How a Derivatives Desk Makes Money

Ask a student how a derivatives desk makes money and you will usually get some version of this:

I think this name is overvalued, so I would short it.

The interview is effectively over at that point. Not because the view is wrong, but because it answers a different question. The candidate has described how a fund makes money, in a room where nobody is running a fund.

A market maker does not get paid for being right about direction. Prop desks inside banks were dismantled after the Volcker rule and its European equivalents, and what is left is a business with a very specific shape: you quote a price, someone trades against it, and you are now holding a risk you did not choose. Everything that happens next is the job.

The bank also borrows money every single day to carry that risk. That is the structural difference between a trading desk and an asset manager. The asset manager already holds its clients' money. You do not. So a position has to clear three hurdles rather than one: it has to be right, it has to cover the financing, and it has to cover the regulatory cost of the balance sheet it consumes.

This article is the long answer to the question. It is the material I teach in the Applied Derivatives course, assembled in one place, and it is the single thing I would want a candidate to have read before walking into a front-office interview.

Part 1 — The three revenue lines

Everything a derivatives desk earns comes from one of three places. Not from prediction, and not from being clever about the market.

• Spread. You quote a bid below fair value and an ask above it. The client crosses one of them. You keep the difference, multiplied by volume.
• Hedging efficiency. When a client buys the call from you, you go and hedge. No individual hedge is perfect. Across thousands of trades, the imperfections are supposed to wash out — and being systematically better at that than the next bank is a real, compounding edge.
• Financing and balance sheet. You hold inventory. It consumes capital, it pays or earns repo and rates, and on some product families the financing line is the largest single source of revenue by a wide margin.

Structured products add a fourth item that is really a packaged version of the first three: an upfront commission at issuance. We come to that.

If you can say those three sentences cleanly, with one concrete number attached to any of them, you are already in a different pile from the person before you.

Part 2 — Revenue one: the spread, and what it is actually pricing

The naive picture of a spread is a toll booth. You quote 99 bid, 101 ask, someone lifts your offer, you made a point.

That picture is wrong in a way that matters, and interviewers probe it deliberately.

The spread is not a fee. It is the price of the risk you are about to warehouse, plus the cost of the hedge you are about to put on, plus a margin for the times you will be picked off. Every one of those three components moves, and a trader who quotes a constant spread is either leaving money on the table or accumulating a loss they have not noticed yet.

The spread is your only protection on an illiquid name

Here is a question I have been asked, and have asked since.

Your salesperson asks for a bid on 1,000 lots of puts on an illiquid stock. You quote. He comes back and asks to double the size. Do you give the same price?

No, and the reasoning is the entire point of the question.

Doubling the size on an illiquid name is not the same trade twice. It is a different trade. Your market impact when you go to hedge roughly doubles. You may already have partially hedged the first thousand, which means the hedge for the second thousand starts from a worse level and is harder to execute. And your concentration in a single name has gone up, which changes your risk limits, not just your P&L.

There is also a question you should be asking yourself out loud: why is he doubling? A salesperson who comes back immediately for more size sometimes knows something about the flow behind him that you do not.

The second lot always costs more than the first. On liquid names you can afford to forget that. On illiquid names, the bid-offer is the only protection you have.

Adverse selection: the spread also pays for being slow

Now the other direction. You are quoting an option. The underlying starts ticking fast, and your option quotes lag behind it by a few milliseconds.

Someone will see the stock move before your pricer does, buy your call at the stale price, and hedge immediately at the new one. That is a risk-free profit for them and a loss for you, and it is called adverse selection, or more bluntly, being picked off.

The fix is partly technical, a low-latency link between the underlying feed and the option pricer so the delta link is effectively instantaneous. But it is also a pricing question: if you cannot be fast, you have to be wide. A spread that does not compensate for the probability of being picked off is a spread that loses money quietly, every day, to people who are better plumbed than you are.

Spreads in a crash

When the market is falling and you are short gamma on the downside, quoting becomes actively dangerous. Three things happen at once, and a desk does all three:

• Widen. A wider quote stays valid longer in a fast market, which also means fewer updates.
• Throttle. Impose a minimum interval between quote updates, ten milliseconds or so. This is not only a risk decision. Exchanges monitor your quote-to-trade ratio, and flooding the gateway with messages earns you messaging fees from Eurex or CME and, in extremis, suspension of a trading ID.
• Hedge before you quote. Short gamma in a crash is lethal. Buy futures or ETFs to reduce the directional exposure first, and worry about the quality of your quotes second.

