Each Archegos lender already held enough data to act

The week

On Friday, March 26, 2021, traders on Wall Street noticed something strange. Huge blocks of shares in well-known companies were hitting the market all at once, sold at a discount, by some of the world’s largest banks. The names included ViacomCBS, Discovery and a number of China-based firms listed in the United States, among them Baidu, Tencent Music and Vipshop. By the weekend, it was clear that the selling was not a market view. It was a fire sale.

The sales followed margin calls that Archegos Capital Management could not meet. A margin call is a lender’s demand for more collateral when the value of a borrower’s investments falls. If the borrower cannot pay, the lender sells the assets to protect itself.

Archegos was the family office of investor Bill Hwang. A family office is a private firm that manages one family’s wealth. As a family office, Archegos was not registered with the SEC as an investment adviser, which meant it faced lighter disclosure requirements than a typical hedge fund.

That exemption came from an SEC rule adopted in 2011 under the Dodd-Frank Act, which excluded family offices from the definition of investment adviser. Without registration, Archegos filed no Form ADV, the public report that describes an adviser’s business, assets, owners and disciplinary history.

Hwang was not new to regulators. His earlier hedge fund, Tiger Asia Management, had pleaded guilty in December 2012 to a wire fraud charge tied to trading in Chinese bank shares, and Hwang and his firms settled related SEC insider trading charges. The firm was then converted into Archegos, which formally took its new name in early 2013 and managed only family money. Every prime broker that later dealt with Archegos could have known this history from public records.

Archegos had built very large, concentrated bets on a handful of stocks, largely through total return swaps with several prime brokers. A prime broker is the division of a large bank that lends to, trades for and holds assets on behalf of professional investors. A total return swap is a contract in which the bank buys the shares and passes all the gains and losses to the client, in return for a fee. The client gets the economic effect of owning the stock without owning it. Because the banks, not Archegos, held the shares, the positions did not appear in Archegos’s name in public ownership filings.

The trigger came early in the week. After the market closed on Monday, March 22, ViacomCBS announced a stock sale of about $3 billion, split between new common shares and mandatory convertible preferred stock, to fund its streaming plans. The shares had more than doubled since the start of the year. Selling new stock after such a run made sense for the company. For a holder with an outsized, borrowed position, it was dangerous.

ViacomCBS shares closed down about 9 percent on Tuesday. The offering was priced at $85 a share, and on Wednesday the stock fell more than 20 percent. Some of the other names Archegos held were sliding at the same time. Each fall eroded the collateral behind Archegos’s swaps and triggered fresh margin demands from every bank that held its positions.

Midweek, according to widely reported accounts, Archegos and several of its prime brokers discussed how its positions might be wound down in an orderly way. No standstill held. By Friday, Goldman Sachs and Morgan Stanley were selling large blocks of the affected shares, and others followed. The race to sell was on.

Banks that sold quickly limited their losses. Goldman Sachs, Deutsche Bank and Wells Fargo were reported to have come through without material losses. Those that moved more slowly took heavy ones: Credit Suisse later put its loss at about $5.5 billion, and Nomura reported a loss of nearly $3 billion. Morgan Stanley and UBS disclosed smaller losses in the hundreds of millions of dollars. Across the banks involved, losses exceeded $10 billion.

For Credit Suisse, the timing was especially bad. Earlier that month it had suspended a family of supply chain finance funds linked to Greensill Capital, which collapsed into insolvency. Two client failures in one month at the same bank would shape how supervisors and investors read everything that followed.

Credit Suisse and Nomura warned of significant losses on Monday, March 29. The same client, the same stocks and the same weekend produced very different outcomes, depending on what each bank knew and what it was prepared to do about it.

The common account of Archegos stresses opacity between banks: each prime broker saw only its slice of a much larger position. That account is true, and it drove later debate about swap reporting. It also lets individual institutions off too lightly.

The independent review Credit Suisse published in July 2021 describes a firm that had the data to see trouble and did not turn it into action. For risk, data and technology leaders, the episode divides into three layers: what a bank could see about its own exposure, what it could infer about exposure elsewhere, and what it did with the signals it had. Only the second was truly a market-structure problem. The other two were problems any organization that lends money or extends credit can have.

