Q2 2026 Mortgage Fraud Risk: Read the Mix

Cotality’s Q2 2026 mortgage fraud risk index rose 9.1% from Q1 as purchases gained share in its application mix, according to the company’s August 4 release. A shift toward higher-risk loan types can lift a portfolio measure even if risk within each type stays unchanged.

For lenders, that is a reason to examine composition before tightening document controls. The increase alone cannot establish a lender’s confirmed fraud incidence or justify denying an individual applicant.

What changed in Q2?

Cotality’s August 4, 2026 release reports these second-quarter results:

What the mortgage fraud index measures

Cotality builds the index from residential applications processed through its LoanSafe predictive scoring technology. Its 2025 report, published September 23, 2025, explains that category trends measure changes in the frequency of risk indicators, rather than the prevalence of each fraud type.

An index reading, a modeled indication and an investigated finding are different measures. An application can trigger an alert that requires explanation without establishing intentional misrepresentation.

Before treating these results as a benchmark, compare your borrower, channel and property composition with the vendor’s application population. As with other housing data and reports, the population behind the number determines which comparisons are useful.

How purchase and refinance mix moves risk

Purchases have historically shown higher modeled risk than refinances in Cotality’s data. Giving purchases a larger weight can therefore raise the combined measure.

Consider a hypothetical illustration, with unchanged indication rates of 1.0% for purchases and 0.5% for refinances:

The first row is 60% × 1.0% plus 40% × 0.5%. The second changes only the weights. Neither group became riskier.

These invented rates illustrate composition; they do not reconstruct Cotality’s proprietary index or estimate how much of its quarterly increase the purchase shift explains. The release supplies no decomposition that would support that calculation.

In a review of lending and origination operations, compare your purchase and refinance results separately, then recalculate the combined result for both quarters using the same shares. Where volume permits, repeat the analysis within occupancy, property type and channel groups. Keep model versions and alert thresholds consistent, or identify changes explicitly. Otherwise, the comparison can mix changes in applications with changes in how they are assessed.

Which risk indicators moved differently?

The annual category comparisons need to be kept separate from Cotality’s report of rising selected transaction, property and occupancy alerts during Q2. Those observations concern particular alerts during the quarter.

For a control decision, identify the discrepancy that needs resolving. An occupancy inconsistency calls for evidence about intended property use; it does not establish that additional income documents would address the concern. An annual category decline likewise cannot establish that a particular flagged file is sound.

Keep category trends, individual alerts and substantiated findings separate in management reporting. Combining them under a single “fraud rate” would blur the distinction between risk signals and review outcomes.

What to examine before expanding document controls

For teams evaluating mortgage technology, build the decision around local review outcomes:

• Establish comparable counts. Count applications and unique flagged files, with the model version, threshold and review rules used in each quarter. Multiple alerts on one file should not become multiple allegedly fraudulent applications.
• Record what reviewers found. Separate cleared discrepancies, unresolved issues and substantiated fraud. Report the denominator for each result. A sample consisting only of flagged files cannot reveal missed cases among unflagged applications; include an independently selected quality-control sample.
• Test the proposed document request. Measure additional substantiated findings alongside cleared alerts, staff time, borrower follow-ups and closing delays. Require evidence that the request resolves the targeted uncertainty before expanding it.

A broader rule might uncover additional problems, but it might also increase review work and document requests. The national index cannot tell a lender which effect will dominate its own pipeline.

Can the results justify automatic denial?

A national risk estimate cannot establish facts about a particular borrower. Measuring a lender’s fraud incidence requires its own defined cohort, case criteria and review outcomes.

Governance provides a separate operational constraint. Fannie Mae’s April 8, 2026 AI/ML letter became effective 120 days later, on August 6, 2026. It covers seller/servicer AI/ML use in originating loans sold to or guaranteed by Fannie Mae, or servicing loans on its behalf. Among its requirements are policies and procedures, accountable policy owners, at least annual policy review and vendor governance.

The letter does not validate Cotality’s model or prescribe denial of flagged borrowers. Have compliance counsel assess any proposed decision rule. Use the portfolio signal to choose where to investigate, then use resolved file evidence to determine which controls need changing.