Stress testing and scenario analysis with every shock cited, every portfolio run each week and every scenario approved by the CRO
New scenarios drafted in hours, every portfolio run weekly, every loss explained.
See one case, screen by screen ↓Stress testing and scenario analysis shows what a market shock would do to every portfolio. In market risk stress testing, a scenario sets moves in rates, currencies, credit spreads, equities and commodities, and each portfolio's loss is measured against its own limit. Stress Testing and Scenario Analysis adds seven agents that draft historical and hypothetical scenarios from central bank texts, test each shock against history, run every portfolio weekly and explain every loss, while the analyst prepares and the CRO approves every scenario.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Stress testing and scenario analysis starts with one question from the CIO or the CRO every time a central bank moves or a market breaks: what would this do to our portfolios? Today a market risk analyst reads the source — often a Bank of Japan, Bundesbank or Banque de France text through browser translation — sets the factor shocks by hand in the risk engine's scenario editor, and argues each shock size in email with no check against history. Then risk operations runs a subset of funds, on positions that may be stale and private assets on old marks.
Reading the results and finding the drivers takes hours per portfolio, and the drivers are explained differently each time. Limit breaches go to portfolio managers by email, with no clock and no record of the answer. The scenario changes between runs, the committee deck is built from pasted numbers weeks later, and when an auditor or examiner asks which version produced a committee figure, it often cannot be reproduced. Meanwhile, a fund that runs a derivatives risk management program under SEC Rule 18f-4 must stress test at least weekly.
Estimated split for one new hypothetical scenario, from the event to an agreed scenario in the committee pack, by hand and with the solution.
Seven agents draft, calibrate, run and explain every scenario; the analyst prepares and the CRO approves.
Quillfield Capital runs 300 portfolios — mutual funds, ETFs, separate accounts, hedge fund vehicles, private funds and wealth models — against 40 official stress test scenarios every week. On Friday the CIO asked: what if the Bank of Japan stops buying bonds altogether? It is Monday, October 19, 2026, three days before the Q3 risk committee. Here is what happens next, screen by screen, in the working solution.
Margaret's home shows four numbers: 12,000 results from Friday's weekly run, 4 portfolios over a limit plus 1 provisional, 3 items waiting for her — one scenario and two breach decisions — and the risk committee on Oct 22 with 6 of 7 pack sections drafted. Under Needs you, each item has one action. Beside it, the agents' overnight work and a conversation she can ask anything.
“SC-2026-041 "Japan policy shock" waits for your approval. The overnight preview ran all 300 portfolios: 1 over its limit, 6 at 90% or more.”
SC-2026-041 “Japan policy shock: BoJ ends bond buying”, v3, hypothetical, 10-day horizon, reads top to bottom as 12 factor shocks: JGB 10-year +80bp, JGB 30-year +95bp, US Treasury 10-year +10bp, USD/JPY −14%, US high yield spreads +140bp, TOPIX −16%, gold +4% and five more. Each shock carries its source and the historical window it echoes. On the right, every portfolio's loss in basis points of NAV is a ranked bar against its own limit, with the three driving positions under it.
“Propagated by the risk engine to 1,840 risk factors, with correlations from the 1990–2026 stress windows.”
The Scenario agent drafted v1 on Oct 14 from four sources: the Bank of Japan's July 31, 2024 statement on its reduction plan for JGB purchases, its June 2025 interim assessment, the July–August 2024 yen carry trade unwind as a market window, and Quillfield's own house view note. The statement opens in its original Japanese beside an English text aligned sentence by sentence, and each highlighted passage links to the shock it supports.
“The original is the record; the English is aligned sentence by sentence.”
The Calibration agent tested the 12 shocks against daily factor history from 1990 to 2026, scaled to the 10-day horizon, and ran three consistency checks: joint plausibility 1-in-150. It sent two back. US Treasury 10-year −15bp contradicted the scenario's own story, since Japanese investors selling foreign bonds push US yields up; Kenji accepted +10bp. JGB 10-year +80bp sits beyond every 10-day move since 1999 — the largest was +63bp in the 2003 “VaR shock” — and the agent suggested +60bp.
Kenji kept it, on the record for the CRO: “Keep +80bp. The scenario removes the Bank’s promise to buy more when yields jump; in 2003 it was still buying.”
The Run agent previewed v3 on all 300 portfolios overnight, full revaluation on the risk engine: one over its limit, six at 90% or more. Asked why Quillfield Global Balanced Fund breaches, the Explainer agent answers with citations: it loses 312bp of NAV ($443.0M) against a 300bp limit, and three positions drive 285bp of it — the short-yen forward overlay −126bp, the global equity sleeve −118bp and the JGB 20Y 1.1% Mar-2046 −41bp. Without the overlay, the fund would lose 186bp.