The sections below start with the vocabulary and mechanics, take each layer in turn, then map the layers onto the people who own them and onto institutions far smaller than a global prime broker. A short case contrasts two versions of the same bank, and the article closes with the questions that test whether the layers are working.

Exposure, collateral and concentration

Four terms come up again and again in this story.

Exposure is how much a bank stands to lose if a client fails. In a swap, the bank holds the underlying shares, so its exposure depends on how far those shares might fall and how much protection it has.

Margin, or collateral, is that protection: money or assets the client posts with the bank as a cushion against losses. If a client posts 20 percent margin on a position, the shares can fall 20 percent before the bank starts losing its own money, assuming it can sell them at that price.

Concentration is how much of a client’s risk is bunched in a few names, and how big the bank’s position is compared with the market for that stock. A large position in a thinly traded stock is dangerous because selling it will push the price down, so the bank cannot get out at the price on the screen.

Counterparty credit risk is the overall discipline of measuring and managing the chance that a trading partner cannot meet its obligations. It is to a trading desk what credit underwriting is to a loan officer.

A pawnbroker who lends against jewelry faces the same problem. If one customer pledges an unusually large collection of a single rare type of ring, the pawnbroker needs to know what the rings would really fetch if all of them had to be sold at once, and whether the customer is pledging the same kind of rings at every other pawnshop in town. Archegos was that customer, and the banks were the pawnshops.

How a total return swap moves risk

The mechanics explain why the banks ended up holding the problem. When a client asks for a total return swap on a stock, the bank’s trading desk typically buys the shares to hedge itself. The client pays the bank a financing charge and posts initial margin. Each day the position is marked to market. If the shares rise, the bank owes the client the gain; if they fall, the client owes the bank the loss, usually settled through variation margin.

The bank is hedged against the share price only while the client keeps paying. Once the client stops, the bank is left owning the shares outright, with whatever margin it collected as its only cushion. Its market hedge turns into its loss.

Swaps also changed how much margin was required. A client buying shares with borrowed money through a U.S. broker-dealer is subject to the Federal Reserve’s Regulation T, which sets a minimum initial margin for stock purchases. Swaps sit outside that regime, and margin on them is set by negotiation between the bank and the client. That flexibility is commercially useful, and it puts the whole burden of judgment on each bank’s own risk methodology.

Why swaps hid the size of the bet

Ownership disclosure in U.S. markets is built around who holds the shares or controls their vote. Large investment managers report their stock holdings each quarter on Form 13F. Anyone who acquires more than 5 percent of a company’s voting shares with an intent to influence it must file a Schedule 13D. Cash-settled swaps, where the client never takes delivery of the shares, generally did not count toward those thresholds.

So the shares Archegos had economic exposure to were spread across the banks’ own books. Each bank’s holding might show up in its own filings, mixed with its other clients and its own trading. None of it carried Archegos’s name. As a family office, Archegos also did not file the adviser reports that would have described it to the SEC.

Potential exposure, the number that should have mattered

Prime brokers measure more than today’s mark-to-market. They also estimate potential exposure: how much the client could owe the bank after a severe but plausible price move, net of the margin held. Credit Suisse measured Archegos this way. Its review records that the potential exposure figure repeatedly breached the limits set for the client.

The useful feature of potential exposure is that it combines all four terms. It asks how far each stock could fall, how concentrated the book is, how much margin cushions the fall and how long a sale would take. A rising potential exposure on a client who is also growing its positions is the clearest single warning a prime broker can get.

Layer one: the bank’s own exposure

Each prime broker knew the size of its swaps with Archegos, the underlying names, the margin it held and the concentration in each stock. It could calculate how much a given price move would cost and whether margin would cover it. That work is basic counterparty credit risk, and it needs no information from competitors.

The Credit Suisse review found that internal limits had been breached repeatedly and that margin terms stayed static even as positions grew more concentrated and volatile. A limit is a ceiling the bank sets on how much risk it will take with a client. A breach is a warning light. A repeated breach that is tolerated is a warning light someone has taped over.

