Eleven Months of Risk Reports on One Profitable Client
One prime brokerage left an unusually detailed internal record: limits, stress tests, margin calculations, and the decisions made as a concentrated client’s exposure kept growing.
By the team at Banyantree Investment Group
At the end of August 2020, the weekly risk report inside Credit Suisse carried four numbers against a single client.
Potential exposure: US$395m, against an approved limit of US$20m.
Severe Equity Crash exposure: US$921m, against a guideline threshold of US$250m.
Potential exposure was the bank's routine measure of counterparty danger: an estimate, run to a 95% confidence level, of what a client's default could cost once the collateral was consumed and the positions closed out. Severe Equity Crash was the bank's scenario for a genuine disaster, assuming developed markets fell 30% within a month, and emerging markets and Japan fell 45%.¹
Numbers of this kind exist to make something happen. The position shrinks, collateral arrives, the client finances the book somewhere else, or somebody with the authority approves a defined exception, with conditions attached and a date on which it expires.
At Credit Suisse, the first response was to re-examine the test.
The case for doing so deserves to be heard the way the risk department heard it. The client's holdings were large listed companies that traded in size every day. The bank held contractual rights to call collateral, terminate the client's swaps, and sell the shares hedging them, and it believed it could be out within days of a default. A crash spread across a month describes the damage to an investor who stays in the market for the whole event, and Credit Suisse had no intention of staying. What it needed to measure was the price move between a default and the exit, a window of days, and a shorter, shallower scenario might measure that more honestly.
The argument was accepted, and the account was measured against a replacement: developed markets down 20% within a week, emerging markets down 30%. The bank's board-commissioned review dates the change from September 2020; its UK regulator, from October.²
The framework had a name for it.
Bad Week.
The US$250m threshold did not move, and on its face that was the conservative half of the change: a milder shock, held to the same tolerance. On 1 September 2020, the account failed Bad Week at nearly twice that threshold, and it went on failing it, at 150% to 240%, through February 2021.² The bank had replaced the scenario with one it believed better matched the risk. The account still failed it.
The client was Archegos Capital Management, the family office of a former hedge-fund manager named Bill Hwang. Seven months separate that August report from the March morning the account collapsed and took about US$5.5bn of the bank's money with it. By August, the file had already been running for four. What happened across the whole eleven was reconstructed twice, once by a law firm working for the bank's own board, and once by its regulator.³
On the bank's paper, the holdings looked close to ordinary: ViacomCBS, Discovery, Baidu, and a Chinese online-education company called GSX Techedu, all listed on major exchanges and trading hundreds of millions of dollars a day. The unusual part sat underneath the names, in how they were held.
Hwang mostly did not own his stocks. He held them through total return swaps, contracts under which Credit Suisse paid him a share's gains and charged him its losses while the share itself never entered his name. To hedge its side of the trade, the bank bought the stock. In the ownership registers, against positions that were economically Hwang's, the name that appeared was Credit Suisse.
The substitution had a consequence, whatever its purpose. Positions of a size that would ordinarily announce themselves did not, because a bank holding shares against a derivative is what the registers show every day. And Credit Suisse was one of roughly a dozen counterparties running similar arrangements, each financing part of the same concentrated book, each hedging by buying many of the same shares, each seeing its own account rather than the whole.
The street-wide book was assembled only afterwards, in the Securities and Exchange Commission's complaint. At its stated peak on 22 March 2021, on the SEC's figures, more than US$36bn of invested capital carried more than US$160bn of gross exposure across Archegos's counterparties. On Archegos's own internal estimates, its swaps and shares together came to more than 70% of GSX Techedu, more than 60% of Discovery's Class A shares, and more than half of ViacomCBS. The SEC alleged that Hwang's buying had itself helped drive the prices against which his gains and his collateral were measured.⁴
Hwang's history was public record when the bank financed him. In December 2012 his previous firm, Tiger Asia, pleaded guilty to wire fraud in the United States over trading on inside information, and Hong Kong's tribunal later banned him and the firm from its market for four years.⁵ Credit Suisse's own systems rated Archegos near the bottom of its hedge-fund book, and the bank's later review found that it financed the account anyway, on margin terms among the most generous it extended to any comparable client.⁶
None of which was, by itself, indefensible. Prime brokerage sells financing to leveraged clients, its price is set in competition with other banks, and some accommodation of a client that six rivals are courting is the ordinary conduct of the trade. The fees were real and banked. The collateral was real, and the swaps were margined. For years the arrangement worked, and a leveraged client who eventually loses is not, on his own, an indictment of the lender.