The lesson generalises. The spread is a risk parameter, not a price list.

Part 3 — Revenue two: hedging efficiency

This is the least visible of the three lines and the one that separates desks.

What actually lands in your book

A client hits your ask on a call. One second later, without you doing anything, you are short the option: short delta, short gamma, short vega, long theta, short rho.

The first question is not how do I hedge this but am I inside my limits. If you are, then keeping some of that risk is a decision you are making, not an accident that happened to you. That distinction is what an interviewer is listening for.

Then you hedge, and where you hedge depends on size. A small clip goes on screen. A large clip goes over the counter, through IB Chat and brokers, because showing size on screen tells the market exactly what you need and the market will charge you for the information.

Long gamma is a business model, not a position

Once you are delta hedged, the P&L from a move is quadratic:

P&L ≈ ½ · Γ · (move)²

One million of cash gamma, spot moves 1%, and you have made five thousand euros — in either direction, because a delta-hedged gamma position does not care which way the market went, only how far.

Because the relationship is quadratic, the path matters enormously. Three separate days of 3% give you 9 + 9 + 9 = 27. One single day of 9% gives you 81. Same total distance travelled, three times the money. Clustered volatility is what makes a long gamma book and destroys a short one.

But you do not get gamma for free. You are paying theta the whole time, and the trade only works if what actually happens beats what you paid for:

P&L ≈ ½ · Γ · S² · (σ²_realised − σ²_implied) · dt

Read it as a business rather than an equation. You bought implied volatility when you bought the option. You sell realised volatility every time you re-hedge, because long gamma mechanically forces you to sell after the market rises and buy after it falls. The desk's edge is the spread between the two, and it accumulates over thousands of small re-hedges rather than arriving in one trade.

Why models exist

A common student question: if the market already quotes a price, why does a bank need a model at all?

Two reasons. To price the thing that is not quoted, and to know what to hedge with.

The second reason is the one that generates or destroys money. Black-Scholes assumes a single volatility for all strikes, which is simply false — invert real market prices and you get a different number at every strike, and on equities that surface skews downward. If you price a barrier option with one flat vol, you will misprice it badly, because the barrier lives at a different point of the smile from the payoff.

That is why local volatility exists: it fits the whole observed surface. It is also why stochastic volatility exists, because local vol produces unrealistic forward volatility dynamics, and why the exotic books run on a hybrid of the two. None of these models is right. Each of them fails somewhere, and the choice is about which failure your particular book can survive.

When you cannot hedge with the underlying

A stock hits limit down, trading is suspended for fifteen minutes, and you are massively long delta.

Your greeks are now blind. There is no mark, your delta and gamma are stale, and the instrument you would normally trade is unavailable. What a desk does: hedge with proxies — index futures, sector ETFs — accept the basis risk, prepare the orders for the reopen, and expect a violent volatility gap when trading resumes, because continuous trading was an assumption of everything in your model and it just broke.

The general point is that hedging efficiency is not a theoretical exercise. It is execution, plumbing, venue choice and market structure, and it is where the difference between two desks running the same book actually shows up in the P&L.

Part 4 — Revenue three: financing, and the turbo as the purest case

If you want to understand how a bank earns money from its balance sheet rather than from its opinions, look at a turbo certificate. Nothing else is as clean.

What it is

Turbo long = long stock − PV(strike)

The client wants leveraged exposure to a stock without paying the full price. Instead of paying 100 for one share, he pays only the value above the strike. The bank finances the rest.

The first thing to say in an interview, because it is a deliberate trap: a turbo is not an option. There is no optionality, no smile, no vega. The delta is 1, times the ratio, for the entire life of the product. Some documentation uses the word “option” and it is misleading. A turbo is synthetic long stock with a knock-out attached.

The mechanics, and where the revenue is

The bank sells the turbo, is immediately delta short, and buys the underlying stock to hedge. Sell 1,000 turbos at a ratio of 0.10 and you buy 100 shares. Then you do nothing. Because delta is constant, there is no dynamic re-hedging during the life of the product. The stock holding stays exactly where you put it.

The revenue arrives every single day through the strike:

ΔK = K × r / 365

The strike creeps upward by the financing cost, which means the turbo price creeps down by the same amount. The client is paying the carry on the money the bank borrowed to buy those shares. Revenue is approximately the risk-free rate times the financed notional, and it accrues whether the stock goes up, down or nowhere.

