This module teaches prime brokerage from zero, but goes deep enough for a professional
A hedge fund manager wakes up with an idea: buy one airline, sell another short. To act on that idea she needs cash she does not have, shares she does not own, somewhere to hold the result, and a daily statement that tells her what it is all worth. A prime broker supplies all four. Strip away the jargon and prime brokerage is that simple, and that hard.
The business is easy to caricature and difficult to understand. Outsiders see a bank lending money to hedge funds. Insiders see a machine that must value a portfolio, compute an exposure, call for collateral, borrow a security, settle a trade, and report the whole thing to a regulator, every single day, without failing once. The gap between those two views is where the interesting work sits.
This module closes that gap. It assumes no prior knowledge of hedge funds, margin, or securities lending, and defines every term on first use, from basis points and repo to manufactured dividends. It then goes far enough that a practitioner will find the mechanics they actually argue about at work: portfolio margin versus Reg T, general collateral versus specials, the re-use chain, the liquidation waterfall, and why a margin engine is the one system a bank will never buy off the shelf.
Two threads run through everything: the business and the technology
The business thread asks who does what, why the money moves, and where the risk sits. The technology thread asks which system holds the record, how the data gets from one place to another, and what breaks when it does not. Prime brokerage is a business where those two threads are the same thread. A financing decision that cannot be computed by the overnight margin run is not a financing decision; it is an aspiration.
This matters for a specific reason. In most of a bank, technology supports the business. In prime brokerage, technology is the product. A fund choosing between two primes is choosing between two reporting portals, two margin methodologies, and two operations teams as much as between two balance sheets. The platform is the pitch.
How to read this module
Read it as a narrative, not a reference card
The five parts build on each other. Part 1 explains what the product is and where it sits. Part 2 walks the entire client lifecycle from first pitch to daily margin call, and is the longest and most important part. Part 3 maps the people, including the fund's own administrator. Part 4 maps the systems. Part 5 maps what goes wrong and what is changing. Read them in order the first time; use the cheat sheet and glossary afterwards.
A single fictional client, Meridian Capital, runs through the whole module. Meridian is a long/short equity hedge fund and every figure attached to it is illustrative only, invented to make the mechanics concrete. Real market figures are flagged as such when they appear.
Each coloured box plays a fixed role
The Key insight box crystallises an idea worth carrying out of the section. Going deeper adds mechanics for readers who want them and can be skipped on a first pass. The IT & data lens box names the systems and data that carry the process just described. Watch out flags a common misunderstanding or a live risk. Worked example puts numbers on the mechanics, always with Meridian. In a nutshell recaps a part in a few sentences. Check your understanding poses questions with answers, and is the fastest way to find out whether a section actually landed.
What you will master
You will understand the prime brokerage relationship end to end
From the first coverage conversation through onboarding, financing set-up, daily trading, stock borrow, settlement, margining and reporting, you will be able to describe what happens at each stage, who does it, and what would happen if it were done badly.
You will be able to map every actor
The fund, the prime, the executing brokers, the securities lending market, the custodians and central securities depositories, and the regulators. You will know what each one wants, what information each one sees, and why the prime sits in the middle of nearly every flow.
You will understand where the financing terms actually come from
Margin is the price of risk in this business, and it is not set by a rulebook alone. You will be able to derive a margin requirement from first principles, explain Reg T, portfolio margin and house add-ons, compute a requirement three ways on the same portfolio, and say which number binds and why.
You will be able to trace data front to back
You will follow a single position from the execution message that creates it, through the books and records, into the margin engine, out to collateral management, and finally into a client statement and a regulatory report, and you will see why an error at the top reaches every system below it.
You will be able to say who owns the asset, and what follows from that
The organising question of the whole module. You will be able to trace the consequences of ownership through capital treatment, public disclosure, voting rights, dividends and their tax, and default mechanics, and explain why the same economic exposure structured as a swap behaves completely differently from one structured as a financed purchase.
You will recognise the risk and transformation landscape
Six risk types, five performance indicators, three real failures that define the business, and five technology shifts that are genuinely changing it, including generative AI and the governance it demands.
1Context and the Big Picture
Before any mechanics, three questions: what a hedge fund actually needs, what a prime broker actually sells, and why this particular bundle of services ended up under one roof.
Prime brokerage is the outsourced back office of a fund that trades with borrowed money
A hedge fund is a pool of capital that is allowed to do things a mutual fund cannot
A fund is money from investors, managed by a firm, invested according to a strategy. A traditional long-only fund buys shares with the cash it has and holds them. A hedge fund is a pool that operates under a lighter rulebook because its investors are professional or wealthy rather than retail, and that lighter rulebook permits two things in particular: it may borrow to invest more than its cash, and it may sell short.
Borrowing to invest more than you have is leverage, and leverage is the reason prime brokerage exists. Think of buying a house with a deposit: put down 20 and borrow 80, and you control an asset worth 100. If the house gains 10 percent you have made 10 on a deposit of 20, a return of 50 percent. If it falls 10 percent, half your deposit is gone. The bank lending the 80 does not share the gain, only the risk of not being repaid, so it insists on the deposit and on the right to seize the house. Prime brokerage is that arrangement, applied to a portfolio that changes every second rather than to a house.
A fund with 100 of investor capital that borrows another 100 controls 200 of assets. If those assets rise five percent, the fund earns ten on its own capital rather than five. If they fall five percent, it loses ten. Leverage multiplies the outcome in both directions, and someone has to lend the extra 100 and make sure they get it back.
Selling short is selling a security you do not own, in the expectation of buying it back cheaper. You cannot deliver what you do not have, so before you sell you must borrow the security from someone who owns it, deliver the borrowed shares to your buyer, and later buy shares in the market to return to the lender. Short selling is therefore not merely a trade; it is a trade that depends on a borrowing arrangement standing behind it for as long as the position is open.
A fund that borrows money and borrows securities needs a counterparty that does both
Put those two permissions together and a picture emerges. The fund needs cash lending for its long positions, securities lending for its short positions, an account to hold everything, a way to settle its trades, and a daily valuation of the whole book. It could assemble those from five different providers. Historically many did, and the reconciliation burden was crushing.
The prime broker bundles them. One institution lends the cash, sources the securities, holds the assets, clears and settles the trades, and produces one consolidated report. In return it takes a financing spread on the money it lends, a fee on the securities it lends, and it gets to hold the client's assets, which it can often re-use to fund itself. That bundle is the product.
The word "prime" is historical rather than descriptive. It came from the idea of a fund's primary broker, the one that consolidates the record even when the fund executes with many others. That consolidation function survives as the core of the business.
The bank is not really selling services; it is selling its balance sheet
Here is the point that separates people who understand this business from people who describe it. The individual services are commoditised. Custody is plumbing. Clearing is plumbing. Reporting is software. What is genuinely scarce, and what the fund is really buying, is the bank's willingness to put its own balance sheet behind the fund's positions, at a price, under conditions, with the risk managed by the bank.
Every conversation in prime brokerage eventually becomes a conversation about balance sheet. How much of it the client consumes, what return the bank earns on it, whether the regulatory capital it attracts is justified by the revenue. Learn to hear that question underneath the service-level discussion and the business becomes legible.
Prime brokerage sits inside markets, but behaves like a lending business
A bank is many businesses, and prime brokerage borrows the logic of two of them
An investment bank splits broadly into advisory work, which sells judgement and takes fees, and markets work, which sells access and liquidity and takes spread. Prime brokerage sits organisationally inside the markets division, usually within equities, and is often grouped with related financing activities under a label such as Equities Financing or Prime Financing and Services.
But it does not behave like a trading desk. A trading desk takes market risk and closes out. Prime brokerage takes counterparty credit risk and holds it open for years. The discipline it most resembles is corporate lending: assess a borrower, set terms, take collateral, monitor continuously, and act decisively when the borrower deteriorates. What makes it harder than corporate lending is that the collateral is a portfolio that changes every second and can become illiquid exactly when you need to sell it.
Why the platform is the pitch
A fund evaluating two primes runs a due diligence process that is largely technical. It asks how quickly trades are booked and how many break. It asks whether the margin methodology can be explained and reproduced, because a requirement the fund cannot predict is a requirement it cannot manage. It asks whether reporting arrives as a portal, a file, or an API that can feed its own systems directly. It asks what happened the last time the prime had an outage.
Those questions are not peripheral to the commercial decision; for sophisticated funds they frequently decide it. A bank with a cheaper spread and a weaker platform will lose to a bank with the reverse, because the operational cost of a poor prime falls on the fund's own team every single day.
Adjacent desks supply the pieces, and the boundaries matter
Prime brokerage is a consumer of other parts of the bank. The securities lending desk sources stock to lend. The delta one desk books the swaps that provide synthetic financing; the name refers to instruments whose value moves one-for-one with an underlying asset, such as swaps, futures and exchange-traded funds, as opposed to options, which do not. Treasury supplies the funding and prices it. Cash equities executes trades. Operations clears and settles. Understanding the boundaries between these teams is how you understand where a problem will surface and who will be asked to fix it.
The relationship with the securities lending desk deserves a note now. In some banks the stock loan book sits inside prime brokerage; in others it is a separate desk that prime brokerage trades with as an internal client. The organisational choice affects who owns the inventory, how transfer pricing works, and whose profit and loss absorbs a squeeze. It is a recurring source of internal friction and is worth asking about in any real institution.
Four financing lines are the core, and three more come attached
Financing is the spine, and the other three attach to it
The prime brokerage bundle decomposes into four financing lines. They are usually sold together, priced together, and reported together, which obscures how different they are. Separate them and the business becomes much clearer.
Before the list, one unit of measurement that recurs throughout. A basis point is one hundredth of one percent, so 100 basis points is 1 percent. Financing spreads and borrow fees are quoted this way because the differences that decide business are small and the balances they apply to are very large: 5 basis points on 10 billion dollars is 5 million dollars a year.
Cash financing, also called margin lending, is the simplest. The fund buys 100 of shares and pays 30 of its own money; the prime lends the other 70 and charges interest, holding the shares as collateral. The mechanics are old, the risk is well understood, and the pricing is a spread over a reference rate. This is the foundation.
Securities lending is the mirror image, and it is a market rather than a loan. To short a stock the fund must borrow it. The prime sources that stock, either from its own inventory, from other clients' holdings that permit re-use, or from the external lending market, and charges a borrow fee. For most stocks the fee is tiny; for hard-to-borrow stocks it can be enormous. This is where the operational sharp edges live.
Custody, clearing and client service are the plumbing that makes the rest usable. Holding the assets, settling the trades, resolving the breaks, producing the statement, answering the query, and introducing the fund to potential investors through capital introduction. It generates modest direct revenue and enormous stickiness.
Synthetic prime brokerage is the most complex, and it is the reference product for this module. Instead of the fund buying a share and borrowing money against it, the bank buys the share and enters a swap with the fund that passes the economic return of that share to the fund in exchange for a financing fee. The fund gets the exposure without ever owning the asset. Everything changes: who holds title, how capital is calculated, which regulations apply, how margin works, and what happens in a default.
Prime services is broader than prime brokerage, and the boundary matters
The four lines above are prime brokerage proper: the financing of a securities portfolio. In most banks they sit inside a wider division usually called Prime Services, which adds three further businesses sold to the same clients by the same coverage teams. Knowing where the boundary falls prevents a common confusion, because an organisation chart labelled Prime Services covers considerably more than this module's core subject.
Listed derivatives clearing is the largest addition. A fund trading futures and listed options does not face the exchange directly; it faces a clearing member, which in the United States is a regulated entity called a futures commission merchant. The clearing broker guarantees the client's performance to the central counterparty, collects margin set by that clearinghouse rather than by its own model, and segregates client funds under a distinct and stricter regime. The economics are thinner than financing and the regulatory treatment of client money is different, which is why it is usually a separate legal entity and a separate platform.
OTC clearing does the same job for cleared swaps. Post-crisis reform pushed standardised interest rate and credit derivatives into central clearing, and the bank acts as clearing member between the fund and the clearinghouse. It is a low-margin, high-capital business that banks retain largely because clients expect the full set.
Middle-office outsourcing sells the bank's operational capability directly. Rather than merely reporting to the fund, the prime runs parts of the fund's own operations: trade support across all its brokers, reconciliation, collateral processing, and data delivery to the fund's administrator. It is sticky, it deepens the relationship, and it turns an internal cost centre into a fee-earning service.
The unifying logic is that a fund wants one relationship, not five. Each of these can be bought separately and often is, but the commercial pitch of prime services is a single counterparty, a single set of documents and a single consolidated report across financing, clearing and operations.
Line them up and the differences become vivid
| Service line | What the fund gets | Prime's revenue | Balance-sheet intensity | Key technology |
|---|---|---|---|---|
| Cash financing (margin lending) | Cash to buy more than its equity allows | Financing spread over reference rate | High — direct funding of assets | Margin engine, books and records |
| Securities lending | Borrowed stock enabling short sales | Borrow fee, from a few basis points to hundreds | Moderate — collateral flows both ways | Inventory management, locate platform |
| Custody, clearing, client service | Safekeeping, settlement, consolidated reporting | Modest fees, high stickiness | Low | Sub-ledger, settlement, client portal |
| Synthetic prime brokerage (swaps) | Economic exposure without ownership | Financing spread embedded in the swap | High — bank holds the physical asset | Swap booking, collateral, capital engines |
| Listed derivatives clearing (FCM) | Access to futures and options markets | Clearing fees per lot, thin | Low, but strict client money rules | Clearing platform, CCP connectivity |
| OTC clearing | Access to central clearing for swaps | Clearing fees, low margin | High capital, low revenue | Clearing member systems, CCP margin |
| Middle-office outsourcing | The bank runs part of the fund's operations | Service fees, very sticky | Minimal | Trade support, reconciliation, data delivery |
The timeline row explains the others, and here that row is "who owns the asset"
In a securities issuance business the organising variable is time. In prime brokerage it is ownership. Ask of any structure: who legally holds the security, and who bears its economic outcome? In cash financing the fund owns the share and the bank has a security interest over it. In synthetic prime brokerage the bank owns the share and the fund has a contractual claim on its performance. In securities lending, title genuinely transfers to the borrower, which is why collateral rather than ownership protects the lender.
That single question determines the regulatory treatment, the capital cost, the default mechanics, and the systems required. Ownership is the variable that the whole business rotates around.
Why synthetic is the hardest case, and therefore the best teacher
Synthetic prime brokerage concentrates every difficulty in one product. The bank holds the asset, so it carries the market risk of a hedge that must track the swap exactly. The exposure is a derivative, so it attracts derivative capital rules and derivative margin regimes. The fund's position is invisible in public ownership disclosures, because the fund does not own anything, which is precisely the feature that made it central to the Archegos failure. And the whole structure depends on legal documentation whose default provisions have to work under pressure.
Master the synthetic case and the cash case becomes easy. That is why it anchors this module.
The numbers that frame the business
A few reference figures make the field concrete
Prime brokerage is a concentrated business. Globally, the industry generates revenue in the high teens of billions of dollars a year, with equities financing and securities lending together forming the bulk of it, and the top handful of banks capturing a disproportionate share. Client balances at a large prime run into the hundreds of billions of dollars. These are order-of-magnitude figures that move with the cycle rather than precise constants.
Concentration is the structural fact worth remembering. A small number of global banks dominate, because the business has punishing fixed costs. You need a margin engine, a settlement platform, a reporting portal, a global custody network and a financing desk before you can serve your first client. Below a certain scale the economics do not work, which is why mid-sized players have repeatedly exited or sold their books.