The description of the holdings as liquid was also correct, on an ordinary day, and incomplete. A dozen lenders could each price a sensible exit against the same market depth without any of them seeing how many other exits had been priced against it. The question that mattered was what the positions would fetch on the day they were forced to move together, into a market where every seller held the same names and every margin call drew on the same finite pool of the client's cash.
No scenario in the bank's framework was built to compute that day, and Severe Equity Crash did not try. What the scenarios could do was price a violent fall against the exposure Credit Suisse had actually written, and they did, every week, whichever scenario was asked. Whether anything more could be seen from inside one bank's slice is a question the file itself would answer in February.
The intervening record survives in unusual detail. Many of its steps are dated and attributed, and the reasoning often survives with them.
From April 2020, the account sat outside its US$20m potential-exposure limit more or less continuously. The machinery for noticing worked. Archegos went onto the standing list of limit excesses. Reminder emails reached the credit team nearly every week. The analyst who owned the account held regular calls with Credit Control. The reports kept documenting a problem the bank was not resolving. At the start of March 2021, the bank's systems recorded that the account had exceeded its limit for more than 170 business days.⁷
The framework did not give every measure the same force. Breaching the potential-exposure limit required action or approval of a formal exception. Breaching a crash-scenario guideline did not. Yet those scenarios came closest to measuring what mattered: the cost of exiting under stress.¹
October 2020 brought the largest accommodation. Breaching a broader multi-factor stress measure as well, the account was given a bespoke scenario appetite of US$900m. The bank's own framework placed that figure US$100m below the appetite for AAA-rated clearing houses and key sovereigns, and described it as comparable to the tolerance reserved for the central banks of Canada, Germany, and the United States, among others. The paper recommending it described Archegos as a "significant relationship" for the prime brokerage and warned that a sudden increase in margin could do "irreversible damage to the client relationship". A senior risk officer approved it without comment.⁸
January 2021 brought the annual counterparty review, and the file's most compact exhibit. The review concluded that the client had become a worse risk: gross leverage was running around six times, against the four to five the rating model had been fed, and the internal rating was cut from BB- to B+. The same review proposed raising the potential-exposure limit from US$20m to US$50m. Under the bank's own guidance, a hedge fund rated B+ qualified for at most US$10m; the existing limit was already double that. Senior risk managers called it, in writing, an "unusual set of facts". After a brief call in early February, the increase was approved.⁹ The review had paired a downgrade to B+ with a higher limit, still far below the exposure the bank was already carrying.
In February, someone in the credit function converted the standing excess into arithmetic. Bringing the portfolio back inside its scenario threshold would require roughly US$1bn of additional margin. The same person asked, in writing, what purpose the bank's termination and margin rights served if it was unwilling to use them against a client of this importance. The reply, dated 9 February, was preserved by the regulator: "Asking for $1bn is pretty much asking them to move their business." The revenue profile, the reply noted, was significant.¹⁰
As a description of the market, the reply was accurate; the credit function had itself recorded that Archegos used six other prime brokers. But a constraint on a leveraged client settles one of two ways: the client brings the risk back inside, by cash or by size, or the client takes it elsewhere. Both are the limit operating as designed. An institution that will accept neither outcome has allowed the constraint to stop operating in practice.