Leverage falls out of the same arithmetic:

Leverage = S / (S − K)

Spot 100, strike 80: the client pays around 20 for exposure to 100, so five times leverage. Move the barrier closer to spot and the turbo gets cheaper and the leverage rises:

Which is more expensive, leverage five or leverage ten? Leverage five, and the reason is worth saying out loud: the barrier sits further away, the product is safer, and the client pays for that safety. A cheap certificate is cheap because it is close to dying.

And now the part they are testing you on

All of the above makes the turbo sound like free money: constant delta, no rebalancing, daily accrual. It is not, and the risk is not where beginners look for it.

When the barrier is touched, the product ceases to exist instantly. Its delta goes to zero. But you are still holding the hedge — 100 shares that are now naked long. You have to sell all of them immediately, into a market that is already falling, because that is why the barrier was hit in the first place.

The residual value paid to the client is:

Residual = max(0, (Spot at KO − K) × ratio)

Spot knocks out cleanly at 85.20 with a strike of 80 and a ratio of 0.10, and the client receives 0.52 per certificate. Fine.

Now let the stock gap overnight on an earnings release, straight from 84 to 78. The barrier and the strike are both behind it. The residual is zero, the client receives nothing, and the bank absorbs the difference between where it thought it would unwind and where it actually can. The buffer between barrier and strike exists precisely to cover part of that, and that buffer is what the desk is charging for.

Scale it. Ten thousand certificates knock out simultaneously and you are selling ten thousand times the ratio in shares, instantly, into a falling market. Your own unwind moves the price against you, which is a liquidity feedback loop rather than a delta problem.

So the “risk-free” financing income is not risk-free at all. It is paid for by warehousing jump risk. Delta hedging protects you against continuous moves. Nothing protects you against a discontinuity, and that is the trade.

Part 5 — Structured products: selling the risk the client wants to own

Here the bank stops being a market maker and becomes an assembler. The revenue is an upfront commission at issuance, but the interesting part is what the bank is left holding afterwards.

The process is always the same four steps: identify the client's need, select the building blocks, construct and hedge, and generate the revenue. Every structured product you will ever see decomposes into stock, bonds and options, and it belongs to exactly one of three families.

Family one: financing

Turbos, leverage certificates, total return swaps. Covered above. The client is renting the bank's balance sheet, and the bank earns the financing spread.

Family two: capital protection

Guaranteed note = long zero-coupon bond + long call

The ZCB returns the principal at maturity, and whatever capital is left over funds the option. A 100 investment might split into 90 of ZCB and 10 of call: the 90 grows back to 100, and the 10 buys the participation.

The trade-off is mechanical and worth stating clearly, because clients frequently misunderstand it. More protection means less capital available for the call, which means less upside. There is no version of this product where the client gets both.

Family three: yield enhancement

Reverse convertible = long zero-coupon bond + short put (or down-and-in put)

This is where the majority of retail structured product volume sits, and it is the family with the most interesting consequences for the desk.

The philosophy is to turn an equity investor into an insurer. The client sells downside protection and collects the premium, expressed as a coupon:

Coupon ≈ put premium / forward

He is not being paid interest. He is being paid an insurance premium for accepting somebody else's downside, and he finds this out when the barrier breaks.

Two products from the same family worth being able to decompose on demand:

Discount certificate = long stock − call struck at the cap

The client buys the stock at a discount, funded by the call he implicitly sold. He is short volatility: higher implied vol means the call is worth more, so the certificate is worth less, so his entry discount is deeper. That is why these products sell well in high-volatility markets, and it is the mechanism behind a good interview question — does a more expensive call make the discount certificate more or less expensive? Less. Write the decomposition and read the sign off it.

Bonus certificate = long stock + long down-and-out put (K at bonus, B at barrier) − call at the cap

Lower the barrier and you can promise a higher cap bonus, because the protection is deeper and the structure can afford more. It beats holding the stock in sideways and moderately bearish markets, and loses only when the stock rallies well past the bonus.

What the desk is left holding

Every one of these products leaves the bank on the other side of the trade, and the aggregate position across the whole street is where a lot of market behaviour comes from.

