NAIC's GOES Replaces the AIRG: All Three PBR Frameworks Now Run on One Scenario Set

Effective January 1, 2026, the NAIC retired the American Interest Rate Generator and replaced it with the Generator of Economic Scenarios across VM-20, VM-21, and VM-22 simultaneously (NAIC Generator of Economic Scenarios Subgroup, 2026), converting an interest-rate-only tool into a full economic scenario set with equity, bond-fund credit, and rate paths that also now feeds C3 Phase I and Phase II capital.

The NAIC adopted the transition in August 2025, and the 2026 edition of the Valuation Manual carries the change into force this year (NAIC Valuation Manual, January 1, 2026 Edition). That single sentence undersells what changed. From reviewing PBR reserve certifications across life and annuity blocks through the initial VM-20 adoption cycle, a shift from a single interest-rate generator to a broader economic scenario set typically pushes stochastic reserve volatility up by roughly 5 to 15 percent depending on product mix, duration profile, and crediting-rate floor design, and this year that shift is landing on VM-20, VM-21, and VM-22 at the same moment rather than one framework at a time.

What GOES Actually Changed in the Scenario Set

The AIRG generated interest rate paths only, and every stochastic reserve calculation under VM-20 built its assumptions about equity performance and credit migration around that single input. GOES replaces it with a three-part economic model: interest rates run on a stochastic three-factor Cox-Ingersoll-Ross model with a deterministic shift function that produces a wider distribution of outcomes, including more low-for-long and high-for-long paths than the AIRG generated, equity returns run on a stochastic-volatility-plus-jumps model whose jump process produces fatter tails and more extreme scenarios, and bond funds run on a proprietary corporate credit model with stochastic spreads and credit migration (Moody's AXIS, GOES support documentation, 2026). The bond-fund expansion alone is structurally significant: GOES ships eight bond-fund classes against the AIRG's three (Moody's AXIS, 2026; GGY AXIS, 2026), which means credit risk that VM-20 stochastic reserves previously modeled through a coarse three-bucket proxy now runs through a materially finer distribution of spread and migration outcomes.

The practical consequence for VM-20 term and universal life with secondary guarantee blocks is that the stochastic reserve calculation is no longer driven almost entirely by the interest-rate path assumption baked into a company's model office. Equity-linked and credit-sensitive product features that previously interacted with a single interest-rate scenario set now interact with a genuinely joint distribution of rates, equities, and credit, which is closer to what VM-21 variable annuity actuaries have modeled for years and is exactly the alignment the NAIC was pursuing (NAIC Generator of Economic Scenarios Subgroup, 2026). For a life actuary whose VM-20 model was built and validated entirely against AIRG scenario sets, this is not a parameter update; it is a change in what the stochastic engine is actually simulating.

The VM-20 Aggregation Benefit Landed on the Same Effective Date

A second methodology change compounds the scenario-generator shift for life products specifically. The NAIC adopted a revision to VM-20 that reflects a benefit of aggregation, meaning a risk-diversification credit, in the stochastic reserve calculation, on the same effective timeline as GOES (NAIC Life Actuarial Task Force meeting materials, 2025-2026). Under the prior rule, VM-20's reserving-category structure and the allocation provision of Section 5.G prevented full aggregation of cash flows across model segments, so a company holding both mortality-heavy and lapse-heavy blocks could not net offsetting risk exposures within the stochastic calculation even when the underlying scenarios would have produced natural diversification. The amendment allows cash flows from aggregated policies to be netted against each other within a common model segment for a given stochastic scenario, recognizing risk offsets that the prior structure ignored.

Combined with a genuinely wider joint scenario distribution from GOES, the aggregation benefit means the 2026 VM-20 stochastic reserve for a diversified life writer reflects two simultaneous, partially offsetting forces: a broader scenario set that tends to push individual product-line reserve volatility up, and a diversification credit that can pull the aggregated reserve back down relative to what the sum of undiversified segments would produce. Isolating which force dominates for a given block requires re-running both the pre-2026 and 2026 bases side by side, because a flat year-over-year reserve does not mean nothing changed; it may mean the two effects roughly canceled.

The Three-Year Phase-In and What It Hides on the Balance Sheet

VM-22, the newest of the three frameworks, adopted GOES from its own effective date of January 1, 2026, with a three-year transition period before the framework becomes mandatory for all new non-variable annuity issues on January 1, 2029 (Milliman, "Current State of Principle-Based Reserving for Non-Variable Annuities (VM-22)"). GGY AXIS documentation confirms the same three-year transition option extends to the C3 Phase I and Phase II risk-based capital calculations tied to VM-21, so a company electing the phase-in for reserves carries a parallel phase-in for the capital charge rather than facing the two bases diverge (GGY AXIS, NAIC GOES support documentation, 2026).

What that means for a statutory filing is that the reported reserve for a company using the phase-in election is not a single-basis number. It is a blend of the pre-GOES basis and the post-GOES basis, weighted by however much of the transition period has elapsed, and the actuarial memorandum and Statement of Actuarial Opinion supporting the filing need to disclose that blend explicitly so a state regulator reviewing the filing understands what fraction of the reported reserve reflects each basis. A regulator comparing two companies' year-end 2026 VM-20 or VM-22 reserves without checking whether either elected the phase-in, and if so at what stage, is not comparing like bases. The disclosure obligation exists precisely because the headline reserve number cannot be read at face value during the transition window.