Nine days later, the same corner of the bank wrote upward, to senior management in the first-line risk function. By then, the relevant pieces were already in the file. The account had been outside its limit for more than ten months and had failed the severe scenario and then its replacement. The credit function could name every issuer in the portfolio. It knew the client ran versions of the same book at six other prime brokers. It had already calculated that roughly US$1bn stood between the portfolio and the bank's own threshold. On 18 February, it added that a forced liquidation could be produced if all of the client's prime brokers raised their margin requirements at the same time.¹⁰ What no scenario in the framework had been built to calculate, one of its users had identified in writing and asked what would happen next.
A reply came the same day, addressed to the other matters in the exchange. On the record the regulator later assembled, the question itself does not appear to have been answered.¹⁰ The file preserves the correspondence on either side of it.
On 8 March, the committee overseeing the account agreed to seek Archegos's migration to dynamic margining, with collateral recalculated as the positions moved, "within the next couple of weeks". Failing that, the client would post an additional US$250m. The bank's own margin work had computed what migration would require on day one: between US$1.27bn and US$1.49bn. The fallback was less than a fifth of that number.¹¹
Dynamic margining required the client's agreement, and Archegos never gave it: the late-February proposal went unanswered, and the client cancelled the calls arranged to discuss it. That is the strongest defence of the missed fortnight. It also shows how little unilateral room the bank had left by then. Neither the migration nor the deposit occurred before the account failed.¹¹
In the third week of March 2021, ViacomCBS decided to raise money, selling new shares into a price that had been rising for months, a rise the SEC would later allege Hwang's own accumulation had helped produce.⁴ The offering landed poorly. The stock fell 6.7% on Monday 22 March, the day it was announced, and kept falling through the week. ViacomCBS was Archegos's largest single position.¹²
The reversal is recorded to the day. On Tuesday 23 March, Archegos still held more than US$600m of collateral at Credit Suisse above what that day's prices required. By Wednesday the cushion was gone; the account owed more than US$175m of variation margin, was called, and paid. The same day, Tencent Music, another large holding, fell about 20%, and the bank calculated what Thursday would require: approximately US$2.7bn. Senior managers, briefed on the figure, expressed surprise that positions capable of producing a call of that size were on the bank's books.¹³ The account producing them had been on those books, outside its limit, for eleven months.
Credit Suisse was not calling alone. Every prime broker financing Archegos ran its own margin arithmetic against its own slice of the same book, and the same falling prices arrived in all of their systems in the same week. Each call was individually correct. Collectively they were addressed to a single pool of cash, and by midweek Archegos was telling Credit Suisse that the US$6bn to US$7bn it had held unencumbered was gone, consumed by the calls that had arrived first.¹⁴ Five weeks earlier, the 18 February message had described this mechanism in a single sentence and asked what would happen next.
On Thursday evening, 25 March, Archegos convened its lenders and showed them, for the first time, the whole book: roughly US$120bn of gross exposure, US$70bn long and US$50bn short, standing on US$9bn to US$10bn of remaining equity.¹⁴ It asked for a standstill. No bank would declare a default; the positions would be wound down gradually. For the group, an orderly exit would probably have preserved more value than a stampede. For each bank, honouring it required something none of them could verify, the restraint of the others, while the advantage of selling first would belong entirely to whichever bank took it. No standstill was agreed.
Morgan Stanley sold blocks that Thursday evening. Goldman Sachs moved about US$6.6bn of stock before New York opened on Friday, and billions more once trading began.¹⁵ On Friday morning, Credit Suisse delivered its default notice, taking up at last the rights it had declined to use in February, and began selling into a market already full of its client's collateral. Falling prices produced fresh calls at the slower banks, and fresh calls produced fresh selling. Contemporaneous reporting put the wider liquidation at roughly US$20bn.¹⁵ By Monday, Archegos had collapsed.
