Interest rate risk management that back-tests every key assumption against what members actually did, puts a number on each gap, and keeps the model unchanged until ALCO approves
Every key assumption checked against what members actually did, with proof for examiners.
See one case, screen by screen ↓Interest rate risk management is how a credit union or bank measures and limits what a change in rates does to its net interest income and the economic value of its balance sheet, and governs the assumptions behind those measures. Interest Rate Risk (IRRBB) Management puts seven agents on that work: they back-test every key assumption against what members actually did, re-run NEV and NII with the observed value, and draft a cited memo, while the ALM Manager, Model Validation and ALCO decide.
AI use cases in banking: a buyer's guide to agentic AI for bank and credit union operations →
Updated
Interest rate risk management at a credit union is only as good as the model's assumptions: how fast regular shares run off, how fast auto loans prepay, how much of a rate rise reaches certificate holders. NCUA's rule for every federally insured credit union over $50 million in assets, 12 CFR 741.3(b)(5) and Appendix B to Part 741, asks management to rely on assumptions that are reasonable and supportable, and to document any changes based on observed information. The 2010 interagency advisory adds back-testing against actual outcomes and independent review.
Today those assumptions come from a deposit study done every few years, sit in a spreadsheet, and are back-tested once a year before the exam. When members change behavior, as they did when certificate rates rose, a broken deposit decay rate or deposit beta can overstate NEV for four quarters before anyone sees it. The change that fixes it travels by email, the second line sees it after the fact, and ALCO minutes say "approved assumptions" without saying which. Interest Rate Risk (IRRBB) Management puts seven agents in front of that work, so the ALM Manager starts the quarter with the evidence and the impact on one page.
Estimated split of the business days between a quarter's back-test and a cited memo in front of the validator, by spreadsheet and email today and with the agents; the 'after' column totals about two days, the solution's design goal. Our arithmetic, kept modest.
Agents back-test, re-run NEV and NII, draft and challenge; the ALM Manager, a validator and ALCO decide.
Norrell Federal Credit Union, a $6.8 billion credit union with 520,000 members, runs assumption set AS-2026.3, approved by ALCO on October 28, 2025. The Q3 2026 back-test ran on October 19. Here is what happens to regular share decay, screen by screen, in the working solution.
The Back-test agent measured all 28 key assumptions on the credit union's own core and loan servicing extracts. Each is one row with two dots on one scale: a hollow dot for what the model assumed a year ago, a solid dot for what members actually did, and the tolerance as a pale band. The September loan servicing extract has not arrived, so HELOC line utilization waits, and the register says so.
Of 28 promises, 20 held, 5 are drifting, 2 broke and 1 is waiting for an extract.
Regular share decay was assumed at 14.0% a year, from the 2024 deposit study; members ran off at 22.6%. Nadia clicks the citation and the core extract opens with the lines the agents used highlighted: 498,300 accounts holding $1,420,612,400, and a runoff of $321,000,000. BSA fields were removed before any agent read it, and member identifiers are masked.
Twelve monthly dots sit against a promise line of 1.17 points a month. March to June stand out, at 2.9, 3.1, 2.6 and 2.4 points: the months of the 13-month certificate special. The Extract Reconciler tied the extract to the general ledger within $291,700, or 0.02%.
41,800 accounts closed or drained under $5 account for 9.8 points; $105.1 million moved into share certificates, $88.0 million of it in the special, for 7.4 points; other withdrawals by members who stayed add 5.4 points. Sample accounts sit behind the number with names hidden and numbers masked.
Why the gap, confidence 0.94.
The Sensitivity agent re-ran the ALM model in a what-if set, never the live one. At +300 bp the NEV ratio falls from 9.1% to 8.2%, still above the 7.0% floor, and NII year one moves from −6.4% to −8.1% against a −15% limit. Limits Watch reads NEV sensitivity at 33.3% against a 40% limit, and the NCUA NEV Supervisory Test stays Low.
A second change is open: certificate dividend beta from 55% to 71%. With both, the +300 bp NEV ratio would be 7.6% and NEV sensitivity 37.7%. Limits Watch flags it before ALCO meets.
With both changes the +300 bp NEV ratio keeps only 0.6 points above the floor, under the 1.0-point line of FIN-04 §4.4: ALCO should discuss strategy before it votes.