The client side is concentrated too. A typical large prime earns most of its revenue from a minority of its clients, and the largest multi-strategy funds are courted intensely because their balances are enormous, and feared quietly for the same reason.
| Dimension | Approximate shape of the market |
|---|---|
| Global prime brokerage revenue | High teens of billions of dollars a year, moving with rates and volatility |
| Market structure | Concentrated: a small group of global banks holds the large majority of balances |
| Client balances at a large prime | Hundreds of billions of dollars |
| Revenue concentration | Most revenue comes from a minority of clients |
| Barrier to entry | Margin engine, settlement, custody network and reporting portal required before the first client |
| Principal revenue lines | Financing spread, securities lending fees, and interest on cash balances |
These are order-of-magnitude observations rather than precise constants, and every one of them moves with the cycle described later in this part.
The revenue comes from four places, and the costs from three
Prime brokerage revenue is often described loosely as "financing income". It is worth decomposing, because the four sources behave differently across the cycle.
The financing spread is the margin earned on cash lent to fund long positions. Securities lending fees are earned on stock lent to cover shorts, and are the most volatile line because a single hard-to-borrow name can generate more than a large book of general collateral. Interest on cash balances is earned on the client cash the prime holds, including the proceeds of short sales, and this line is close to zero when policy rates are near zero and substantial when they are not. Commissions and clearing fees come from execution and from the clearing businesses described above.
Against those sit the bank's own funding cost, the charge for the regulatory capital the business consumes, and its operating cost, which is dominated by technology and operations headcount.
Where the money actually comes from: the funding chain
A question worth asking plainly: when the prime lends a client 70 to buy a share, whose 70 is it? Rarely the bank's own equity. The money is raised in three ways, and understanding this explains why financing is priced as it is.
Client cash funds a large part of it. Some clients hold credit balances and short sellers generate cash proceeds. That cash sits with the prime and can fund other clients' long positions. It is the cheapest funding the business has, which is precisely why the interest line above matters so much.
Re-use of client securities funds another part. Assets the client has pledged can, within limits, be lent or pledged onward to raise cash. This is the rehypothecation mechanism examined in Part 2, and it is a funding technique before it is a risk topic.
Repo funds the remainder. A repurchase agreement, or repo, is the sale of a security with a simultaneous agreement to buy it back later at a slightly higher price. Economically it is a secured loan: the seller gets cash now, the buyer holds the security as collateral, and the price difference is the interest. Banks fund large securities inventories this way continuously. A book of offsetting borrowing and lending run for the spread between them is called a matched book.
Treasury prices all of this internally through funds transfer pricing, the mechanism by which a bank charges its own businesses for the money they use and pays them for the money they raise. It is an internal price, but it is not cosmetic: it determines whether a desk is profitable, and a change in transfer pricing can make a client relationship uneconomic without anything changing outside the bank.
League tables and client surveys keep score, and they distort behaviour
Prime brokerage has no public deal volumes to rank, so the industry keeps score in two other ways. Market-share estimates rank banks by client balances and revenue, compiled by research firms from surveys and disclosures. Client satisfaction surveys, run annually by trade publications, rank primes on service, technology, financing and capital introduction, often segment by segment.
Both distort behaviour, and the second more than the first. Because funds consult these rankings when selecting a prime, and because allocators ask managers which primes they use, a poor survey result has commercial consequences. Banks therefore manage to the survey: they invest in the dimensions that are scored, they encourage satisfied clients to respond, and they treat a ranking decline as an escalation. The healthy version of this is genuine attention to service quality. The unhealthy version is effort spent on measured dimensions at the expense of unmeasured ones, of which the most important is risk discipline, since no survey has ever ranked a prime on the quality of its margin methodology.
Return on balance sheet, not revenue, is the number that governs behaviour
Since the post-crisis capital rules, a prime broker cannot simply chase revenue. Every client consumes leverage exposure and risk-weighted assets, both of which are scarce. Risk-weighted assets, universally shortened to RWA, are a bank's assets restated according to how risky each one is — a government bond might count for almost nothing, an unsecured loan to a volatile counterparty for its full value or more — and regulators require the bank to hold capital in proportion to that total. RWA is therefore the currency in which a bank pays for taking risk. The internal metric that matters is the return earned per unit of balance sheet consumed.
This metric quietly distorts behaviour in the same way league tables distort issuance businesses. A client that generates good revenue but consumes a great deal of balance sheet may be repriced or asked to leave, no matter how long the relationship has run. Coverage teams spend real energy steering clients toward products that are cheaper in capital terms. When you see a prime pushing a client from cash financing toward a synthetic structure, or vice versa, the driver is usually this ratio rather than client preference.
The bundle is a default, not a law of nature
Three alternatives to the single-prime model come and go with conditions
The multi-prime model splits the fund's business across two or more primes. After 2008, when funds discovered that assets held at a failing broker could be frozen for years, almost every fund of consequence added a second prime. The cost is real: positions are fragmented, margin is computed separately at each prime with no netting benefit across them, and the fund's operations team must reconcile several sets of records. The benefit is survival. Most institutional-quality funds now run at least two primes, and the largest run several.
The mini-prime or introducing-broker model serves funds too small for a global prime. A smaller firm holds the relationship with a global prime and sub-allocates access to a group of small funds. The small fund gets the plumbing it could not otherwise obtain; it accepts an extra intermediary and the credit risk that comes with it.
Self-clearing and custody-only arrangements strip the bundle back. A very large fund may execute through many brokers and hold assets at an independent custodian, buying financing separately as a commodity. This maximises asset safety and negotiating leverage, and requires an internal operations capability most funds do not have.
| Model | Who uses it | Main benefit | Main cost |
|---|---|---|---|
| Single prime | Small and emerging funds | Best pricing, full cross-margining, one reconciliation | Total dependence on one institution |
| Multi-prime | Most institutional funds since 2008 | Asset safety, competitive tension, continuity if one prime fails | Fragmented margin, no netting across primes, duplicated operations |
| Mini-prime / introducing broker | Funds too small for a global prime | Access to infrastructure otherwise unavailable | Extra intermediary and its credit risk, higher all-in cost |
| Self-clearing / custody-only | Very large, operationally sophisticated funds | Maximum asset safety and negotiating leverage | Requires an internal operations capability most funds lack |
The choice is a trade between price, safety, and operational burden
There is no correct answer, only a position on a curve. Concentrating with one prime buys better pricing, cross-margining benefits and a single reconciliation. Spreading across several buys safety and competitive tension at the cost of efficiency and complexity. Where a fund sits on that curve tells you a great deal about its size, its memory of 2008, and the sophistication of its chief operating officer.
The business breathes with rates, volatility, and regulatory capital
Prime brokerage is cyclical, but not in the way issuance businesses are
An equity issuance desk lives or dies with the window: when markets close, revenue goes to zero. Prime brokerage is steadier, because financing balances persist. But it has its own three-variable cycle, and each variable pulls differently.
Interest rates drive financing revenue directly. The prime earns a spread on cash it lends and, importantly, on the cash balances it holds. The rise in rates from 2022 onward substantially improved prime brokerage economics, after more than a decade in which near-zero rates had compressed exactly that income.
Volatility drives both revenue and risk, in tension. Volatility increases trading activity, borrow demand and margin requirements, all of which raise revenue. It also raises the probability that a client cannot meet a call. The business likes moderate volatility and fears the sharp, correlated kind.
Regulatory capital sets the ceiling. Leverage-ratio rules and the ongoing implementation of the Basel III endgame determine how much balance sheet the business may consume for a given return. Regulatory change can shrink or reshape the business without any change in client demand whatsoever.
The recent cycle tells the story cleanly
The years from 2021 onward compressed an unusual amount of instruction into a short period. In early 2021 the meme-stock episode showed how a crowded short position becomes an operational crisis, as borrow fees spiked and recalls forced covering. In March 2021 the collapse of Archegos Capital Management demonstrated what happens when several primes each extend heavy synthetic leverage to a single concentrated client without seeing the whole picture, producing losses of roughly ten billion dollars across the industry and the effective destruction of one major bank's prime business.
The response was immediate and industry-wide: tighter concentration limits, far more attention to synthetic exposures, more frequent margin recalculation, and much sharper questions about what a client holds elsewhere. From 2022 rising rates lifted revenue substantially. Through 2023 and 2024 the pressure shifted to capital efficiency as banks prepared for tighter requirements.
The cycle matters because it sets the rhythm of technology investment
Each phase of that cycle produced a technology programme. The squeeze produced investment in real-time inventory and recall management. Archegos produced investment in intraday margin, cross-product exposure aggregation, and concentration analytics. The rate rise produced investment in treasury and funding optimisation. The capital agenda is producing investment in collateral optimisation and capital attribution.
Read the technology roadmap of any prime broker and you are reading the last decade's incidents. This is not a criticism; it is how the business learns. But it explains why these platforms accumulate layers, and why the architecture question in Part 4 is rarely a clean-sheet question.
Why does a hedge fund need a prime broker at all, in one sentence?
Because it trades with borrowed money and borrowed securities, and it needs a counterparty willing to supply both, hold the resulting assets, and value the whole book daily.
What is the single organising question that distinguishes the financing lines?
Who owns the asset. In cash financing the fund owns it; in synthetic prime brokerage the bank owns it and the fund holds a contractual claim; in securities lending title transfers to the borrower and collateral protects the lender.
Why is synthetic prime brokerage treated as the reference product?
It concentrates every difficulty: the bank holds the asset and must hedge it, the exposure attracts derivative capital and margin rules, the fund's position is invisible in ownership disclosures, and default depends entirely on documentation.
Why did the multi-prime model become standard after 2008?
Because Lehman showed that assets held at a failing broker can be frozen for years, so funds accepted fragmented margin and duplicated reconciliation in exchange for survivability.
Which internal metric governs client decisions, and why is it distorting?
Return per unit of balance sheet consumed. It can push a bank to reprice or exit a long-standing, revenue-generating client purely because that client is capital-inefficient.
Name the three variables in the prime brokerage cycle and how each acts.
Interest rates, which lift financing revenue directly; volatility, which raises both revenue and the risk of client default; and regulatory capital, which caps how much balance sheet the business may consume.
What separates prime brokerage from the wider Prime Services division?
Prime brokerage is the financing of a securities portfolio. Prime Services adds listed derivatives clearing through a futures commission merchant, OTC clearing for cleared swaps, and middle-office outsourcing, each with different economics, client money rules and platforms.
When the prime lends a client 70 to buy a share, whose money is it?
Rarely the bank's own equity. Mostly other clients' credit balances and short sale proceeds, plus cash raised by re-using pledged securities, topped up with repo, all priced internally through funds transfer pricing.
Why does a rise in interest rates improve prime brokerage economics so markedly?
Because the business holds large client cash balances, including short sale proceeds, and the interest earned on them is close to zero when policy rates are near zero and substantial when they are not.
How do client surveys distort behaviour, and what do they fail to measure?
Funds and allocators consult them, so banks manage to the scored dimensions of service, technology and financing. No survey ranks a prime on the quality of its margin methodology, which is the dimension that determines whether it survives a client default.
1
Why does a hedge fund need a prime broker?
Not the core reason.
They can.
No such rule.
Right.
What single question distinguishes the financing lines?
Right. The fund in cash financing, the bank in synthetic, the borrower in stock lending.
No.
No.
No.
Why did multi-prime become standard after 2008?
No.
Not the reason.
Right. Survivability beat efficiency.
It is more expensive: fragmented margin.
In synthetic prime brokerage, who owns the asset?
No.
Right.
No.
That is cash financing.
2The End-to-End Client Lifecycle
An issuance business has a deal with a beginning and an end. Prime brokerage has a relationship that starts once and then runs every day until someone ends it. This part walks that lifecycle from the first coverage call to the daily margin run, and then to the ending nobody plans for.
Hold the whole arc in your head before the detail
The lifecycle runs through eight stages, and only the first three happen once
A prime brokerage relationship is set up over roughly three months and then operated indefinitely. The distinction between set-up and operation is the most useful structural idea in this part. Stages one to three are a project with a completion date. Stages four to eight are a cycle that repeats every trading day for the life of the relationship.
| Stage | Timing | Core activity | Key output | Who leads |
|---|---|---|---|---|
| 1. Origination and mandate | Ongoing, months | Coverage, pitch, commercial terms | Term sheet, indicative pricing | PB sales / coverage |
| 2. Onboarding and legal | 6–12 weeks | KYC, credit assessment, documentation | Signed PB agreement, approved limits | Onboarding, credit, legal |
| 3. Financing set-up | 1–2 weeks | Margin methodology, account structure | Live accounts, margin parameters | Risk, client service |
| 4. Execution and give-up | Daily | Trades executed anywhere, booked to the prime | Trade records in the sub-ledger | Client, executing brokers, ops |
| 5. Locate and stock borrow | Daily | Sourcing securities for short sales | Confirmed locates, borrow contracts | Securities lending desk |
| 6. Clearing and settlement | T+1 | Moving securities against cash | Settled positions, custody records | Operations |
| 7. Margin and collateral | Daily, overnight and intraday | Valuing the book, computing exposure, calling collateral | Margin requirement, collateral movements | Risk, collateral management |
| 8. Reporting, service, cap intro | Daily and monthly | Statements, queries, investor introductions | Client reporting, capital raised | Client service, cap intro |
The pressure concentrates in two places, and neither is where newcomers expect
Intensity spikes twice. The first spike is onboarding, where the credit decision is taken and the legal documentation is negotiated. Get this wrong and every subsequent day inherits the error. The second and larger spike is the daily margin and collateral run, which is the moment the bank finds out whether its risk assessment was correct and whether the client can pay.
The daily margin run is the pricing night of prime brokerage. In an issuance business there is one irreversible evening when the price is struck. Here that evening happens every night, forever. It must complete before the market opens, it must be right, and there is no version of the business in which it can be skipped.
One fictional client carries the whole part
Step 1 — Winning the mandate is about balance sheet appetite, not charm
The pitch is a claim about three things, and only one of them is price
A prime brokerage relationship begins in coverage: a salesperson who has known the fund's chief operating officer for years, a fund that is launching or is unhappy with an incumbent, a conversation about what the fund needs. When it becomes concrete, the bank pitches on three dimensions.
Financing terms are the visible dimension. What spread over the reference rate will the bank charge on cash lent, what margin will it require, what will it charge to borrow hard-to-find stock, and how much leverage will it permit. This is what gets negotiated hardest and what the fund's investors will ask about.
Service and technology are the dimension that actually retains clients. The quality of the reporting portal, the speed of the operations team, whether breaks are resolved in hours or weeks, whether the margin methodology is explicable, whether the fund can get an answer at seven in the morning. Funds leave primes over service far more often than over price.
Balance sheet commitment is the dimension that decides the outcome. Will this bank still be financing the fund when markets are difficult? A prime that withdraws leverage in a stress is worse than useless. Funds probe this question relentlessly, and the honest answer is usually visible only in a crisis.
The bank is choosing too, and its criteria are unsentimental
The pitch runs both ways. Before committing, the bank assesses whether the client is worth having, and the assessment is quantitative. It models the expected revenue, the balance sheet the client will consume, the regulatory capital that balance sheet will attract, and the resulting return. It examines the strategy for concentration and liquidity. It asks how many other primes the fund uses, and what share this bank would hold.
A wallet-share conversation happens early and is more revealing than it sounds. A bank that will be the fund's third prime with fifteen percent of the business earns less, sees less of the portfolio, and has weaker information for risk purposes. Banks price accordingly, and sometimes decline.
The four commercial terms you will meet constantly
The financing spread is the margin over a reference rate charged on cash lent to fund long positions, and the rate paid on credit balances. The short rebate is the interest the prime pays the fund on the cash proceeds of short sales that it holds, less the borrow cost, and it can be negative for hard-to-borrow names. The margin requirement is the percentage of a position's value the fund must fund from its own capital, which sets the maximum leverage. The minimum revenue commitment is an undertaking by the fund to generate a floor level of revenue, which the bank uses to justify committing balance sheet.
