GSE investors better beware underlying risks

Investors taking part in the federal government’s efforts to transfer mortgage risk from its balance sheet to the private sector should know that it might not be a fair trade.

New analysis of data released jointly by the American Enterprise Institute and the Federal Housing Finance Agency shows that the government has only selectively released the data at the core of these deals, making assets appear substantially less risky than what’s actually being undertaken.

Andrew Harrer/Bloomberg

As a result, the private sector may be severely underpricing risk. The only way to solve this dilemma is for the government to release all its loan-level data to bring greater transparency to the market place.

The story starts with the 2008 bailout and conservatorship of Fannie Mae and Freddie Mac, the two government-sponsored mortgage giants. Backed by taxpayers and under the control of their conservator, the FHFA, Fannie and Freddie have remained the $5 trillion behemoths that securitize half of all mortgages.

One of FHFA’s strategic goals (from 2014) for the two government-sponsored enterprises has been to “reduce taxpayer risk through increasing the role of private capital in the mortgage market.”

To this end, FHFA required the GSEs to undertake the credit risk transfer (CRT) program in 2013. The CRT program allows private parties to purchase credit-risk investments from Fannie and Freddie, and assume the interest rate risk and a portion of the credit risk.

Ostensibly, to help investors participate in CRTs and appropriately price the risk, the FHFA required the GSEs to release historical loan-level data. These datasets, known as the loan performance, or LP, data, contain origination characteristics and a performance history on millions of loans originated between 1999 and today.

The loans are “plain vanilla” loans. Meaning, they are fully amortizing, fully documented, fixed-rate mortgages. However, this data excludes loans with risky product features, such as low or no documentation, loans with an adjustable-mortgage rate, and many affordable housing loans “to make the dataset more reflective of current underwriting guidelines,” according to Fannie.

Yet investors and even researchers who continue to rely on the LP data to analyze the run-up to the financial crisis never seemed to have suspected that Fannie’s and Freddie’s censoring would go much further than as described by the FHFA. That was a mistake.

The LP data excluded nearly half of the 30-year primary owner-occupied, plain vanilla loans originated in 2006 and 2007. These cohorts are of particular interest to modelers due to the extreme levels of stress these loans were exposed to during the Great Recession and collapsing home prices.

To compound the issue, the level of censoring rises with risk. For example, a whopping 93% of 30-year POO plain vanilla loans were never disclosed in the two crucial stress years, according to internal analysis. And these were for the riskiest loans: those with a combined loan-to-value ratio exceeding 95% and credit score below 660.

Despite Fannie’s and Freddie’s assertions, there is reason to doubt that the LP data is fully reflective of today’s lending standards.

It’s true that in 2012, Fannie and Freddie were no longer securitizing loans with a 97% combined loan-to-value ratio and have reversed course since then at the behest of FHFA. Yet Freddie’s LP reporting guidelines were never amended to reflect this change, even though these loans are included in the CRT pools.

Furthermore, after controlling for risk, my analysis shows that the 30-year POO plain vanilla loans (from 2006 to 2007) in the LP data have much lower default rates on average than the full book of loans originated in those years.

For the riskiest loans, the LP data show an aggregate default rate through 2018 of 34%, while the full book shows a rate of 50%. Similar relationships hold for 316 out of 320 risk buckets, where loans are bucketed into groups according to their combined loan-to-value ratio, credit score and debt-to-income ratio. This cannot be a coincidence.

Over six years after the LP’s initial release, the full extent of its censoring is finally known through a new working paper undertaken by researchers from AEI and the FHFA. Crucially, the paper’s data appendix provides a small window into the full book of Fannie and Freddie loans, which allows for the aforementioned comparisons to the LP data.

Today, the CRT business is booming. Since 2013, Fannie and Freddie had transferred part of the credit risk on $3.1 trillion of loan acquisitions. Some observers have called it a major success, and others are working to expand it further.

Fannie and Freddie have gotten away with their data censoring for far too long, especially since there is now real money at stake.

Fortunately, the Treasury Department’s 2019 Housing Reform Plan calls for the release of the entire Fannie and Freddie loan-level data.

FHFA should swiftly release this data, with the appropriate privacy safeguards, to bring real transparency to the CRT deals so that investors — not Fannie and Freddie — can decide which loans actually reflect today’s lending standards.

To ensure that investors know which loans would be eligible for inclusion in CRTs, Fannie and Freddie should add a CRT flag in the dataset.

From a larger policy perspective, one question remains: If the goal is to transfer credit risk from the taxpayer to the private sector, why do so indirectly through CRTs that are untested during a housing downturn?

A far superior solution would be to gradually, but decisively, shrink Fannie’s and Freddie’s footprint so that private capital directly owns the loans and the risk.