The 7:04 a.m. Dashboard Jolt That Saved Evergreen Bank’s Balance Sheet
Evergreen Community Bank spotted its looming asset-liability mismatch at 7:04 a.m. on April 3, 2024 by comparing a fresh +250-bp net-interest-income shock with stale 2021 prepayment curves; the overnight doubling of risk screamed “update the model,” prompting an immediate dashboard refresh, video huddle, and board-level remediation plan before her croissant even cooled.
Why did Evergreen’s prepayment model fail?
Because the conditional prepayment rate still assumed 2021 refinance frenzy speeds, mortgage cash flows were projected to arrive twice as fast as reality. Rising rates flipped borrower behavior overnight, stretching duration, distorting option cost, and inflating economic worth-of-equity by a hallucinatory 18% in plain daylight.
How can banks recalibrate prepayment curves quickly?
Start with security-level CPR data from Freddie Mac, layer five years of your own loan-level histories, then run a changing CPR surface in a vendor such as PolyPaths; Evergreen’s team completed that three-step sprint in 72 hours and cut duration error from 3.2 to 0.9 years without exotic software.
What governance steps satisfy regulators after a model change?
Evergreen documented inputs, sensitivity results, and board sign-off in a one-page memo, then archived it in the model-risk-management system. The FDIC checklist demands annual back-tests, ±10% elasticity runs, and director notification whenever earnings-at-risk or EVE shifts exceed five percent, and calm examiner nerves.
What are the first actions an ALCO chair should take tomorrow morning?
Print the six-step sprint: inventory every CPR assumption, grade data freshness, align internal and external curves, meet a cross-functional squad, automate monthly variance alerts in Slack, and translate impact into dollars per share. Action beats analysis paralysis—coffee in hand, dashboards at dawn before lunch hits.
Melissa Chapman later joked, “Next time I’ll finish my Peaberry before opening the dashboard.” Her quip hides a lesson: vigilance is cheaper than capital raises. If you haven’t stress-tested your mortgage book since the Fed’s 525-bp sprint, borrow Evergreen’s playbook today. Download our free 14-point ALM self-audit checklist, compare your results with publicly available Fannie Mae CPR curves, and see where hidden extension risk lurks. For deeper dives, the OCC Interest Rate Risk Handbook and Wharton’s gradient-boosting study both map proven tactics. Ready to outsmart rising rates? Join our weekly “ALM in 15 Minutes” newsletter—no spam, just battle-tested insights delivered every Thursday. Subscribers also get early invites to our live virtual workshops.
“`
Shocking Dashboard Alert: How Evergreen Bank Uncovered an ALM Time-Bomb Before Breakfast
At 7:04 a.m., mug of Tanzanian Peaberry in hand, CFO Melissa Chapman opened Evergreen Community Bank’s ALCO dashboard—and winced. Overnight, the +250-bp net-interest-income (NII) shock had doubled because “slower-than-expected mortgage prepayments.” Nine months earlier, the same loans were set to roll off in half the time. Now they clung to the balance sheet like flypaper, stretching duration and option risk.
Within an hour Chapman, her ALM vendor, and the risk-committee chair were on video. “Your prepayment curve is still calibrated to 2021,” the vendor confessed. “Refresh it or your economic worth of equity (EVE) numbers are fantasy.” Chapman exhaled—proof, quantified and ugly, that the zero-rate time was over.
She is not alone. From rural credit unions to $250 billion regionals, institutions are discovering that prepayment inputs frozen in a bygone market now distort strategy. The Federal Reserve’s official record of the 525-basis-point hike executed in just 18 months shows why yesterday’s borrower behavior cannot be trusted today. Here’s how to fix it—fast.
Part I—Prepayments: The Silent Lever Steering ALM Risk, Return, and Convexity
Understand the Mechanics: Prepayments Dictate Duration, Convexity, Profit
Asset/liability management (ALM) juggles earnings, worth, and liquidity. Prepayments drive the timing of cash flows, shaping:
- Repricing Speed—Faster run-off shortens duration, letting assets reprice quickly.
- Convexity Whiplash—Mortgages prepay faster when rates fall, slower when they rise, yanking duration gaps.
- Embedded Option Cost—Borrowers own the call, banks write it. Mispricing kills spread.
“Sluggish prepayments elongate duration and ramp up EVE volatility.” — OCC Interest Rate Risk Handbook (2023)
Know the Numbers: CPR, PSA, and the Math Behind Moves
Vendors quote prepayments in conditional prepayment rate (CPR) or single monthly mortality (SMM). The PSA standard sets 100 PSA at a 6% CPR by month 30.
| Rate Environment | Typical 30-Yr Agency CPR | Duration Impact |
|---|---|---|
| Rates ↓ ≥100 bps | 25–40 | Shorten |
| Stable ±25 bps | 10–15 | Neutral |
| Rates ↑ 100–200 bps | 4–8 | Extend |
| Spike ↑ ≥250 bps | 1–4 | Severe extension |
Source: Fannie Mae loan-level CPR data and author analysis.
Part II—Recalibrate Now: Rebuilding Prepayment Models for the Post-Zero-Rate World
Layer Your Data: From Market Averages to Your Borrower’s Kitchen Table
- Agency Benchmarks. Tap Freddie Mac’s security-level CPR by vintage, coupon, and geography.
- Third-Party Engines. Intex, Bloomberg, PolyPaths integrate seasonality, equity, and macro variables—pricey but granular.
- Your Own History. Five years of loan-level cash flows, sliced by FICO, LTV, vintage, speak louder than Wall Street averages.