Step 2 — Before a bank lends, the client runs an internal gauntlet
Three gates stand between a term sheet and a live account
This is where the credit decision is actually taken, and it is the most consequential moment in the relationship. The bank is about to extend leverage to a counterparty whose portfolio can change completely within a day. It vets that counterparty through three distinct gates, each owned by a different function.
The know-your-customer and anti-money-laundering gate establishes who the client is. Fund structure, domicile, the management company, ultimate beneficial owners, the source of investor capital, sanctions screening, and the regulatory status of the manager. Offshore master-feeder structures are standard in this industry and entirely legitimate, but they mean the legal entity opening the account is often not the entity everyone talks about. Getting the entity tree wrong at this stage propagates into every downstream record.
The credit and counterparty risk gate establishes whether the client can be lent to. Analysts examine the strategy, the liquidity of the instruments traded, the concentration of positions, the fund's own investor base and its redemption terms, the track record of the principals, and the operational infrastructure of the manager. Redemption terms matter more than newcomers expect: a fund whose investors can withdraw at short notice may be forced to liquidate at the worst moment, which makes it a worse credit regardless of its performance.
The new business committee gate establishes whether the bank wants this business. Representatives from risk, legal, compliance, operations, finance and the business line consider reputational questions, capital consumption, operational readiness, and whether the bank can actually service the strategy. A fund trading instruments the bank's systems cannot process will be declined regardless of its credit quality.
The documentation is not paperwork; it is the mechanism that works in a default
Three agreements typically govern the relationship, and their default provisions are the entire point. The prime brokerage agreement covers cash financing, custody and clearing, and sets out margin terms, the bank's right to demand collateral, its rights over the client's assets, and the events that permit it to close out. The ISDA master agreement with its credit support annex governs the swap relationship for synthetic financing, defining collateral thresholds and, critically, close-out netting. A securities lending agreement, commonly on the GMSLA framework, governs stock borrowing.
Close-out netting is the single most important legal concept in this part. Picture a running tab at a bar where you have bought some rounds and been bought others. Netting says that when the evening ends you settle one net figure. Without netting, the other party could insist on collecting every round you owe while disclaiming every round owed to you. It means that on default, all outstanding transactions between the parties collapse into one net amount owed in one direction, rather than the defaulting party's administrator cherry-picking profitable contracts to enforce while disclaiming unprofitable ones. Without enforceable netting, the bank's exposure is the gross sum of everything, and the capital treatment changes drastically. Lawyers obtain formal netting opinions per jurisdiction, and the risk systems must know which agreements net with which.
Onboarding
Onboarding is a workflow and data problem before it is a legal one. A client lifecycle management platform orchestrates the case, collecting documents and routing approvals. Screening tools check sanctions and politically exposed persons against external data. A legal entity master creates the entity records and links them to LEIs, and this hierarchy becomes the spine that every downstream system inherits. A limits system stores the approved credit lines and concentration caps. Documentation terms are digitised into the margin and collateral engines as parameters.
The failure mode is duplication. Large funds have dozens of entities with similar names, and creating a second record for an entity that already exists fragments exposure across two views. The bank then believes it has two moderate exposures rather than one large one. Entity deduplication is a genuinely hard, genuinely mission-critical problem.
Step 3 — The margin methodology is where the bank prices its risk
Margin is the answer to one question, asked in three ways
Margin is the amount of its own capital the fund must commit against a position, and therefore the limit on how much it can borrow. The question behind it is always the same: if this client defaulted tomorrow and we had to liquidate this portfolio, how much could we lose before we got out? Margin is the buffer sized to that loss. Three methodologies answer the question differently.
First, derive the number from nothing
Before meeting three methodologies, it is worth deriving what the answer ought to be, because the methodologies are just different ways of estimating the same quantity.
Suppose the bank finances a position worth 100. The client defaults today. The bank cannot sell instantly: it must work out of the position over some number of days without crashing the price. Call that liquidation horizon d days. Over those days the price will move, and the size of a typical daily move is the stock's volatility, call it s per day. Losses over several days do not simply add up, because moves partly offset; they grow with the square root of time. So the expected size of the move over the horizon is roughly s × √d.
But margin sized to a typical move would be wrong half the time. The bank wants to survive a bad move, not an average one, so it multiplies by a confidence factor k, commonly between 2 and 3 for the sort of confidence levels regulators expect.
That gives the whole logic in one line: margin ≈ k × s × √d × position value.
Put numbers in it. A liquid large-cap stock with a daily volatility of 2 percent, liquidatable in one day, at a confidence factor of 2.5, needs roughly 2.5 × 2 percent × 1 = 5 percent. A concentrated small-cap position with 4 percent daily volatility that would take nine days to exit needs 2.5 × 4 percent × 3 = 30 percent — six times as much, from the same formula.
Everything else in this section is a refinement of that expression. Portfolio margin estimates it across a whole book rather than position by position. Concentration and liquidity add-ons are corrections to the horizon d. Gap-risk add-ons are corrections to the assumption that s describes the move at all, since a takeover rumour is not a normal day. Reg T abandons the expression entirely and uses a fixed number, which is why it is a floor rather than a measure.
Hold this formula and you can reconstruct any margin discussion in the business from first principles.
Reg T is rules-based, simple, and mostly wrong
The United States Federal Reserve's Regulation T sets an initial margin requirement of 50 percent for equity purchases in a standard margin account. Buy 100 of stock, post 50. The rule is easy to compute, easy to audit, and takes no account of risk. A diversified portfolio of large-cap stocks and a single concentrated position in an illiquid small-cap attract the same requirement, which is plainly wrong as a risk measure. It functions as a regulatory floor rather than a risk model.
Portfolio margin is risk-based, and it is what sophisticated clients want
Portfolio margining computes the requirement by stressing the entire portfolio. The engine applies a set of market scenarios, typically a grid of price moves up and down combined with volatility shifts, revalues every position under each scenario, and takes the worst aggregate loss as the requirement. Hedges are recognised: a long position offset by a short in a correlated name produces a smaller requirement than either alone, because they do not lose money simultaneously.
The grid is worth seeing rather than imagining. A simplified version applied to a single equity position looks like this, with each cell holding the profit or loss of the portfolio under that combination.
| Price move | Volatility down | Volatility unchanged | Volatility up |
|---|---|---|---|
| −15% | −14.2 | −15.0 | −16.1 |
| −8% | −7.4 | −8.0 | −8.9 |
| 0% | +0.3 | 0.0 | −0.6 |
| +8% | +7.6 | +8.0 | +8.3 |
| +15% | +14.4 | +15.0 | +15.4 |
The requirement is the worst cell, here a loss of 16.1. For a single long position this is unremarkable. The power appears when the grid is applied to the whole portfolio at once: every position is revalued under the same scenario, so a long and a short that genuinely offset produce a small combined loss in the cells that matter, and the requirement collapses toward the residual risk rather than the gross exposure. Real grids are far larger, adding interest rate and dividend shifts and running many more price points, and a large prime runs hundreds of scenarios per portfolio.
For a genuinely hedged book, portfolio margin can cut the requirement dramatically compared with Reg T, which is why funds pursue it and why regulators require the bank to demonstrate model competence before offering it. The catch is that the model is only as good as the correlations it assumes, and correlations break precisely in the stresses that matter.
House add-ons are where the bank's own judgement enters
No prime relies on a model alone. On top of the computed requirement the bank layers add-ons for the risks the model handles poorly. A concentration add-on applies when a position is large relative to the fund's capital or to the stock's daily volume. A liquidity add-on reflects how many days it would take to exit. A gap-risk add-on covers the possibility that price moves discontinuously through a stop, as it does around earnings or bid rumours. A correlation add-on penalises books where apparent hedges rest on unstable relationships.
The binding requirement is the greater of the regulatory floor and the bank's own risk-based number including add-ons. This is the practical answer to how margin is set, and the add-on layer is where post-Archegos tightening concentrated.
| Methodology | How it computes | Recognises hedges | Typical outcome on a hedged equity book | Role |
|---|---|---|---|---|
| Reg T | Fixed percentage of position value | No | Roughly half of long market value | Regulatory floor |
| Portfolio margin | Stress grid applied to the whole portfolio | Yes | Often mid-teens percent of gross exposure | The risk model |
| House add-ons | Judgemental overlay for concentration, liquidity, gap | Partly | Adds to the model number | The bank's own protection |
| Binding requirement | The greater of the floor and model plus add-ons | — | Whichever is highest | What the client actually posts |
The same trade, done two ways: cash against synthetic
One position, two structures, and almost everything differs
The reference product for this module is synthetic prime brokerage, and the clearest way to see what it is doing is to run one position through both routes.
In the cash structure, Meridian wants exposure to 100 million dollars of a listed share. It buys the shares, posts margin of 28 million from its own capital, and borrows the remaining 72 million from the prime. Meridian owns the shares. They sit in its account, it votes them, it receives the dividends, and if its stake crosses a disclosure threshold it appears in public ownership filings. The prime holds a security interest and charges a financing spread on the 72 million.
In the synthetic structure, Meridian instead enters a total return swap with the prime. The prime buys the 100 million of shares itself, on its own book, as a hedge. Meridian owns nothing. The swap says that the prime will pay Meridian any appreciation and any dividends on those shares, and Meridian will pay the prime any depreciation plus a financing fee. Economically Meridian has the same exposure. Legally it has a contract, not a share.
| Dimension | Cash (physical) | Synthetic (total return swap) |
|---|---|---|
| Who owns the share | The fund | The bank |
| What the fund holds | Shares, financed | A contractual claim on performance |
| Public ownership disclosure | Fund appears once thresholds are crossed | Fund does not appear; the bank does |
| Voting rights | The fund votes | The bank holds them; typically votes are not passed through |
| Dividends | Received directly | Passed through the swap as a cash flow |
| Capital treatment for the bank | Secured lending exposure | Derivative exposure under SA-CCR |
| Margin regime | House and Reg T style rules | Derivative margin, potentially uncleared margin rules |
| On default | Seize and sell the collateral | Close out the swap, unwind the bank's own hedge |
| Access to markets the fund cannot hold directly | No | Yes, a significant driver of demand |
Why a fund chooses synthetic, and why that choice mattered in 2021
Funds choose synthetic exposure for four honest reasons. It gives access to markets where the fund cannot easily hold securities directly, whether for tax, regulatory or operational reasons. It can be cheaper in capital terms for the bank and therefore cheaper in price for the fund. It avoids the operational burden of holding and settling the physical asset. And it simplifies the fund's own reporting to a single swap valuation.
There is a fifth reason that is legal but consequential: the fund does not appear in ownership disclosures. A fund building a large stake through swaps is not visible in the filings that would reveal a physical holding, because the shares belong to the bank. Archegos used precisely this. Each of its primes held part of its concentrated exposure and no public record aggregated it, so neither the market nor the other banks could see the total. The subsequent regulatory response, tightening reporting of large derivative positions, was aimed directly at this gap.
Step 4 — Trades are executed anywhere and end up in one place
The fund executes where it gets the best price, then gives the trade up to the prime
A fund does not have to trade with its prime. It executes wherever execution is best, with any number of brokers, and then instructs each of those executing brokers to transfer the trade to the prime for clearing, settlement and financing. That transfer is the give-up, and it is the mechanism that makes the consolidated record possible.
The give-up runs on a standing agreement among three parties: the fund, the executing broker and the prime. The executing broker sends the trade details to the prime, the prime accepts it against the client's account, and from that moment the prime carries the settlement obligation and the position enters the client's margin calculation. The executing broker is paid commission and steps out of the risk.
It works rather like shopping in many different stores but having everything delivered to one address: each shop knows only what you bought there, while the single delivery address accumulates the complete picture. This is why the prime sees the whole portfolio while each executing broker sees only its own slice. The consolidation is not a courtesy; it is the structural reason a prime can margin a portfolio on a net basis at all.
Trade capture is where data quality is won or lost
Every give-up must be matched, validated and booked. Details are compared between the executing broker's version and the client's, discrepancies in price, quantity, settlement date or account are flagged, and mismatches become breaks that operations must resolve before settlement.
A trade that fails to book correctly does not merely sit in a queue. It is invisible to the overnight margin run, so the client is margined on a portfolio that is not the portfolio it holds. Under-booking understates exposure and leaves the bank short of collateral; over-booking calls for collateral the client does not owe and damages the relationship. Trade capture accuracy is a risk control, not an operational nicety.
Execution and capture
Give-up instructions arrive as FIX messages, the same protocol that carries electronic orders, with allocation and give-up message types identifying the target account. A middleware layer normalises formats across dozens of executing brokers, each with its own conventions. The trade then lands in books and records, the sub-ledger that is the authoritative position record for the client.
Matching platforms compare the two sides and flag exceptions. Straight-through processing rates, the share of trades that book without human intervention, are a headline operational metric; leading platforms exceed the high nineties in percentage terms and the remainder consumes most of the operations team's time. Each unmatched trade is both a settlement risk and a margin risk.
Step 5 — A short sale is only possible if someone finds the stock first
The locate comes before the sale, and it is a regulatory requirement
Before a fund sells a security short, it must have reasonable grounds to believe the security can be borrowed and delivered. In the United States this is codified in Regulation SHO as the locate requirement; European short-selling rules impose a comparable obligation. A locate is the prime confirming that it has identified stock available to borrow.
The locate is not the borrow. It is an indication of availability, given before the trade. The actual borrow is contracted around settlement. The gap between the two is a real operational risk: a locate given in the morning on stock that has been lent to someone else by the afternoon produces a failed delivery.
The securities lending market prices scarcity, and the range is enormous
The fee to borrow a security is set by supply and demand, and the market divides sharply in two.
General collateral describes securities that are widely available. Large-cap stocks with broad institutional ownership can be borrowed for a few basis points a year. The fee is trivial and availability is effectively unlimited.
Specials are securities in short supply relative to borrowing demand. The distinction is the difference between renting a common hand tool, available anywhere for almost nothing, and renting the only ladder in a town where everyone suddenly needs to reach a roof. When many funds want to short the same name, or when the free float is small, or when an index event or corporate action removes lendable stock, the fee rises. Fees of several percentage points a year are routine for genuine specials, and in extreme squeezes annualised borrow costs have reached levels that make maintaining a short position ruinous regardless of whether the investment thesis is correct.
The short rebate: the most counter-intuitive mechanic in the business
Follow the cash of a short sale and something surprising appears. The fund sells borrowed stock for 100. That 100 in cash does not go to the fund; it goes to the lender of the stock as collateral, and it sits in the system earning interest. The fund is owed something for the use of its sale proceeds, and what it receives is called the short rebate.
Work the arithmetic on a general collateral name, with illustrative figures. The prevailing short-term rate is 4 percent. The stock is easy to borrow, costing 0.3 percent. The fund receives roughly 4 percent minus 0.3 percent, so about 3.7 percent a year on the cash — a genuine income stream simply for having a short position open.
Now run the same arithmetic on a special. The rate is still 4 percent, but the borrow costs 7 percent. The rebate is 4 minus 7, which is negative 3 percent: the fund now pays 3 percent a year to keep the position open. The rebate has flipped from income to cost.
This is why short sellers talk about the rebate rather than the borrow fee. It is the net carry, and it determines whether time is working for the position or against it. A short in a general collateral name can be held almost indefinitely at no cost, or at a small profit. A short in a special is a position with a clock on it, and the clock speeds up as more people crowd in.
