Keys, Mukherjee, Seru & Vig (2010)

Citation: Keys, B. J., Mukherjee, T., Seru, A., & Vig, V. (2010). “Did Securitization Lead to Lax Screening? Evidence from Subprime Loans.” Quarterly Journal of Economics, 125(1), 307-362.

This paper provides causal evidence that securitization adversely affected lenders’ screening incentives in the subprime mortgage market—a central question in understanding the 2008 financial crisis.

The identification problem:

The fundamental challenge in studying securitization’s effects is endogeneity: loans that are securitized may differ systematically from those retained. Correlations between securitization and poor loan performance could reflect selection rather than causation.

The identification strategy:

The authors exploit an arbitrary rule of thumb in the subprime market: loans to borrowers with FICO scores of 620 or above were substantially easier to securitize than loans just below 620. This creates a discontinuity:

• At FICO 620+, securitization is ~100% higher than at 620-
• Loan terms (interest rates, LTV ratios) are identical on either side
• Borrower demographics are identical on either side

If lenders screen similarly around this threshold, default rates should be smooth (or actually lower above 620, since higher FICO should mean lower risk). Any discontinuity in defaults must reflect the causal effect of differential securitization on screening.

flowchart TD
  subgraph Key Finding
    E[Easier to securitize at 620+]
    E --> L[Lenders screen less]
    L --> D[620+ loans default MORE]
  end
  
  subgraph The Paradox
    P1[Higher FICO should mean lower risk]
    P2[But 620+ defaults 10-25% MORE]
    P1 --> W[Something is wrong]
    P2 --> W
    W --> S[Securitization caused lax screening]
  end

Key findings:

1. Doubling securitization volume is associated with 10-25% higher defaults. Around the 620 threshold, loans with slightly higher credit scores default significantly more—the opposite of what credit models predict.

2. The effect is driven by soft information. For “low documentation” loans (where soft information like income verification is missing), the screening differential is stark. For “full documentation” loans, where hard information substitutes for soft screening, the effect disappears.

3. Natural experiment confirms causality. When Georgia and New Jersey passed anti-predatory lending laws that temporarily reduced securitization in those states, the screening differential around 620 vanished. When the laws were amended and securitization resumed, the differential returned immediately.

4. Lenders responded to incentives, not blindly following rules. The natural experiment shows that when securitization became harder, lenders screened more carefully even for 620+ loans.

Implications for securitization design:

The findings highlight a fundamental tension in securitization:

• Distance creates moral hazard. Securitization increases the distance between borrowers and ultimate investors. Information loss occurs at each step.

• Hard information is not enough. Relying solely on quantifiable metrics (FICO scores, LTV ratios) misses “soft information” that human judgment captures. When lenders know they won’t hold the loan, they don’t bother collecting this soft information.

• Skin in the game matters. The results suggest that existing securitization practices did not require originators to hold sufficient risk. “If lenders were in fact holding on to optimal risk where it was easier to securitize, there should have been no differences in defaults around the threshold.”

On the role of models:

The paper concludes with a prescient warning about overreliance on default models:

“Our research suggests that by relying entirely on hard information variables like FICO scores, these models ignore essential elements of strategic behavior on the part of lenders… The formation of a rule of thumb, even if optimal, has an undesirable effect on the incentives of lenders to collect and process soft information. As in Lucas (1976), this strategic behavior can alter the relationship between observable borrower characteristics and default likelihood.”

This is a Lucas Critique for credit markets: when models based on historical data are used to make lending decisions, agents’ strategic responses can break the very relationships the models depend on.

Key contribution:

This paper provides some of the cleanest causal evidence that securitization, as practiced in the subprime market, created moral hazard that led to worse loan quality. It distinguishes between:

• The potential benefits of securitization (risk transfer, liquidity)
• The actual implementation that failed to maintain screening incentives

The lesson is not that securitization is inherently bad, but that its design matters enormously. Loans that depend heavily on soft information require mechanisms to preserve screening incentives—either through risk retention, reputation, or improved observability.