Jane Street.
Twenty-five years. From an ETF arb desk to the world's largest market maker. The combinator isn't always an ensemble. Sometimes it's a quoting engine with risk baked in.
A market maker that refuses to look like one.
Founded 2000 by a group of ex-Susquehanna traders. The opening business was simple and unglamorous: arbitrage the price difference between an ETF and its underlying basket. The interesting decision was structural: the firm chose not to grow by adding products, but by making the pricing engine smarter at each asset it already touched.
The bet was on the combinator: a quoting engine that prices everything at once, with the risk of each leg already reflected in the quote you offer the next leg. Buy a junk bond ETF from a customer, the system instantly re-quotes the underlying bonds, the swaps that hedge them, and the index futures that hedge those. Every leg is a function of every other leg, continuously.
That choice (combinator-first, products-second) is why a firm with ~3 000 people prints more trading revenue than the trading divisions of multi-trillion-dollar banks. The moat isn't a signal. It isn't an ensemble. It's the latency between a customer trade arriving and every related quote updating to reflect it.
From ETF arb desk to global quoting engine.
Each step was a deeper investment in the combinator (the pricing engine), not in adding products or hiring star traders.
Founded · ETF arbitrage desk
Tim Reynolds, Marc Gerstein, Rob Granieri leave Susquehanna with the bet that ETF/basket arb is a structurally-mispriced corner of the market. Headcount ≈ 30. No outside capital.
OCaml decision · the architectural bet
Migration off ad-hoc C++/Python begins. The case: a typed functional language lets the combinator (a giant graph of cross-asset pricing functions) refactor safely at scale. Will turn out to be the single most important non-trading decision.
eventually >25M lines of OCaml in productionGFC · the quoting engine holds
While Wall Street's prop desks blow up on directional bets, Jane Street's combinator keeps quoting because it has no directional view, only relative-pricing risk that nets across products. Profitable through 2008.
no directional bets · combinator is the strategyExpansion · options, futures, FX
Each new asset class is added through the combinator, not alongside it. The pricing graph grows. Latency between a customer fill and every related quote updating drops below ten milliseconds.
COVID · $8.4 B in net trading revenue
March 2020 ETF dislocations are exactly the regime the combinator was built for: extreme spreads between ETFs and their underlying as liquidity fragments. The pricing graph absorbs the dispersion and prints a record year.
$8.4 B · record · pre-AI hype$14.2 B net revenue · ~2× Goldman FICC
The combinator now spans equities, options, fixed-income, ETF, FX, crypto, and select credit. A firm with ~3 000 people generates 2× the trading revenue of Goldman's flagship trading division. Headcount-adjusted, it's not close.
$4.7 M revenue / employee · industry-best by ~6×Quiet · still private · still no IPO
The firm refuses outside capital, refuses an IPO, refuses the brand. The reason is structural: once the combinator is the moat, the only thing you have to protect is the combinator. You don't sell shares in it.
We don't really have signals in the sense that other quants do. We have a model of what every related thing should be priced at, given what just traded. The edge is the time between those two things.Paraphrasing the Jane Street pricing-engine thesis · per public interviews & OCaml dev talks
The combinator can wear different shapes.
Medallion's combinator is an ensemble: many weak directional signals fused into a portfolio. Jane Street's combinator is a quoting engine: many related prices linked into a single instantaneous view. Same architecture, different shape.
The model most banks use.
Each desk prices its own product. Risk is netted at end-of-day, sometimes end-of-hour. When a customer trades, the next quote on a related product is stale until the desk talks to the other desk.
One pricing graph · everything moves at once.
Equity, options, ETF, futures, FX: every quote is a function of every other quote. A fill in one corner re-prices the whole graph in under 10 ms. Risk is embedded in the quote, not added later.
Three things Jane Street teaches.
The combinator isn't always an ensemble.
A combinator is anything that fuses many signals/prices into one decision faster than the market can. For Medallion that's a portfolio engine. For Jane Street it's a quoting engine. The architectural pattern is the same.
↪ Read the taxonomyArchitecture choice dictates ceiling.
OCaml looked exotic in 2005. It turned out to be the only language with the typing discipline to refactor a 25-million-line cross-asset pricing graph without breaking it. The combinator's ceiling is set by the language it's written in.
↪ See the architectureOnce the combinator is the moat, growth shape changes.
Jane Street grew revenue 10× without growing headcount proportionally. That spread between revenue and people is what tells you the firm is selling work the combinator does, not work people do. A signal of a real moat.
↪ View the infographicFour firms. One pattern.
Build the graph, not the quote.
The combinator can wear any shape.