Asked why +80bp was kept: “At +60bp, Global Balanced would lose 302bp, still 101% of its limit: the breach comes from the yen, not the JGB shock.”
The fund page for QF-0007 — USD, NAV $14.2B, uses derivatives under Rule 18f-4 — breaks the 312bp into factors and positions as of the Oct 16 close: the USD/JPY forward overlay, short ¥199.7 billion against the dollar, 9.0% of NAV, costs $178.9M. Below sit all 40 official scenarios against the fund's limit. The what-if halves the overlay to 4.5% of NAV: the fund would lose 249bp, 83% of its limit.
“The Explainer never recommends a trade. Trades are decided by the PM and placed in the order management system.”
The decision sheet sums up the case: drafted by the Scenario agent, then Kenji as maker; calibration 10 within range, 1 changed, 1 kept with a reason; the preview's results; and the effect on the official set — it replaces SC-2024-017 “BoJ yield-curve band widening”, retired because yield-curve control ended in March 2024, so the set stays at 40. Margaret picks a reason code — A1 approve, A2 approve with a condition, R1 return to the analyst, X1 reject — and writes her comment on the record.
Kenji drafted it, so he can never approve it: the bar reads “Maker: Kenji Watanabe · checker: you”.
SC-2026-041 v3 joins the weekly run on Fri Oct 23. The provisional breach becomes case BR-2026-122 for Global Balanced, and the Escalation agent notifies Elena Marchetti, who has two business days to respond — by Wed Oct 21. SC-2024-017 is retired, and the official set stays at 40.
“SC-2026-041 v3 is official · BR-2026-122 opened · Elena Marchetti notified”
Breach BR-2026-118: Quillfield Global Macro Fund loses 538bp in the US term premium spike scenario, 108% of its 500bp limit, driven by long 10-year Treasury futures (−301bp). Daniel Okafor answered on Oct 19: cut those futures by 40% by the close on Oct 21, taking the loss to about 420bp. Margaret's decision is due Oct 26, with reason codes D1 to D5; unresolved cases reach the risk committee at 10 business days.
D2, accept within tolerance: “Not available: 108% is over the 102% tolerance.”
The Q3 2026 pack was drafted from signed runs WR-2026-30 to WR-2026-42: 13 weekly runs, 156,000 results. Every underlined number opens its run, scenario version and as-of date. The regulatory extracts use the same signed runs: Rule 18f-4 weekly stress results for the funds that use derivatives, the Form PF Question 42 factor effects for the Global Macro Fund, and the quarterly UCITS stress summary under ESMA guidelines.
“Factor sizes per the firm’s Form PF methodology; amended Form PF compliance date July 1, 2027.”
The same scenario, seen by five of the people who carry it — what the weeks looked like, and what they look like now.
Seven agents draft, calibrate, run and explain every scenario; the analyst prepares and the CRO approves.
Drafts factor shocks from central bank statements, financial stability reports and market events in any language, in the risk engine's own factor set; every shock cited to a source passage and page, or to a historical window.
Tests each shock against daily factor history from 1990 to 2026 — percentile, nearest analogs, horizon scaling — and the shocks against each other, and sends back anything outside the plausible range with a suggestion.
Runs previews and the weekly run: pulls positions from portfolio accounting and the order management system, prices from two market data vendors, maps instruments to risk factors, sends batches to the risk engine, retries and reconciles.
Names the positions and factors behind each loss in plain words, each cited, and answers why questions from the CRO, portfolio managers and the committee.
Compares losses to limits, opens breach cases, starts the clocks, notifies and chases portfolio managers, and escalates to the risk committee at 10 business days.
Drafts the quarterly risk committee pack and the regulatory extracts — Rule 18f-4, Form PF Question 42, UCITS — from signed runs only, and flags what changed since last quarter.
Seals every scenario version, run and decision with a hash and builds evidence packs for auditors and regulators.
Approves every scenario. The agents propose; a named person decides.
Ask about any scenario, portfolio, breach or run. The agent that knows answers, and every answer cites the run, the scenario version or the source it came from.
Why does Global Balanced breach?
Global Balanced loses 312bp of NAV ($443.0M) in SC-2026-041 v3, 12bp past its 300bp limit. Three positions drive 285bp: the short-yen forward overlay −126bp, the global equity sleeve −118bp and the JGB 20Y 1.1% Mar-2046 −41bp. The overlay was opened on Sep 14; without it the fund would lose 186bp.
Why keep JGB +80bp?
JGB 10-year +80bp over 10 days is beyond every 10-day move since 1999; the largest was +63bp in Jun–Jul 2003. The Calibration agent sent it back and suggested +60bp. Kenji Watanabe kept +80bp with a reason: the scenario removes the Bank's promise to buy more when yields jump, and in 2003 the Bank was still buying. At +60bp, Global Balanced would still lose 302bp, 101% of its limit.
