Reading Goldman's Apple Card loss expectations from its CECL reserve

It has been almost a year since Goldman Sachs launched its credit card in partnership with Apple. By most accounts the launch has gone well: balances reached $2B per Q1 2020 call report filings, and total open credit lines are an even more impressive $18.5B.

People in the industry know how hard it is to grow card balances that fast. With growth like that, the natural question is whether the company took on too much credit risk to get there.

Using CECL as a window

There isn't much loss history to go on, so we have to look elsewhere. The CECL adoption by Goldman and other US commercial banks in Q1 2020 is a rich source. It reveals each bank's internal loss expectations.

Most banks adopted CECL in Q1 2020, and almost all used the Moody's baseline for their macroeconomic inputs. That gives a fairly standardized basis for comparing card portfolios across banks.

CECL's detailed rules are complex, but at its core it estimates lifetime losses on the portfolio given what's known about conditions. A useful simplification:

CECL reserve ≈ expected annual loss rate × weighted-average life (WAL)

For example, a portfolio with a 5% annual loss rate and a 2-year WAL would need roughly a 10% reserve.

The comparison

Comparing annual card loss rates with CECL reserves for banks with card portfolios over $1B shows a very high correlation between the two. Most large issuers sit close to the best-fit line.

Goldman set its card CECL reserve at 11.5%. Reading the implied annual loss rate off the best-fit line gives ~5.75%.

Caveats

This assumes all banks use similar CECL methods and similar levels of conservatism. That's almost certainly not true, so the estimate carries some variance. Still, the signal is clear: Goldman appears to expect its card portfolio to look less like Amex and more like Citi in risk content.

That has implications for the economics. Amex-like losses support a rewards-heavy, low-fee proposition. Citi-like losses on a no-fee, low-APR-positioned card leave much less margin.

Postscript (2023)

Apple Card and Goldman's Marcus have been in the news a great deal since. My guess is that actual annual losses on Apple Card came in even higher than the ~6% I projected in 2020. A few years later, credit card underwriting continues to humble new entrants, which makes the long-run consistency of Capital One and Discover look all the more impressive.