DRW trades its own capital. No outside clients, no fund to market, which is why the interview cares about one question above all: can you find an edge and hold your nerve when money is moving against you. Don Wilson started the firm on the floor of the Chicago Mercantile Exchange in 1992, and today it runs positions across fixed income, options, energy, and crypto through its Cumberland desk. That heritage shows up in how the rounds are built. The loop doesn’t reward resume polish. It rewards speed, calibration, and whether you actually reason under uncertainty instead of reciting a memorized answer.
Before anything else, figure out which track you applied to, because the loops diverge hard after the first screen.
Quant, research, or engineering: pick your loop
DRW hires quantitative trading analysts (the desk-facing role that grows into a trader), quantitative researchers (signal and model work), and software engineers (the low-latency systems and tooling the desks run on). A trading analyst spends the interview proving fast mental math and market intuition. A researcher gets pushed harder on probability, statistics, and how you would validate a model without fooling yourself. An engineer writes real code in a shared editor and then defends the design choices. There is overlap, and a strong candidate can flex across it, but walking in assuming every DRW interview is a mental-math game will sink you on the engineering side, and the reverse sinks you on the desk.
The stages line up like this.
| Stage | Quant track (trading analyst / researcher) | Software engineering track |
|---|---|---|
| At-home screen | Quantitative challenge: linear algebra, Markov chains, distributions, probability brainteasers | Two algorithmic problems on HackerRank, roughly 90 minutes |
| First live round (45-60 min) | Probability and statistics interview with a working trader; answers expected out loud and fast | Live coding in a shared editor in Python, plus complexity and testing follow-ups |
| Superday (in person, 4-6 sessions) | Python coding challenge, a market-making trading simulation, more brainteasers, senior-trader culture fit | Coding and design sessions, code review, computer-systems questions, senior-engineer culture fit |
| Locations | Chicago, New York, or London | Chicago, New York, or London |
| Decision | Offer typically within 1-2 weeks | Offer typically within 1-2 weeks |
The take-home isn’t a formality
Most candidates start with an assessment done from home. For engineers, that is usually two algorithmic problems on HackerRank inside a 90-minute window: clean data-structure work, not obscure competitive-programming tricks. The bar is real, and something like a quarter to a third of people clear it, so passing the visible tests isn’t enough. Your code should read like you would ship it. Handle the empty input, name your variables, and don’t leave an O(n^2) scan where a hash map belongs.
For the quant tracks, the take-home is a quantitative challenge: linear algebra and matrix manipulation, a Markov chain or two, common distributions, and probability puzzles. It is less about one brutal problem and more about whether the basic machinery is loaded and ready. If you have to rederive the variance of a binomial from scratch mid-question, you are already too slow for what comes next.
The probability call, where a trader wants the number now
Clear the take-home and you get one or two calls, usually 45 to 60 minutes. Quant candidates land in a probability and statistics interview run by an actual trader, and the part that catches people off guard is the pace. They want a number, out loud, with your reasoning exposed as you build it. Ninety seconds of silent thinking reads as a red flag even when the answer you eventually give is correct.
Typical territory: expected value of a betting game, a conditional probability that fights your intuition, a dice or coin setup with a twist, a quick Bayesian update. A representative one goes like this. You flip a fair coin until the first head appears; what is the expected number of flips? Let E be that expectation. Then E = 1 + (1/2)E, because you always spend one flip, and half the time you land tails and start over. Solve it and E = 2. Nobody cares about the number 2. They care that you set up the recursion aloud in fifteen seconds.
Engineers get a live coding round instead, in a shared editor with someone watching you type. Expect a mid-difficulty problem in Python, plus follow-ups on time and space complexity and how you would test it. Some loops include a code-review segment where they hand you a messy function and ask what is wrong with it, which rewards people who have maintained production code rather than only solved puzzles.
Superday and the trading simulation
The final round is in person, in Chicago, New York, or London, and runs four to six back-to-back sessions of about 45 minutes each. Quant candidates do a Python coding challenge and, the part most people underprepare, a trading simulation. This is a live market-making game. You are asked to make a two-sided market on some quantity, an integer sealed in an envelope, the number of countries in Africa, the population of a random city, while other players trade against your bid and ask and new information trickles in. The interviewer is not grading whether you nail the true value. They are watching whether you widen your spread when you are uncertain, tighten it when you have an edge, and move your quotes when someone lifts your offer three times in a row, which usually means they know something you don’t.
A concrete way to hold it in your head: if you would buy at 40 and sell at 42, you are claiming fair value sits near 41 with a two-point cushion for your uncertainty. If a counterpart keeps hitting your bid, your fair-value estimate should drift down, and refusing to move it is the single most common way candidates blow this round. Adverse selection is the entire lesson, and they want to watch you feel it in real time rather than recite the definition.
The rest of the Superday mixes brainteasers, more probability, systems questions for the engineers, and at least one senior trader or engineer reading for culture fit. The DRW floor is collaborative and blunt, and they are checking whether you can hold a position in an argument without being a pain to sit next to. Offers tend to land within one to two weeks of the day.
What actually moves the needle in prep
For the quant tracks, drill speed over depth. Work a probability problem book (the trading-interview crowd leans on Heard on the Street and the green book of quant questions) until setups like the coin-flip recursion are automatic, then practice narrating your reasoning on a timer, because thinking silently is a habit you have to break before the call. Play a few market-making games with a friend acting as the adversary. Reading about adverse selection is nothing like feeling your imaginary P&L bleed because your spread was too tight.
For engineering, keep your Python sharp and write as if code review is watching, because it is. Practice mid-level data-structure and graph problems, then go back and clean each answer: better names, guard clauses, a one-line note on complexity. Be ready to walk through a system you built and the one tradeoff you would revisit today. DRW runs serious low-latency infrastructure, so having something specific to say about where time goes in a hot path, cache behavior, allocation, the cost of a lock, gives the systems conversation somewhere real to go.
Comp, and where to find the real numbers
DRW pays at the top of the Chicago prop-trading tier, which sits above standard big-tech software offers for comparable years of experience once the bonus is in the picture. New-grad quant and trading roles carry a base in the low-to-mid six figures with a bonus that can rival or exceed it in a good year, plus a sign-on. Engineering comp is strong but shaped more like a tech package, weighted toward base and bonus rather than equity, since there are no public shares to grant. Exact figures move year to year and hinge on desk performance, so pull current ranges from levels.fyi, Wall Street Oasis, and Glassdoor rather than trusting a figure from a blog, and treat any single data point as noisy.
One failure mode has nothing to do with math. When a trader asks something you can’t answer, the worst move is to bluff, because their whole day is spent pricing uncertainty and they will smell it instantly. Say what you know, mark where your confidence runs out, and give your best estimate anyway with the caveat attached. At a firm that prices uncertainty for a living, being calibrated about your own is part of the job description.