Yield enhancement means the desk is long puts, in enormous quantity, accumulated over years of issuance. Long puts means long dividends. To hedge that, the desk sells dividend futures — and so does every other desk, all hedging the same inventory. Which is why the dividend futures curve typically slopes downward even though companies grow their dividends over time. The curve is telling you about dealer inventory, not about corporate policy.

That is the kind of observation that ends an interview well, because it can only come from someone who has thought about flow rather than about formulas.

The funding spread is a product input

One more link that students almost never make. The zero-coupon bond inside these structures is the bank's own paper, so it is discounted at the bank's own credit spread.

When the bank's spread widens, the ZCB is cheaper, which frees capital inside the same 100 of client money to fund a bigger coupon or a larger option. A spread 100 basis points wider on a 10 million note frees roughly 100,000 to spend.

So a bank with a worse credit standing can offer a better-looking coupon. The client is being paid more partly because he is taking more issuer risk, and if the bank defaults before maturity the coupon was irrelevant. Anyone who lived through Lehman certificates understands this line viscerally.

Part 6 — Where the money leaks out

The three revenue lines are gross. Several things eat them, and a candidate who can name the leaks sounds like someone who has sat on a desk.

• Funding and regulation. Liquidity coverage and net stable funding requirements have a cost, and it is charged to the desk. So does counterparty risk, through CVA and DVA on every OTC trade. A position that is profitable before these charges is not necessarily profitable after them.
• Market impact on hedges. Every hedge you execute moves the price you are hedging at. On liquid names this is noise. On illiquid names it is the whole trade.
• Discontinuities you cannot replicate. The clearest example is a digital option. A three-month at-the-money digital call is worth about 0.50, because it pays N(d₂), the risk-neutral probability of finishing in the money, and at the money that is a coin flip.

But a digital cannot be hedged. Its gamma is infinite at the strike at expiry. So desks over-hedge it with a real, tight call spread — long the 99 call, short the 101 — which necessarily over-prices the digital slightly. That over-hedge is the margin. It is also why digitals always trade above their Black-Scholes theoretical value: you are not paying for the payoff, you are paying for the fact that nobody can replicate it.

The price is the hedging cost. That sentence generalises to the whole business. Whenever you cannot work out where a margin is coming from, look for the thing in the payoff that cannot be replicated cleanly.

Correlation. As soon as a product has more than one underlying, you are trading correlation whether you meant to or not. The holder of a worst-of put is short correlation and long dispersion: three stocks at 96, 95 and 95 give a payoff of 5, while 96, 95 and 80 give a payoff of 20. The outlier creates the value.

And the hedging of these is deeply unintuitive. In a worst-of on three names where two are 40% below strike and the third is 2% above, the highest delta sits on the third. The two that already collapsed have their downside captured and contribute almost linearly. The one near the strike can still become the new worst-of and rewrite the entire payoff. Delta follows uncertainty — you hedge what can still flip the product today, not what already moved.

Part 7 — The P&L is a mark, not cash

This is the most misunderstood part of the business, and there is one interview question that tests it perfectly.

Your internal model marks a vol at 20%. The market is quoting 22%. A colleague is short that option, and marking it at 22% would show a 20,000 loss in his book. You buy the option in the market at 22%. What is your initial P&L?

The answer is zero.

You bought at the market price, so your entry equals the mark. P&L is (mark − entry) × quantity, and that is zero by construction. The blotter might briefly display a theoretical loss, because you paid 22 while the surface still says 20, and that display disappears the moment the surface is updated to the market.

The important sentence is the second one. You did not create your colleague's loss. You exposed it. The 20,000 was already sitting in his book, hidden behind a stale mark, and your trade forced the surface to be updated to reality.

Two things follow, and both matter more than the arithmetic.

Mark-to-market is not the same as cash. A desk's daily P&L is an opinion about the value of positions that are still open, and that opinion depends on a volatility surface that somebody has to mark. Where a surface is thin or illiquid, the mark is a judgement, and judgement is where P&L disputes live.

And marking is a governance question, not a trading one. This is why banks separate the people who trade from the people who validate the marks, and why an “unexplained” P&L line in the daily attribution gets attention immediately. It usually means either a hedge that behaved differently from the model, or a mark that was not what somebody believed it was.

Part 8 — The money you must not make

Every conversation about how a desk makes money should include the boundary, because the boundary is closer than students expect and interviewers do sometimes ask.