The Governance agent drafted memo P-2026-15 in five sections with 11 citations: what the model assumes, what members did, why, the impact, and a recommendation to adopt 22.6% in AS-2026.4 and re-test on January 19, 2027. Nadia proposes the value, picks reason code SB-1, observed behavior and a three-quarter trend, and submits it to the validator at 10:21 AM.
The Challenge agent reads the raw extract, never the first-line memo, and recomputes 22.58%, a difference of 0.02 points. It then tries to explain the gap away. Without March to June, runoff would still be 17.4%, 3.4 points over the assumption, and the 2027 budget plans two certificate specials, so the effect recurs.
Samuel accepts with reason code MV-AC, accept with a condition. The ALCO Packet agent queues the change as item 2 for the October 27 ALCO, with his condition attached.
Re-test in the Q4 2026 back-test on January 19, 2027. If runoff without certificate specials is under 17%, bring a revised value to ALCO.
The October 27 agenda holds only changes Model Validation has reviewed, each with its NEV and NII impact, its maker and its validator decision. The deferred certificate dividend beta change goes to November 24. Until ALCO approves and Nadia releases the new set, the model keeps using 14.0%.
The ALM model is using assumption set AS-2026.3. Nothing on this agenda reaches the model until ALCO approves it and the ALM Manager releases it.
The same change, seen by the people who propose it, check it, vote on it and oversee it: what their day looked like, and what it looks like now.
Agents back-test, re-run NEV and NII, draft and challenge; the ALM Manager, a validator and ALCO decide.
Pulls the quarter's share, certificate and loan extracts, ties balances to the general ledger within 0.10%, strips BSA fields and masks member data.
Measures each of the 28 assumptions on the cohort, compares it with the promise, and attributes the gap to causes with citations.
Asks the ALM model to re-run NEV and NII in a what-if set with the observed value, alone and together with every other open change.
Reads every run against the board's interest rate risk policy and the NCUA NEV Supervisory Test, and flags headroom under 1.0 point.
Drafts change and re-affirmation memos from the back-test, the re-run, the policy and prior memos, with every figure cited.
Works for Model Validation: recomputes the observed value from the raw extract and tests other explanations, such as one-time events, data errors and cohort choice.
Builds the ALCO item and the board-portal packet, records each member's vote, and stages the next assumption set.
Maker proposes, a second line checks, the committee approves. The agents propose; a named person decides.
The assistant answers with the source behind every figure, and can open the screen it is talking about.
Why did regular shares run off faster than we assumed?
Three causes, from the core extract: 41,800 accounts closed or drained (9.8 points), $105.1 million moved into certificates (7.4 points, $88.0 million in the 13-month special), and $76.7 million of other withdrawals (5.4 points). The 2024 study assumed 14.0% on low-rate years.
What happens at +300 if ALCO accepts both open changes?
The NEV ratio at +300 bp would be 7.6%, 0.6 points above the 7.0% floor. That is under the 1.0-point early-warning line, so FIN-04 §4.4 asks ALCO for a strategy discussion. Sensitivity would be 37.7%, still under 40%.
Which assumptions drifted two quarters in a row?
Four: regular share decay and certificate dividend beta drifted in Q1 and Q2 and broke in Q3, so both were escalated to Carla Benning on October 19. Certificate renewal rate and reinvestment spread have drifted for two quarters and need a comment by November 2.
Build the NCUA examiner pack for Q3
Ready for item 14 of the exam request list: the register, every back-test with its extract and tie-out, the what-if runs, memos P-2026-14 and P-2026-15, the Challenge agent's work and the validator's decisions, ALCO minutes and the limits report, with Appendix B mapped section by section.
13 screens from the working solution, on its sample data. Pick one to see it large.