Some other banks reportedly used dynamic margining, in which collateral requirements rise automatically as a position grows more concentrated or less liquid. Static margin stays where it was set, however much risk the client piles up. That design choice shaped who lost what.

According to the review, proposals to move Archegos onto dynamic margining were discussed for months before the collapse, and the client resisted. The bank did not force the change. Static margin was a commercial concession, and every week it stayed in place, the gap between the margin held and the risk carried grew wider.

Architecture mattered as much as culture. Exposure to one client often lived across several systems: swaps in one booking platform, cash equities in another, margin calculations in a third, credit limits in a fourth. A booking platform is the system where trades are recorded. Large banks often have many, built at different times for different businesses, and they rarely speak to each other natively.

A single current view of a client’s total exposure, stressed for concentration and for how many days it would take to sell each position, meant joining data never designed to join. Where that view was produced weekly or by hand, it lagged the market. Where it ran daily and automatically, risk managers could argue from current numbers.

A useful single figure is days to liquidate: the bank’s position in each name divided by a cautious share of normal daily trading volume. If a bank holds shares equal to five full days of a stock’s normal trading, and it assumes it can sell only a modest fraction of each day’s volume without crashing the price, it may take weeks to exit. That single number turns an abstract concentration worry into a concrete figure a non-specialist can understand: how long it would take the bank to get out.

What the data model has to join

Producing one view of a client is mostly a reference data problem. The bank first needs a reliable client hierarchy: every legal entity, fund and account that belongs to the same economic client, linked to one parent. Professional clients often trade through several entities. If the hierarchy is wrong, exposure is split before any calculation starts.

Next comes instrument mapping. A swap on a stock and a cash holding of the same stock must resolve to the same underlying security, so concentration adds up across products. Different booking systems often identify securities differently, and the mapping tables are maintained by hand more often than anyone admits.

Then the market data: prices, volatility and average daily volume for every name, refreshed at least daily. Days to liquidate is meaningless if the volume figure is a month old, and volume in a stock like ViacomCBS changed sharply in the weeks before the collapse.

Last, the terms: the margin schedule, any notice period before margin can be changed, and any cross-default or netting provisions. Contract terms often live in document management systems, not in the risk engine, so the calculation may assume rights the bank does not actually have.

Margin terms are a design decision

Margin methodology has several moving parts, and each one is a choice with a commercial price. A bank can set a flat percentage per position. It can scale margin by volatility. It can add a concentration charge when one name dominates the book, and a liquidity charge when the position is large relative to daily volume. It can reserve the right to change margin at short notice, or agree to keep terms fixed for a set period.

Clients compare these terms across prime brokers. A bank that demands more margin, or insists on the right to raise it quickly, may lose the business to a competitor that does not. That is the commercial pressure the risk function has to withstand, and it is why margin methodology belongs in documented policy, not in a negotiation between a salesperson and a client.

The evidence a supervisor would ask for

An examiner reviewing this layer will usually want to see a short list of things. The limit framework for counterparties, including how potential exposure is calculated. The breach log, with dates opened and closed and who approved any temporary increase. The margin policy, and the list of clients on terms that depart from it. The reconciliation between the risk view of a client and the positions in each booking system.

Each item is cheap to produce if the systems are built to produce it, and expensive to assemble after the fact. A bank that needs a week and a team of analysts to answer “what is our total exposure to this client?” has already answered the more important question about its data.

Layer two: inferring the rest of the market

The harder problem lay beyond each bank. Swap positions did not show up in the public filings that normally reveal a concentrated holder. Still, there were signals. A client holding a very large share of a company’s float, the shares available for public trading, through one bank probably holds more elsewhere. Unusually large positions in thinly traded names, the stock’s own trading patterns, and the client’s answers about its other relationships all carried information.

Using that information means treating inferred data as an input to limits and margin, alongside what the bank can prove. A practical method is to assume the client holds a multiple of the bank’s own position in each name, and to stress the liquidation scenario on that basis. Back in the pawnshop, it is the pawnbroker saying: I cannot see what the other shops hold, but I will price this loan as if they hold just as much.

It also means asking clients direct questions and writing contract terms that allow margin to change when the answers are unsatisfactory or absent. Some banks did this better than others, and it showed in their outcomes.

