Why is Jane Street, a so-called anarchist commune, so collaborative, and yet profitable, compared to the silos of the multistrats? How do they do it with such little headcount?

So I’ve been thinking about the world of mechanism design for a while—not super deeply, but at Roughgarden’a first and second lectures

Note that I worked at a heavily siloed, PnL attributed, game based firm (I will call this a “trader-driven” firm). DRW’s DNA is conquest, trading, and heros. Pretty glamorous, a lot of stress. I’ve only talked to one or two insiders at JS, but have a good idea of how much more relaxed the trading org is at that firm (call it “quant-driven”). And as we’ve seen, there’s a repeatable process they do to move into new markets. Most of this is my guess, but it lines up with what I’ve heard from within the firm.

I’m not gonna get into “pod shops” which are merchant firms that fund specific strategies at specific times (pods are essentially contractors with a specific focus, renting technology).

Hopefully this is helpful to people considering a trading career. and this is obviously a simplification.

In trader-driven firms, decision making and PnL is associated with desk or a solo trader. The amount of money a (partner, desk head, trader, analyst) contributes to the firm is a direct reflection of how much value they contribute to the firm and how much allocation they get. Therefore the percentage of PnL you contribute is how you get a bigger chunk of allocation next year.

By contrast, a quant-driven firm is much more egalitarian, and less hierarchical (one desk head has a lot more direct reports, talks directly to the committee of partners above them), and compensation is correlated to firm-wide performance, and estimate of % contribution. So fewer layers of hierarchy, and more team-wide communication.

Jane Street is known for being relatively efficient with low turnover, while some place like CitSec in the same market churn a lot more people yearly. Trader-driven firms tend to have a bunch of redundant software and intra-firm competition. The quant-driven firms tend to socialize technology and efficiency gains pretty well. Therefore a lot of technologists stay for a long time.

Now let’s bring in some game theory. Why does the trader-driven nature of the first firm lead to more strategic, cutthroat behavior? Note that I’m not necessarily making a value judgement, competitive pressures within trader driven firms allow for more risk taking. When no more money is there to be made in TradFi, trader driven firms tend to be the first to conquer new lands like crypto and exotics.

Trading firms have a lot of rational actors, so let’s assume every actor is perfectly rational. A lot of this might seem intuitive at first, but having a little theory is really helpful.

U_i represents individual utility to the firm. Compensation is a function of U_i. I’m guessing most individuals in trading would derive utility linearly based on compensation, but as we will see, downstream of the incentive structure we will see how firms can attract individuals focused less on utility.

Trader-driven: individual P&L

U_i \approx PnL_i

If I discover a useful signal and give it to you, your book gets better. Maybe mine doesn’t. If promotion and compensation depend heavily on individual attribution, information sharing can literally reduce my relative payoff.

Then the firm has accidentally created a little market inside itself:

U_i(a_i,a_{-i})

My payoff depends on outperforming my coworkers. Hoarding information can become rational.

Quant-driven: firm contribution

U_i \approx f(PnL_{\text{firm}},\ \text{contribution}_i)

Now if I tell you about a bug in your strategy and you make another $20 million, that’s potentially good for me too. If you discover an execution improvement and spread it across five desks, everyone benefits. Since my utility is no longer dependent on the rest of the organization performing.

Let’s now consider 3 efficiency gains, each creating the same amount of value to the firm:

1. Trader Alice discovers alpha worth $10m.

2. Quant Bob has an execution improvement that increases captured alpha by 20%.

3. Engineer Carl makes the system fast enough to deploy both globally.

What happens downstream of this

Trader-driven

In a trader-driven firm, Alice would be incentivized to keep the alpha to themselves. Revealing the alpha could leak the PnL to another desk. 10m gained is 10m for their desk this year, meaning more opportunity for Alice to get budget (and headcount) next year. Alice would definitely not tell Aaron about her PnL because Aaron could then look for those opportunities (hunting for other firm’s alpha is super common)

Alice would also not get the efficiency gains of Bob, unless she hired another quant Barbara who found the same improvement as Bob. Now the firm has two quants who found the same improvement. Quants are pretty solitary creatures who don’t really talk much, so unless they are on the same discord, let’s assume Bob and Barbara have not talked to each other. Note that in a trader-driven firm, Bob and Barbara are cost centers (unless they are “quant traders”), so hiring Barbara when Alice has bankroll to hire would probably mean getting rid of Bob.