The losses did not distribute evenly, and no single variable explains their order: exposure size, collateral held, initial margin terms, the particular securities, and the timing of each exit all bore on the outcome. Goldman Sachs, Deutsche Bank, and Wells Fargo described their losses as immaterial. Morgan Stanley lost about US$911m. UBS lost US$774m in the first quarter and a further US$87m in the second. Nomura lost about US$2.9bn. Credit Suisse lost approximately US$5.5bn, more than half of the roughly US$10bn the episode cost its lenders combined.¹⁶
Nomura's loss matters because the mechanism was not unique to Credit Suisse. Fragmented visibility affected every lender carrying material exposure, and no single institution could reconstruct the street-wide book from its own slice alone. The narrower Credit Suisse record concerns what the bank did with the warnings, calculations, and authority already inside its own account.
The board's investigation, run by an outside law firm, concluded that the losses were the result of a fundamental failure of management and controls, and was precise about which failure. The ratings, the limits, the scenarios, the escalation paths, and the committees at the top of them had all been present. The risk systems had identified the relevant risks, and identified them in time. The warnings and measurements were not acted upon.¹⁷
One remedial measure shows where the bank's own review placed part of the vulnerability. Any attempt by the business to appeal a risk decision to a more senior risk manager would thereafter be escalated and reported to the chair of the board's risk committee.¹⁷ The second conversation, the one held after risk says no, remained available, but it would now be visible at board level.
The Swiss regulator closed its proceedings in July 2023 with a finding that Credit Suisse had seriously and systematically violated financial market law in the Archegos relationship, and its notice fixed on one payment from the final fortnight. In mid-March 2021, Archegos asked Credit Suisse to return US$2.4bn of collateral. Staff understood the contract to require the payment, and the bank paid. The regulator recorded how: without any documented examination of whether it could decline, delay, or hold the money against the exposure it would face a fortnight later.¹⁸
Credit Suisse was entitled to bet on Archegos, and entitled to lose. It was also deceived: the criminal case the United States brought against Hwang turned on what his firm told its lenders about what stood behind their collateral, and a jury convicted him of it.¹⁹ The narrower finding in the institutional record survives all of it: every number the bank declined to act on was produced by its own models, circulated on its own lists, and archived in its own systems. None depended on the client's honesty to carry its warning.
In February, Credit Suisse examined a US$1bn margin demand and judged it too dangerous to the relationship. In March, it returned US$2.4bn without a documented examination of the alternatives.
Status at publication
28 January 2026 July. Hwang was convicted on 10 July 2024 and sentenced on 20 November 2024 to 18 years' imprisonment. He remained free on bail pending appeal at publication.
General information only. Not personal advice. Past performance is not indicative of future performance. This material is intended for wholesale and professional investors.
Related Practice Note
Continue with What a Position Ceiling Rests On, and Who Can Move It, a practice note on deriving a position boundary from tolerable loss, recording the assumptions behind it, and governing later revisions and breaches.
Notes
1. Prudential Regulation Authority, Final Notice: Credit Suisse International and Credit Suisse Securities (Europe) Limited, 24 July 2023 (the "PRA Final Notice"), Annex A: the potential-exposure limit of US$20m and Severe Equity Crash guideline threshold of US$250m for Archegos; potential exposure of US$395m and scenario exposure of US$921m at the end of August 2020; the parameters of both scenarios (Severe Equity Crash: a one-month fall of 30% in developed markets and 45% in emerging markets and Japan; Bad Week Equity Crash: a one-week fall of 20% and 30% respectively); the definition of potential exposure as a non-stressed calculation assessing, to a 95% confidence level, the bank's maximum exposure in the event of a counterparty default; and the framework's classification of potential exposure and the multi-factor scenario as limits, and the single-factor crash scenarios as guidelines. The US$250m threshold was not amended when the scenario was substituted.
