What a trading firm sells decides how it treats you

Two firms advertise the same role, with the same title, at similar headline compensation. One of them will pay you well for a decade and promote you internally. The other will make you overhead on somebody else's P&L and cut you the quarter their drawdown limit trips. Nothing in the job description distinguishes them.

What distinguishes them is what the firm actually sells. Every trading business monetises something specific, and that thing determines which people are assets and which are costs. Comp structure, retention policy, hiring bar and firing speed all follow from it mechanically. They are not culture. They are arithmetic.

Six models

1. Proprietary market making at scale

Jane Street, Optiver, IMC, SIG, Citadel Securities, Jump, Flow Traders, DRW. Revenue is spread capture across enormous volume, plus fee tiers, exchange liquidity programmes and market-share rebates.

The edge is the system. Infrastructure, venue relationships, tier economics, risk architecture, and the accumulated calibration of a thousand small decisions. None of it fits in a departing employee's head. A trader who leaves takes very little, because the trader was one component in a machine that will keep running without them.

So: high base, deferred bonus, firm-wide profit sharing rather than individual attribution, long graduate programmes, promote from within, garden leave and non-competes on the way out. They pay generously to keep the machine intact and to make sure nobody assembles a competing one. The person is fungible; the system is not, and the compensation structure says exactly that.

2. Pod / multi-manager platform

Millennium, Balyasny, Point72, ExodusPoint, Schonfeld, Tower. The platform supplies capital, infrastructure, financing and risk oversight. The portfolio manager runs an independent book with their own P&L and takes a stated share of it.

Here the PM's alpha is the product, and it is portable. A PM with a working strategy can walk it across the street. So the payout has to be explicit and formula-based, because a discretionary bonus would not survive contact with a competing offer.

The same portability explains the discipline that outsiders find shocking. Drawdown limits are hard and the response is automatic: roughly 5% and your capital is cut, roughly 7.5% and you are gone. That is not cruelty, it is portfolio management applied to people. The platform holds a portfolio of pods and runs it the way any portfolio is run — cut the losers quickly, allocate to the winners.

And it has a consequence that rarely appears in the recruiting pitch: support staff are charged against a pod as cost. When the pod is cut, the engineers attached to it are cut with it, regardless of how good their work was.

3. Land grab / new-venue integration

Common in crypto through the last cycle and largely invisible in the literature, because it is not a strategy business at all. Revenue comes from deal flow and integration speed: token issuers and new exchanges pay for liquidity, usually as a market-making agreement with a token loan and a call option attached.

The trading strategy is deliberately commodity — a basic maker with inventory limits. Nobody is trying to win on quoting logic. What is scarce is being early: first on venue N+1, first to sign the next issuer, first to have working connectivity when a market opens.

Engineers in this model are valued for throughput, not for edge. How fast can you integrate the next venue? It is a genuinely good place to learn: you see more exchange APIs, settlement models and failure modes in two years than you would in ten anywhere else.

But the model has an expiry date built into it. When the new venues stop appearing and the issuer deals dry up, there is nothing left to integrate, and engineering is the first line cut — not because the work was poor, but because the thing being sold no longer exists. People who lived through that reliably interpret it as a judgment on their performance. It usually isn't. It is the business model reaching the end of its runway.

4. Single-manager systematic fund

Renaissance, Two Sigma, DE Shaw, AQR, Winton. Management and performance fees on external capital, or own capital in the rare closed cases.

The signal library is the firm asset, and it is built cumulatively over decades by many people. That produces the observable behaviour: heavy secrecy, siloing so that no individual sees the whole, long non-competes, and compensation that is discretionary and opaque because contribution to a pooled result genuinely cannot be attributed cleanly.

If you want your work to be individually attributable and portable, this is the wrong building. If you want to do deep research on large data with a long horizon and be paid well without carrying personal drawdown risk, there is nowhere better.

5. OTC / principal client flow

Cumberland, B2C2, GSR, and every bank trading desk. Revenue is the spread on client flow plus whatever the desk makes hedging it.

The client relationship is the asset, and it is semi-portable. A salesperson can take a book of clients with them; a trader mostly cannot. So salespeople are paid on their book, traders on desk P&L, and engineers are cost.

The bank version has a further wrinkle worth stating plainly: technology is structurally a cost centre in an organisation whose revenue lines are elsewhere, and cost centres are the first thing trimmed in a bad year. In banking there is a bad year most years.

6. Vendor and infrastructure

CoinRoutes, Talos, Kaiko, Architect. Not trading firms, but they employ a large share of the people who call themselves quant developers, and they belong in the taxonomy because the incentives are cleanest here: the software is the product, so the engineers are the revenue line rather than an expense against someone else's.

Which seat, in which model

The counterintuitive one

A quant developer is better paid and better treated at a market-making firm than at a hedge fund. The instinct runs the other way — funds sound more prestigious, the work sounds more intellectual — and the instinct is wrong for a structural reason.

At a market maker, the system is the edge. The people who build and maintain it sit directly next to the revenue, so they are compensated as revenue-adjacent and retained accordingly. That is why Optiver and Jane Street pay engineers numbers that surprise people who assume the money is all on the trading side.

At a fund or a pod, the same person is overhead charged against someone's P&L. Their work is necessary, their cost is visible, and their contribution is unattributable. When the year is bad, the attributable thing gets protected and the unattributable thing gets cut.

Same title, same skills, opposite position in the business.

How to read a firm before you join it

The taxonomy is only useful if it converts into questions you can ask in an interview. Four do most of the work:

• What does the firm sell? Spread capture, allocated alpha, deal flow, fees on external capital, or software. Everything else follows.
• Is my output attributable? If nobody can point at a number and say that was yours, you will be paid by discretion. That is fine if the firm is structured to reward system-building; it is dangerous if the firm is structured to reward individual P&L and you are not producing one.
• Am I a cost against someone else's book? Ask literally where your seat sits in the cost allocation. The answer predicts what happens to you in a bad quarter better than any conversation about culture.
• How long does this model have? Land-grab businesses run until the land runs out. That is not a hypothetical failure mode, it is the base case, and the timing is usually visible from outside.

None of these are hostile questions and all of them are answerable. A firm that cannot answer the third one has told you something useful anyway.

What this does not cover

This is a taxonomy of revenue models, not a ranking. Every one of these businesses is a good place for somebody. The pod platform that treats a quant developer as disposable is the best possible home for a PM with a working strategy and no desire to raise a fund. The market maker that pays engineers exceptionally will never give a researcher an attributable book.

The mistake is not choosing the wrong model. It is choosing a model whose economics are pointed away from the thing you actually produce, and then reading the consequences as a verdict on your work.