Fed provides details on changes to stress-testing models

The Federal Reserve has disclosed the details of how it conducts and calculates various parts of its stress tests, in an effort to “improve public understanding” of its regime.

The 80-page report provides information on how the central bank calculates certain figures, such as operational risk losses, pre-provisional net revenue (PPNR) and regulatory capital ratios. It also describes the supervisory framework and policies surrounding model risk management.  

“Using this additional information, a firm would be better able to evaluate the risks in its own portfolio or compare the losses from its own models to losses from the board’s models,” the Fed said in an earlier statement regarding its push for greater stress-test transparency.

Vice-chair for supervision Randal Quarles said in a speech in November: “Enhanced transparency goes to the very core of democratic accountability and the rights of all US citizens – including the management and shareholders of the institutions that are subject to the stress tests – to understand the requirements to which they are subject. It also helps ensure the continued credibility of our regime.”

Changes for 2019 tests

This year the Fed made some adjustments to the stress-testing models that project some loan losses, the report reveals. Specifically, they have refined models for estimating auto loan losses, credit card losses, corporate loan losses, fair value for debt securities and commercial real estate loan losses.

The model for projecting auto loan losses, for example, has been adjusted to include changes to the way certain risks are captured. This reduces the volatility from historical macroeconomic movements, the Fed says. It also made an adjustment to the model that aims to better reflect higher credit risk of newly originated accounts. 

“Collectively, the enhancements are expected to result in a small increase in overall projected auto loan losses,” the central bank says. “However, for firms with large domestic auto loan portfolios, the changes may result in materially higher projected losses.”

The commercial real estate loan model has been improved by using a common data source and framework. One problem with the previous model, the central bank says, was that individual firms reported data, which led to “idiosyncratic reporting differences”.

The Fed also simplified the process for calculating specific risk drivers in the commercial real estate model. For example, “a single conceptual framework is used to project auxiliary risk drivers, which increases consistency and decreases complexity”.

“The refinements are expected to result in a slight increase in overall projected [commercial real estate] loan losses with modestly larger increases and decreases for firms depending on the risk characteristics of their portfolio,” the report says.

This year, the Fed also aims to increase the flexibility of models that are used to project a fair value for debt securities by allowing risk drivers to vary across the planning horizon. And it will adopt a new model to project certain yield spreads, such as the option-adjusted spread, it says.

What else is in the report?

The rest of the report contains detailed information on a range of stress-testing models and frameworks.

For example, it gives details on how the Fed estimates loss rates for each specific loan type, such as non-investment and investment grade or secured and unsecured loans. It also includes additional details on the approach to model development and validation, capital ratio calculations, model risk committees and data use.

Data use

One section of the report takes a closer look at where and what data the Fed collects ahead of its stress-testing programme, as well as detailing its procedure for deficient or erroneous data.

To develop and implement the models, for example, the Fed takes the data it collects from the financial accounts of the United States statistical release and the consolidated financial statements for holding companies report. However, most of the data comes from the capital assessments and stress testing (FR Y–14) information collection, it says.

“While firms are responsible for ensuring the completeness and accuracy of data reported in the FR Y–14 information collection, the Federal Reserve makes efforts to validate firm reported data and requests resubmissions of data where errors are identified,” it says.

If the data remains poor after resubmission, the Fed applies “conservative assumptions” for the information.

“If the Federal Reserve deems the quality of a firm’s submitted data too deficient to produce a supervisory model estimate for a particular portfolio, then the Federal Reserve assigns a high loss rate (eg 90th percentile) or a conservative PPNR (eg 10th percentile) to the portfolio balances based on supervisory projections of portfolio losses or PPNR estimated for other firms,” the report says.