How a CFD broker manages its risk

Every client position has a mirror 

A CFD is a contract between two named parties, so the moment a client's position opens, the firm holds its exact opposite: the same instrument, the same size, the other direction. That is not a figure of speech or an accounting convention. It is what a bilateral contract is. Where the client's side records a credit of a given amount, the firm's side records a debit of the same amount, because the two are one piece of arithmetic read from opposite ends.

Nothing about this is unusual or concealed, and it follows directly from the structure covered in foundations. The contract is written with the firm and closed with the firm, and no exchange or clearing house stands in between to take the far end of it. What differs between firms, and between instruments and clients within a single firm, is only what happens next.

A firm that did nothing at all would finish each day holding the sum of every position its clients had opened, in whichever direction they happened to lean. Exposure of that kind is not elected. It arrives continuously as a by-product of writing contracts, so managing it is a permanent operational function rather than a decision taken once.

The book nets before anything leaves it 

The first thing that happens to that exposure is that most of it disappears against itself. Clients do not all lean the same way, and seen from the firm's side the two groups already offset: its mirror of one client's long is a short, its mirror of another's short is a long, and the two cancel to the extent that they overlap. What survives the cancellation is the net exposure, and it is smaller, frequently very much smaller, than the total size of the contracts written.

The residual is the only part of the book with any market risk in it. The offsetting portion carries none, because whatever the price does, the two sides move by the same amount in opposite directions and the firm's result on them is zero either way. Matching client orders against each other in this manner is called internalisation, and it is the single largest thing that happens to retail order flow, at brokers and at banks alike.

Why internalisation happens 

The reason is cost, and the arithmetic is not subtle. Passing every client order outward means dealing on a liquidity provider's own bid and offer and paying whatever commission or brokerage the relationship carries, on every order, in both directions. Two clients dealing opposite ways in the same instrument at the same moment would generate two such round trips, at two costs, in order to arrive at a combined position of nothing. Matching them against each other arrives at the same combined position without either cost.

A second reason is capacity. A wholesale counterpart deals in wholesale sizes, and retail orders are routinely smaller than the minimum a provider will quote for, so an order sent out on its own is not merely expensive but sometimes not sendable at all. Aggregating flow into the sizes the wholesale market recognises is part of what standing between a client and that market consists of.

Internalisation is therefore a structural feature of the market rather than a deviation from it, and the first lesson in this module noted that large banks do the same thing on a far greater scale. The objection to it has never been that it happens. The objection is that the price at which it happens is set by the firm doing it, which is why execution quality, the subject of a later lesson here, is the measurable question rather than the booking arrangement.

The three things that can be done with the residual 

That leaves the net exposure. There are three broadly recognised arrangements for dealing with it, they are not mutually exclusive, and most firms run more than one at once, in different proportions for different instruments and different groups of clients.

1. Pass it out order by order. Each client order is met by an equal and opposite deal with a liquidity provider at the moment it is filled, so the firm's own book is flat by construction and holds no market exposure at all. This is the most operationally expensive of the three, and the one least sensitive to how any client's position turns out.
2. Net it, then cover the residual. The book is allowed to offset internally and only the leftover is dealt externally, either continuously or once it passes a stated size. The firm holds market exposure in the interval between the residual forming and the residual being covered, and that interval is where its own risk lives.
3. Retain it inside limits. The residual is held rather than covered, and the firm's result on it is the mirror of its clients' aggregate result on the same positions. Firms operating this way generally do so within stated exposure limits per instrument, with hedging triggered automatically when a limit is breached.

The third arrangement produces most of the argument, so it is worth being exact about what it does and does not mean. It does not mean a firm is dealing against a named client, or that anybody is watching one account: exposure is measured on the aggregate of an instrument, so what is held is a net position arrived at by arithmetic across everyone dealing in it at that moment. Nor does it mean the position is costless in either direction. A retained residual has the same properties as any other position, and it moves against the firm as readily as it moves for it.

Reading the last two rows against the first six is the whole of the mechanism. Covering converts an uncertain result into a known cost. Retaining converts a known cost into an uncertain result. Neither is a clever arrangement and neither is a scandalous one. They are two answers to a single question, which is who carries the price risk on the residual, and firms answer it differently, revise the answer by instrument, and revise it again as market conditions change.

