Jane Street booked $39.6bn in net trading revenue in 2025, the largest single-year trading result recorded by any firm, roughly double the $20.5bn it earned the year before. Millennium returned 10.5% over the first half of 2026 on approximately $89bn in assets. Both firms sit under the same three-letter label in every industry survey published this year.
The label carries almost no information. Holding periods across the strategies inside it range from microseconds to eighteen months. Capacity ranges from a few hundred million dollars to well over a hundred billion. The economic service being sold differs completely from one end to the other.
A market maker earns a fee for holding inventory it did not choose. A trend follower earns compensation for holding a directional position that other participants abandon during drawdown. A factor manager earns a premium for owning exposures that most allocators find uncomfortable to hold through a full cycle. These are separate businesses with separate research stacks, separate cost structures, and separate ways of failing.
What follows is the working map, organised by what each strategy actually gets paid to do, with the capacity limits and the failure modes attached to each.
For daily Quant articles and market breakdowns:
Most taxonomies get built around instrument type or holding period. Both are downstream variables. The primary variable is the identity of the counterparty and the reason they are willing to pay you.
There are four payments available in public markets, and every systematic strategy in existence collects one of them or blends several.
The first is immediacy. A participant needs to trade now and accepts a worse price to do so. The spread they concede funds market making, and the revenue scales with volume rather than with forecast accuracy.
The second is convergence. Two related instruments have drifted apart for reasons unconnected to fundamentals, usually flow or index mechanics. Statistical arbitrage and relative value collect this payment when prices normalise.
The third is risk transfer. A participant wants less exposure to a persistent systematic risk and pays a premium to be relieved of it. Factor investing, carry, and volatility selling all sit here, and the return is compensation for accepting something genuinely unpleasant.
The fourth is convexity. A participant wants protection against a large move, or lacks the tolerance to hold a position through the drawdown required to capture the eventual trend. Trend following and long volatility collect this payment, and both accept long stretches of negative carry in exchange.
The blend between these four is where correlation surprises originate, which explains why platform risk officers spend more time on aggregate exposure than on any individual signal.
The business is straightforward to describe and difficult to run. Post a two-sided quote, capture the difference between bid and offer, hedge the resulting inventory, and repeat several million times per session. Profit accumulates on flow regardless of market direction, and the binding constraint is the share of counterparties who know something you do not.
The scale figures now exceed the entire hedge fund side of the industry. Non-bank trading firms generated $114bn in combined revenue during 2025, a 45% increase year on year. Market-making revenue within that total rose 20% to $30.2bn, while proprietary trading and investment activity climbed almost 60% to $84.3bn. Banks still produced a larger overall pool at $260.7bn, up 13%, though their contribution has shifted toward structuring and complex institutional execution while the non-bank firms absorbed retail flow, exchange-traded products, and high-frequency quoting.
Concentration inside that pool is extreme. Citadel Securities, Jane Street, and Hudson River Trading collectively produced around $27bn in trading revenue during the first quarter of 2026 alone.
Citadel Securities closed 2025 with a record $12.2bn in net trading revenue, 25% above the $9.7bn posted in 2024. Hudson River Trading delivered $3.7bn in the third quarter of 2025, an 81% increase on the prior year. Jane Street produced $16.1bn in trading revenue and $10.3bn in net income during the first quarter of 2026, more than double the equivalent quarter, from roughly 3,000 employees across five offices.
Revenue behaves as a derivative on realised volatility. A single quarter delivering more than 40% of the prior full year, which is what Jane Street’s first quarter of 2026 represented, demonstrates how tightly these earnings track market movement. Quiet sessions compress spreads and reduce capture per contract, and no amount of technology investment changes that relationship.
Regulatory exposure has become the largest single-line risk. Jane Street generated more than $2.3bn in equity-derivatives revenue in India during 2024, and the Securities and Exchange Board of India opened an investigation into the index-options strategies behind those gains. The case turns on the boundary between supplying liquidity and influencing an index level, and the eventual finding will shape how regulators treat algorithmic positioning well beyond that jurisdiction.
Flow acquisition also carries a real cost that rarely appears in strategy descriptions. Citadel Securities paid $388m for order flow in the first quarter of 2025, 45% above the same period a year earlier. Spread capture only counts as alpha net of what the firm paid for the privilege of seeing the order first.
Statistical arbitrage runs thousands of positions simultaneously, each carrying a small expected return, aggregated into a portfolio that holds close to zero net exposure to the broad index. Holding periods run from a few hours to a few weeks. Turnover is high enough that transaction cost modelling matters as much as signal quality.
The edge sits in research throughput and execution efficiency rather than in any individual predictive relationship. Teams that test more hypotheses per quarter, and that lose less to slippage on each rebalance, outperform teams with better ideas and worse infrastructure.
The failure mode has been documented for nearly twenty years and has never been engineered away. Positions overlap across managers because the underlying signal families overlap, and the overlap only becomes visible during a forced unwind.
The reference episode remains the first week of August 2007. Quantitative equity market-neutral books took severe simultaneous losses between the sixth and the ninth, followed by a sharp partial recovery on the tenth. The accepted explanation involves one large participant deleveraging a crowded market-neutral portfolio, which pushed widely held long-short positions against every other holder of the same positions, many of whom then reduced into the same move. Losses ran between 5% and 30% depending on leverage, and several funds never recovered their prior high-water mark.
Statistical arbitrage saw intense crowding through the first half of this year, and the overlap contributed directly to muted returns at the larger multi-strategy platforms. Increasing similarity across momentum, mean reversion, and relative value books was identified as the cause of the performance disturbances that ran through systematic equity during the period.
The first half of January 2026 produced the weakest ten-day stretch for systematic long-short equity managers in more than three months, with losses of around 1% concentrated almost entirely in US equities. That figure sounds small until it is scaled by the leverage typically applied to a market-neutral book.
Crowding cannot be measured directly, since competitor positions are unobservable. The usable proxies are spreads that once took several days to revert and now revert within hours, elevated short-term reversal following earnings announcements, and correlated drawdown patterns visible in prime brokerage data.
Faster convergence looks like improved execution and usually indicates additional participants trading the same relationship.
Capacity in this bucket is genuinely bounded by market impact. Doubling the size of the book does not double the return, because the incremental dollar moves the price against the dollar that preceded it. Any pitch describing an uncapped stat arb strategy has either not been tested at size or has not been costed honestly.
The factor bucket covers value, momentum, carry, quality, and low-beta exposures applied systematically across asset classes. Turnover is modest and holding periods run from several months to several years. The return represents compensation for bearing a risk that most participants prefer to avoid, or for exploiting a behavioural pattern that persists because arbitraging it requires tolerating extended underperformance.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.