Academy Shows Why a Thin Baa3 CLO Tranche Carries 12.52% C-1

The American Academy of Actuaries' C-1 Subcommittee has published the model behind the CLO capital factors life insurers will apply at year-end 2026, and traces the table's largest jump to one regression coefficient: a 0.0980 premium on any tranche rated Baa3 or lower that spans 4% or less of its deal (American Academy of Actuaries, October 2026). That premium turns a 2.73% after-tax Baa3 charge into 12.52%.

Posted October 22, the 122-page report documents the factors the NAIC's RBC Investment Risk and Evaluation (E) Working Group adopted on June 23 as proposal 2026-12-IRE MOD, the Capital Adequacy (E) Task Force on June 30 and the Financial Condition (E) Committee on July 8. Plenary ratified it in August.

Those factors land on $276.8 billion of insurer CLO holdings at year-end 2024, 82% of it in life companies, according to the NAIC Capital Markets Bureau. Only 12% of that book is rated BBB and 2% sits below it.

Key Takeaways

Thickness enters the factor only from Baa3 down: a thick Baa3 tranche carries 2.73% after tax, a thin one 12.52%, against 2.70% at Baa2.
The Academy's tail measure is a CTE90 built from 17 scenarios between the 90th and 99.99th percentiles, where the bond model uses a single 96th percentile.
Regulators set the portfolio adjustment factor at 1.0 after the modeled schedule, 1.3907 for one CLO, would have cost one outlier 0.4% of surplus.
The NAIC's adopted surcharge is a flat 11.77% pretax on thin broadly syndicated tranches at 2.C or below; the Academy's range ran 11.77% to 11.78%.

Why thickness counts only from Baa3 down

The regression put raw C-1 against rating dummies and years to reinvestment end, with interaction terms by rating bucket, plus a single dummy for tranches at Baa3 or below with thickness of 4% or less. That dummy's coefficient was 0.0980 with a t-value of 35.23, and the model's adjusted R-squared reached 83.25%.

Data stops the cutoff at Baa3. Only 15 thin Baa1 tranches and 14 thin Baa2 existed in the universe, "too infrequent to be credible," the report said. Across Baa3 and below the thin premium averaged about 10 points, so one premium covers every rating in the range.

A thin Baa3 is a rating-methodology artifact. Table 8 compares a traditional CLO, Baa3 tranche at 12.6% to 18.5%, with a "Split-BBB CLO" carrying two Baa3 tranches at 11.8% to 17.9% and 10.8% to 11.8%. S&P and Fitch rate on first-dollar-of-loss, Moody's and KBRA on expected loss. "If a CLO is rated with 'first-dollar-of-loss' methodology, then the severity does not drive the charge; the probability of penetration does," the subcommittee wrote. A 12% collateral loss wipes out the thin tranche; the thick one needs 18%.

No Baa3 tranche in the data had thickness between 3.68% and 4.29%, and the Academy picked 4% as "a round number in that range between the two clusters."

A $10 million thin Baa3 tranche carries $1.25 million of after-tax C-1 against $273,000 for a thick one. Pretax, the thin tranche's 3.28% base plus the 11.77% surcharge reaches 15.05%, against a 2.168% bond factor before year-end 2026 (NAIC LR002, proposal 2026-12-IRE MOD). Sidley notes thickness is measured from the most recent trustee report at the reporting date. Tranche par is fixed while the collateral balance falls after reinvestment ends, so a thin tranche can cross the 4% line with no rating action.

A CTE90 over 17 tail scenarios replaces the bond model's VaR96

The CLO model starts from the ACLI and Moody's Analytics bond model that has set corporate bond C-1 since 2021: Moody's default rates from 1983 to 2020, a Markov chain of economic states, a Gaussian copula with 10% correlation and a 96th percentile value-at-risk. It kept the default rates, copula and correlation and changed four things.

It removed the economic-state chain, which Appendix B shows moves bond factors by at most 0.56 points. It swapped bond recoveries for senior and non-senior loan recoveries from S&P ultimate recovery deciles. It simulated all 1,957 loan issuers at once so that cross-rating diversification flows into senior tranches. And from 10,000 simulations it ran 17 tail scenarios, from the 90th to the 99.99th percentile, through the NAIC's CDOnet cash-flow engine, averaged with Riemann sums into a CTE90 for each of 2,074 CLOs.

