Jane Street's trading interview has a reputation: rounds of probability that start friendly and end somewhere past the edge of what you can hold in your head, plus games where you quote prices on things you have never once thought about. The reputation is mostly earned. It is also more learnable than the folklore suggests, because the firm tests a small set of skills with unusual consistency.
One thing before the breakdown. QuantPit is independent and not affiliated with Jane Street in any way. Everything below describes the process as candidates have publicly and repeatedly reported it, not an official syllabus, and nothing here reproduces a real question. Details shift by role, office, and year, so read this as a map of the terrain rather than a script. Our Jane Street firm page keeps the same picture next to practice sets built in the firm's style.
The shape of the funnel
The funnel candidates describe has three broad phases. Each phase tests the same core skills at higher resolution.
First contact: an assessment or a probability phone screen
Depending on role and region, candidates report either a timed online assessment of math and logic puzzles or a direct jump to a first phone interview. The phone screen is typically 30 to 45 minutes with someone from the desk. Expect probability, expected value, and quick arithmetic, starting at a level most STEM students find comfortable and ramping until it bites.
The interviewer is not grading against a checklist of right answers. From the first minute, they push on the reasoning: why that approach, how confident are you, what would change your mind. Silence followed by a bare answer scores worse than a wrong start corrected mid-flight.
The middle rounds: the same skills, sharper
Candidates who clear the screen report one to three further phone or video rounds with the difficulty turned up. The questions stay in the same family, conditional probability, multi-step expected value, games with choices, but each now carries a second layer: a follow-up that breaks your first method, a twist on the setup, an invitation to bet on your own answer.
The escalation is deliberate. The interviewer wants to find the level where you stop being sure, because that is where the real evaluation starts. Everyone eventually reaches a question they cannot finish cleanly. What happens next is the data point.
The final round: markets and games
The final stage, onsite or virtual, is several interviews back to back. Reported staples include market-making style games where you quote two-sided prices and trade against the interviewer, harder probability with layered follow-ups, and, for some roles, group exercises where candidates play trading games against each other. The firm even publishes Figgie, a card game it built to recreate open-outcry trading, which tells you how central game play is to how it thinks about evaluation.
There are usually conversations over lunch or between rounds. Treat them as professional. They may not be formally scored, but you are still at the interview.
What the interview actually selects for
Strip away the brainteasers and four traits explain most of what candidates infer about the rubric.
Probabilistic reasoning out loud. They want the live feed of your thinking, not a polished final answer. Narrating a model as you build it is the skill, and it is trainable.
Calibrated confidence. Knowing the answer matters less than knowing how sure you are. A candidate who says 70% and is right about 70% of the time is demonstrating the exact skill a trading desk runs on.
A betting mindset. Beliefs are cheap. The interview keeps converting them into action: how much would you pay, which side do you take, what odds make this fair. If your numbers never touch a decision, you are answering a different question than the one asked.
Comfort being wrong. You will be wrong at some point, by design. Candidates report that interviewers respond well to fast, ego-free updates and badly to defended wreckage. The correction is not the failure; fighting it is.
The question style, with two worked examples
Both examples are ours, written for this guide and tuned to the style candidates describe. Work them out loud, exactly as you would in the room.
Example 1: price this bet
You flip a fair coin until you get heads, with a maximum of three flips. Heads on the first flip pays $2, heads arriving on the second flip pays $4, on the third pays $8. Three tails pays nothing. What should you pay to play?
Decompose by where the game ends:
- Heads immediately: probability , pays $2, contributes $1.00.
- Tails then heads: probability , pays $4, contributes $1.00.
- Tails, tails, heads: probability , pays $8, contributes $1.00.
- Three tails: probability , pays nothing.
Sanity check before announcing: the four probabilities sum to 1, and each paying branch contributing exactly $1 is a pattern worth pointing out as you go. Fair value $3.
Then take the step that separates candidates: turn the value into decisions. You would happily pay $2.80 for $0.20 of expected edge per play, you would pass at $3.40, and if asked to quote a market you might say 2.75 bid, 3.25 offered, the tight width reflecting low uncertainty about a fully specified game. Expect the follow-up: what if the flips are uncapped and payouts keep doubling? Now the expectation diverges, which is exactly why real bets have caps, bankrolls, and variance limits, and saying so is worth more than the original answer. Build that reflex at /topics/expected-value/.
Example 2: a conditional probability trap
I roll two fair dice where you cannot see them. I look at both and tell you, truthfully: at least one die shows a 4. What is the probability the sum is 8?
Two wrong answers arrive fast. ignores the information entirely. reasons that the other die just needs to be a 4, which feels airtight and is not, because "the other die" is not a well-defined object when either die could be the 4.
Condition properly. Of the 36 ordered outcomes, those containing no 4 number , so 11 equally likely outcomes contain at least one 4. Among those 11, the only one summing to 8 is (4,4). Pairs like (2,6) and (3,5) reach 8 without a 4, so the announcement eliminated them.
