This allows economists to dynamically update the Loan-to-Value (LTV) ratios of individual mortgages within an MBS pool.

Combining the Models to Value MBS

To value an MBS across interest-rate shocks, servicers and economists merge the outputs of the AI and the hazard models into a unified valuation engine.

First, the regional home-price AI projects future home values, which determines each borrower's updated equity position (LTV). This equity metric is then fed directly into the prepayment-speed hazard model. For example, if interest rates plummet, the hazard model might initially predict a massive wave of refinancing. However, if the AI indicates that a specific region is experiencing a localized housing downturn, borrowers in that area will be "underwater" (owing more than the home is worth) and unable to refinance. The hazard model adjusts the prepayment speed downward for those specific loans.

Finally, financial analysts run Monte Carlo simulations—generating thousands of potential future interest-rate paths (shocks). For each path, the combined AI-hazard framework projects the exact monthly cash flows (principal and interest) the MBS will generate. By discounting these projected cash flows back to the present day and calculating the Option-Adjusted Spread (OAS), mortgage servicers and economists arrive at a highly accurate valuation for the security, allowing them to hedge their risks effectively.