Recall is the risk that defines short selling operationally
The lender of a security can generally demand it back. Beneficial owners recall stock to vote at a shareholder meeting, to sell the position, or because the lending agent has reallocated it. When stock is recalled and the prime cannot source a replacement, the fund's short position must be closed by buying the stock in the market, whatever the price and whatever the fund thinks of the investment case.
A forced buy-in is how a crowded short becomes a squeeze. Recalls force covering, covering pushes the price up, the rising price triggers margin calls on other short sellers, and their covering pushes the price further. The 2021 meme-stock episode was this mechanism running at full speed. Note that it is an operational and financing mechanism, not only a sentiment story.
Securities lending
An inventory management system tracks what is available to lend across the bank's own holdings, client assets that permit re-use, and external lending relationships. Locate platforms handle the high volume of availability requests, increasingly automated, with algorithmic approval for general collateral and manual review for specials. Trading and post-trade platforms such as EquiLend and Pirum automate matching, billing and reconciliation of stock loans across the market.
The hard problem is that availability is dynamic and distributed. Stock that appears available in one system may already be committed in another, and the same underlying shares can be counted twice across a re-use chain. Real-time, accurate inventory is the difference between a locate that holds and a delivery that fails.
Step 6 — Settlement moves securities against cash, and failure is expensive
Delivery versus payment removes the risk that one side performs and the other does not
Settlement is the exchange of securities for cash that makes a trade final. The governing principle is delivery versus payment: the two legs move simultaneously and conditionally, so neither party can end up having delivered without being paid. Central securities depositories operate this mechanism, holding securities in electronic form and moving them against cash across linked accounts.
Settlement cycles have shortened. North American markets moved to one business day after trade date in 2024, and other markets are following. The compression is not a minor operational adjustment. It removes an entire overnight window that operations teams historically used to resolve breaks, chase confirmations and arrange funding, which forces automation of processes that previously tolerated manual repair.
For a prime broker the shortened cycle bites hardest in securities lending and funding. A borrow that must be arranged, matched and settled by the next business day leaves no room for a slow locate or an unmatched instruction. This is why the settlement-cycle change drove a wave of investment in automation.
A failed settlement is a chain of consequences, not an isolated event
When a security is not delivered on the due date the trade fails. The buyer does not receive the security and may in turn fail on its own onward delivery, propagating the failure. European rules under the Central Securities Depositories Regulation impose cash penalties for settlement fails, which converts an operational inconvenience into a direct financial cost. Persistent fails can trigger mandatory buy-in, where the counterparty purchases the securities in the market and charges the difference to the failing party.
For the prime, a fail in a client's short position is particularly awkward: the client has sold stock it must deliver, the borrow has not arrived, and the prime is the party on the hook to the market.
Step 7 — The daily margin and collateral run is the center of gravity
This is where everything built so far is tested, every night
At the end of each trading day the prime revalues the client's entire portfolio at official closing prices, recomputes the margin requirement under the agreed methodology, compares it with the collateral the client has posted, and determines whether more is owed. If it is, a margin call goes out, and the client must meet it within a contractually specified window, typically the next morning.
Understand this cycle and you understand prime brokerage. Everything upstream exists to make it possible: onboarding established the terms, financing set-up fixed the methodology, trade capture created the positions, settlement made them real. Everything downstream reports its results. The margin run is the moment where the bank's risk is measured, and the only moment where it can act on what it finds.
The run is a genuine computational undertaking. A large prime revalues millions of positions across dozens of markets and currencies, applies hundreds of stress scenarios per portfolio, aggregates across accounts subject to netting agreements, applies house add-ons, and produces requirements per client, all in the hours between the last market close and the first market open. There is no version of this that can be late.
Intraday margin is the post-Archegos change that mattered most
Overnight margining assumes the world is stable between closes. In a violent move it is not. A client can be comfortably collateralised at Monday's close and deeply short of collateral by Tuesday lunchtime, and the bank does not find out until Tuesday night.
Archegos exposed this precisely. The positions moved violently within days, and the primes' ability to recognise and act on deteriorating exposure was slower than the deterioration. The industry response was to move toward intraday recalculation for concentrated or volatile exposures, with the ability to issue an intraday call. This is technically demanding: it requires live prices, live positions, and a risk engine that can run in minutes rather than hours.
Collateral is not one thing, and choosing what to post is an optimisation
The client meets a call by posting collateral, and it usually has choices: cash in various currencies, government bonds, equities, or existing positions already held at the prime. Each is valued with a haircut, a discount reflecting how much its value might fall before it could be sold. Cash takes no haircut; a government bond might take two percent; a volatile equity might take twenty-five percent or be ineligible entirely. The logic is the same one a pawnbroker applies: lend freely against gold, cautiously against a used car, and not at all against something that may be unsellable next week.
Because assets have different haircuts, different funding costs and different alternative uses, choosing what to post is an optimisation problem worth real money. Post your cheapest eligible asset, keep the valuable and reusable ones free. Large funds run collateral optimisation continuously, and so do banks on their own side. This is the prime brokerage equivalent of a cheapest-to-deliver calculation, and it is one of the clearest places where better technology converts directly into basis points.
| Collateral type | Typical haircut | Why the haircut is set there |
|---|---|---|
| Cash in major currency | None | No price risk; immediately usable |
| Government bonds, high quality | Low single-digit percent | Price risk is small and liquidity is deep |
| Investment-grade corporate bonds | Moderate | Wider spreads and thinner liquidity in stress |
| Large-cap listed equities | Substantial | Volatile, and correlated with the exposure being secured |
| Small-cap or illiquid equities | Very large, or ineligible | May be impossible to sell in the window that matters |
Haircuts are indicative and vary by agreement, market and counterparty. The pattern to remember is that the haircut rises with price volatility and with the time it would take to sell.
Re-use and rehypothecation: the mechanism that funds the business and endangers the client
When a client posts assets or holds financed positions, the agreement often permits the prime to re-use those assets: to lend them, to pledge them, or to fund itself with them. This is rehypothecation. It is central to the economics of prime brokerage, because it lets the bank fund client positions with client assets rather than its own cash, and it is the reason financing is as cheap as it is.
The pawnbroker analogy extends here, and becomes uncomfortable. You pawn your watch as security for a loan; the pawnbroker then pledges your watch to raise money of his own. While everyone is solvent, nothing is wrong and your loan is cheaper because of it. If the pawnbroker fails, the watch is somewhere down a chain of other people's claims, and you are holding a piece of paper.
It is also the mechanism that hurt clients most in 2008. Assets that have been rehypothecated are no longer sitting identifiably in the client's name; the client has a contractual claim rather than a segregated holding. When Lehman Brothers failed, hedge fund assets that had been rehypothecated through its London entity became claims in an insolvency process that took years to resolve, and clients that believed they held assets discovered they held creditor positions.
Regulation now constrains this. United States customer protection rules under Rule 15c3-3 limit rehypothecation to a percentage of the client's indebtedness and require segregation of fully paid securities. United Kingdom and European client asset rules impose segregation and disclosure obligations. Sophisticated clients negotiate re-use limits explicitly and monitor them.
Margin and collateral
The margin engine is almost always proprietary. It ingests positions from books and records, prices from market data providers, and parameters from the documentation, then runs the stress grid and produces requirements. Banks build rather than buy because the methodology is a competitive differentiator, because it must handle the bank's specific product mix, and because regulators expect the bank to own and explain its model.
Collateral management systems track what has been posted, by whom, against which agreement, with which haircut, and whether re-use is permitted. Triparty collateral agents, operated by major custodians, hold and allocate collateral between counterparties, automating selection and substitution against eligibility rules.
The reconciliation burden is severe. The bank's view of the client's collateral must match the client's view, the custodian's view and the triparty agent's view. Portfolio reconciliation services compare positions between counterparties daily and flag differences, which is how disputes are caught before they become defaults.
Corporate actions are where lending and financing get genuinely complicated
A share is not a static object, and events on it have to be passed to the right party
Everything so far has treated a security as a thing that has a price. It is also a bundle of rights: to a dividend, to a vote, to participate in a rights issue, to be exchanged in a takeover. Corporate actions are those events, and they are the operational domain where prime brokerage most often breaks, because every event has to reach whoever is economically entitled to it, and lending and financing have separated economic entitlement from legal ownership.
Start from the simplest case and build. If a fund owns a share outright, the dividend arrives and the vote is theirs. Nothing is difficult. Now introduce the mechanics of this module and the question becomes genuinely hard.
The manufactured dividend: paying a dividend on a share you lent away
Recall what a stock loan does. Title genuinely transfers to the borrower, who sells the share to a third party. When the company pays a dividend, it pays whoever holds the share on the record date, which is now that third party. But the beneficial owner who lent the stock has not agreed to give up their income; they lent the share, not the dividend.
The borrower therefore has to make them whole, and the payment that does this is called a manufactured dividend, or a payment in lieu. It is a contractual cash payment from borrower to lender, economically replacing the real dividend the lender would have received.
For the short seller the mirror applies, and it surprises newcomers every time. A fund that is short a stock pays the dividend rather than receiving it. It borrowed the share and sold it; the buyer receives the real dividend from the company; the fund must fund the manufactured dividend passed back up the chain. A high-yielding stock is therefore expensive to stay short in, entirely apart from the borrow fee, and dividend dates are a recurring reason short positions get closed.
Voting rights travel with the share, and that has consequences
Whoever holds legal title on the record date votes. A lender who has lent stock across a shareholder meeting has, in effect, lent away their vote, which is why lending agents recall stock before contested meetings and why some beneficial owners restrict lending around them entirely. This is the mechanism behind the recalls described earlier: they are frequently not about the fund at all, but about somebody upstream wanting to vote.
The abusive version has a name. Empty voting is acquiring votes without economic exposure, by borrowing shares purely to vote them. It is legal in many jurisdictions and widely regarded as illegitimate, and it is a live governance debate.
In the synthetic structure the position is different again: the bank holds the shares and therefore the votes, and swap documentation generally does not pass voting rights through to the fund. A fund with large synthetic exposure has economic ownership and no voice, which is one of the honest arguments for the disclosure reforms that followed Archegos.
Rights issues, mergers and the elective problem
Some corporate actions require a choice: take shares or cash in a merger, subscribe to a rights issue or sell the rights, accept a tender offer or not. These are elective events, and they force the prime to ask the client what they want, apply the answer before a deadline set by someone else, and handle the consequences of a client who does not respond.
Elective events on financed and lent positions are one of the most reliable sources of loss in operations, because the deadline is external, the default action if nobody responds may be economically wrong, and a missed election cannot usually be reversed.
The tax dimension is not a footnote, and it has shaped regulation
Different holders of the same share face different withholding tax on its dividend. A domestic pension fund may receive it gross; a foreign investor may suffer a deduction at source. Because securities lending moves shares between holders across a dividend date, it can move the dividend to a holder with a better tax position, and the parties can share the benefit. This is dividend arbitrage, and in its ordinary form it is a well-known and much-scrutinised feature of the lending market. Lending demand visibly spikes around dividend dates in some markets for exactly this reason.
The abusive extreme is worth knowing because it reshaped European regulation. The cum-ex schemes exploited settlement timing around dividend dates to obtain refunds of tax that had been paid once but was reclaimed more than once, costing several European treasuries billions of euros and resulting in criminal prosecutions. The consequences for this business were direct: far stricter scrutiny of dividend-period lending, documentation of tax status at position level, and internal policies at most banks that prohibit lending structured around a tax outcome.
The synthetic structure has its own tax history. Section 871(m) in the United States exists because total return swaps could replicate the economics of holding a dividend-paying share while avoiding the withholding tax a foreign holder would face on the dividend itself. The rule treats certain swap payments as dividend equivalents and taxes them accordingly, which is a clean illustration of the module's central theme: change who owns the asset and every consequence downstream changes with it, including the tax.
Corporate actions
Corporate actions are a data problem before they are a processing problem. Announcements arrive from multiple sources in inconsistent formats, and the first task is to build a single golden record of the event, its dates and its terms. Vendors supply cleansed feeds precisely because reconciling raw announcements is unreliable, and firms commonly take two sources and compare them.
The processing engine must then apply the event to every affected position, which requires knowing not only who holds what but under what arrangement: owned outright, financed, lent out, borrowed, pledged as collateral, or referenced by a swap. Each produces a different entitlement and a different cash flow. Manufactured payments have to be generated, matched and billed, and this is a substantial part of what platforms such as Pirum automate in the lending market.
Two failure modes recur. Missing an elective deadline creates a direct and usually unrecoverable loss. And applying an event to the wrong entitlement, for instance paying a real dividend to a client whose stock was lent out and who was therefore due a manufactured payment with different tax treatment, creates a break that is discovered later and is painful to unwind.
When a client cannot pay, the waterfall is the only thing that matters
Default management is prime brokerage's version of the irreversible moment
If a client fails to meet a margin call, the bank enters a sequence that its documentation has anticipated. First the bank declares an event of default under the relevant agreements. Close-out netting collapses all transactions into a single net amount. The bank then applies the collateral it holds, liquidates the client's positions to recover the remainder, and pursues any shortfall as an unsecured claim.
The liquidation is where the theory meets the market. The bank now holds a portfolio it did not choose, must sell it into a market that frequently knows something is wrong, and competes with other primes liquidating the same client's positions at the same time. Every assumption in the margin model is now being tested at once: the liquidity assumption, the correlation assumption, and the days-to-liquidate assumption.
| Step | What the bank does | What can go wrong |
|---|---|---|
| 1. Declare default | Serve notice under the relevant agreements | Trigger is disputed, or documentation is ambiguous |
| 2. Close out and net | Collapse all transactions to one net amount | Netting is unenforceable in a jurisdiction |
| 3. Apply collateral | Seize and value the collateral held | Collateral has fallen in value, or is illiquid or ineligible |
| 4. Liquidate positions | Sell the portfolio into the market | Market knows, prices gap, other primes are selling the same names |
| 5. Pursue the shortfall | Claim as an unsecured creditor | Recovery is slow, partial, and often negligible |
Coordination between primes is the problem Archegos exposed
When a multi-prime client defaults, each prime holds part of the portfolio and each must decide whether to sell immediately or coordinate. Selling first gets the best price; coordinating gets a better outcome for everyone but requires trust and, in some interpretations, raises legal questions. In the Archegos case, banks that liquidated quickly limited their losses substantially, while those that moved more slowly absorbed far more damage. The lesson banks took was not primarily about coordination; it was that the margin should have been higher and the concentration recognised much earlier.
Step 8 — Reporting and service are what the client actually experiences
The statement is the product, from the client's side of the desk
The prime produces a daily consolidated report: positions, valuations, profit and loss, cash and stock balances, financing charges, borrow fees, margin requirement and collateral. The fund's own operations team reconciles this against its internal books daily, and any difference must be investigated.
This report is the client's window into the relationship, and its quality drives satisfaction more than pricing does. A fund that must call to understand a financing charge, or that finds recurring breaks, will move business regardless of spread. Reporting quality is a genuine competitive dimension, which is why primes invest heavily in client portals and increasingly in data feeds that let sophisticated funds pull raw data into their own systems.
Capital introduction is the service that looks least like banking and matters commercially
Capital introduction is the prime's function of connecting hedge fund clients with potential investors: pension funds, endowments, funds of funds, family offices and sovereign wealth funds. The prime hosts events, makes introductions and advises managers on positioning.
The bargain is straightforward and rarely stated aloud. A fund that raises capital through its prime grows, and a larger fund generates larger balances and more revenue for that prime. The prime is investing in its clients' growth because it captures a share of the result. Capital introduction teams are compensated on the relationships they build rather than directly on assets raised, and the strongest primes treat this as a genuine differentiator.
There is a conflict to name here. The prime knows a great deal about its clients' portfolios and performance, and it is introducing those clients to investors. The information barrier between what the prime knows as a financier and what it says as an introducer must be real, and it is governed by policy and monitored by compliance.