Tolerances, limits, clocks and data access are settings. Every change is a change request that a second administrator approves before it applies, and the limits mirror the board-approved policy.
| Setting | Default | Choose from |
|---|---|---|
| Tolerance, regular share decay | ±3.0 points | Dividend betas ±10 points; drifting at two thirds of the tolerance |
| Back-test calendar | Quarterly, the third Monday after quarter end | Escalate to the CFO after two quarters outside tolerance |
| NEV ratio floor (IRR Policy FIN-04 §4.2) | At least 7.0% at +300 bp | At least 5.0% at +400 bp; changes only by a board vote |
| NII year-one change | No worse than −15% at ±300 bp | No worse than −20% at +400 bp |
| Early warning | Under 1.0 point of headroom, to ALCO | A breach goes to the CFO the same day |
| Outside tolerance | Memo or documented re-affirmation within 15 business days | And before the next ALCO; drifting: comment in 10; missing extract: 2 |
| Back-test attribution confidence | 0.85 | Below it, the ALM Manager decides |
| Member data | Masked; every unmasked view logged | Aggregate only for ALCO members and the board; no member-level export |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
Sources: 12 CFR 741.3(b)(5) and Appendix B to Part 741, interest rate risk policy and program for credit unions over $50 million (ecfr.gov) · NCUA Supervisory Letter 22-01, NEV Supervisory Test (ncua.gov) · Interagency Advisory on Interest Rate Risk Management, SR 10-1 (federalreserve.gov) · Risk.net IRRBB benchmarking, NII sensitivity as a primary metric. “demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our arithmetic (the business-day split) · “typical” = published figures and regulatory thresholds. Norrell Federal Credit Union, its members and the people named on this page are fictional: characters and sample data in the working solution.
Interest rate risk management is how a credit union or bank measures and limits what a change in rates does to its net interest income and the economic value of its balance sheet, and governs the assumptions behind those measures. Interest Rate Risk (IRRBB) Management puts seven agents on that work: they back-test every key assumption against what members actually did, re-run NEV and NII with the observed value, and draft a cited memo, while the ALM Manager, Model Validation and ALCO decide.
IRRBB is interest rate risk in the banking book: the risk to earnings and capital from rate changes on loans, deposits and investments a bank or credit union holds rather than trades. It is usually measured two ways, net interest income over the next year and the economic value of equity, which credit unions call net economic value (NEV).
NII sensitivity shows how net interest income over the next year changes under a rate shock; EVE, or NEV at a credit union, shows how the value of assets less liabilities changes. The solution re-runs both at every shock from −300 to +400 bp, and reads each against its board limit, for example an NEV ratio of at least 7.0% and an NII change no worse than −15% at +300 bp in the demo policy.
The Back-test agent takes the accounts open at the start of the year, ties their balance to the general ledger, and measures how much of it ran off. In the demo, 498,300 regular share accounts held $1,420.6 million and $321.0 million ran off, 22.6% against 14.0% assumed, with the gap split into cited causes. It is measured by balances, as the 2024 deposit study was; counted by accounts, runoff would read 8.4%.
12 CFR 741.3(b)(5) requires every federally insured credit union with more than $50 million in assets to have a written interest rate risk policy and an effective program, and Appendix B to Part 741 asks for reasonable, supportable assumptions and documented changes based on observed information. NCUA Supervisory Letter 22-01 reads a post-shock NEV ratio above 7% with sensitivity below 40% as Low risk, and below 4% or above 65% as High.
Yes. The working solution shows a credit union, where the measure is called NEV and the owner is the ALM Manager; at a bank the same measure is EVE and the work usually sits with the Corporate Treasurer. The 2010 interagency advisory on interest rate risk management sets the same expectations for both: documented assumptions, back-testing against actual outcomes and independent review.
No. The agents measure, re-run in what-if sets, draft and challenge; they never change an assumption, a limit or the live model, and never submit, vote or release. The ALM Manager proposes with a reason code, Model Validation accepts or challenges with its own, and the model uses a change only after ALCO approves and the ALM Manager releases the new set.
The Agentic Solution Engine builds and deploys it from your requirements and documents: your assumption set, your interest rate risk policy, your last deposit study and a quarter of extracts. It goes live once every quality gate has passed, and we walk you through it on your own assumptions first.
Every morning, each liquidity move explained by the accounts behind it, before 7:30 a.m.
1-min demoContingency Funding PlanA plan that works on the day it is needed, with every line tested and every loan counted.
1-min demoDeposit PricingEvery rate exception answered while the member waits, priced against other funding.
1-min demoWe'll run Interest Rate Risk (IRRBB) Management on your own assumption set, a quarter of extracts and your interest rate risk policy.