What if the yen move is −12%?
At USD/JPY −12% instead of −14%, the overlay loses 108bp instead of 126bp, and Global Balanced loses 294bp, 98% of its limit: a warning, not a breach. The house view note puts the move at 12% to 15%; −14% sits inside that range, at the 99.7th percentile.
Is the committee pack ready?
The Q3 pack has 7 sections; 6 are drafted from signed runs WR-2026-30 to WR-2026-42. Section 5, on the Japan scenario, waits for your approval of SC-2026-041; until then it is marked provisional. Robert Lindqvist has asked one question on section 4.
13 screens from the working solution, on its sample data. Pick one to see it large.
How scenarios are approved and breaches are handled, set by the firm. Every change is versioned and needs a second person; roles enforce segregation of duties.
| Setting | Default | Choose from |
|---|---|---|
| Warning threshold | 90% of limit | share of limit that raises a warning; breach at 100% |
| Tolerance band | 102% | the most of a limit the CRO may accept |
| PM response clock | 2 business days | from the case opening |
| CRO decision clock | 5 business days | from the PM's response; committee escalation at 10 |
| Official set size | 40 | kept constant; additions retire or replace |
| Calibration | 1990–2026 daily history | shocks rarer than 1-in-100 need a written reason; rarer than 1-in-500 are blocked |
| Review every official scenario | 12 months | re-approved by the CRO |
| Weekly run | Fri 10:00 p.m. ET | every portfolio against every official scenario; previews on demand |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published requirements (SEC Rule 18f-4 — stress testing at least weekly for funds with a derivatives risk management program, results reported to the board; SEC/CFTC extension of the amended Form PF compliance date to July 1, 2027, Aug 31, 2026). Quillfield Capital and every person, fund and client named on this page are fictional — characters and sample data in the working solution. Real central banks, indices and historical market windows appear only as sources and analogs.
Stress testing and scenario analysis shows what a market shock would do to every portfolio. In market risk stress testing, a scenario sets moves in rates, currencies, credit spreads, equities and commodities, and each portfolio's loss is measured against its own limit. Stress Testing and Scenario Analysis adds seven agents that draft historical and hypothetical scenarios from central bank texts, test each shock against history, run every portfolio weekly and explain every loss, while the analyst prepares and the CRO approves every scenario.
Used together in risk management, scenario analysis defines the event and the stress test measures what it would cost. In the working solution, a scenario is a named set of factor shocks: historical, replaying a window such as the August 2024 yen carry unwind; hypothetical, an event that has not happened, such as the Bank of Japan ending bond buying; or regulatory. The stress test runs that scenario against every portfolio on the risk engine and compares each loss with the portfolio's own limit.
The weekly run puts every portfolio against every official scenario each Friday night. From the same signed runs, the Committee pack agent drafts the Rule 18f-4 weekly stress results for funds that use derivatives, the Form PF Question 42 factor effects for the hedge fund vehicle and the quarterly UCITS stress summary under ESMA guidelines, each figure linked to its run. The CRO signs the pack; under Rule 18f-4, the derivatives risk manager reports the results to the board.
The Scenario agent reads the statement in its original language, aligns an English text sentence by sentence, and proposes each factor shock with the passage and page it rests on, or the historical window it echoes. The Calibration agent, a separate agent, then tests every shock against 1990–2026 history and sends back what is implausible. The analyst changes it or keeps it with a written reason the CRO sees.
No. The Run agent drives your existing risk engine: it pulls positions from portfolio accounting and the order management system, prices from your market data vendors, maps instruments to risk factors and sends the batches. It reconciles the totals to NAV and holds a portfolio rather than guessing a price.
No. The agents never change positions or prices, never recommend a trade, never make a scenario official, never close a breach and never send the committee pack. What-ifs are labeled as illustrations; trades are decided by the portfolio manager and placed in the order management system.
Yes. The analyst who drafts a scenario can never approve it; only the CRO approves, returns or rejects, with a reason code. Portfolio managers answer breaches on a two-business-day clock, the CRO decides each one, limit changes need the CRO and the committee chair, and every step is on a sealed audit trail.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your scenario set and rules, limit register, roles and the systems your risk engine already uses — and it goes live once every quality gate has passed. We will walk you through it on your own portfolios first.
Every holding bucketed monthly, every override explained, no filing trigger missed.
1-min demoLiquidity Stress TestingEach fund's breach day found and evidenced, and the quarterly pack signed in a day.
1-min demoDerivatives Risk Management ProgramEvery daily VaR test explained in plain words, every escalation signed, no late N-RN.
1-min demoWe'll run Stress Testing and Scenario Analysis on your own scenarios, limits and risk engine.