NAIC Generator of Economic Scenarios (GOES)

Overview

The NAIC's change from the American Interest Rate Generator (AIRG) to the Generator of Economic Scenarios (GOES) for U.S. reserve and capital models was adopted by the NAIC in August 2025. The 2026 version of the Valuation Manual has been updated to reflect this change for VM-20, VM-21, and VM-22, with an effective date of January 1, 2026.

Moody's is committed to supporting current and future actuarial modeling needs in AXIS under the prescribed GOES regulatory environment. We plan to update and refine our articles when necessary, as changes or updates are made to the GOES models or calibrations. Please do not hesitate to reach out to us with any questions or concerns about our software solutions.

(1) The underlying models for interest rates, equity, and bond funds are all different than what was implemented under the AIRG

Interest
GOES is based on a stochastic 3-factor CIR model with a deterministic shift function. Some of the major differences from the AIRG include:

  • A wider distribution of interest rate outcomes, including more "low for long" and "high for long" type scenarios
  • Initial yield curve fitting that generally provides a better fit to the entire yield curve, not just the 1-year and 20-year tenors
  • The ability to produce negative interest rates
  • A "dynamic generalized fractional floor" is applied to suppress low rates (i.e. below 0.4%) and prevent the model from producing excessively negative rates
Equity
GOES is based on a stochastic volatility + jumps (SVJ) model. Some of the major differences from the AIRG include:

  • The inclusion of the jumps process leads to more extreme tail scenarios
  • The distribution of total return "gross wealth factors" has fatter tails - this is partially due to being calibrated to acceptance criteria with slightly fatter tails than was used to calibrate the AIRG equity model

Bond Funds
GOES is based on a proprietary corporate bond yield model. Some of the major differences from the AIRG include:

  • Modeling of stochastic credit spreads (with jumps) and credit migration leads to wider and more realistic distributions of results than the formulaic simplifications used in the AIRG
  • 8 bond funds are available, compared to only 3 funds under the AIRG
  • Limited documentation of the underlying proprietary model makes public discussion or in-depth understanding of the model impossible

(2) There is a shift from a well-defined "prescribed generator" (with transparent underlying code) under the AIRG to "prescribed scenario files" provided by the NAIC's vendor (with limited documentation available). This introduces a lot of uncertainty about scenario delivery and how the underlying models will perform under future conditions. The NAIC has drafted a model governance framework and is currently seeking guidance and input on best practices.

(3) Non-prescribed proprietary models were previously only allowed for VM-21. They will now be allowed across the board for VM-20, VM-21, and VM-22, subject to the testing requirements in the Valuation Manual about not materially understating reserves for VM-20 and VM-22 or materially understating the Total Asset Requirement (TAR) for VM-21. This may become a more attractive option for companies seeking additional flexibility, transparency, efficiency, and model governance control than what is offered through the prescribed scenario files but may require additional reconciliation work.

(4) The AIRG scenario picking method (i.e. significance values based on the path of 20-year US Treasury rates) was previously prescribed for VM-20. Moving forward, any scenario selection or stratification method will be allowed, provided that adequate documentation and justification is provided, and subject to the testing requirements in the Valuation Manual about not materially understating reserves or TAR. Additionally, for those opting to use the NAIC's scenario selection tool, a new "gross wealth factor" option based on Large Cap equity total returns is provided as a secondary alternative to the original UST significance value option.

(5) The 16 Stochastic Exclusion Ratio Test (SERT) scenario definitions were adjusted to work with the new underlying GOES model structure:

Interest
The prescribed shocks used in the AIRG were translated into similar shocks to apply to the 3 CIR stochastic processes in GOES, to produce scenarios with stylistically similar behavior. Note that the distribution of interest rate outcomes in GOES SERT scenarios is wider than under the AIRG, primarily due to a different underlying model calibration with much greater interest rate volatility. This often leads to more disperse SERT outcomes and higher SERT ratios, making it generally more difficult to pass the exclusion test. Please see Knowledge Base article 2484 for further information.

Equity
A prescribed shock approach is no longer used in GOES due to complications with the presence of jumps in the model. Instead, the NAIC's vendor utilizes a targeted percentile method (e.g. to maintain equity total return gross wealth factors at a cumulative 90%, 10%, or 50% level, for high, low, or baseline equity scenarios, respectively).

Bond Funds
Little or no documentation has been provided yet by the NAIC's vendor about how SERT scenarios are produced under the proprietary corporate bond model.

In partnership with consultants, the NAIC performed model office testing in AXIS in 2024 and early 2025 to test the impact of the proposed scenario changes on VM-20, VM-21, VM-22, and C3 Phase II results. Some of their presentations can be found in the "Documents" section of the GOES Subgroup webpage and on the VM-22 Subgroup webpage. Note that these results were not using the GOES scenarios finalized later in 2025, so results may vary with the current finalized GOES paradigm.

Additionally, some companies opted to participate in field testing in 2024. Because the individual company field test results were presented in regulator only sessions and not aggregated or shared publicly, it is difficult to assess the overall financial impact to the industry. However, regulators have given companies the opportunity to gradually phase-in the financial impact over a 3-year transition period.

Risk based capital measures under C3 Phase I and C3 Phase II for year-end 2026 are expected to use GOES and offer a 3-year transition. Initial updates were published in 2025 and are expected to be refined before being finalized in early 2026.

For valuation models that will be externally reported:

  1. The simplest option is to download the very large 10 GB prescribed scenario files from the NAIC website and import them into AXIS. The NAIC's scenario selection tool can optionally be pre-run to reduce the number of scenarios and the size of the CSV files imported into AXIS. An example DataLink import process is demonstrated in Knowledge Base article 2322.

To dynamically generate customized GOES scenarios for internal pricing, projections, or risk sensitivity analysis, some additional options exist in AXIS:

  1. Knowledge Base article 2540 contains example Scenario Sets that use native AXIS formula table code to generate GOES type scenarios. The code implemented in these examples is transparent and customizable. The interest and equity in these samples provide a very close replication of the GOES prescribed scenarios (with validation statistics included) for any initial valuation date par yield curve. An American Academy of Actuaries (AAA) corporate bond fund model is implemented as a substitute for the proprietary corporate bond model used in the prescribed scenarios. Corporate bond fund returns in the AAA model are well aligned to the targeted acceptance criteria regarding average excess returns, with validation and documentation of this provided in the KB article. Dynamic SERT scenario generation is also supported in the examples.
  2. The AXIS dynamic link to Moody's Scenario Generator software through an embedded API can be used to take advantage of a similar replicator for prescribed interest and equity combined with a non-prescribed Simple Bond Model. Additionally, licensed Moody's Scenario Generator users can take advantage of their functionality to model other economic variables that are not included as part of the basic prescribed scenario set.

Companies can also choose to use the 2nd or 3rd option for their reported valuation runs, but this would be subject to additional documentation and testing to satisfy regulatory requirements regarding the use of non-prescribed models. Since effectively only the corporate bond model is being substituted in these examples, this is expected to be a simpler exercise to demonstrate compared to starting with a completely different proprietary model for interest and equity.

For scenario selection, AXIS natively offers the two standard picking options provided in the NAIC scenario selection tool (UST significance values and Large Cap gross wealth factors). Additionally, advanced optimization tools such as Principal Component Analysis (PCA) are available for customizable stratified sampling techniques. Please see Knowledge Base article #2538 and #2540 or reach out to Moody's for more information.