The relationship has a lifecycle of its own, and it rarely ends cleanly
Balances grow, terms get repriced, and the conversation repeats
A successful relationship compounds. The fund grows, balances rise, the prime earns more, and both sides become more entangled. Periodically the terms are revisited, usually because something changed on the bank's side: a new capital rule, a shift in funding costs, a strategic decision to reduce exposure to a strategy. Repricing conversations are the normal texture of a mature relationship.
Funds respond by adjusting wallet share rather than by leaving. A fund unhappy with one prime's pricing shifts balances toward another. Because most institutional funds run multiple primes, this reallocation is continuous and is a far more common expression of dissatisfaction than a full termination.
Offboarding is slow, and exit risk runs in both directions
Ending a prime relationship means moving positions, collateral and stock loans to another provider without disturbing the market or breaking the fund's strategy. It takes weeks and is operationally delicate. Banks also exit clients, and occasionally whole segments, when capital rules change or risk appetite shifts. A bank withdrawing from a strategy gives notice and works the positions down, and the affected funds scramble for replacement capacity.
The exit that matters most is the one nobody controls: the client that deteriorates. Watching a client's risk profile drift, raising margin, reducing limits, and eventually declining to renew is a normal part of the discipline. Doing it early is the difference between a difficult conversation and a loss.
What is a give-up, and why does it make prime brokerage possible?
It is the transfer of a trade from the executing broker to the prime for clearing, settlement and financing. It lets the fund execute anywhere while the prime holds the consolidated position record, which is what allows margining on a net portfolio basis.
Why is close-out netting the most important clause in the documentation?
Because on default it collapses all transactions into a single net amount, preventing an administrator from enforcing profitable contracts while disclaiming unprofitable ones. Without it, exposure is gross rather than net and the capital treatment changes drastically.
A hedged book attracts a much lower requirement under portfolio margin than under Reg T. What is the hidden assumption, and when does it fail?
That the correlations recognised as hedges will hold. They break down precisely in the correlated stress scenarios where the margin is actually needed, which is why house add-ons exist on top of the model.
Distinguish a locate from a borrow, and explain the risk in the gap.
A locate is a pre-trade confirmation that stock appears available; the borrow is the actual contract arranged around settlement. Stock located in the morning can be committed elsewhere by the afternoon, producing a failed delivery.
Explain the mechanism by which a recall turns a crowded short into a squeeze.
A beneficial owner recalls stock, the prime cannot source replacement, the fund must buy in the market to cover, that buying raises the price, the higher price triggers margin calls on other short sellers, and their covering pushes the price higher still.
Why is rehypothecation both essential and dangerous?
It lets the bank fund client positions using client assets rather than its own cash, which is why financing is cheap. But re-used assets are a contractual claim rather than a segregated holding, so in an insolvency the client is a creditor, as Lehman's prime brokerage clients discovered.
Why did the industry move toward intraday margining after 2021?
Because overnight margining leaves the bank blind between closes, and Archegos showed that a concentrated portfolio can deteriorate far faster than a once-daily recalculation can detect or act on.
Derive the margin requirement from first principles, and say what each add-on corrects.
Margin must cover a bad price move over the time needed to liquidate, so it is roughly a confidence factor times daily volatility times the square root of the liquidation horizon, applied to position value. Concentration and liquidity add-ons correct the horizon; gap-risk add-ons correct the assumption that ordinary volatility describes the move at all.
A fund is short a general collateral name and a special, both at a 4 percent short-term rate. Describe the carry on each.
On the general collateral name, borrowing at roughly 0.3 percent, the rebate is about positive 3.7 percent, so the fund earns income for holding the short. On the special, borrowing at 7 percent, the rebate is negative 3 percent, so the fund pays to keep the position open and time works against it.
Why is a fund with large synthetic exposure invisible in ownership filings, and why did that matter?
Because the bank owns the shares as its hedge and the fund holds only a contractual claim, so the fund never crosses a disclosure threshold. Archegos spread concentrated synthetic exposure across several primes, so no public record and no single bank aggregated the total.
Why does a fund that is short a stock pay the dividend rather than receive it?
It borrowed the share and sold it, so the buyer receives the real dividend from the company. The lender must still be made whole, so the short seller funds a manufactured dividend passed back up the lending chain, which makes high-yielding stocks expensive to stay short in.
Read the lifecycle backwards from a client default. What purpose does each earlier stage serve?
Onboarding assessed whether default was likely, documentation set out what the bank may do when it happens, the margin methodology sized the buffer against the expected liquidation loss, and collateral made that buffer real and available.
2
What is a give-up?
No.
No.
No.
Right. The fund trades anywhere; the prime holds the whole position.
Why is close-out netting the most important clause?
No.
No.
Different clause.
Right.
What is the hidden assumption of portfolio margin?
Portfolio margin replaces Reg T.
Right. Hence house add-ons.
No.
No.
What is the risk between a locate and a borrow?
No.
A locate is only a pre-trade check.
No.
Right.
3Roles, Actors, and Interactions
A prime brokerage relationship looks bilateral and is not. Five distinct groups touch every position, and the prime sits between all of them, holding information that none of the others can see in full.
Five camps surround the relationship, and the prime sits in the middle of every flow
The camps want different things, and the prime is the only party talking to nearly all of them
The client is the hedge fund, represented day to day by its chief operating officer and operations team rather than by the portfolio managers who get the attention. The fund wants cheap financing, reliable borrow, accurate reporting, and confidence that leverage will still be there when markets turn. Its operations team is the prime's real counterpart, and its satisfaction usually determines whether the relationship grows.
The bank is the prime broker, itself a collection of teams whose interests do not align perfectly. Sales wants the client, risk wants the buffer, treasury wants the funding cost covered, operations wants clean trades, and finance wants the return on balance sheet. The internal negotiation among these is constant.
The market counterparties are the executing brokers and the securities lending market. Executing brokers trade with the fund and give the trades up. Beneficial owners, lending agents and other borrowers form the market from which stock is sourced. Neither group sees the fund's full portfolio, and neither carries the fund's credit risk once the give-up completes.
The fund's own service providers are the camp most often forgotten, and the administrator is the important one. A hedge fund does not mark its own book for its investors. An independent fund administrator maintains the official books, computes the net asset value — the NAV, being the fund's total assets minus liabilities, and therefore what an investor's stake is actually worth — and strikes it monthly or more often. The administrator receives position and valuation data from the prime and reconciles it against the fund's own records.
This makes the administrator the independent check on what the prime reports, and it is the reason a discrepancy between prime and fund is not merely an internal irritation: it can delay a NAV, which is visible to the fund's investors. Alongside the administrator sit the fund's auditor, who verifies annually, and its investors, usually institutional allocators whose operational due diligence explicitly asks which primes the fund uses, how assets are protected, and how NAV is verified. A prime with weak reporting damages its client's standing with its own investors, which is a commercial consequence banks sometimes underestimate.
Market infrastructure is the custodians, central securities depositories, central counterparties and triparty agents. A central counterparty, or CCP, is a clearinghouse that interposes itself between buyer and seller so that each faces the clearinghouse rather than each other, collecting margin from both. These are the utilities that hold securities, move them against cash, and allocate collateral. They are neutral, systemically important, and largely invisible until something fails.
Every major interaction, mapped
| From | To | What flows | Stage |
|---|---|---|---|
| Hedge fund | Executing brokers | Orders | Execution |
| Executing brokers | Prime broker | Give-up instructions, trade details | Execution and capture |
| Hedge fund | Prime broker | Locate requests, collateral, instructions | Daily cycle |
| Prime broker | Hedge fund | Financing, borrow, margin calls, statements | Daily cycle |
| Prime broker | Securities lending market | Borrow requests, collateral, returns | Stock borrow |
| Prime broker | Custodian and CSD | Settlement instructions, custody records | Settlement |
| Prime broker | Triparty agent | Collateral allocation and substitution | Collateral |
| Prime broker | Regulators | SFTR reporting, client asset returns, capital reporting | Continuous |
| Prime broker | Fund administrator | Position and valuation feed | Daily, for NAV |
| Hedge fund | Fund administrator | Its own books, for reconciliation | Daily and monthly |
| Fund administrator | Fund investors | Official NAV | Monthly |
| Prime broker | Investors (for the fund) | Capital introduction | Service |
Notice how many arrows have the prime at one end. That is the structural insight of this part. The prime is the only participant with a complete view of the client's activity, and that informational position is simultaneously its commercial advantage, its risk management capability, and its most sensitive compliance exposure.
Hedge fund clients are not one client type but several
Strategy determines the financing profile, and therefore the whole relationship
The phrase "hedge fund client" hides more than it reveals. Different strategies consume balance sheet differently, generate different revenue, and fail in different ways. A prime's client mix is a risk position in its own right.
Long/short equity funds are the archetypal prime brokerage client
These funds hold long positions and short positions in equities, aiming to profit from the difference rather than from market direction. They are the natural client: they need cash financing for longs, stock borrow for shorts, and they use both continuously. Meridian is one. They generate balanced revenue across financing and lending, and their risk is generally manageable because equity books can be liquidated relatively quickly, at a price.
Multi-strategy funds are the largest, the most profitable, and the most feared
Multi-strategy platforms run many independent teams across equities, credit, macro and events, often with substantial leverage. They are enormously valuable clients: their balances are large, they trade across every product, and they use several primes simultaneously. They are also the concentration risk that keeps senior risk officers awake, because a single such client can represent a material share of a prime's exposure, and because their internal complexity makes them harder to model.
Quantitative and systematic funds trade constantly and are operationally demanding
Systematic funds run algorithmic strategies with high turnover and often thousands of small positions. They generate strong commission and financing flow, and their books are usually well diversified, which makes them attractive credits. What they demand is technology: low-latency data, precise reporting, high straight-through processing rates, and operations that can handle enormous trade volumes without breaks. They will leave a prime whose systems cannot keep up.
Macro and credit funds bring different instruments and different liquidity
Global macro funds trade rates, currencies and index exposure, often through derivatives, and consume balance sheet through margin on those derivatives rather than through equity financing. Credit funds hold instruments that can be far less liquid than equities, which lengthens days-to-liquidate assumptions and raises the margin required. Both extend the prime beyond its equity core and require different systems and different risk models.
| Client type | Financing profile | Revenue quality | Balance-sheet intensity | Why the prime wants or fears them |
|---|---|---|---|---|
| Long/short equity | Cash financing plus heavy stock borrow | Balanced, recurring | Moderate | Core client; borrow demand is the profitable part |
| Multi-strategy | All products, large balances, multi-prime | Very high, diversified | Very high | Most valuable and the single largest concentration risk |
| Quantitative / systematic | High turnover, diversified, low concentration | High volume, thin per trade | Moderate | Attractive credit; demands the best technology |
| Macro / credit | Derivative margin, less liquid instruments | Lumpy, product-specific | High per unit of revenue | Extends the platform; longer liquidation horizons |
The multi-prime model creates an information asymmetry that no single bank can resolve on its own, and it is the structural condition behind the largest loss in this business.
A fund using four primes shows each of them roughly a quarter of its activity. Each prime margins the slice it can see, and each slice may look diversified and well collateralised in isolation. The aggregate position can be concentrated, enormous, and invisible to every one of them. Public disclosure would normally provide a partial correction, since large equity holdings must be reported, but synthetic exposure held through swaps sits with the bank rather than the fund and therefore never appears in the fund's name.
The available defences are imperfect and worth knowing. Banks ask clients directly about total leverage and exposure elsewhere, and can make honest answers a condition of the relationship. They monitor for signs that a position is crowded, such as unusual borrow costs or their own hedging difficulty in a name. Regulators receive transaction reports that in principle allow aggregation, though not in a form that reaches banks in time. And a bank can simply decline to extend concentrated leverage to a client that will not explain its full position.
That last defence is the only reliable one, and it costs revenue. This is why the honest description of the post-Archegos change is not that the industry solved an information problem, but that it became more willing to lose business rather than lend blind.
The prime knows too much, and information barriers turn that into an architecture problem
The problem stated plainly: one institution holds knowledge it must not share with itself
A prime broker sees its clients' complete portfolios, in real time. Elsewhere in the same bank, traders take proprietary and market-making positions, research analysts publish opinions that move prices, advisory bankers hold inside information about transactions, and capital introduction teams talk to investors about the same funds whose books the prime can see.
The situation resembles a law firm that finds itself acting for both sides of the same dispute: the knowledge is legitimately held, the institution cannot un-know it, and the only workable answer is to make sure the two halves of the building genuinely cannot talk to each other. Each of those adjacencies is a conflict that regulation and policy must contain. If a market-making desk learned that a large client was building a short position, it could trade ahead of it. If capital introduction learned that a fund was performing badly, its introductions would be compromised. If advisory bankers working on a takeover could see which funds held the target, the information would be extraordinarily valuable and extraordinarily illegal to use.
Four separations do the work
Client position confidentiality separates the prime's client-facing systems from the bank's trading activity. Position data is restricted to those who need it for financing, risk and operations. This is the most important barrier in prime brokerage and the one most specific to it.
The traditional information barrier separates the public side from the private side, keeping advisory teams holding inside information away from trading and sales. Prime brokerage sits on the public side but interacts with clients whose activity may relate to transactions the private side knows about.
Research separation keeps analysts independent of the financing relationship, so that a research view is not shaped by the bank's exposure to funds positioned in that name.
Capital introduction separation limits what the introduction team may know and say about a client's performance and positioning, so that introductions rest on the fund's own disclosures.
Wall-crossing is the controlled exception, and it must be logged
Sometimes a person must legitimately cross a barrier: a risk officer needs position detail, or a senior manager must be briefed on a deteriorating exposure. This is done through a formal wall-crossing procedure. The person is named, the reason recorded, the time stamped, and the individual becomes restricted from trading the affected names. The log is auditable and reviewed.
The essential point is that a wall-crossing is an event with a record, not a conversation. Regulators examining a suspected leak will ask who had access, when, and under what authorisation, and the answer must come from a system rather than from memory.
Enforcing the barriers
Barriers are enforced technically, not by policy alone. Entitlement and access control systems restrict position data at the level of individual clients and individual fields, so that a user sees only what their role requires. Restricted lists, maintained centrally and fed automatically into order management and personal-account-dealing systems, block trading in affected names. Communication surveillance monitors email, chat and voice for indications that information has crossed a barrier, increasingly with language models that can detect intent rather than keywords.
Wall-crossing logs are append-only records with full audit trails. Personal account dealing systems require employees to pre-clear trades against the restricted list. Data-loss prevention tools flag exports of position data to unmanaged destinations.
The moving target is the channel. Barriers designed for email now have to cover chat platforms, mobile messaging, video calls and collaboration tools, some of which sit outside the bank's control entirely. Enforcement actions in recent years have focused heavily on unmonitored messaging applications, and the remediation has been as much cultural as technical.
Seven teams run the relationship, and the barrier runs through them
Each team owns a stage and lives in a system
| Team | Division | Core role | Primary technology |
|---|---|---|---|
| Prime sales and coverage | Markets, front office | Wins and manages the client relationship, negotiates terms | CRM, pricing and revenue analytics |
| Client service and onboarding | Operations | Onboards the client, resolves queries and breaks daily | Client lifecycle management, case tools, client portal |
| Securities lending desk | Markets, front office | Sources and prices stock borrow, manages inventory | Inventory and locate platforms, EquiLend, Pirum |
| Counterparty credit risk | Risk, independent | Sets limits, approves clients, owns the margin methodology | Margin engine, limits and exposure systems |
| Collateral and treasury | Finance and operations | Manages collateral, funding and re-use, optimises balance sheet | Collateral management, triparty, funding systems |
| Asset servicing (corporate actions) | Operations | Processes dividends, votes, elections and manufactured payments across owned, financed, lent and swapped positions | Corporate actions engine, vendor event feeds |
| Operations and settlement | Operations | Books trades, settles, reconciles, produces reporting | Books and records, settlement, reconciliation tools |
The independence of counterparty credit risk is the structural control. Risk does not report to the business it constrains. When sales wants to onboard a client that risk finds unattractive, or wants an add-on reduced, the disagreement escalates through a governance structure rather than being settled by whoever is more senior in the revenue line. Every prime brokerage failure in the historical record involved that independence being weaker in practice than on the organisation chart.