At 20% correlation the Baa3 bond factor would rise from 2.17% to 3.47%, which the Academy said "would have dramatically changed the C-1 Factors adopted for bonds."

Above Baa1 the data gave no curve. Aa2 had one non-zero C-1 in 1,879 observations, and A1 at 0.41% sat above A2 at 0.10%. An isotonic regression pooled A1 and A2 by observation count into 0.14%. A3 grouped with A2 produced 0.16%; grouped with Baa1 and Baa2 it produced 1.45%. Rejecting a 44/56 Akaike weighting as noise, the subcommittee took a 50/50 average of 0.80%. That is the revised 1.G figure, 0.97% pretax, that the Working Group adopted; the site's July coverage recorded the 1.743% first proposed.

NAIC staff convert the after-tax table at a 21% rate with 80% recovery, and filled the one gap, NAIC 6, with the mean of 100% and the 5.C factor, 92.56% pretax.

NAIC designation	Bond factor	CLO, thick	CLO, thin
1.A (Aaa)	0.158%	0.040%	0.040%
1.D (Aa3)	0.523%	0.050%	0.050%
1.G (A3)	1.016%	0.970%	0.970%
2.A (Baa1)	1.261%	2.180%	2.180%
2.C (Baa3)	2.168%	3.280%	15.050%
3.A (Ba1)	3.151%	15.140%	26.910%
3.C (Ba3)	6.017%	27.990%	39.760%
4.C (B3)	12.428%	56.880%	68.650%
5.C (Caa3)	30.000%	85.120%	96.890%
6	30.000%	92.560%	92.560%

Pretax C-1 factors for year-end 2026 from the LR002 instructions in proposal 2026-12-IRE MOD. Thin-column figures add the 11.77% surcharge, applied only to broadly syndicated loan CLOs at 2.C or below. Residual tranches stay at 45% pretax.

Charges fell for tranches rated A and above and rose from A-minus down, J.P. Morgan Asset Management said. A reconciliation in the report puts the new factors on a vertical slice of an average deal at 4.17% of face, against 2.67% under the old bond factors. Tax and residual-tranche book-value accounting explain the gap to the 9.73% collateral loss, it says.

The portfolio factor that was modeled and then set aside

Correlations between deals would have needed more CDOnet simulations than the Academy had, so it treated one tranche as a claim on the deal's vertical slice and counted unique loan issuers across sampled portfolios of two or more CLOs. Median relative factors came out at 1.3907 for a single CLO, 1.2222 for two, 1.0661 for ten and 1.0 at eleven or more.

In 2024 year-end group filings, the median insurer held 40 unique CLOs and the mean 99.5, pulled up by six groups holding more than 500 deals. Fifty-six insurers held ten or fewer, 35 of them only Aa3-and-above paper, where the schedule would move the factor by at most 0.015%. Insurers with three or fewer CLOs and any credit below Aa3 put no more than 0.9% of assets in CLOs, and those with seven to ten no more than 2.2%, with one exception at 8.4%.

That outlier owns three 2.A CLOs, four thick 2.C and two thin 2.C, for a base factor of 4.22%. A 1.0704 factor would add 0.30 points, or 0.4% of its total capital and surplus. "Changes that have a small impact on RBC ratios may not justify a change to the RBC formula," the report said. Regulators took the flat 1.0, let CLO issuer vehicles count toward the non-CLO bond issuer total as an interim step, and said the method will be reviewed for 2027 (NAIC, June 2026).

Scope is the other open item. The model is calibrated to broadly syndicated loan CLOs only, yet regulators applied the base factors to middle-market CLOs as well, with the surcharge on BSL alone, and the E Committee declined a request to commission middle-market work for 2027, Mayer Brown reported. Committee chair Nathan Houdek said a 2027 plan for further data collection would follow.

Footnote 4 cuts the other way. The model finds tail risk falling as a CLO approaches maturity. "CLO practitioners have advised that this relationship may be backwards," the footnote says. Practitioners see a longer reinvestment horizon as less risky because the manager can buy discounted loans to offset poor credit. A model that books every loan at par and gives no value to active management cannot see that effect either way.

Academy Shows Why a Thin Baa3 CLO Tranche Carries 12.52% C-1

Key Takeaways

Why thickness counts only from Baa3 down

A CTE90 over 17 tail scenarios replaces the bond model's VaR96

The portfolio factor that was modeled and then set aside

Further Reading

Sources