2. Credit Suisse Group Special Committee of the Board of Directors, Report on Archegos Capital Management, 29 July 2021 (the "Paul, Weiss report"): use of Bad Week from September 2020; exposure at nearly twice the unchanged US$250m threshold on 1 September 2020; and repeated readings at 150% to 240% of the threshold through February 2021. PRA Final Notice, Annex A: the formal change recorded from October 2020, and the internal rationale (the liquidity of the holdings and the expected close-out period under the bank's margin and termination rights). The body preserves the difference in dating; the 1 September reading follows the Paul, Weiss dating.
3. PRA Final Notice: total Credit Suisse losses following the Archegos default of approximately US$5.5bn. The two reconstructions on which this account draws throughout are the Paul, Weiss report and the PRA Final Notice, supplemented where identified below.
4. Securities and Exchange Commission v Sung Kook (Bill) Hwang et al, complaint, US District Court for the Southern District of New York, 27 April 2022, paras 32, 52, 56–58, and 67–70: trading arrangements with about a dozen counterparties; more than US$36bn of invested capital against more than US$160bn of gross exposure as at 22 March 2021; combined cash-equity and swap exposure exceeding 70% of GSX Techedu, 60% of Discovery Class A, and 50% of ViacomCBS, on Archegos's own estimates; and allegations that Archegos's trading placed upward pressure on and artificially inflated the prices of concentrated holdings, including ViacomCBS. The complaint establishes the SEC's allegations and figures attributed to Archegos; it is not treated here as an independent finding of every alleged act. Hwang's subsequent conviction in the related criminal case is recorded at note 19.
5. United States Department of Justice, District of New Jersey: Tiger Asia Management's guilty plea to wire fraud, December 2012, with the related SEC settlement. Securities and Futures Commission of Hong Kong and Market Misconduct Tribunal proceedings, 2013 to 2014, including the four-year ban imposed on Hwang and Tiger Asia from trading in Hong Kong markets.
6. Paul, Weiss report: the bank's internal credit assessment of the account, and the findings on Archegos's margin terms relative to comparable Prime Services clients.
7. PRA Final Notice, Annex A: the account's inclusion on the standing list of limit excesses; near-weekly reminder emails to the credit team from April 2020 to February 2021; the responsible analyst's regular calls with Credit Control; and the recording, at the start of March 2021, that the account had exceeded its potential-exposure limit for more than 170 business days.
8. PRA Final Notice, Annex A: the October 2020 bespoke scenario appetite of US$900m, US$100m below the allowance for AAA-rated central counterparties and key sovereigns, and described in the entity's framework as comparable to the appetite for the central banks of Canada, France, Germany, Sweden, Switzerland, the UK, and the US. Paul, Weiss report: the "significant relationship" description, the warning that a sudden margin increase could cause "irreversible damage to the client relationship", and approval without comment.
9. PRA Final Notice, Annex A: the January 2021 annual review recording gross leverage of about six times against a rating-model input of four to five times; the downgrade from BB- to B+; the proposal to raise the potential-exposure limit from US$20m, already double the US$10m guidance maximum for a B+ rated fund, to US$50m; the characterisation in writing as an "unusual set of facts"; and approval following a brief call in early February 2021.
10. PRA Final Notice, Annex A: the February 2021 calculation that around US$1bn of additional margin was required to bring the portfolio within its scenario limit; the written query on the purpose of unused termination and margin rights; the 9 February response that "asking for $1bn is pretty much asking them to move their business", with the observation that the revenue profile was significant; the credit function's separate record that Archegos used six other prime brokers; and the 18 February exchange, in which the credit function told senior management in the first-line risk function that it knew the identities of all issuers in the portfolio and that a liquidation could be forced if all prime brokers increased margin requirements simultaneously. A reply was sent the same day to the last message in the exchange; the notice records that this question does not appear to have been answered.