Mapping teams to the lifecycle makes accountability concrete
Origination belongs to sales. Onboarding belongs jointly to client service, credit risk and legal, and the credit gate is where risk exercises its veto. Financing set-up belongs to risk with client service implementing. Execution and give-up belong to operations. Locate and borrow belong to the securities lending desk. Settlement belongs to operations. The margin run belongs to risk, with collateral and treasury executing the calls. Reporting and service belong to client service, and capital introduction sits separately with its own information constraints.
Corporate actions belong to asset servicing, and this team is worth naming explicitly because its work cuts across every other stage rather than sitting at one point in the lifecycle. A dividend touches financed longs, lent stock, borrowed stock and swap positions simultaneously, and each requires a different entitlement and a different payment. Where the clearing businesses exist, they are usually run by a separate clearing operations team in a separate legal entity, because client money rules for listed derivatives differ from those for securities and cannot be commingled.
When something goes wrong, the useful diagnostic question is which stage failed, because that identifies both the team and the system. A margin call that was too small points at risk and the margin engine. A settlement fail points at operations and the settlement platform. A locate that did not hold points at the lending desk and inventory management. This mapping is the fastest way to make sense of an incident report.
Why does the prime broker sit at one end of nearly every interaction?
Because the give-up model concentrates all of the fund's activity into one consolidated record at the prime, while each executing broker, lender and infrastructure provider sees only its own slice.
Which client type is simultaneously the most valuable and the most feared, and why?
Multi-strategy funds. Their balances and cross-product activity generate the most revenue, but a single such client can represent a material share of the prime's total exposure and their internal complexity makes them harder to model.
What is the hidden correlation in a prime's client book?
Position overlap. Funds that are independent as counterparties may hold the same crowded positions, so they deteriorate together and try to exit simultaneously, which is the unwind scenario that margin models handle worst.
Which information barrier is most specific to prime brokerage, and what does it protect?
Client position confidentiality, separating the prime's view of client portfolios from the bank's own trading. It prevents the bank from trading on knowledge of what its clients hold or are building.
Why is a wall-crossing a system event rather than a conversation?
Because a regulator investigating a suspected leak will ask who had access, when, and under what authorisation, and that must be answerable from an append-only log rather than from recollection.
What role does the fund administrator play, and why does a reporting break matter beyond the bank?
The administrator maintains the fund's official books and strikes the NAV that investors rely on, reconciling the prime's data against the fund's own. A discrepancy can delay a NAV, which is visible to the fund's investors and damages the fund's standing, so poor prime reporting has consequences outside the bilateral relationship.
Why does counterparty credit risk report outside the business line?
So that limits, client approvals and margin add-ons cannot be overruled by the revenue line. Every historical prime brokerage failure involved that independence being weaker in practice than on the organisation chart.
3
Which client is both the most valuable and the most feared?
Not a prime client.
Not a typical prime client.
Right. Huge revenue, but a material share of exposure.
Smaller exposure.
What is the hidden correlation in a prime's client book?
Right.
No.
No.
No.
Which information barrier is most specific to prime brokerage?
No.
Right.
General, not specific to PB.
No.
Why does the prime sit at one end of nearly every interaction?
Executing brokers execute.
Right.
No.
No.
4Systems, Data, and IT
In most of a bank, technology supports the business. In prime brokerage the platform is the product, and a financing decision that the overnight run cannot compute does not exist. This part takes the machine apart.
The functional architecture is a chain of eight blocks
Ask what functions the business needs before asking who sells them
Vendor names change and mislead. Start instead with the functions any prime brokerage platform must perform, regardless of who builds them, and the architecture becomes stable and comparable across institutions.
| Functional block | What it must do | Why it is hard |
|---|---|---|
| Onboarding and KYC | Create the client and entity hierarchy, screen, capture legal terms | Entity deduplication across similar names and structures |
| Books and records | Hold the authoritative position, cash and stock record | Must be right in real time across markets and currencies |
| Margin engine | Value the portfolio, run stress scenarios, produce the requirement | Enormous computation in a fixed overnight window |
| Securities lending and inventory | Track availability, price borrow, manage locates and recalls | Availability is dynamic, distributed and double-countable |
| Collateral management | Track posted collateral, haircuts, eligibility, re-use | Must reconcile against client, custodian and triparty views |
| Corporate actions (asset servicing) | Capture events, compute entitlements, generate manufactured payments and elections | Entitlement depends on arrangement, not just holding; deadlines are external |
| Settlement and custody | Instruct and confirm delivery versus payment, hold assets | Compressed cycles, global market variation, fails and penalties |
| Client reporting | Produce statements, portal views and data feeds | The client's whole experience of the relationship depends on it |
Four capabilities cut across every block. Compliance and information barriers govern who may see what. Regulatory reporting extracts and submits data to several regimes. Reference data supplies instruments, prices, entities and identifiers to everything. Counterparty credit and limits monitors exposure against approved lines continuously.
The books and records layer is the one that must never be wrong
Among the seven, one deserves emphasis. Books and records, the sub-ledger holding each client's positions, cash and stock balances, is the authoritative source from which margin, collateral, settlement and reporting all derive. It is the single point of truth.
Everything downstream inherits its errors without question. The margin engine does not sanity-check the positions it is given; it margins what it receives. The client statement reports what the sub-ledger holds. This is why reconciliation between the sub-ledger and custodian records, exchange records and client records is run daily and treated as a control rather than as housekeeping.
Where banks buy and where they build tells you where the advantage is
The vendor landscape fills most blocks, but not the important one
A functioning prime brokerage platform mixes purchased components with proprietary ones, and the pattern of that mix is a map of where the bank believes its competitive advantage lies.
| Block | Typical approach | Representative providers |
|---|---|---|
| Onboarding and KYC | Buy, then configure heavily | Client lifecycle and screening vendors, Fenergo-class platforms |
| Books and records | Buy a core, extend substantially | Broadridge and comparable securities processing platforms |
| Margin engine | Build, almost always | Proprietary; vendor models used for validation and benchmarking |
| Securities lending | Buy connectivity, build inventory logic | EquiLend, Pirum, FIS and similar |
| Collateral management | Buy, integrate with triparty | Vermeg, Murex, Calypso, triparty agents at major custodians |
| Corporate actions | Buy the event data, build the entitlement logic | Vendor event feeds, with proprietary entitlement engines |
| Settlement and custody | Buy or use the bank's own custody network | Core securities processing platforms, SWIFT connectivity |
| Client reporting | Build the portal, buy the plumbing | Proprietary portals over purchased data infrastructure |
The margin engine is built, and the reason is instructive. The methodology is a competitive differentiator, because it determines what the bank can offer a client and what risk it retains. It must handle the bank's specific product mix and netting agreements. And regulators expect the bank to own, explain and defend its model rather than point at a vendor. Where a bank builds, it believes the function distinguishes it. Where it buys, it has concluded the function is plumbing.
The client portal is the second place banks build, and for the opposite reason: not because it manages risk but because it is what the client experiences every morning.
The clearing businesses run on a separate stack, and that is deliberate
Where the bank also offers listed derivatives or OTC clearing, those businesses generally sit on their own platform in their own legal entity. The reason is regulatory rather than technical: client money in futures clearing is governed by a distinct and stricter segregation regime, and commingling those balances with securities financing balances is not permitted. Margin is set by the clearinghouse rather than by the bank's own model, which removes the single most differentiating component of the prime brokerage stack and explains why this business is far more standardised and far less profitable.
The consequence for architecture is a recurring integration problem. A client wants one consolidated view across financing and clearing; the systems are deliberately separated and cannot simply be merged. The usual answer is a reporting layer that aggregates across both without commingling the underlying records, which is exactly the sort of requirement that sounds trivial in a meeting and consumes a programme.
Data lineage stitches the blocks into one machine
Follow one position from execution to regulatory report
The clearest way to understand the platform is to trace a single object through it. Take one of Meridian's positions from the moment it is created to the moment a regulator reads about it.
| Stage | System | Data object created |
|---|---|---|
| Execution and give-up | Order management, FIX gateway | Trade record with instrument, quantity, price, account, UTI |
| Capture and booking | Books and records | Position and stock record, cash movement |
| Valuation and risk | Margin engine | Marked value, stress results, margin requirement |
| Collateral | Collateral management | Collateral requirement, pledge and re-use instruction |
| Settlement | Settlement platform, CSD link | Settlement instruction, confirmation, custody entry |
| Reporting | Client portal, regulatory reporting | Client statement, SFTR transaction report |
The cascade principle is the reason data quality is a risk topic
Each object depends on the one before it. Get the instrument identifier wrong at booking and the position is valued against the wrong security, margined incorrectly, settled to the wrong place, and reported inaccurately. It behaves like a wrong figure entered at the top of a spreadsheet: nothing looks broken at the point of entry, and every formula below it quietly returns a confident wrong answer. An error does not stay where it was made. It reaches every system downstream and surfaces at the point of greatest consequence, which is usually the margin call or the regulatory submission.
This is why reconciliation runs at every boundary, and why the industry's operational metrics focus on breaks: unmatched trades, position differences, collateral disputes and settlement fails. Each is a symptom of a lineage break, and each has a downstream consequence that is larger than the break itself.
The margin run is a computing problem before it is a risk problem
Multiply the numbers and the engineering constraint becomes obvious
The overnight run is described casually as "revaluing the book". Put arithmetic on it and the scale of the engineering appears.
A large prime might hold two million client positions. Each must be revalued under a scenario grid of, say, five hundred scenarios once interest rate, dividend and volatility dimensions are combined with the price grid. That is a billion position revaluations. Those results must then be aggregated up netting sets, which is not a simple sum, because the whole point of portfolio margining is that offsets are recognised at the portfolio level and therefore aggregation must respect which positions may legally offset which. Add-ons are applied, results are compared with collateral held, and calls are produced.
The entire sequence has to fit between the last market close and the first market open, with enough slack to rerun it if something fails. That is the constraint that shapes the architecture.
Four engineering responses, and why intraday changes the problem rather than shrinking it
Parallelism is the first and most obvious. Position revaluation is embarrassingly parallel: scenarios and positions can be distributed across a large compute grid and combined afterwards. This is why prime brokerage risk was an early and heavy user of grid computing, and why it is a natural fit for elastic cloud capacity, since the requirement is enormous for six hours and near zero for eighteen.
In-memory computation removes the database as the bottleneck. Reading two million positions and a billion intermediate results from disk repeatedly is slower than the calculation itself, so modern engines hold the portfolio and market data in memory across the run.
Incremental recalculation avoids redoing unchanged work. If ninety percent of a portfolio has not changed and the market data for it has moved only slightly, recomputing everything from scratch is wasteful. Engines increasingly recompute only what has changed and reuse the rest, which is what makes frequent recalculation affordable.
Approximation is used deliberately and carefully. Full revaluation of every instrument under every scenario is sometimes replaced by sensitivity-based approximation, which is far faster but less accurate, particularly for options where the relationship between price and value is curved. The engineering judgement is where approximation is acceptable and where full revaluation is required, and that judgement is itself subject to model validation.
Intraday margining is not the same problem run more often. The overnight run may take hours; an intraday run must complete in minutes to be actionable, against live prices and positions that are still moving. This is why it depends on the event-streaming architecture described earlier rather than merely on faster hardware: you cannot recompute exposure at noon from a file that arrives at midnight, and you cannot wait for a batch to assemble the portfolio when the point is to act before the market moves again.
Where prices come from
Every margin number rests on a price, and prices are less straightforward than they look.
The overnight run uses official closing prices from each exchange, which are struck by a defined auction procedure rather than being simply the last trade, and which arrive at different times across time zones. A global portfolio therefore has no single moment of valuation, and the engine must define its own consistent snapshot.
Three cases break the simple picture. An illiquid instrument may not have traded at all, so its price is a stale quote, a model value, or a broker mark, each carrying its own uncertainty. A suspended instrument, halted pending an announcement, has a last price that is known to be wrong and a real value that is unknown, which is precisely the situation where margin matters most. And a stale price that quietly fails to update is the most dangerous of the three, because the system reports a number with full confidence and nobody is alerted.
Controls therefore sit around the data as much as around the model: cross-checking multiple vendors, flagging prices that have not moved when comparable instruments have, tolerance checks against the previous day, and explicit escalation for positions carrying model or broker marks rather than traded prices. A margin engine validated to the highest standard, fed a stale price, produces a confident and wrong answer.
The model must be validated, not merely built
Because the margin methodology determines the bank's protection against loss, it falls squarely inside model risk management, the governance framework applied to any model influencing a risk or capital outcome. Four obligations follow, and they explain why changing a margin model is slow.
The model must be documented so that an independent reader can reproduce its logic. It must be independently validated by a team that did not build it, testing its assumptions and challenging its limitations. It must be backtested, comparing what it predicted against what actually happened: if a model sized to cover the worst outcome in ninety-nine days out of a hundred is breached far more often than that, it is wrong and the breaches are counted and reported. And material changes must be approved and often run in parallel with the existing model before replacing it, so their effect on client requirements is understood before clients feel it.
This is why the post-Archegos tightening took the form of add-ons rather than new models. An add-on can be applied to a concentrated exposure within days. Rebuilding a validated portfolio margin methodology takes quarters.
Anatomy of a position, and why deduplication is genuinely hard
The fields of the core record, and what each one drives
The position record is the atom of prime brokerage. Its fields are not administrative; each one determines a downstream behaviour.
| Field | What it holds | What it drives downstream |
|---|---|---|
| Client and account identifier | Which legal entity holds the position | Netting set, margin aggregation, legal agreement applied |
| Instrument identifier (ISIN, FIGI) | Which security | Valuation, borrow eligibility, settlement venue |
| Quantity and direction | Size, long or short | Exposure, borrow requirement, settlement obligation |
| Trade and settlement date | When it was struck and when it settles | Settlement instruction timing, fails monitoring |
| Price and currency | Value and denomination | Valuation, FX exposure, funding requirement |
| Booking entity and venue | Which bank entity holds it | Regulatory regime, client asset protection, capital treatment |
| Collateral and re-use flags | Whether the asset may be re-used | Rehypothecation limits, segregation obligations |
| Arrangement type | Owned, financed, lent, borrowed, pledged or swap-referenced | Corporate action entitlement, manufactured payment, voting rights |
| Tax status of holder | Domicile and treaty position of the holding entity | Withholding on dividends, manufactured payment treatment |
| Agreement reference | Which master agreement governs | Close-out netting scope, margin methodology |
The arrangement type field is the one most often underestimated. Two positions in the same security, of the same size, in the same account, can be economically entitled to completely different things depending on whether the stock is owned outright, lent out, or referenced by a swap. A platform that records only the holding and not the arrangement cannot compute a corporate action correctly, and will discover this on a dividend date rather than in testing.
Two other fields deserve special attention. The booking entity determines which regulatory regime protects the client's assets, which is why sophisticated clients ask about it before they ask about price. The agreement reference determines what nets with what, and an error here means the bank believes it is protected across positions that are legally separate.