Cross-firm opacity was real, and no single bank could fix it. But a bank can decide how to price what it cannot see. Treating the unknown as zero was a choice, and an expensive one.

Signals hiding in plain sight

Several kinds of evidence were available without any data from competitors. The bank’s own position as a share of each company’s float. The movement of each stock compared with its sector, since a stock rising far faster than its peers on no news may be driven by a single large buyer. The client’s own growth: a family office whose positions grow much faster than its capital is borrowing heavily somewhere.

Market structure carried signals too. When several banks hedge swaps for one client in the same names, they all buy the same shares. A bank that sees its own hedge holdings approaching a filing threshold in a company it has no view on is seeing, indirectly, the size of the client’s bet.

The trigger itself was public. Early that week, ViacomCBS announced a large stock offering, and its share price fell sharply over the following days. A bank that knew its client held a large swap position in ViacomCBS could see the problem in market data it already licensed. The client’s collateral was eroding in the very names where it was most concentrated.

Joining internal exposure with external market events is a modest data engineering task. In practice, it means a system that knows which clients hold big positions in which stocks, watches news and price moves in those stocks, and alerts the right person when something significant happens. It is also the kind of alert that turns a weekend fire sale into a Tuesday margin call.

Why banks did not compare notes

Prime brokers had good reasons not to share client positions with each other. Client agreements impose confidentiality. Sharing position data among competitors raises antitrust concerns. And a bank that told its rivals a client was overextended would hand them a reason to sell first.

That last point shaped the endgame. Once several banks knew they held the same positions for the same client, each faced a choice between an orderly joint unwind and being first out the door. An orderly unwind protects everyone only if everyone trusts the others to wait. Selling first protects the seller and pushes the price down for everyone else. The banks that sold first on Friday lost least.

A bank cannot rely on a collective arrangement to rescue it in a crisis. It has to be ready to act alone, which means its margin, data and authority must already be in place before the meeting with other lenders begins.

The strongest counter-argument

The best case against this article’s framing is that inference is easy in hindsight. Many clients run large, concentrated positions without failing. A bank that assumed every concentrated client was secretly levered across the street would overcharge its good clients and lose them.

The answer is proportion. Inferred exposure does not have to drive every client’s margin. It should drive margin for the few clients whose positions are large relative to float, concentrated in a handful of names, growing quickly and resistant to disclosure. Archegos met all four tests. A rule that applied only to clients like that would have touched very little revenue and covered most of the risk.

Layer three: from signal to action

The final layer is governance. Risk data is worth only the decisions it triggers. At the banks that lost most, warnings were raised and then deferred, often because the client was profitable and the relationship valued. Several banks that exited quickly appear to have given risk teams clearer authority to act on their own data, including on a Friday afternoon.

The Basel Committee’s principles for risk data aggregation, published as BCBS 239 in 2013, were written for exactly this kind of failure. The Basel Committee is the international forum where bank supervisors agree common standards. BCBS 239 asks banks to produce complete, timely, accurate risk data that reaches decision-makers, including under stress. The aim is simple: when things go wrong, the people in charge should be able to see the whole picture quickly, and trust it.

Archegos showed that compliance on paper differs from a risk function able to force a decision about a valued client. Useful tests include how long a limit breach can remain open before it escalates automatically, who can override it, and whether overrides for high-revenue clients are reported to the board risk committee.

That last test matters most. The pressure to keep a lucrative client happy is constant and human. A rule that says “breaches for our biggest clients go to the board” changes the conversation, because no one wants to explain to directors why a warning about the firm’s most profitable relationship sat open for months.

How overrides become the norm

Limit breaches rarely get ignored outright. They get managed. A temporary increase is approved for a few weeks while the client is asked to add margin. The client agrees in principle. The deadline passes, and another temporary increase follows. Each step is documented and each looks reasonable on its own.

The Credit Suisse review describes this pattern, including temporary increases in the risk the bank was willing to take with Archegos. The cumulative effect was a limit that existed on paper and moved whenever the client’s positions did.

Two controls break the pattern. The first is a hard cap on how many times, or for how long, a temporary increase can be renewed before it must go to a more senior committee. The second is reporting that shows the age of each open breach as well as its size. A breach open for ninety days tells a different story from one opened yesterday.