The names the industry uses, and what they hide 

Anyone reading around this subject meets two pieces of jargon immediately, A-book and B-book, used to mean roughly passing exposure out and roughly retaining it. They are trade shorthand rather than regulatory categories, they carry no agreed definition, and a firm described as one or the other is almost always doing both. Their real weakness is that they name where a trade is booked, an internal matter, rather than the two things a client is affected by: the price the order was filled at, and whether the firm's own interest bore on it.

The neighbouring labels are looser still. Straight through processing describes an order reaching a counterparty without manual handling, which is a statement about automation rather than about whether exposure was retained. No dealing desk states that no person intervenes, a distinction that mattered when dealing desks were manual. Neither term says what became of the net position of the book, and both are commonly used as though they did.

Where the revenue comes from in each case 

Following the money keeps these arrangements straight far better than following the jargon. Where exposure is passed out, the firm's revenue is the difference between the price at which it deals with its provider and the price at which it deals with its client, plus any commission it charges, and it earns that whether a client's position ends in a credit or a debit. Where exposure is retained, revenue is that same spread and commission plus the result of the net residual, and the result of the net residual is the mirror of the aggregate client result on those positions.

The second arrangement contains a conflict of interest, and no rephrasing removes it. A firm holding the other side of an exposure records a credit on that position in the circumstances where the aggregate of its clients records a debit. Regulators do not treat that conflict as prohibited, and they do treat it as one that has to be identified, managed and disclosed. The obligations that apply are those governing conflicts of interest, order execution, and the treatment of client money, and the last two lessons of this module deal with the supervisory and client money side of them directly.

What hedging does not change for the client 

One point deserves stating flatly, because the two arrangements are often presented as though one of them made a client's contract safer than the other. A hedge is a second contract, between the firm and its provider. The client is not a party to it, holds no claim on it and cannot enforce it. The client's contract remains with the firm, and the firm remains the party that owes the difference whenever that contract settles in the client's favour, exactly as before.

What hedging changes is the firm's own risk rather than the client's counterparty. A firm that covers its exposure is less likely to be damaged by a single large market move, and that is not nothing, since a counterparty's financial condition is part of what any CFD holder carries. The protections that bear on that condition, though, are structural rather than contractual: capital requirements imposed by a supervisor, and the requirement that client money be held apart from the firm's own. Both are examined in the two lessons that close this module.

What is published, and what this page does not state 

This is where the subject stops being theory. In every jurisdiction that supervises this business the answers sit in published documents, and four of them carry the substance.

• The conflicts of interest policy, which sets out which conflicts the firm has identified and what it does about each of them.
• The order execution policy, which sets out where orders go, on what basis a venue or a provider is selected, and what the firm weighs when it judges whether a fill was a good one.
• The terms of business, which state who the counterparty to the contract is, and the circumstances in which the firm may decline an order, requote it, or close a position.
• The regulatory disclosures, which state the legal entity, the licence it holds and the authority that supervises it.

None of them will state a proportion of flow retained, and such a figure would be worth little if one did, since it changes by the minute. What they state is the process, which is both the stable part and the part a supervisor examines.

This page does not state which arrangement YAL runs. A firm's account of its own model is a regulatory disclosure rather than a piece of teaching copy: it belongs in the documents above, in wording its compliance function has approved, and it renders here only once that wording exists. [YAL execution and conflicts of interest disclosure. TBD]

Where practitioners disagree 

The first argument is whether a retained residual is a conflict management can contain or one only separation can remove. One position holds that a firm quoting a price to a client whose aggregate debit it stands to receive can never be wholly indifferent to that price, and that the structural answer is an arrangement in which it cannot hold the other side at all. The opposing position holds that passing everything outward is not free of conflict either, since the firm selects the providers, negotiates its own spread against theirs and earns the difference, and that supervised disclosure with measurable execution quality is what constrains behaviour in both cases. Neither has won.

The second concerns internalisation and price formation, and the first lesson in this module introduced it. Matching two clients against each other is cheaper for both than sending each of them out, and every such match is a transaction that never reaches an external venue or contributes to a public quotation. Whether that reads as efficiency or as erosion depends on what share of total flow it accounts for, and that share cannot be measured from outside the firms holding the data. The disagreement is about a quantity nobody can observe, which is why it persists rather than resolving.