Deduplication is a mission-critical problem that sounds trivial
A large fund family has dozens of legal entities with similar names across multiple jurisdictions, and it opens accounts over many years through different channels. If the bank creates two entity records for what is legally one counterparty, exposure fragments: the bank sees two moderate positions rather than one large one, concentration limits are not breached because the aggregation is wrong, and the margin engine may fail to net positions that the documentation permits it to net.
This is not a data hygiene issue; it is a risk aggregation failure with a data cause. Entity resolution combines identifier matching using LEIs, fuzzy name matching, hierarchy mapping to a parent, and manual stewardship for the difficult cases. Banks maintain dedicated data stewardship functions for exactly this, and post-Archegos reviews found aggregation gaps of precisely this kind.
Identifiers, integration, and immutable logs
Three identifiers answer three different questions
| Identifier | What it names | Where it bites |
|---|---|---|
| ISIN | The security, globally, twelve characters | Valuation, settlement, borrow eligibility, reporting |
| LEI | The legal entity that is a party to a transaction | Client hierarchy, netting sets, SFTR and derivative reporting |
| FIGI | The instrument in a vendor-neutral way, at venue level | Mapping across systems and data providers without licence constraints |
The LEI carries particular weight in this business because so much depends on identifying the counterparty precisely. Transaction reporting regimes require it, netting depends on it, and the entity deduplication problem described above is fundamentally an LEI-mapping problem where the LEI alone is insufficient.
A fourth identifier matters for derivatives: the unique transaction identifier, which names an individual transaction so that both sides report the same trade consistently and regulators can pair the reports.
Three integration patterns, and the migration between them changes more than speed
Batch is the legacy default. Files move overnight, systems reconcile in the morning, and the architecture assumes a daily rhythm. It is robust, well understood, and structurally incompatible with intraday risk management, because a batch platform cannot tell you your exposure at eleven in the morning.
APIs make data available on demand. A risk system can request a current position rather than waiting for tonight's file. This is the modernisation most banks are in the middle of, and it is what makes client data feeds and self-service reporting possible.
Event streaming pushes changes as they happen. A trade booking publishes an event, and every interested system, margin, collateral, reporting, limits, consumes it immediately. Kafka-style platforms are the common implementation.
Why streaming is not just faster batch
The temptation is to read the batch-to-streaming migration as a speed upgrade. It is more than that: it removes the need for reconciliation.
The difference is that between emailing photocopies of a document to five colleagues and having all five work in one shared document. Under batch, each system holds its own copy of the position, built from a file. Copies diverge, so reconciliation exists to find and fix the divergence, and an entire operational function is built around that repair work. Under event streaming, systems consume the same ordered event log, so they converge on the same state by construction. Reconciliation does not become faster; it becomes largely unnecessary between systems that share the log.
That is why this migration justifies its cost. It does not merely accelerate a process; it eliminates a category of work and the errors that category was created to catch. It is also why intraday margining became feasible: you cannot recompute exposure at noon from a file that arrives at midnight.
Compliance-critical records must be append-only
Margin calls, collateral movements, wall-crossing authorisations, limit approvals and client instructions must be recorded so that they cannot be altered after the fact. Append-only storage with full audit trails means a record is added rather than overwritten, and the history remains reconstructible.
The reason is evidential. When a regulator asks why a margin call was not made on a particular date, or a defaulted client's administrator disputes the bank's close-out, the bank must reconstruct exactly what it knew, when it knew it, and what it decided. A system that overwrites state cannot answer those questions, and an unanswerable question is treated as an admission.
Three messaging worlds carry a prime brokerage relationship
Speed falls and legal finality rises as a trade moves right
FIX carries the front office. The Financial Information eXchange protocol is a lightweight tagged message standard for orders, executions, allocations and give-ups. It is fast, measured in milliseconds, and it is how trades reach the prime from executing brokers. It carries no legal finality; a FIX message is an instruction, not a transfer of title.
SWIFT carries post-trade settlement. The securities message types in the MT5xx series handle settlement instructions and confirmations between the prime, custodians and depositories. These messages are slower, measured in minutes, and they are legally weighty: they instruct the movement of assets. The industry is migrating this traffic to ISO 20022, a richer XML-based standard with structured data that supports far more granular information than the older format.
Financing and collateral run on their own rails. Triparty collateral platforms operated by major custodians allocate and substitute collateral automatically against eligibility rules. Securities lending platforms such as EquiLend and Pirum handle trade matching, billing, mark-to-market and reconciliation across the lending market. These operate intraday.
| Layer | Standard | Carries | Timescale |
|---|---|---|---|
| Front office | FIX | Orders, executions, allocations, give-ups | Milliseconds |
| Post-trade | SWIFT MT5xx, migrating to ISO 20022 | Settlement instructions and confirmations | Minutes to T+1 |
| Financing | Triparty, EquiLend, Pirum | Collateral allocation, stock loan lifecycle | Intraday |
Resilience: the margin run cannot fail, because there is no second chance
Some moments have no recovery window
Most systems can fail and be repaired. A few cannot, because the deadline is external and immovable. In prime brokerage there are three: the overnight margin run, which must complete before markets open; settlement cut-offs, which are set by depositories and do not move; and the liquidation of a defaulted client, which happens once, into a falling market, with no opportunity to retry.
Resilience engineering exists for exactly these moments, and four terms are used precisely rather than interchangeably.
| Concept | What it guarantees | Why it matters on the critical day |
|---|---|---|
| Redundancy | A duplicate component exists and can carry the load | A failed server does not stop the margin run |
| Automatic failover | The switch to the duplicate happens without human action | Nobody has to be awake and correct at 3am |
| Disaster recovery | Processing can resume at another site within a defined time | A site-level failure does not miss the market open |
| Business continuity | People and processes continue, not just systems | Calls still get made when the office is inaccessible |
Redundancy alone is insufficient, and this is the point most often missed. A duplicate that requires a human to notice the failure and trigger the switch is only as reliable as the person on call at three in the morning. Automatic failover is what converts redundancy into availability. And systems continuity without people continuity is useless, because a margin call is ultimately a person making a decision and a client being contacted.
Recovery objectives make this measurable: recovery time objective is how long the business can be down, and recovery point objective is how much data may be lost. For the margin run, both are close to zero, and firms rehearse failover rather than merely documenting it. Regulatory frameworks now require this: the European Union's Digital Operational Resilience Act and the United Kingdom's operational resilience regime both mandate identifying critical services, setting impact tolerances, and testing against severe but plausible scenarios.
Five regulatory regimes shape every system in the chain
Read each one as a system specification rather than as law
Client asset protection rules dictate segregation and re-use. United States Rule 15c3-3 requires broker-dealers to segregate fully paid customer securities and limits rehypothecation relative to customer indebtedness, with a reserve computation performed on a defined cycle. United Kingdom CASS rules impose parallel segregation, record-keeping and reconciliation obligations. As a specification: the platform must know, per position, whether an asset is fully paid, segregated or eligible for re-use, and must perform and evidence a periodic computation.
Margin rules for uncleared derivatives govern the synthetic book. The uncleared margin rules require initial and variation margin to be exchanged on non-centrally-cleared derivatives, with initial margin held in segregated accounts and computed by an approved model. As a specification: the platform must compute margin by an approved methodology, segregate it, and support dispute resolution against the counterparty's own calculation.
Securities financing transaction reporting creates a reporting obligation on the financing book itself. The European SFTR requires both parties to report repo, stock loan and margin lending transactions with extensive field-level detail, including collateral and re-use. As a specification: every financing transaction must carry a UTI, be reportable within a day, be reconcilable with the counterparty's report, and re-use must be tracked at position level.
Capital and leverage rules set the price of the balance sheet. The Basel framework, including the standardised approach for counterparty credit risk and the leverage ratio, determines the capital a prime must hold against client exposures. As a specification: the platform must attribute capital consumption to individual clients and transactions, or the business cannot manage return on balance sheet.
Short selling and market conduct rules govern the short book. Regulation SHO in the United States imposes the locate requirement and close-out obligations; the European Short Selling Regulation adds position disclosure thresholds. As a specification: the platform must evidence a locate before a short sale, monitor aggregate short positions against disclosure thresholds, and produce records on demand.
Which functional block is the single point of truth, and what follows from that?
Books and records. The margin engine, collateral system, settlement platform and client statements all derive from it and none of them independently validate it, so its errors propagate everywhere.
Why do banks build the margin engine but buy collateral management?
Because the margin methodology determines what the bank can offer and what risk it keeps, and regulators expect the bank to own and defend the model. Collateral management is largely standardised plumbing where a vendor is cheaper and adequate.
Explain why entity deduplication is a risk problem, not a data problem.
Two records for one legal counterparty fragment exposure, so concentration limits are measured against the wrong aggregate and netting permitted by the documentation may not be applied. The bank believes it has two moderate exposures when it has one large one.
What is the difference between FIX and SWIFT in this business, beyond speed?
FIX carries provisional front-office instructions in milliseconds and transfers no title. SWIFT MT5xx and ISO 20022 carry post-trade settlement messages that instruct actual movement of assets and carry legal finality.
Why does event streaming remove reconciliation rather than merely accelerate it?
Under batch, each system builds its own copy of the position from files and the copies diverge, so reconciliation exists to repair divergence. Under streaming, systems consume one ordered event log and converge on the same state by construction.
Why is redundancy alone insufficient for the overnight margin run?
Because a duplicate that requires a human to detect the failure and trigger the switch is only as reliable as the person on call. Automatic failover converts redundancy into availability, and business continuity is needed because a margin call ultimately requires people.
Read SFTR as a system specification. What must the platform do?
Assign a UTI to every securities financing transaction, report both sides within the required window with full collateral detail, track re-use at position level, and reconcile its report against the counterparty's.
Why is intraday margining an architectural change rather than a scheduling change?
An overnight run may take hours, but an intraday run must complete in minutes against live prices and moving positions. That requires event streaming and incremental recalculation, because exposure cannot be recomputed at noon from a file that arrives at midnight.
Which price problem is the most dangerous, and why?
A stale price that quietly fails to update, because the system reports it with full confidence and raises no alert, whereas an illiquid or suspended instrument is at least visibly uncertain. A perfectly validated engine fed a stale price produces a confident wrong answer.
Why did the industry respond to Archegos with add-ons rather than better models?
Because a margin model falls under model risk management and must be documented, independently validated, backtested and usually parallel-run before replacement, which takes quarters. An add-on can be applied to a concentrated exposure within days.
Why can two identical-looking positions be entitled to different things on a dividend date?
Because entitlement depends on the arrangement, not the holding. Stock owned outright receives the real dividend, stock lent out receives a manufactured payment with different tax treatment, and a swap-referenced position receives a contractual cash flow. A platform recording only the holding cannot compute this.
Why does listed derivatives clearing sit on a separate platform and entity?
Because client money in futures clearing is governed by a stricter segregation regime that forbids commingling with securities financing balances, and because margin is set by the clearinghouse rather than the bank's model, removing the component that differentiates the prime brokerage stack.
Why do sophisticated clients ask about booking entity before they ask about price?
Because the booking entity determines which regulatory regime governs client asset protection and re-use, and therefore what happens to their assets if the prime fails. Lehman made that question consequential.
4
Which block is the single point of truth?
No.
No.
Right. Margin, collateral, settlement and statements derive from it.
No.
Why do banks build the margin engine but buy collateral management?
No.
Right.
No.
Not the reason.
Why is entity deduplication a risk problem?
Minor issue.
No.
No.
Right.
Beyond speed, how does FIX differ from SWIFT here?
No.
No.
No.
Right.
5Risks, Pain Points, and Transformation
Prime brokerage has produced three of the most instructive failures in modern finance. Each one killed or maimed an institution, and each one maps precisely to a stage of the lifecycle in Part 2.
A MECE map of where prime brokerage risk lives
Six buckets, and each has an owner and a named defence
| Risk type | What it is | Who owns it | Main defence |
|---|---|---|---|
| Counterparty and credit | The client cannot meet a call and defaults | Counterparty credit risk | Margin, concentration limits, close-out netting |
| Market and liquidity | Positions gap or cannot be liquidated at modelled prices | Risk and the financing desk | Haircuts, liquidity add-ons, days-to-liquidate limits |
| Operational and settlement | Mis-booking, failed settlement, reconciliation breaks, missed corporate action elections | Operations and asset servicing | Straight-through processing, daily reconciliation, event deadline controls |
| Collateral and re-use | Collateral is insufficient, ineligible, or unavailable when needed | Collateral and treasury | Haircuts, eligibility rules, re-use limits, triparty |
| Conduct and conflict | Client position information is misused, barriers fail | Compliance | Entitlements, surveillance, wall-crossing logs |
| Regulatory and capital | Capital consumed exceeds return; reporting fails; client money segregation breached | Finance and compliance | Capital attribution, limits, reporting controls, segregation computations |
The taxonomy is complete in the sense that matters: every loss this business has suffered fits one of these six, and most large losses involve two or three interacting. Archegos was counterparty risk detonated by concentration, amplified by market and liquidity risk during the unwind, with a data aggregation failure underneath it.
Two of these deserve a note that the table cannot carry. Corporate actions sit inside the operational bucket but behave unusually: the loss from a missed elective deadline is immediate, quantifiable and almost never recoverable, because the deadline belonged to somebody else and has passed. And client money segregation sits inside the regulatory bucket but is existential rather than financial, since a segregation breach in a clearing business is the kind of failure that ends firms rather than costing them money.
How to tell a good client relationship from a lucky one
Five indicators, and the first one is the one that misleads
| Indicator | What it measures | Why it matters |
|---|---|---|
| Revenue | Fees, financing spread and borrow income earned | The headline number, and the least informative alone |
| Return on balance sheet | Revenue per unit of leverage exposure and RWA consumed | Whether the relationship is worth the capital it eats |
| Peak exposure coverage | Whether collateral covered exposure in the worst observed moves | Whether the margin model held under real stress |
| Days to liquidate | How long the book would take to exit at modelled prices | The assumption every margin number depends on |
| Operational quality | Straight-through processing rate, breaks, settlement fails, disputes | Predicts both client retention and loss events |
Revenue is the first-day-pop fallacy of prime brokerage. A relationship generating outstanding revenue while consuming enormous balance sheet at inadequate margin is not a good relationship; it is an unrecognised risk position that happens to be paying well for now. Archegos generated substantial revenue for several banks right up to the point where it destroyed considerably more than it had ever produced.
Three real failures cover the three ways this business dies
Archegos, 2021 — the counterparty and concentration failure
Archegos Capital Management was a family office running highly concentrated, heavily leveraged positions in a small number of equities, held largely through total return swaps with several prime brokers simultaneously. Because the exposure was synthetic, Archegos did not appear in public ownership disclosures, and because it used multiple primes, no single bank could see the total position.
When the underlying stocks fell in March 2021, Archegos could not meet margin calls. The banks moved to liquidate. Those that sold quickly limited the damage; those that moved more slowly absorbed far larger losses. Industry losses totalled roughly ten billion dollars, with Credit Suisse alone losing in the region of five billion, a blow that contributed materially to its eventual collapse, and Nomura and others taking substantial hits.
The failure was not the default; it was everything before it. Margin was too low for the concentration, the risk was not aggregated properly across products, concentration limits were not enforced against a client generating attractive revenue, and the risk function's warnings did not override the commercial relationship. It maps to Steps 2 and 3 of the lifecycle: the credit gate and the margin methodology.
Lehman Brothers, 2008 — the client asset and re-use failure
When Lehman Brothers failed, its hedge fund clients discovered that the distinction between assets held and assets protected was not academic. Assets that had been rehypothecated through Lehman's London entity, where re-use limits were far more permissive than in the United States, were no longer identifiable client property. Those clients became unsecured creditors in an administration that took years to resolve, and many funds were unable to access assets they had believed were theirs, some of them failing as a result.