Carl, meanwhile, is attached to a desk , and so his findings are built on the platform of his desk and he’s not allowed to share them with other engineers, and definitely not allowed to share them externally. Carl as an engineer wants to have an external profile so he’s not stuck to an organization when bad times hit. Funds have started to create separate tech teams (seen often in multistrat orgs) but still handicapped relative to big tech organizations.

Quant-driven

Alice here would report to the partner committee, and this $20m is then added to the firm’s PnL. Alice’s team would then have a higher share of firmwide PnL, and she would have ownership over the signal, and that would be associated with her book. She wouldn’t have an incentive to hoard the info, since the firm isn’t paying for her PnL directly.

Bob’s improvement is passed to the committee as well, and added to everyone’s PnL. Bob is rewarded across the alpha verticals and stays longer and is more loyal to the firm.

Carl’s improvements would be spread firmwide. And since everyone is sharing in the profit, Carl becomes pretty well known within the firm (lots of credit). And eventually once the improvement is commodity, Carl can appear on a podcast and get his name out there.

So as we can see the Quant driven firm attracts happier and better quants and engineers. Traders have less of an incentive to hoard info infra-firm, and efficient improvements are quickly shared firm wide.

I’m not gonna argue if one is better than the other, because new markets always exist. So a trader who is upset at not getting attribution in equities can rush to be the first in crypto, or prediction markets, or swaps, or Asia (aka “be entrepreneurial”). Markets are big and complex. But the system does incentivize intra-firm behavior as well as how a firm collaborates. And that impacts retention and employee quality downstream of that.

Actual Game theory: Setup

Two colleagues, Alice and Bob, each discover something valuable with probability baked into the payoffs below. Each simultaneously chooses Share (S) or Hoard (H) their finding.

• If both share, total value created = $20m (synergy: full information flow, no duplicated effort)
• If one shares and one hoards, the sharer generates $10m firm-wide but the hoarder still gets their own private $10m
• If both hoard, each only realizes their own $10m in isolation, no synergy = $20m total but zero information transfer benefit and duplicated effort costs, say −$2m each in redundant work → $8m each

Let c = private cost of sharing (competitive risk: the colleague could use your info against you, or take partial credit) — this is the key variable that differs between regimes.

Trader-driven payoffs: U_i ≈ PnL_i (individual attribution)

Here your payoff is your own attributed PnL, minus the cost of sharing.

	Bob: Share	Bob: Hoard
Alice: Share	10−c, 10−c	10−c, 10
Alice: Hoard	10, 10−c	8, 8

With individual attribution, sharing your alpha lets the other person use or claim partial credit for it without reciprocating — this is the source of c. As long as c > 0 (any leak risk, credit dilution, or “they use my info without giving me a cut”), Hoard strictly dominates Share for both players. The unique Nash equilibrium is (Hoard, Hoard) → (8, 8), even though (Share, Share) → (10−c, 10−c) is Pareto-superior whenever c < 2. This is a genuine prisoner’s dilemma — the individually rational move destroys joint value.

Quant-driven payoffs: U_i ≈ f(PnL_firm, contribution_i)

Now suppose compensation is contribution-weighted share of firm PnL, so U_i = (contribution_i / total_contribution) × PnL_firm. Sharing no longer transfers value to a rival at your expense — it grows the pool you both draw from, and your contribution share is preserved (or even increases, since visible sharing is itself credited as contribution).

	Bob: Share	Bob: Hoard
Alice: Share	10, 10	5, 15*
Alice: Hoard	15*, 5	4, 4

*In the mixed case, the sharer’s info still boosts firm PnL to $20m total, but the hoarder captures a disproportionate share of that $20m by contributing “more” (their $10m is legible; Alice’s shared portion gets diluted across the team that used it), while Alice’s own book looks smaller in isolation — this is the free-rider risk I flagged earlier.

This is the crux: quant-driven regimes don’t eliminate the incentive problem, they just change its shape. If contribution attribution is done well (sharing is itself rewarded, e.g., Carl’s firm-wide deployment credit), (Share, Share) becomes the equilibrium. If contribution attribution is done poorly (hoarders’ individually-legible PnL crowds out sharers’ diffuse contribution), you get a second prisoner’s dilemma, just with the payoffs relabeled — now hoarding your specific number is safer than trusting the committee to correctly value your diffuse contribution.