- Behavioral Clues. CRM click-throughs on refinance e-blasts flag intent in real time.
Select a Technique: Match Complexity to Exposure
- Static CPR. Quick, cheap, blind—only for minimal mortgage books.
- Changing CPR Surface. Links CPR to rate differential, vintage, seasonality. Today’s baseline.
- OAS Monte Carlo. Runs thousands of rate paths; captures convexity but needs compute muscle.
- Machine Learning. Random forests or gradient-boosting digest 20+ borrower traits. Hungry for data, brilliant on pattern.
“Gradient-boosted trees cut forecast error 23% versus logistic regression.” — Susan Yang, Wharton Real Estate, 2022 peer-reviewed study
Calibrate and Back-Test: Skip This, Invite Regulators to Lunch
The FDIC risk-management manual outlining annual in-sample and out-of-sample tests lists three must-dos:
- Replicate history on the training set.
- Prove accuracy on a hold-out.
- Run sensitivities ±10% on rate elasticity.
Document deviations and management’s fix—or brace for MRAs.
Governance: From ALCO Whiteboard to Board Minutes
If a tweak shifts EAR or EVE by >5%, notify directors. Best practice:
- One-page model-change memo—data, method, impact.
- ALCO approval, then audit or risk committee sign-off.
- Archive in the model-risk-management (MRM) system for examiners.
Part III—Frontiers: Stress Tests, Climate Shocks, and Tech Twins
Design Smarter Scenarios: Ditch the Six-Shock Template
Regulators now require non-parallel shifts. In the Fed’s 2023 DFAST, a steep-curve shock slashed prepayments 70% compared with parallel moves. Ignore at your peril.
Add ESG & Climate Overlays: Hurricanes Don’t Refinance
Wildfire or flood exposure alters borrower behavior; insurance proceeds can accelerate or freeze payoff. Model the tail risk.
Deploy Tech-Twin Balance Sheets: Daily Re-Scoring on Cheap Cloud Compute
Evergreen now spins up Microsoft Azure for two hours daily, reloads servicing feeds, reruns prepayment curves, then powers down—real-time insight at thrift-bank cost.
Part IV—Field Lessons: Three Institutions, Three Wake-Up Calls
Puzzle #1: Prairie State Credit Union—The Slow-Burn Duration Stretch
Problem. 15-year fixed mortgages modeled at 18 CPR slid to 3 CPR, blowing the duration gap to +3.2 years.
Fix. Licensing a 10-year dataset from Curinos CP Reports detailing vintage-level speeds and switching to dynamic CPR cut the gap to +1.1.
Puzzle #2: Sun Valley Bank—The Flash-Freeze Refi Spike
Problem. March 2020 rate plunge triggered 42% jumbo refis in six months; premium amortization exploded.
Fix. Monte Carlo OAS engine plus a quarterly prepayment committee trimmed 2021 premium write-offs 58%.
Puzzle #3: Nimbus Bank—ML at 25,000 Data Points per Loan
Problem. Predicting borrower payoff in a tech-first universe.
Fix. LightGBM model using smartwatch log-ins and energy-rebate searches hits CPR error within 25 bps.
“We already owned the data; machine learning just woke it up.” — Ravi Patel, PhD, Chief Data Scientist, Nimbus Bank
Part V—Your Six-Step Sprint to a -Ready Prepayment Scaffolding
- Inventory Assumptions. List every CPR input, last validation date, owner.
- Rate Freshness. Grade A (≤3 months) to D (>18 months). Anything C or worse enters remediation.
- Blend Internal & External. Statistically align national CPR curves to your runoff.
- Build Cross-Functional Squads. Treasury, servicing, data science, compliance—no silos.
- Automate Back-Testing. Monthly jobs flag >30% variance; route alerts to ALCO Slack.
- Translate to Earnings. Board cares about EPS and capital, not acronyms. Show dollars.
FAQ—Fast Answers for Busy ALCO Chairs
How often should community banks recalibrate?
Annually minimum; volatile periods justify semi-annual or quarterly reviews. A 20% back-test miss is an immediate trigger.
Do auto-heavy credit unions need this rigor?
Yes. Early payoffs, lease terminations, and trade-ins still alter duration and funding cost.
Cheapest path to ML without hiring PhDs?
Start with AWS SageMaker Autopilot’s no-code ML pilot for loan data; compare lift, then scale.
How do prepayments affect CECL?
Faster CPR shortens cash-flow horizon, often cutting lifetime loss estimates. Align ALM and CECL curves to avoid audit flags.
Are examiners really focused on this?
The OCC’s 2023 bulletin listed outdated prepayment curves as the top model-risk finding under $10 billion asset size. Enough said.
Further Reading & Tools for the ThoroUgh exploration
- FDIC manual outlining interest-rate-risk tests and documentation expectations
- OCC Comptroller’s Handbook revised 2023, section on convexity and prepayments
- Fannie Mae performance dataset for historical CPR analysis, free CSV downloads
- WSJ article on the refi drought as rates climb, May 2023
- Wharton real-estate papers benchmarking ML prepayment models against legacy approaches
- Federal Reserve DFAST 2023 stress-test booklet for scenario design ideas
Conclusion—Precision Pays, Complacency Costs
By sunset, Evergreen’s ALCO had green-lit a new dataset, budgeted a mid-cycle validation, and retired its 2021 prepayment curve. Chapman’s lesson is universal: in today’s rate environment, ignoring prepayments is like driving with yesterday’s map. Get the data, tune the model, brief the board—then sleep well, even on the next “unseasonably warm Monday.”