The failure maps to Step 2, the documentation, and to the re-use mechanics in Step 7. The lesson reshaped the industry: multi-prime became standard, chief operating officers began asking about booking entity and re-use limits before pricing, and regulators tightened client asset rules substantially.
Long-Term Capital Management, 1998 — the leverage and systemic failure
LTCM was a fund of exceptional pedigree running convergence trades at extreme leverage, financed by essentially every major bank on terms that reflected the founders' reputations rather than the risk. When Russia defaulted and correlations that the models treated as stable broke down simultaneously, the positions moved against the fund catastrophically, and its leverage meant there was no buffer.
The problem was systemic: LTCM's positions were so large relative to the markets it traded that liquidating them would have moved prices against every bank holding similar exposure. The Federal Reserve Bank of New York convened a private-sector recapitalisation because the disorderly alternative was judged too dangerous.
It maps to the margin methodology and the liquidity assumption. Banks lent on reputational terms, models assumed correlations that failed exactly when needed, and days-to-liquidate assumptions ignored the fund's own size relative to the market.
A fourth case deserves a mention without displacing these three, because it belongs to the clearing side rather than to financing. When MF Global failed in 2011, a shortfall appeared in segregated customer funds in its futures clearing business, and customers waited a long time for money that rules said should never have been at risk. It is the clearing-business analogue of the Lehman lesson: segregation is a control that has to be operationally real, not merely documented, and it is the reason those balances sit in a separate entity on a separate platform.
| Case | Failure type | Maps to lifecycle stage | Lesson |
|---|---|---|---|
| Archegos, 2021 | Counterparty concentration and data aggregation | Steps 2 and 3, credit gate and margin | Concentration must be aggregated across products and enforced against revenue |
| Lehman, 2008 | Client asset protection and re-use | Step 2 documentation, Step 7 re-use | Held is not protected; entity and re-use terms decide the outcome |
| LTCM, 1998 | Leverage and liquidity assumptions | Step 3, margin methodology | Correlations break in stress, and size is itself a liquidity constraint |
Five technology shifts genuinely change the business
Each one attaches to a specific stage
| Trend | Attaches to | Why it matters |
|---|---|---|
| Real-time and intraday margining | Step 7, margin and collateral | Closes the blind window between closes that Archegos exploited |
| AI-driven counterparty and concentration analytics | Steps 2 and 7, credit and monitoring | Detects position overlap and deteriorating profiles earlier than periodic review |
| Collateral optimisation automation | Step 7, collateral | Converts a manual choice into basis points, at scale, continuously |
| Cloud and API-first architecture | Whole platform | Supplies elastic compute for stress runs and enables client data feeds |
| Straight-through processing in securities lending | Steps 5 and 6, borrow and settlement | Mandatory under compressed settlement cycles; removes manual repair time |
A sixth deserves a mention because it follows directly from the settlement compression already under way: automation of corporate actions and manufactured payments, where vendor platforms now match and bill lending-related payments across the market that were historically reconciled by hand. It attaches to the asset servicing function and it moves cost and operational risk together.
Real-time margining is the trend with the clearest economic justification. It directly addresses the mechanism by which the industry lost ten billion dollars, and it is expensive precisely because it requires live positions, live prices and a risk engine that runs in minutes. This is why the batch-to-streaming migration in Part 4 is a risk programme wearing an infrastructure costume.
Collateral optimisation is the trend with the clearest revenue justification. Choosing which asset to post, across thousands of positions and multiple agreements with different eligibility rules, is a constrained optimisation that humans perform badly and software performs well. The gains are measured in basis points on very large balances, which is a real number.
Apply one filter to every claimed transformation in this business: does it change economics or risk, or does it change the appearance of things?
A client portal redesign improves satisfaction and may aid retention, but it moves neither cost nor risk. Intraday margining changes risk materially and measurably. Collateral optimisation changes cost measurably. Straight-through processing in securities lending changes both, by removing manual effort and by eliminating the fails that compressed settlement cycles now penalise in cash.
The filter is useful in the other direction too. When a programme cannot articulate which of cost or risk it moves, and by roughly how much, it is usually a modernisation project in search of a justification. That is not always wrong, since technical debt is real, but it should be argued as such rather than dressed as transformation.
Generative AI arrives, and governance arrives with it
Three uses where the gain is real
Client service and query resolution is the clearest. A prime's client service team answers a high volume of repetitive questions about balances, charges, borrow rates and breaks. A language model with access to the client's own data can draft accurate answers, summarise a statement, or explain a financing charge in plain language, with a human reviewing before it reaches the client. The gain is capacity, and capacity in client service is directly linked to retention.
Onboarding and document review is the second. KYC involves reading large volumes of unstructured material: fund documents, ownership structures, offering memoranda. Language models extract entity relationships, flag inconsistencies and pre-populate files, compressing a process that routinely takes six to twelve weeks. Given that onboarding delay is a genuine commercial disadvantage, this is worth real money.
Legal agreement analysis is the third and most specific to this business. A prime holds thousands of agreements whose terms, margin thresholds, eligible collateral, re-use limits, netting scope, default triggers, must be represented accurately in the systems. Extracting those terms and comparing them against system parameters is exactly the reconciliation that Part 2 identified as the dangerous gap, and it is work that language models do well and lawyers find tedious.
Three risks, and the first is unique to this business
| Use of GenAI | The gain | The governing risk |
|---|---|---|
| Client service and query resolution | Capacity, faster response, better retention | Client position data leaking into models or across clients |
| Onboarding and KYC document review | Weeks removed from a commercial bottleneck | Hallucinated entity relationships entering the client hierarchy |
| Legal agreement extraction | Closes the gap between documentation and system parameters | Misread terms silently mis-parameterising margin or netting |
Client position confidentiality is the risk that makes prime brokerage different. The data most useful to a model is precisely the data most tightly restricted: what clients hold. A model trained or prompted on one client's positions must not surface them to another, and the information barriers of Part 3 must extend into model architecture, retrieval scope and prompt logging. This is not a policy question that can be answered with a memo; it is a design constraint.
Hallucination is dangerous here because the outputs enter systems rather than documents. An invented entity relationship that enters the client hierarchy corrupts exposure aggregation, which is the exact failure mode of Archegos. A misread netting clause changes what the margin engine believes it may offset.
Explainability and model risk governance apply because regulators treat these as models. Any model influencing a risk or capital outcome falls under model risk management: documented, validated independently, monitored for drift, with a named owner. A language model that helps set a margin parameter is inside that perimeter, and firms that deploy without recognising this are creating a supervisory problem alongside a risk one.
Why is the six-bucket taxonomy a map of controllability?
Operational and conduct risk are genuinely reducible through better systems and surveillance; market and liquidity risk can only be buffered through haircuts and add-ons; counterparty risk sits between, since the bank cannot control client failure but fully controls how much it lends and against what collateral.
What is the prime brokerage equivalent of the first-day-pop fallacy?
Judging a relationship by revenue. A client can generate excellent revenue while consuming enormous balance sheet at inadequate margin, which is an unrecognised risk position that pays well until it does not, as Archegos did.
Why was Archegos invisible to each individual prime?
Its exposure was synthetic, held through total return swaps, so it did not appear in public ownership disclosures, and it used several primes simultaneously so no single bank saw the total concentrated position.
What did Lehman teach about the difference between held and protected?
Assets that had been rehypothecated, particularly through the more permissive London entity, were no longer identifiable client property, so clients became unsecured creditors in a multi-year administration rather than owners of segregated assets.
Apply the transformation filter to a client portal redesign and to intraday margining.
The portal improves satisfaction and may aid retention but moves neither cost nor risk materially. Intraday margining changes risk directly and measurably by closing the blind window between closes, which is why it justifies its considerable expense.
Which GenAI risk is specific to prime brokerage rather than generic, and why?
Client position confidentiality. The data most useful to a model is exactly the data most restricted, so the information barriers between clients must be built into retrieval scope, model architecture and prompt logging rather than asserted in policy.
Why does a missed corporate action election behave differently from other operational losses?
Because the deadline was set externally and has passed, so the loss is immediate, quantifiable and almost never recoverable, unlike a settlement break or a reconciliation difference which can usually be repaired.
What does MF Global add to the Lehman lesson?
That segregation must be operationally real rather than merely documented, and that the same principle applies on the clearing side, where a shortfall in segregated customer funds is existential rather than merely expensive.
Why does hallucination matter more here than in a drafting context?
Because outputs enter systems rather than documents. An invented entity relationship corrupts exposure aggregation, reproducing the Archegos failure mode, and a misread netting clause changes what the margin engine believes it may offset.
5
Why was Archegos invisible to each individual prime?
Not the mechanism.
Right. No one saw the total.
It was highly leveraged.
No.
What did Lehman teach about held versus protected?
Right.
The opposite.
The London entity was more permissive.
Not the lesson.
What is the prime brokerage version of the first-day-pop fallacy?
Right. It may hide balance sheet at inadequate margin.
Not the parallel.
No.
No.
Which risks can only be buffered, not removed?
Reducible through surveillance.
Right.
Reducible through systems.
Market and liquidity risk are buffered only.
Module Summary — The Six Ideas That Matter Most
If everything else fades, these six should remain.
The client defines the platform, not the size of the balance
A prime broker's architecture is determined by what its clients trade, not by how much they trade. A book of long/short equity funds needs stock borrow, locate management and equity margining. Add multi-strategy clients and the platform needs cross-product aggregation, derivative margin and netting across agreements. Add synthetic financing and it needs swap booking, hedge management and derivative capital attribution. Widen the lens to Prime Services and clearing and outsourcing add their own platforms again. Complexity comes from the product mix, and the platform must be built for the most complex client the bank intends to serve, not the average one.
Data cascades, so errors are never contained
The position record is the atom, and everything downstream inherits it without checking. An instrument identifier that is wrong at booking is valued wrongly, margined wrongly, settled wrongly and reported wrongly, and it surfaces at the point of maximum consequence. Entity deduplication is the sharpest case: two records for one counterparty fragment exposure so that concentration limits are measured against the wrong aggregate. When a bank discovers after a loss that it did not know its true exposure, the cause is almost never inattention. It is aggregation.
The daily margin run is where the risk is actually priced
Everything upstream exists to make the margin run possible and everything downstream reports its results. The requirement is the greater of a regulatory floor, a risk-based portfolio model, and house add-ons for concentration, liquidity and gap risk, and the negotiation over those add-ons is the commercial heart of the business. The client's strategy determines how much that model can be trusted: a diversified systematic book behaves as the model expects, and a concentrated position in an illiquid name does not. Margin is where client type, risk model and commercial pressure meet, every single night.
Client asset protection is core architecture, not an afterthought
Whether an asset is segregated or re-used determines what happens to a client if the prime fails, and that outcome is decided by the booking entity, the regulatory regime and the documentation, all of which must be represented correctly in the systems. Rehypothecation funds the business and endangers the client simultaneously, which is why re-use limits, eligibility rules and segregation computations are platform functions rather than compliance paperwork. Corporate actions belong here too: because lending and financing separate legal ownership from economic entitlement, the platform must know not just who holds a security but under what arrangement, and route dividends, manufactured payments, votes and elections accordingly. The tax consequences of that separation, from cum-ex to section 871(m), are the clearest proof that ownership is an architectural variable rather than a legal footnote. The three messaging worlds and the resilience standards around the margin run belong to the same category: compliance-driven requirements that shape the architecture from the beginning.
Balance sheet scarcity quietly distorts behaviour
Return per unit of balance sheet consumed is the metric that governs internal decisions, and it distorts behaviour exactly as league tables distort issuance businesses. It pushes coverage teams to steer clients toward capital-efficient products, justifies repricing long-standing relationships, and occasionally ends profitable ones. It also creates the pressure that competes with risk discipline, because a client generating strong returns on balance sheet is a client the business fights to keep, including when risk wants tighter margin. Naming this tension is the beginning of managing it.
Real transformation moves cost or risk
Intraday margining changes risk measurably and directly addresses the mechanism that cost the industry ten billion dollars. Collateral optimisation changes cost measurably across very large balances. Straight-through processing in securities lending changes both, and compressed settlement cycles have made it mandatory rather than optional. Generative AI belongs on this list where it meets the same test, in client service, onboarding document review and legal agreement extraction, governed as a model with a human reviewer, because its outputs enter risk systems rather than documents.
Process Cheat Sheet — The Whole Relationship on One Page
A single reference for the entire prime brokerage lifecycle.
| Stage | Timing | Core activity | Key output | Lead and systems |
|---|---|---|---|---|
| 1. Origination | Ongoing | Pitch on terms, service, balance sheet commitment | Term sheet | Sales; CRM, pricing analytics |
| 2. Onboarding | 6–12 wks | KYC, credit assessment, legal documentation | Signed PBA, ISDA/CSA, GMSLA; approved limits | Onboarding, credit, legal; CLM, screening, limits |
| 3. Financing set-up | 1–2 wks | Fix margin methodology and account structure | Live accounts, margin parameters | Risk; margin engine |
| 4. Execution and give-up | Daily | Trade anywhere, give up to the prime | Trade records in the sub-ledger | Ops; FIX, books and records |
| 5. Locate and borrow | Daily | Source stock for short sales, manage recalls | Confirmed locates, borrow contracts | Sec lending desk; inventory, EquiLend, Pirum |
| 6. Clearing and settlement | T+1 | Move securities against cash | Settled positions, custody records | Ops; settlement platform, SWIFT, CSD |
| 7. Margin and collateral | Daily, plus intraday | Revalue, stress, call collateral | Margin requirement, collateral movements | Risk, collateral; margin engine, triparty |
| 8. Reporting and service | Daily / monthly | Statements, queries, capital introduction | Client reporting, capital raised | Client service; portal, data feeds |
| Throughout: corporate actions | Event-driven | Dividends, votes, elective events on financed and lent stock | Manufactured payments, elections | Ops; corporate actions engine |
Key numbers to remember
| Metric | Rule of thumb |
|---|---|
| Onboarding duration | 6 to 12 weeks, with KYC the long pole |
| Reg T initial margin | 50 percent of long market value, a regulatory floor rather than a risk measure |
| Portfolio margin on a hedged equity book | Often in the mid-teens as a percentage of gross exposure |
| Binding requirement | The greater of the regulatory floor and the house risk-based number plus add-ons |
| General collateral borrow fee | A few basis points a year |
| Specials borrow fee | From low single-digit percent to extreme levels in a squeeze |
| Settlement cycle | T+1 in North America since 2024, with other markets following |
| Margin call deadline | Typically next business morning, commonly around 11am |
| Typical long/short equity gross exposure | Roughly 200 to 400 percent of capital |
| Number of primes used by an institutional fund | Two or more since 2008; the largest funds use several |
| Archegos industry loss, 2021 | Approximately 10 billion dollars |
| Margin from first principles | Roughly confidence factor x daily volatility x square root of liquidation days x position value |
| Short rebate | Short-term rate minus borrow fee; negative on specials, so the position costs carry |
| Manufactured dividend | The short seller pays the dividend; a high-yielding stock is expensive to stay short in |
| Basis point | One hundredth of one percent; 5bp on 10 billion is 5 million a year |
| Margin run recovery tolerance | Minutes, because the market open is an immovable deadline |
Figures in this table are market rules of thumb and regulatory reference points, not precise constants; the Meridian Capital figures used throughout the module are illustrative only.
The one-sentence version
A prime broker lends a hedge fund cash for its longs and stock for its shorts, holds and settles everything that results, revalues the whole book every night to decide how much collateral it must hold against the possibility that the fund cannot pay, and makes its money on the spread between what that leverage costs and what the balance sheet behind it is worth.