Jane Street Probability Questions: What They Actually Test
Jane Street's interview process is famous for probability questions, and infamous for how differently it treats them compared to other firms. Understanding that difference is worth more than memorizing another fifty problems.
The style: estimation, updating, and "are you sure?"
Most firms ask a probability question to check you can compute. Jane Street asks to watch you think. Three patterns recur across their phone screens and onsites:
1. Questions with a cheap answer and a deep answer. Take the classic: "I flip two coins. At least one is heads. What's the probability both are?" The cheap answer is 1/3. The deep answer engages with how you learned that at least one is heads — if you saw one coin land heads, the answer is 1/2, not 1/3. Jane Street interviewers reliably probe this second layer. If you state 1/3 without qualifying the information structure, expect "how did I find out?" as the follow-up.
2. Making markets on quantities. Rather than "what's the expected number of X," you'll often get: "Make me a market on the number of times a die shows 6 in 100 rolls." Now you need not just the mean (16.67) but a sense of the spread (standard deviation ≈ 3.7) to quote a bid-ask you'd be willing to trade against. They may then trade with you and ask you to update your quote. This is a probability question wearing trading clothes — and it's testing whether your uncertainty estimate is calibrated, not just your point estimate.
3. "Are you sure? Want to bet?" After your answer, the interviewer offers a wager at some odds. This is a calibration check. Taking a bad bet to look confident is worse than declining; declining a great bet suggests you don't actually believe your own reasoning. The right response is to compute whether the offered odds beat your credence, out loud.
A worked example at interview depth
You roll a fair die repeatedly. What's the expected number of rolls until you've seen every face at least once?
The setup is the coupon collector problem. Having seen $k$ distinct faces, the probability the next roll shows a new one is $(6-k)/6$, so the wait for the next new face is geometric with expectation $6/(6-k)$:
$$E = 6\left(\frac{1}{6} + \frac{1}{5} + \frac{1}{4} + \frac{1}{3} + \frac{1}{2} + 1\right) = 6 \times \frac{49}{20} = 14.7$$
At most firms, saying "14.7" ends the question. At Jane Street, it starts the interesting part:
- "Is the distribution symmetric around 14.7?" No — it's right-skewed; the last face can take a long time. The median is below the mean.
- "Make me a market on it." You want a sense of spread. The variance is dominated by the last one or two coupons; a reasonable quote is wider than intuition suggests.
- "Now the die is biased. Direction of the effect?" Any bias increases the expected time — the rare face dominates the wait. Being able to argue the direction without recomputing is exactly the skill being tested.
How to prepare for this specific flavor
Do fewer problems, more deeply. For every problem you solve, force yourself through three extensions: What if the parameter changes — which direction does the answer move? What does the distribution look like around the expectation? Would I bet on my answer at 2:1?
Practice saying probability out loud. The interview is conversational. Candidates who compute silently for ninety seconds and emit an answer do poorly even when the answer is right. Narrate your decomposition: "I'll condition on the first flip… the two states are…"
Get comfortable being wrong mid-flight. Interviewers deliberately introduce doubt after correct answers. The skill is distinguishing "I made an error" from "I'm being tested" — which comes only from having genuinely understood, rather than recalled, your solution.
The Quant Ladder question bank is built for this style of preparation: every problem carries a worked solution written to interview depth, including the standard follow-ups, and the adaptive engine keeps you at the difficulty where you're learning rather than reciting.