11. PRA Final Notice, Annex A, paras 15.5–15.7, and Annex B, para 3.21: the 8 March 2021 decision to seek migration to dynamic margining with liquidity and concentration add-ons "within the next couple of weeks", or alternatively to request US$250m of additional margin, against proposed day-one dynamic-margining step-ups of approximately US$1.27bn to US$1.49bn; and the record that neither the migration nor the additional margin had been obtained before the default. Paul, Weiss report: the dynamic-margining proposal sent to Archegos on 24 February 2021 and ignored despite repeated follow-ups, and the three follow-up calls scheduled in the five business days before the default, all cancelled by Archegos at the last minute. The observation that migration required the client's agreement reflects the bilateral renegotiation of margin terms described in both reconstructions.
12. Paul, Weiss report: ViacomCBS as Archegos's single largest position, its 6.7% fall on 22 March 2021, the day the share offering was announced, and its continued decline in the days that followed. The magnitude of the preceding rise is not quantified in the body; doing so would require a named price source, dated endpoints, and a stated adjustment convention, which neither reconstruction supplies.
13. Paul, Weiss report: on 23 March 2021 Archegos held over US$600m of excess variation margin at Credit Suisse; by 24 March that excess had been eliminated by market movements and Archegos owed, was called for, and paid more than US$175m of additional variation margin; the same day Tencent Music fell approximately 20%, and the bank determined it would issue a variation margin call of approximately US$2.7bn the following day. PRA Final Notice, Annex A: senior management's surprise that positions capable of producing a call of that size were on the books.
14. Paul, Weiss report: Archegos's statements that the US$6bn to US$7bn it had held unencumbered had been consumed by margin calls earlier in the week; the disclosure on the evening of 25 March 2021 of approximately US$120bn of gross exposure (US$70bn long, US$50bn short) against US$9bn to US$10bn of remaining equity; and the proposed standstill and its rejection. The SEC complaint, para 52, places gross exposure above US$160bn at its stated peak on 22 March 2021. Archegos disclosed approximately US$120bn to its lenders on 25 March, after three days of market declines and margin calls. Neither source reconciles the change line by line.
15. Financial Times and related contemporaneous reporting, 29 March 2021: Morgan Stanley block sales on the evening of 25 March, and the broader Archegos-linked liquidation estimated at approximately US$20bn; contemporary estimates varied. Bloomberg News, "Goldman Sold $10.5 Billion of Stocks in Block-Trade Spree", 28 March 2021: approximately US$6.6bn sold before the US market opened on 26 March and further sales that day.
16. PRA Final Notice: Credit Suisse losses of approximately US$5.5bn, the largest share of total counterparty losses of approximately US$10bn. Morgan Stanley, Form 10-Q for the quarter ended 31 March 2021; Nomura Holdings, consolidated financial results for the year ended 31 March 2021; UBS, first- and second-quarter 2021 reports: the respective disclosed losses. Goldman Sachs, Deutsche Bank, and Wells Fargo described their losses as immaterial. The body attributes the dispersion of losses to no single variable; the ranking is not attributed to exit speed alone.
17. Paul, Weiss report: the conclusions that the losses resulted from a fundamental failure of management and controls rather than from missing risk architecture, and that the risk systems identified the relevant risks, which were not acted upon; and the post-Archegos remediation measures, including the requirement that any effort by the business to appeal a Risk decision to a more senior Risk manager be escalated and reported to the chair of the Board Risk Committee.
18. Swiss Financial Market Supervisory Authority (FINMA), "FINMA concludes 'Archegos' proceedings against Credit Suisse", 24 July 2023: the finding that Credit Suisse seriously and systematically violated financial market law in the context of the Archegos relationship; the payout of US$2.4bn to Archegos approximately two weeks before the collapse, made without documented examination of alternatives; staff's understanding of the contractual position; and corrective measures ordered from UBS as legal successor.
19. US Department of Justice, Southern District of New York: Hwang was convicted on 10 July 2024 of racketeering conspiracy, securities fraud, market manipulation, and wire fraud, including on the misrepresentations made to counterparties; he was sentenced on 20 November 2024 to 18 years' imprisonment. United States v Hwang, US Court of Appeals for the Second Circuit, No. 25-1941, filed 11 August 2025. Current status is recorded in the dated status note above.