Suppose you are 1% away from a barrier, and if it breaks you realise a very large gain. Every one of the following is market abuse and prohibited by ESMA, the AMF and the SEC:

• Banging the close. Selling the underlying aggressively yourself to push spot through the level. This is price manipulation, full stop.
• Collusion. Asking a contact at another bank, or a client, to do the selling for you. Concerted action, and the fact that you did not press the button yourself is not a defence.
• Painting the tape. A series of small sell orders designed to trigger momentum algorithms into doing the work for you.
• Wash trading. Buying and selling between accounts you control to manufacture artificial pressure and an artificial price.

The same category covers a much more mundane situation. If you must sell a stock for book A and buy the same stock for book B, you do not simply cross them yourself: that is a wash trade. You use an internal crossing desk with the match documented at mid for compliance, or self-match prevention on the trading IDs so the exchange cancels the self-matching orders, or you deliberately separate the two legs in time.

Know the line clearly enough to describe it. This is criminal, not clever, and the candidates who treat the question as an invitation to be creative do not get called back.

Part 9 — Why the model is under pressure

Everything above still works. It works on thinner margins every year, and you should be able to say why.

• Commoditisation. Every bank now sells the same bonus certificates, the same discount certificates, the same autocalls. When the product is identical everywhere, the only differentiator left is price. Structures that carried 3% margins twenty years ago carry a fraction of that.
• Transparency. MiFID II and PRIIPS force full cost disclosure to retail clients, who can now compare issuers in seconds. There is nowhere left to hide a margin.
• Automation. Pricing and hedging a vanilla structure used to occupy a junior for an afternoon. It is now milliseconds of compute. Headcount followed the work.
• Shorter settlement chains. Cash equity settled in a week plus a day twenty years ago, settles at T+2 in Europe today, and will settle instantly at some point. Every compression of that chain removes a piece of intermediation, and intermediation was a revenue line.

The consequence for anyone entering the industry is specific rather than gloomy. The value of doing standardised work has gone to roughly zero, and the remaining margin sits in bespoke institutional flow, in the management of aggregate risk across a book, and in the products where the hedging is genuinely hard. Those are the seats worth wanting.

It is also worth being able to name the moments when the whole machine has been tested. Eurostoxx futures have gone into volatility breaks around the Lehman crash in 2008, the Flash Crash in 2010, the eurozone debt crisis in 2011, the Chinese devaluation in 2015, Volmageddon in 2018, the Covid crash in 2020 and the rate shock of 2022. If somebody asks you how many times, they are testing whether you know the events, not whether you memorised a number.

Part 10 — Saying it in sixty seconds

You will not get ten minutes for this in an interview. You will get about a minute, and here is a version that fits.

A derivatives desk earns in three ways. It quotes a bid and an ask and captures the spread, which is really the price of the risk it is about to warehouse. It hedges that risk, and being systematically better at hedging than the next bank is an edge that compounds over thousands of trades — a delta-hedged long gamma book earns half gamma times the move squared, and profits when realised volatility beats the implied it paid. And it earns on its balance sheet: on a turbo certificate, the strike accrues by K times r over 365 every day, so the client is paying the carry on shares the bank bought with borrowed money. Structured products bundle all three into an upfront commission, and leave the desk holding whatever the client did not want — which is why the whole street is long puts and therefore long dividends, and why the dividend futures curve slopes down.

That is roughly fifty seconds and it contains three formulas, one product, and one flow observation. It is not a script to copy. It is a demonstration of the shape a good answer has: mechanism, then number, then consequence.

The three questions to ask of anything you are shown

Whatever product lands in front of you, in an interview or on a desk, run the same three:

• What is it made of? Decompose it into stock, bonds and options. If you cannot draw the payoff, you cannot price it.
• Who holds which risk? Say it from both sides. The client is long the stock and short the call. Therefore you are short the stock and long the call, long gamma and paying theta.
• Where does the money come from? Spread, commission, or financing. And what would have to happen for that money to turn into a loss.

Three questions, and they work on a vanilla call, a bonus certificate and a worst-of autocall equally well.

Nobody expects a candidate to price an exotic in their head or to have memorised a closed-form solution. What is expected is that you understand that the desk is not in the prediction business.

It is in the business of taking risk that somebody else wants to get rid of, charging correctly for it, hedging it well enough to survive, and financing the whole thing on borrowed money.

Say that, attach one number to it, and you will have answered the question better than most of the people ahead of you in the queue.