VM-22 became effective January 1, 2026 for new non-variable annuity business, replacing the prescribed factors of the Commissioners Annuity Reserve Valuation Method with company-specific prudent estimate assumptions modeled stochastically at a CTE70 tail standard.

Insurers have until January 1, 2029 to apply it to everything. Two valuation bases will therefore sit on the same balance sheet for three annual statement cycles, against US annuity sales running near a record $464.1 billion.

Key Takeaways

  • CTE70 replaces a factor table, setting the reserve at the average of the worst 30% of stochastic outcomes rather than the greatest present value of guaranteed benefits under prescribed assumptions.
  • Field testing has produced reserves above CARVM, inverting the working assumption that principle-based reserving releases capital.
  • The cash surrender floor is the structural driver, which RGA describes as essentially requiring a 100% mass-lapse reserve in every modeled path.
  • RILA sales grew 21% to $21.2 billion in a 30th consecutive quarter of growth, so the share of a book on the new basis rises fastest exactly where the stochastic tail bites hardest.
  • GOES removes the prior generator's bound on negative rates, and a tail-weighted standard gives those paths real weight in the reserve rather than in a stress test alongside it.

What CARVM Did and What Replaces It

CARVM's defining feature was uniformity. Every company took the greatest present value of future guaranteed benefits using the same prescribed mortality tables and discount rates, whatever it actually experienced on lapses, credited rates or asset risk. Reserves were comparable across companies and disconnected from the economics of any block.

VM-22 substitutes prudent estimate assumptions the company sets for mortality, policyholder behavior and expenses, run through a Stochastic Exclusion Test and a Single Scenario Test to determine whether the filed reserve is deterministic, stochastic, or in narrow cases a continuation of CARVM. Where the stochastic reserve governs, it is the cumulative tail expectation above the 70th percentile of the company's generated scenarios.

Anyone who lived through VM-20 will recognize the mechanical shift: a table lookup becomes a full asset-liability projection, each in-force cell run forward under hundreds of paths.

What is not familiar is the direction of the result. RGA reports field testing producing "over-conservative reserves with redundancies, in some cases with resulting reserves higher than those required by the outgoing CARVM methodology." The driver is the cash surrender floor inside the stochastic calculation, which RGA calls "essentially equivalent to requiring an insurance company to reserve for a 100% mass-lapse scenario" in every path. On a fixed indexed annuity block with a guaranteed lifetime withdrawal benefit, that floor can set the reserve even where the company's own lapse studies show persistency far above any plausible mass-lapse trigger.

Two Bases on One Balance Sheet, Reweighting Every Quarter

VM-20's phase-in ran against term and universal life books where new issue volume was flat to declining, so the parallel-regime distortion stayed modest in dollar terms. VM-22 phases in against the fastest-growing part of the annuity shelf.

Product (Q1 2026) Sales YoY Change VM-22 Reserve Sensitivity
Total U.S. annuity sales $104.6B -2% 10th consecutive $100B+ quarter
RILA $21.2B +21% High: index-linked crediting, cap/floor optionality
Fixed indexed annuity (FIA) $26.6B -4% High: GLWB mass-lapse floor exposure
Fixed rate deferred (FRD) $34.0B -16% Moderate: primarily target spread and lapse assumption

Total US annuity sales were $104.6 billion in the first quarter of 2026, 2% below the prior year and a tenth consecutive quarter above $100 billion. RILA reached $21.2 billion, up 21% and a 30th straight quarter of growth. FIA came in at $26.6 billion, down 4%, and fixed rate deferred fell 16% to $34.0 billion as its credited-rate advantage narrowed.

That mix is what makes the transition period hard to read. A company writing $2 billion of new FIA and RILA premium a year under VM-22 while its 2024-and-prior in-force runs off under CARVM reports a blended reserve growth rate mixing two valuation bases, and the blend ratio moves every quarter as new business accumulates on the VM-22 side.

So year-over-year reserve trend for non-variable annuity lines is not a clean signal of experience between now and 2029. An analyst, rating agency or reinsurance counterparty comparing 2026 statutory reserves with 2025 is comparing a partly CARVM number with a partly stochastic one.

RILA is where that shifts fastest: its cap-and-floor crediting is path-dependent, exactly what stochastic reserving captures and formula factors could not, so a carrier growing RILA 21% a year moves onto the new basis faster than a fixed-rate-weighted writer regardless of any change in risk written.

The subgroup has seen the edge of this. A retrospective application exposure draft extending VM-22 back to business issued from January 1, 2017 closed its 90-day comment period on June 22, 2026, because leaving older blocks on CARVM widens the gap the longer the transition runs.

The Reserve Now Depends on a Generator and an Assumption Set Nobody Shares

VM-22 runs on the Generator of Economic Scenarios, which VM-20 and VM-21 also now draw on, and it carries a requirement companies have not had to build for: models must handle negative interest rates. The prior generator was bounded such that near-zero and negative paths were a tail curiosity. GOES removes the bound, and because CTE70 is a tail-weighted average, paths spending time at or below zero on the discount curve feed the reserve itself rather than a stress test beside it.

For a company whose asset-liability infrastructure was calibrated to the old generator, that is not a parameter update. Asset cash flow projections, hedge cost models and crediting rate algorithms built against a generator that never produced sustained negative rates can behave unpredictably when one does, particularly for floating-rate and option-embedded assets whose logic was never exercised in that regime.

The same company-specificity dismantles the shared reference point reinsurance was built on. Flow treaties and captive cessions across the fixed and indexed market were structured around a CARVM reserve credit both parties could compute from a published table. The ceded credit is now a stochastic output of the ceding company's own assumption set.

The direction is not uniform, which is what makes renewal harder than repricing. Where VM-22 lands below CARVM, the letters of credit and captive structures that existed to fund CARVM redundancy lose their purpose.

Where the cash surrender floor pushes the stochastic reserve above the old level, likeliest on FIA blocks with lifetime withdrawal guarantees, strain rises and the case for reinsurance strengthens, but the treaty has to be priced against a number the assuming party cannot independently verify. RGA's own conclusion is that the company-specific nature of the guidelines requires each solution to be customized. A treaty written in 2023 against a predictable credit does not renew onto this basis; it gets renegotiated before the block crosses over.

Further Reading

Sources

  1. NAIC, Valuation Manual (A) VM-22 Subgroup, content.naic.org, 2026
  2. NAIC, PBR Data and Valuation Manual, Jan. 1, 2026 Edition, content.naic.org, 2026
  3. Milliman, “VM-22 Readiness: Key Areas for Consideration,” Milliman.com, 2026
  4. Milliman, “Current State of Principle-Based Reserving for Non-Variable Annuities (VM-22),” Milliman.com, 2026
  5. RGA, “VM-22 Is Here: Where Does Reinsurance Fit In?” RGAre.com, 2026
  6. LIMRA, “U.S. Annuity Sales Notch Tenth Consecutive $100 Billion+ Quarter,” LIMRA.com, May 2026
  7. LIMRA, “Final U.S. Retail Annuity Sales Set New Sales High, Totaling $464.1 Billion in 2025,” LIMRA.com, 2026