In Summary
- Private credit losses stay contained, but each hike costs more than the last. After the 25bps hike the Fed delivered in September, the first since July 2023, our baseline assumes three more, taking the terminal range to 4.50%-4.75%. That is a cumulative +75bps since September and implies an expected loss of about 1.2% on US direct lending over the next year, a little above the 21-year realized average of 1.0%. For sensitivity, +50bps gives 1.05% and +100bps gives 1.33%, so the second 50bps costs about three times as much as the first. In severe stress (+200bps) losses reach 3.5% if rescued loans recover as badly as outright defaults, as they tend to in deep stress. If they keep their lighter treatment, the figure is 1.6%.
- Rate stress moves through the loan in stages, and defaults come last. These loans are mostly floating rate, so the short end of the curve raises the borrower’s interest bill within a quarter, by contract. The long end works more slowly through refinancing terms and fair value marks. For leveraged borrowers, spreads have historically widened by about 1.3x the policy move, so beyond the coupon reset, each hike also raises the spread borrowers pay on new and refinanced lending. Thinner interest coverage shows up first in payment in kind (PIK) and amendments, then in non-accruals, then in reported defaults.
- PIK is a leading indicator when it is switched on after origination, not when it is built in by design. PIK means the borrower adds interest to the loan balance instead of paying cash. Loans carrying PIK are about 2.8x as likely to go on non-accrual, the point at which the lender stops recognizing interest income. The type matters. PIK agreed at origination is a structuring choice and carries little signal once other loan characteristics are controlled for. PIK introduced later on a loan not structured with it (“bad PIK”) can signal a borrower that was paying cash and no longer can, and it predicts non-accrual in the following quarters. Such loans are 55% of PIK today. PIK overall is 11% of interest income, twice its late 2022 low.
- Reported defaults run high while losses stay modest, because most defaults are rescues. Non-accruals lead the default rate by about a quarter. Our policy rate baseline implies a broad default rate of roughly 6%, counting every rescue, of which only about a quarter are hard defaults (missed payment or bankruptcy). The rest are soft defaults: distressed exchanges, PIK conversions and maturity extensions, where the lender takes a small loss and the borrower keeps paying on restructured terms. Between 25% and 35% of rescued loans later default again, which is the risk behind our severe case. It is also why losses climb faster at higher rate levels, since deeper stress prices more rescues at hard default severity.
- Software is the main concentration risk. It is about a fifth of direct lending books and has the highest leverage and thinnest interest coverage of any major sector. These loans were largely underwritten at near-zero rates and peak growth multiples, and AI disruption now threatens the enterprise value that sponsor support depends on. Stressed through the 2001-02 telecom failure pattern, software produces an expected loss of about 3.3%, more than three times the long-run average. We chose telecom as an analog for its mechanism (capacity built against revenue that never arrived). No current data can calibrate an AI-driven revenue shock, so this is an illustration, not a forecast.
- The pickup over high yield is real but shrinks with every hike. At +50bps, direct lending nets +131bps over BB/B high yield in the US and +182bps in Europe after expected losses. Each further 25bps takes about 14bps off the US figure through higher losses. Two things limit the headline. The high-yield index is stronger than it was, because its weakest issuers have moved into loans and private credit. And the premium relies on the same investor demand that keeps high-yield spreads tight, so if flows reverse, both narrow together.
- Dispersion matters more than the average. The share of loans priced below 90% of par rose from 4% to 7% in a year, and to 12% for borrowers with USD10–20mn of EBITDA. A stable average conceals a widening tail, which puts manager selection and workout capability ahead of the asset class call.