Home › Guides › NAIC GOES Interest-Rate Scenarios, Explained
Official "what if" futures for interest rates - the paths insurance regulators make insurers test against.
Nobody knows where interest rates will be in 10 or 30 years. But life insurance companies make promises that stretch that far, so US insurance regulators require them to test their finances against a standard set of possible futures for interest rates. Those futures are generated by GOES - the Generator of Economic Scenarios - and published for free every month by the NAIC (the National Association of Insurance Commissioners, the body that coordinates the 50 state insurance regulators).
Official "what if" futures for interest rates - the paths insurance regulators make insurers test against.
Nobody knows where interest rates will be in 10 or 30 years. But life insurance companies make promises that stretch that far, so US insurance regulators require them to test their finances against a standard set of possible futures for interest rates. Those futures are generated by GOES - the Generator of Economic Scenarios - and published for free every month by the NAIC (the National Association of Insurance Commissioners, the body that coordinates the 50 state insurance regulators).
Each line on this page is one scenario: one internally-consistent possible future for Treasury yields, month by month, decades ahead. One scenario might show rates climbing steadily; another shows them falling and staying low; another spikes then reverses. They are not predictions, and no single one is expected to happen - the point is the range. If an insurer (or your own retirement plan) only works in the futures where rates cooperate, that's worth knowing.
This page shows the 16 scenarios from the set's standard test battery, chosen to span the plausible extremes: rising, falling, whipsawing, and steady rate environments. The full regulatory set behind them contains 10,000 computer-generated scenarios; the 16 shown here are the prescribed stress paths regulators use as a quick screen.
Why it's on this site: everything else here shows where rates have been. This is the one page showing a regulator-grade view of where rates could go - useful context for anyone deciding between locking a mortgage rate, buying long-term bonds, or just wondering what "rates might do anything" actually looks like drawn out.
GOES replaced the AIRG on 1/1/2026 for statutory reserves and capital; this page shows its 16 SERT paths.
GOES (Generator of Economic Scenarios) is a custom NAIC calibration of Conning's GEMS economic scenario generator. Effective January 1, 2026, the Valuation Manual made it the prescribed generator for US statutory work, replacing the Academy Interest Rate Generator (AIRG) that had served since the 2000s. Insurers use the scenarios for principle-based reserves - VM-20 (life insurance) and VM-21 (variable annuities) - and for risk-based capital (RBC) calculations on economically-sensitive products.
The NAIC publishes new scenario files on the first business day of each month, free, at naic.conning.com. Each release is anchored to the actual Treasury curve on the as-of date (which is why scenario paths on this page all start from the same point - that's the real curve at month-end) and projects monthly steps roughly 100 years forward across Treasury yields and other economic variables.
The 16 scenarios shown here are the SERT set - the Stochastic Exclusion Ratio Test. VM-20 lets a company skip full stochastic modeling for products with little interest-rate sensitivity, but only if it passes this test: compute reserves under 16 prescribed rate paths (a baseline plus rising, falling, and whipsaw variants), and if the worst result is close enough to the baseline, the product is demonstrably rate-insensitive and qualifies for exclusion. That makes these 16 paths a deliberately wide, regulator-designed spread of rate futures - which is exactly what makes them a good visual summary of rate uncertainty.
What this page does not show: the full 10,000-scenario stochastic set (the file is about 10 GB) and the non-Treasury variables (equity returns, other yields) that GOES also generates. Yields shown are spot rates - the rate for a single payment at that maturity - which differ slightly from the par yields on this site's other Treasury pages.
Anchoring, mean reversion, and acceptance criteria; SERT ratio mechanics; spot vs par comparability.
Generator mechanics: each monthly GOES release re-anchors to the month-end Treasury curve, so scenario dispersion starts at zero and fans out with horizon - near-term paths cluster tightly around the actual curve, long-horizon paths spread according to the generator's volatility and mean-reversion structure. The calibration is governed by the NAIC's published acceptance criteria and stylized facts (developed through the GOES (E/A) Subgroup with extensive field testing before go-live): long-run distributions of the short rate and the 10-year, curve-shape frequencies (including how often the curve inverts), and bounds on extreme outcomes. Watching the 30-year horizon on this page shows the mean-reversion target region directly - paths converge toward the generator's long-run distribution regardless of the starting curve.
SERT mechanics, precisely: the test computes a ratio - (largest adjusted deterministic reserve across the 16 scenarios minus the baseline scenario reserve) divided by the baseline (with prescribed adjustments) - and compares it to a threshold (6% in VM-20). Below threshold, the product group passes the exclusion test and can use deterministic-only reserves; above, full stochastic modeling on the big set is required. The 16 paths are therefore not a sample from the stochastic distribution - they are prescribed stress paths, and reading them as "equally likely draws" is a category error. For distributional statements (percentiles, tail behavior) you need the 10,000-scenario set or the NAIC's published summary statistics.
Comparability cautions: (1) spot vs par - GOES files publish spot (zero-coupon) yields; the yield-curve pages on this site show par yields; at typical curve slopes the difference is a few basis points at the long end, larger when the curve is steep. (2) This page thins monthly data to annual steps for display; the underlying file is monthly. (3) One of the 16 paths is the test's baseline, but the file does not label which, and none of them is an official rate forecast - the Fed's own projection is the dot plot, a different animal entirely.
GOES vs AIRG in practice: reserve impact channels, scenario reduction, and what the SERT view can't tell you.
For practitioners, the AIRG-to-GOES transition changes results through several channels: re-anchoring each month to the live curve (AIRG anchored quarterly with interpolation conventions), a recalibrated long-run mean-reversion framework, integrated equity/rate dynamics from the GEMS engine rather than AIRG's separate equity model, and different tail behavior at both ends (the acceptance-criteria negotiations centered on low-for-long persistence and fast-rate-rise frequencies - both directly visible in this page's fan of paths). Companies running VM-20/VM-21 stochastic reserves (CTE70) and C-3 capital saw level shifts at transition that had to be explained to regulators and auditors; the monthly re-anchoring also makes reserve sensitivity to the starting curve more mechanical and more frequent.
Workflow reality: almost nobody runs all 10,000 scenarios in production. The NAIC publishes a scenario-picking tool (the Scenario_Picking_Data_Set in each monthly release) supporting prescribed reduced subsets (e.g., 1,000 or 200 scenarios) selected to preserve the tail metrics that drive CTE calculations. The additional-statistics workbook in each release carries Conning's official distributional summaries - the right source for percentile fan charts, and a planned addition to this page.
Honest limits of this view: 16 SERT paths under-represent the stochastic set's interior (they were designed to bound, not to sample), annual thinning hides intra-year whipsaw, spot quoting differs from par pages, and the non-rate variables that dominate VM-21 results (separate-account equity paths) aren't shown at all. Treat this page as a shape-of-uncertainty visual and a regulatory artifact viewer, not as a substitute for running the files. The files themselves are free - the source link downloads exactly what this page was built from.