Metrics that show whether warnings lead to decisions

Most risk dashboards report exposure. Few report how the organization responded to it. A board that wants to know whether layer three works needs a different set of numbers: open breaches by age, renewals of temporary limits, margin calls issued compared with margin calls proposed, and the time between a risk alert and a documented decision.

These measures are easy to compute from data the bank already holds, once breach and decision records are captured in a system rather than in email. They also change behavior, because people know the measures will be read.

Incentives and the cost of acting early

Acting on a warning has a visible cost and an invisible benefit. Raising a valued client’s margin may cost revenue, and the client may move to a competitor. The loss avoided is hypothetical until the day it is not.

Compensation can tilt that balance. Pay that rewards revenue without adjusting for the risk taken to earn it pushes decisions toward deferral. Deferred pay subject to clawback, where bonuses can be reduced or recovered after losses come to light, gives revenue-generating staff a stake in the downside. Credit Suisse said in July 2021 that it had clawed back pay from staff connected to the losses. A clawback after the fact matters less than the expectation of one beforehand.

How the work happens day to day

Each layer belongs to different people inside a bank.

The front office is the business side: the salespeople and relationship managers who win clients and earn revenue from them. They know the client best, and they have the strongest incentive to keep the client happy.

The risk function sets limits, calculates exposure and decides when collateral should change. Its people are meant to be independent of the revenue the client brings in. Their effectiveness depends entirely on two things: whether they can see the numbers in time, and whether they can make a decision stick.

The technology and data teams build and run the systems that record trades, calculate margin and produce risk reports. They decide, often without much fanfare, whether a client’s exposure appears in one place or four, and whether the report runs every morning or once a week.

Senior management and the board are above all three. They set the bank’s appetite for risk and, in principle, settle disagreements between the business and the risk team.

Seen this way, the three layers map onto three groups. Layer one is mostly a data and technology question: whether the systems can assemble a current picture. Layer two is a judgment for the risk team about what to assume regarding what it cannot see. Layer three is a governance question for senior leaders: which side prevails, and how quickly, when the risk team and the business disagree about a valued client. A weakness in any one group can undo the strength of the other two.

Three lines of defense, tested under pressure

Banks describe their control structure as three lines of defense. The business owns its risks. An independent risk and compliance function sets limits and challenges the business. Internal audit checks that both are working. Archegos tested the second line most, and it tested audit too, because audit is meant to notice when the second line keeps losing arguments.

A practical check for audit is to sample the largest limit breaches of the past year and trace each to its resolution. If most were closed by raising the limit rather than reducing the risk, the second line is advising, not controlling.

Who owns the data that holds the view together

Data ownership is often the gap between the first two lines. The client hierarchy may belong to onboarding, the instrument master to operations, volume data to a market data team and contract terms to legal. No one of them owns the combined view of a client. A named owner for that combined view, with authority to fix upstream errors, is one of the cheapest controls a bank can add.

Technology leaders should also plan for the moment of stress itself. A margin call on a Friday needs current data on a Friday. If the exposure report depends on an overnight batch that fails, or on a spreadsheet only one analyst knows how to refresh, the bank’s ability to act depends on that analyst being at their desk.

The same three layers at a regional bank

Archegos matters to leaders outside investment banking, because the three layers apply well beyond prime brokerage.

Wealth managers see concentrated client positions and margin loans against them. Regional banks see commercial borrowers with debts at other lenders. Credit unions see members with obligations elsewhere. In each case the questions are the same: can you aggregate what you already know, can you reason about what you do not, and does anyone have the authority to act?

The costs of getting it wrong are the same too, only at a different scale. Direct loss comes first, when a borrower or client fails. Regulatory attention follows when supervisors conclude that risk data or governance fell short. Reputational damage arrives when investors and customers wonder what else the institution failed to see. And staff pay a cost too: the risk officers who raised concerns and were overruled, and the relationship managers who are left explaining losses to clients and colleagues.