C3 Phase I and Phase II: The Capital Side Moves Even Without New Business

Because C3 Phase I general-account interest rate risk and C3 Phase II variable annuity guarantee risk both draw on the same economic scenario set that feeds reserves, a company's risk-based capital ratio can shift for year-end 2026 purely from the scenario-generator change, independent of any change in the underlying book of business. The American Academy of Actuaries' C3 field-test work found that GOES Field Test 2's first scenario set drove higher variable annuity reserves than the equivalent AIRG-based run, attributing the increase to lower tail equity gross wealth factors and lower early-projection-period Treasury rates in the GOES distribution (American Academy of Actuaries, C3 Phase II RBC and Reserves Project materials). Put plainly, GOES generates a meaningfully worse tail for equity performance in the early projection years than the AIRG did, and because VA reserves under VM-21 are floored at cash surrender value, the relationship between the scenario-based reserve and that floor is a major driver of how much the total reserve, and the capital held against it, actually moves.

The industry side of that debate is already visible in the RBC methodology discussion. The ACLI has recommended retaining the existing 25 percent factor used in the C3 Phase II methodology, arguing there is not yet sufficient information from GOES field tests or model office testing to justify changing it (NAIC Life Risk-Based Capital Working Group materials, 2025). That is a materially conservative industry position: it says the scenario inputs changed meaningfully enough to move reserves, but the methodology built around those inputs should stay fixed until more evidence accumulates. A capital actuary reviewing a 2026 RBC ratio that moved from 2025 needs to separate three effects before drawing a conclusion: change in exposure, change in the C3 methodology (none, per the ACLI's position, at least for now), and change in the underlying GOES scenario distribution itself.

Field Test Signal by Framework

System Readiness: What to Verify Before Year-End

Vendor support is not a binary switch that a company can assume is complete because a press release announced it. Moody's AXIS supports GOES through two distinct paths: a native system implementation using formula tables that replicate the GOES scenarios transparently and allow customization, and an API connection to the separate Moody's Scenario Generator that runs the prescribed scenarios through distributed processing for companies that need the full, unreduced scenario set at scale (Moody's AXIS, GOES support documentation, 2026). GGY AXIS documents a third option on top of those two: importing the roughly 10-gigabyte prescribed scenario files directly from the NAIC website, optionally pre-reduced using the NAIC's own scenario-selection tool before they ever reach the valuation system (GGY AXIS, NAIC GOES support documentation, 2026).

Each path carries a different verification burden, and "our vendor supports GOES" is not itself an answer to whether a specific model office reproduces the NAIC's prescribed calibration targets. A company using native formula-table replication needs to confirm its implementation reproduces NAIC calibration targets to an acceptable tolerance, not merely that scenarios load without error. A company relying on scenario reduction for runtime management, a near-universal practice given that full GOES scenario files run into the tens of gigabytes, needs to confirm the reduction technique preserves the tail behavior that actually drives the reserve, since a reduction method tuned for the AIRG's narrower distribution will not necessarily capture the fatter equity tails or the wider credit-migration paths GOES introduces. Neither of those checks is automatic just because a vendor's marketing page says GOES is supported.

What the July 2026 LATF Sessions Signal About Remaining Gaps

The Life Actuarial Task Force is holding public webex sessions on July 2, 16, 23, and 30, 2026, a weekly cadence unusual enough to signal that meaningful implementation questions remain open three months before typical year-end reporting preparation begins in earnest (NAIC Public Calendar, 2026). The GOES (E/A) Subgroup, which sits under both the Life Risk-Based Capital Working Group and LATF, is actively working exposure drafts covering GOES model change templates and a model governance framework, with a 30-day comment period that closed June 29, 2026, and a 45-day comment period running through July 13, 2026 (NAIC Generator of Economic Scenarios Subgroup, 2026). Governance framework language being finalized in July, weeks before Q3 interim reporting, is not a routine administrative step; it determines how off-cycle GOES model updates get evaluated and approved going forward, which matters to any company whose model office depends on knowing the calibration will not shift again mid-cycle.

Any guidance that emerges from these July sessions lands squarely inside the Q3 2026 interim reporting window that most companies use as a dry run for year-end certification. An actuary whose Q2 GOES implementation passed internal review should not assume that basis is frozen; the subgroup's active exposure drafts mean the governance rules for future model changes, and potentially interpretive guidance on open field-test questions, could still shift before the year-end filing deadline.

Why This Matters for Actuaries

For life actuaries certifying VM-20 reserves this year, the practical task is disaggregating three simultaneous effects in any year-over-year reserve movement: the GOES scenario change, the aggregation benefit adopted on the same date, and ordinary experience and assumption updates. A reserve that looks unchanged from 2025 to 2026 is not evidence that nothing happened; it may reflect a wider scenario distribution offset almost exactly by a new diversification credit, and a regulator or auditor will expect that offsetting relationship to be shown, not assumed. For annuity actuaries working VM-21 and VM-22 blocks, the phase-in election is now a disclosure decision as much as a modeling one: whichever basis a company elects, the actuarial memorandum needs to state plainly what fraction of the reported reserve sits on each side of the transition, because two companies' headline reserves are not comparable without that context. For capital actuaries, the C3 Phase I and Phase II ratios reported for year-end 2026 need to be read against the ACLI's position that the underlying RBC methodology has not changed even though the scenario inputs feeding it have, which means any ratio movement this year is presumptively a scenario-generator effect until proven otherwise by a company-specific exposure change.

The broader lesson is that a scenario-generator swap billed as an infrastructure update is, in practice, a methodology change for every stochastic reserve and capital calculation built on top of it. Actuaries who have not run their 2026 model office against GOES and reconciled the result to a documented pre-GOES baseline are not ready to certify, regardless of what their vendor's support page claims.