Abstract. The index really did get safer, partly by losing its worst borrowers to private credit, so its spread history is not a constant-quality series. Quality is up, compensation is near record lows, and the risk that left the index still exists.

1. The claim, verified

A recent KKR piece argues high yield has outgrown its junk-bond reputation: a record BB share, duration near a 15-year low, call protection as a bonus. The composition checks out against independent index data, and the duration claim is conservative. The facts are right. The conclusion does not follow.

Composition, and where the risk went
MeasureValue
BB share of the US index57 to 58%, a record; KKR cites 57%
BB share in 200038%, roughly the level held through the pre-Lehman era
BB share of the European index (KKR)68%
Index durationroughly 3.0 years against a long-run average near 4; arguably a record low, not the 15-year low claimed
CCC share of the US indexdown from a post-crisis peak near 30% to about 10%
Private credit assetsroughly $158 billion (2010) to about $2.1 trillion (IMF)
Public high yield marketabout 11% smaller than its 2021 peak
Ratings migration since 2021rising stars exceed fallen angels; Ford back in investment grade

2. What the quality costs

Five FRED series, pulled 19 August 2026, data as of 18 August, percentiles over the freely available three-year window. The B row was missing from the version first published, and B is roughly a third of the index by weight.

FRED data as of 18 August 2026; percentiles over the freely available three-year window
SeriesLevelThree-year percentile
High yield index yield7.09%51st
High yield index spread2.75%17th
BB spread1.63%11th
B spread2.94%30th
CCC spread10.27%98th

The yield looks normal only because most of it is Treasury yield wearing a credit costume. The spread, the only part that pays for credit risk, sits near its floor, and the three-year window understates how near.

The price of the quality, on the longer record
MeasureValue
Yield decomposition7.09% = 4.34% Treasury plus 2.75% credit; the marketed 7% yield is mostly Treasury
Index spread, 1996 to 2025 window2.75% at the 4th percentile, against the 17th on three years
Long-run median index spread4.56%
All-time low index spread2.41%
CCC spread1,027 basis points, 98th percentile
Gradient by quality11th percentile at BB, 30th at B, 98th at CCC

The average conceals not a two-sided split but a monotone gradient by quality: the high-quality majority is priced for perfection while the remnant reprices violently. A gradient holding at every rung is more demanding than the two-extreme reading first published here.

3. Where the junk went

The riskiest borrowers did not deleverage. They changed address, to leveraged loans and to private credit. Only the second is genuinely dark: the loan market prints a daily secondary mark, and the Morningstar LSTA US Leveraged Loan Index publishes a daily weighted average bid. Federal Reserve and BIS research finds private credit borrowers smaller, more leveraged and riskier than public issuers. The IMF and the Financial Stability Board have warned about that opacity.

4. The strongest counterargument, answered

The sophisticated bull case adjusts for duration and finds spreads only modestly rich. The arithmetic is correct and partly circular, because duration is this short partly because bonds trade to their call dates at tight spreads. The adjustment uses the symptom to excuse the price.

The bull case, tested against the public record
ClaimWhat the record says
Spreads per year of duration are only modestly richabout 90 basis points per duration-year today against roughly 100 historically
Call protection is a bonusnot verifiable from public sources; no index provider publishes the share of the market trading to call
The BB-heavy mix justifies today's tightnessMarty Fridson, the analyst who tested it, is the named sceptic in this literature, not its cheerleader
The improvement is deleveragingone asset manager titled the phenomenon exactly: addition by subtraction

5. What this does not establish

No source, this note included, cleanly separates genuine improvement from migration; both are real, and the 2020 fallen-angel wave genuinely upgraded the index. The percentile table uses the three-year window freely reproducible from FRED. The 1996 to 2025 anchors are computed here from a cached daily history of the index spread, not taken from a published source. This is a measurement note about compensation, not a forecast, a recommendation, or a view on any fund or firm.

Provenance, bs-prov/1.0

Method
Spreads and yields from FRED (ICE BofA series BAMLH0A0HYM2, BAMLH0A1HYBB, BAMLH0A2HYB, BAMLH0A3HYC, BAMLH0A0HYM2EY), pulled 19 August 2026 (data as of 18 August). Every percentile in the table is computed deterministically over that three-year window, 786 daily observations from 21 August 2023 to 18 August 2026, because FRED’s own ICE window is three years. The 1996 to 2025 anchors in section 2, the 4th percentile, the 4.56% median and the 2.41% all-time low, are not cited published figures. They are computed here from a cached daily history of BAMLH0A0HYM2 at state/cache/wayback_BAMLH0A0HYM2.csv, 7,530 daily observations from 31 December 1996 to 3 November 2025.
Sources
Morgan Stanley IM High Yield Market Monitor (ICE Indices data); VanEck and SSGA index tables; Polen Capital; IMF Global Financial Stability Report, April 2024; BIS Quarterly Review, March 2025 (Avalos, Doerr and Pinter); Federal Reserve FEDS Note, February 2024 (Cai and Haque); Chernenko, Erel and Prilmeier, Review of Financial Studies 2022; PitchBook LCD; Osterweis; Fridson. Prompted by KKR, High Yield’s Second Act, August 2026.
Honesty
No source, this note included, cleanly separates genuine quality improvement from migration; both are real. A measurement note, not a forecast or a view on any fund or firm. The declared control has now been run (21 August 2026). The ticket behind this paper specified holding BB, B and CCC weights constant so that composition drift could not masquerade as spread compression, the paper’s own central worry. An earlier edition said the control could not run for lack of the weight history; that was wrong on both legs: the ticket declares fixed 40/40/20 weights precisely so no history is needed, and the three bucket series are in the cache. At the cache’s last common date (24 July 2026) the actual changing-composition index spread of 2.79% sits at the 18th percentile of three years, while the fixed-mix 40/40/20 spread of 3.85% sits at the 62nd: composition drift flattered the aggregate by 1.06 points. The near-record-lows reading holds for the aggregate and for the BB and B buckets on their own; at constant quality the market is priced near its three-year middle, which strengthens rather than weakens the paper’s claim that the improvement is partly a change of address. The finding survives, now quantified until the control runs.
Revision
19 August 2026. Four corrections. (i) A duplicated clause in the lede, live since publication, is removed. (ii) The B bucket, roughly a third of the index by weight, was missing from the table; adding it turns a two-extreme "split market" into a monotone gradient, 11th percentile at BB, 30th at B, 98th at CCC, which is the stronger claim. (iii) The whole table is recomputed on a single fresh pull dated 18 August rather than adding one row from a different vintage. (iv) The paper cited long-run figures secondhand while a 1996 to 2025 history of the index spread sat cached on disk; used directly it puts the current spread at the 4th percentile against the 17th on the three-year window, so the original presentation was conservative rather than aggressive. No conclusion reversed; two were strengthened.
Related
Working Paper No. 9 (the AI-CDO), No. 11 (the risk teardown), No. 27 (a fitted value is not a measurement).