For boards in particular, Archegos is a reminder that the most important risk reports show whether warnings lead to decisions, whatever the data volume or system sophistication behind them. A board that receives tidy dashboards but never hears about overridden limits is not seeing the risk that matters.

Broker-dealers and wealth managers

A broker-dealer that lends to retail customers against securities works under rules that set minimum margin, including Regulation T and FINRA’s margin rule. Those minimums are floors, and firms can and do set higher “house” requirements for concentrated or volatile positions. Layer one for such a firm is whether house requirements actually rise with concentration, and whether the firm can see a customer’s total exposure across accounts held in different names.

Wealth managers offering securities-based lines of credit face the Archegos problem in miniature. A founder borrowing against a large holding of one company’s stock is concentrated by definition. The layer two question is whether the client has pledged the same shares or borrowed against them elsewhere, which a wealth manager can ask at onboarding and at each renewal.

Regional and community banks

For a regional bank, the parallel to a prime broker’s client is a commercial borrower with obligations at several lenders. National banks face a legal lending limit on how much they can lend to one borrower, and that limit depends on correctly aggregating related borrowers, which is the client hierarchy problem again. Banks that join syndicated loans see only their share of the credit, and depend on the lead bank and the borrower’s reporting for the rest.

The layer two habit translates directly. If a borrower’s reported debt at other lenders seems low for its size and business, the bank can price and structure the loan as if it were higher, and write covenants that require disclosure of new borrowing.

Credit unions and fintech lenders

Credit unions lending to member businesses face NCUA rules on member business lending and their own concentration limits, and often see members whose obligations to other lenders are only partly visible through credit reports. Their constraint is usually staff, not data. A small credit union may have one or two people covering commercial credit. For them, layer three is about making sure a single credit officer can escalate a concern to the board without first winning an argument with the lending team.

Fintech lenders and buy-now-pay-later providers face a version of layer two at enormous scale. Many short-term obligations do not appear in traditional credit reports, so a lender may see little of a borrower’s debts elsewhere. The Archegos principle applies: when you cannot see a borrower’s other debts, decide on purpose what to assume, and write that assumption down.

One borrower, two versions of the bank

Suppose a regional bank has a growing wealth management arm. One of its clients, a successful entrepreneur, has borrowed heavily against a large holding in a single small-cap stock, the shares of a company with relatively light trading.

In the weaker version of this bank, the loan is booked in one system and the investment account in another. A monthly report shows the loan is within policy. Nobody calculates how long it would take to sell the shares if needed. When the stock falls sharply after the company announces it will issue new shares, the alert reaches the relationship manager days later. He is reluctant to call his best client with bad news, so he waits. By the time the bank acts, the shares have fallen further, and selling the bank’s collateral pushes the price down again.

In the stronger version, a daily report joins the loan and the investment account into one view of the client. It shows that the collateral would take more than two weeks to sell at a cautious pace. The bank’s policy, written in advance into the loan agreement, raises the collateral requirement automatically as concentration grows. The client was asked at onboarding whether he borrows against the same shares elsewhere, and declined to answer, so the bank priced the loan as if he did. When the share issue is announced, an alert goes the same morning to both the relationship manager and the credit officer. The credit officer has clear authority to call for more collateral, and any decision to wait is logged and reported upward. The call is uncomfortable. It is also on a Tuesday, not a weekend.

The stronger version costs something. The client pays more for the loan and may be irritated by the questions. The bank invests in a data feed and a daily job. The credit officer spends time on a relationship the business would rather handle itself. Those costs are real and recurring, and they are small next to a single loss on a concentrated loan.

The weaker bank’s failings line up with the three layers:

• Fragmented data. Exposure to one client is split across systems and only assembled by hand, weekly or monthly.
• Static terms. Collateral requirements set at the start of a relationship never change as the risk grows.
• Treating the unknown as zero. Assuming a client has no exposure elsewhere because you cannot see it.
• Ignoring public signals. Failing to connect market news about a stock to the clients most exposed to it.
• Deferring to revenue. Letting limit breaches stay open because the client is profitable, with no escalation to senior management or the board.