About 9%, not 17%. And there is a deeper layer the strongest candidates reach: the answer depends on how I decided what to tell you. If I had instead picked one die at random and read it out, and it happened to show a 4, the remaining die would be uniform by symmetry and the answer really would be . Asking "how did you choose what to reveal?" is precisely the kind of move the later rounds exist to find. Drill this family until it is boring at /topics/probability/.
How "make me a market" questions are graded
At some point an interviewer will ask you to quote a market on something you do not know: the weight of a commercial aircraft, the number of windows in a skyscraper, the total length of paved roads in a country. You name a bid, the price at which you buy, and an offer, the price at which you sell, and the interviewer trades against you.
Candidates consistently report that the final number barely matters. What gets graded is the reasoning:
- Structure. Decompose the unknown into pieces you can estimate, multiply through, and check the order of magnitude before you quote anything.
- Width that matches uncertainty. A tight market on something you genuinely know, a wide one on something you do not. Quoting tight to look confident is the classic blowup.
- Two-sided honesty. If you would never actually sell at your offer, it is not your offer. You should be close to indifferent about which side of your quote the interviewer takes.
- Updating on flow. If the interviewer keeps buying from you, your fair value is probably too low, so raise it. Refusing to move reads as stubborn. Lurching wildly reads as uncalibrated. Smooth, reasoned adjustment is the target.
A workable script: estimate out loud, state a range, quote around the middle, then adjust as the trades come in. The mechanics turn natural with repetition, which is what our structured tracks and market games are for.
Confidence intervals and the calibration habit
A reported favorite: give a 90% confidence interval for some quantity, meaning a range you believe contains the truth with 90% probability. Most untrained people produce intervals that contain the truth barely half the time. Overconfidence is the human default, and this question is a calibration meter wearing a trivia costume.
Calibration is built, not willed. The habit: for every estimate in practice, force the interval first, log it, and score yourself over dozens of attempts. If fewer than nine in ten of your 90% intervals contain the truth, widen until honesty returns. Timed estimation screens automate exactly this loop, and one question a day at /daily/ keeps the sample growing.
Calibration also lives in your sentences. "I am about 70% on this" is a statement an interviewer can grade later. "Probably" is not.
The classic mistakes
Rushing to a number
Five seconds of thought followed by a blurted answer signals that you optimize for looking fast. The interview rewards brisk structure: a moment to choose an approach, then visible, steady progress. Fast and wrong with no recovery plan is the worst quadrant.
Refusing to commit
The opposite failure is hedging forever and ending on "it depends" with no number attached. Trading is deciding under incomplete information, so "I would need more data" is an anti-signal. Give the number, state your uncertainty around it, and let the interviewer attack it. That is the job.
Ignoring base rates
The most common technical leak. Suppose a pattern flags profitable trades: only 5% of opportunities are genuinely good, the flag catches 90% of good ones, and it falsely fires on 10% of bad ones. Given a flag, the probability the trade is actually good is
A "90% accurate" signal is wrong two times out of three here, because good trades were rare to begin with. Interviewers bury this structure inside dice, cards, and coins. Whenever you hear "given that," locate the base rate before you touch anything else.
Preparing for each stage
Before the first screen. Make the probability core reflexive: counting, complements, conditioning, independence. Work through /topics/probability/ until the easy tier feels insulting, and keep arithmetic sharp with a short daily session at /daily/.
For the middle rounds. Shift weight to expected value and decision framing at /topics/expected-value/, pull harder material from the full question bank, and add time pressure with company-style screens.
For the final round. Simulate the room. Run full mock interview loops speaking every step aloud, play market-making games until quoting feels mechanical, and follow a structured track so coverage is deliberate rather than random. If your timeline is measured in weeks, our 4-week prep roadmap sequences all of this day by day.
FAQ
Do I need a finance background for the Jane Street trading interview?
No, and candidates without one are reportedly common. The interview tests probability, expected value, and decision-making, not markets trivia. Curiosity about how the games connect to trading helps, but nobody expects you to price options on a phone screen.
How much mental math is actually required?
Quick, accurate arithmetic under pressure: fractions, percentages, and expected values without paper. Nobody needs party tricks with six-digit numbers. Speed comes from doing small computations every day until they stop costing attention, which is what the daily set is built for.
What happens if I get a question completely wrong?
By most accounts, nothing fatal, provided the error was honest and the update was fast. Interviewers reportedly care whether the reasoning was sound, whether your confidence matched your accuracy, and whether you absorbed the correction. Several wrong answers defended stubbornly is a different story.
Train the instinct, then walk in
Reading about the process buys you orientation; the interview pays out on reflexes. Start with the firm-style sets on our Jane Street page, drill probability and expected value until the standard traps look like old friends, then pressure-test yourself with timed screens and full mock loops. The candidates who sound calm in round three are the ones who made the room feel familiar before they entered it.