Where the next margin call will be decided

1. For our largest clients or borrowers, can we produce a single view of everything we are exposed to, across all systems, today? How old is the data in that view?
2. Do we know how many days it would take us to sell the collateral behind our largest positions without moving the price against ourselves?
3. When we cannot see a client’s exposure at other firms, what do we assume? Is “nothing” our default?
4. How many limit breaches are open right now, how long has each been open, and who approved leaving it open?
5. If our risk team wanted to call for more collateral from a valued client late on a Friday, would they have the authority to do it without waiting for the business to agree?

Each question maps to one layer. The first two test what the institution can see of its own exposure. The third tests what it assumes about everything it cannot see. The last two test whether anyone acts.

Weak answers have a recognizable sound. They name a report that exists but runs monthly, a policy that allows dynamic margin but is applied to few clients, or an escalation path that has never been used. Strong answers name a system, a person and a date. They can be checked in an afternoon.

The cost of the strong answers is mostly engineering and governance time, plus some revenue from clients who prefer looser terms elsewhere. The cost of the weak ones arrives all at once, usually on a Friday.

With hindsight

Archegos turned out to be an early chapter in Credit Suisse’s decline. The bank was taken over by UBS in March 2023 after a crisis of confidence. Archegos was not the only problem the bank faced in those two years, but a loss of about $5.5 billion on a single client became one of the clearest examples of the risk management weaknesses that came to define its story.

The bank’s own response came fast. In April 2021 it announced the departures of its chief risk officer and the head of its investment bank. The independent review it published in July 2021 found failures of management and control, not fraud by the bank’s staff, and the bank said it would move hedge fund clients to dynamic margining and sharply reduce the size of its prime services business. Those steps came after the loss, not before.

Supervisors drew the same connection. In July 2023, the Federal Reserve and the UK Prudential Regulation Authority fined Credit Suisse entities over the risk management failures that Archegos exposed. The Fed’s penalty was about $268.5 million, and the PRA’s £87 million fine was the largest it had imposed. Switzerland’s supervisor, FINMA, which cannot levy fines, concluded that the bank had seriously breached its obligations. The Fed’s order also required improvements to counterparty credit risk management in the bank’s U.S. operations, so UBS inherited the remediation along with the bank. The penalties were a formal statement that the problem lay in what the bank did with what it could see, as well as in what the market could not.

The man at the center of the story faced his own reckoning. Bill Hwang was convicted of fraud and market manipulation in 2024 and sentenced to 18 years in prison. Archegos’s chief financial officer was convicted at the same trial. The prosecution’s case included evidence that Archegos misled banks about its positions, which makes the layer two point sharper: the banks that assumed the worst about what they could not see did better than those that relied on the client’s answers.

On the market-structure side, the picture is mixed. Regulatory reporting of security-based swaps to registered repositories began in November 2021, so regulators gained more visibility than they had in March of that year. In December 2021 the SEC proposed public position reporting for large security-based swap holdings, an attempt to bring the kind of hidden positions Archegos built into public view. That proposal was formally withdrawn in June 2025, along with more than a dozen others. The SEC’s 2023 amendments to beneficial ownership reporting shortened filing deadlines and offered guidance on cash-settled derivatives, but stopped short of treating them as ownership. Cross-firm visibility stayed partial. The layer of the problem that no single bank could fix remained, in large part, unfixed.

International standard setters returned to the other two layers. In November 2023, a Basel Committee progress report found that only two of 31 global systemically important banks were fully compliant with the BCBS 239 data principles. In December 2024 the Committee issued new guidelines for counterparty credit risk management, citing Archegos and replacing sound practices first written after the Long-Term Capital Management collapse of 1998. Their emphasis on due diligence, ongoing monitoring of non-bank clients and governance reads like a checklist drawn from the Credit Suisse review.

That makes the inside-the-bank findings the ones that aged best, and the most useful. Concentration data was again available and underused in the regional bank failures of 2023, where uninsured deposit concentration was visible in each bank’s own records well before the runs. Uninsured deposits are balances above the federal insurance limit, and depositors holding them have strong reasons to move quickly at the first sign of trouble. Those banks did not need anyone else’s data to see the risk. They needed to look at their own, and to act.

The most dangerous gap is rarely the information you cannot get. It is the information you already have and do not use, and the warning you hear and decide to wait on.