HackTheRounds Interview Experiences

Jane Street Quant Researcher Interview Experience (2026) - Optimal Stopping, Market Making & Stat Arb, Pending

Jane Street QR full loop: three phone screens (dice stopping, Avellaneda Stoikov, Bayesian + order book), five round VO with stat arb and probability, result pe

By Anonymous · 2026-03-16

Background

I spent more time prepping for Jane Street than for any other loop on my board, and after three phone screens plus a full-day virtual onsite, I still do not know my result. Writing this the day after the final round, while the detail is fresh. I come from a stats background with an applied-math masters and two summers of quant internships, and Jane Street's bar for probability and strategy depth was visibly above every other shop I interviewed at. The rooms are dense, the pace is fast, and nearly every interviewer had a PhD. If you are preparing for Jane Street, here is the full loop structure and the specific flavor of each question I got.

Timeline

Phone Screen 1: Probability and Strategy (60 min)

The opener was an optimal stopping problem framed as a dice game. The setup: you roll a 20-sided die up to 100 times. After each roll, you can take the current face value as your payoff and stop, or continue rolling. Design the strategy that maximizes expected payoff.

My first instinct was to pick a threshold and take on any roll above it. The interviewer immediately pushed: "Why that threshold?" The right framing is backward induction. With one roll remaining, your expected value is 10.5 (mean of uniform 1 through 20), so on the penultimate roll you should take only if the current face is strictly above 10.5, meaning 11 or higher. That gives you a new expected value for having the penultimate roll, which becomes the threshold for the roll before that, and so on. The thresholds converge quickly, which is the insight the interviewer was checking for. I worked it out live to around 14 for the early rolls and 11 for the very last.

The follow-up reframed it as a rerolling game: each time you take, the die auto-rerolls, and the interviewer and I compete. That folded into a Stackelberg-style analysis where my threshold and the interviewer's threshold become coupled. I did not get to a closed form in the time, but I framed the equilibrium correctly.

This question is coming soon to HackTheRounds.

Phone Screen 2: Market Making and Python (75 min)

Live coding in Python, plus a market-making modeling conversation underneath. The prompt: "You are the market maker for an asset with volatility sigma . Given current inventory I , quote a bid and ask price around the mid."

I opened with the Avellaneda-Stoikov framing: reservation price equals mid minus an inventory-skew term, spread equals a volatility-linked base plus a risk-aversion-linked adjustment. The interviewer wanted working code, so I wrote a skeleton function that took mid, sigma, inventory, and a risk parameter, and returned (bid, ask) . No numeric calibration, just the structure.

The follow-up: "Add a parameter for the remaining time in the trading session. How does the spread behave as T - 0 ?" The correct answer is that spread narrows to the base volatility term because inventory risk vanishes as the horizon closes. I added the T factor to the inventory term, explained the intuition, and we discussed adverse selection scenarios.

This question is coming soon to HackTheRounds.

Phone Screen 3: Bayesian + Order Book Design (60 min)

Two problems back-to-back. First was a straight Bayesian update: two boxes, one with two white balls and one with one white and one black. Pick a box uniformly at random, pull a white ball. What is the probability the pulled ball came from the two-white box? Standard Bayes, answer is 2/3. I wrote the full conditional.

Second was a design problem: implement an order book supporting insert, cancel, and best-price queries. I used two heaps, a max heap for bids and a min heap for asks, plus a hashmap from order id to order details so cancels are O(log n) lazy deletions. Insert is O(log n) , best-price query is O(1) amortized after popping stale entries off the heap top. The follow-ups were about HFT-specific optimizations: memory pools for order structs, aligned cache lines, avoiding malloc in the hot path, and timestamp consistency across feeds.

Practice it: [[problem/564?company=34|Simple Order Book]]

Virtual Onsite

Five rounds, spread across a single day, with short breaks.

Round 1 — Statistical Arbitrage

The prompt: trade a price divergence between two historically correlated stocks. Outline: test cointegration (Engle-Granger or Johansen), estimate the long-run mean and deviation half-life, set entry thresholds at a multiple of the spread's standard deviation, exit at mean reversion, stop-loss on regime breaks, size positions to cap drawdown, backtest out-of-sample. The pushback was regime change. My answer: monitor the rolling half-life and exit when it extends past a threshold.

Round 2 — Probability Brain Teaser Sequence

Five probability questions back-to-back. The one I remember clearest: three points uniform on a unit circle form an obtuse triangle with probability 3/4, because they form one if and only if all three lie on some semicircle. Others were Penney's game, a Bayesian double-headed coin, and a bus waiting-time paradox.

Round 3 — Game Theory Coding

Implement a game where two players alternate taking from a pile of integers. It reduces to DP on subarrays, each entry the current player's optimal score minus the opponent's. Python with memoization, O(n^2) time and space. Follow-up was whether sorted values admit a simpler closed form.

This question is coming soon to HackTheRounds.

Round 4 — Market Microstructure Discussion

No coding. A conversation on how limit order book dynamics generate adverse selection, what toxic flow looks like, and how to detect it. I leaned on the "informed trader vs noise trader" decomposition and named trade imbalance and order arrival clustering as signals. The interviewer pushed on the queue-position race, and I made sure I could articulate price-time priority versus pro-rata.

Round 5 — Open-Ended Research

Pick a published market anomaly, critique it, propose an implementation. I picked momentum and walked through its decay since 2000, transaction-cost erosion, and how regulation and microstructure shift the anomaly's shape. This was a discussion, not a right-answer test.

Behavioral Wrap-Up

Four short conversations afterwards: "Why quant," "Describe a high-pressure decision," "Describe a complex problem you solved," and one I liked, "What is something you changed your mind about in the last year?" Jane Street reads the behavioral rounds seriously. Vague answers get dinged.

Result

Still waiting. Timeline from the recruiter is "10 business days" after the final behavioral call.

Tips

  1. Memorize backward induction for optimal stopping. The 20-sided die problem is Jane Street's favorite opener. Practice it so you can derive the thresholds live in under five minutes. Cold-reciting the answer 11 without explanation gets you cut.
  2. Know Avellaneda-Stoikov well enough to code a stub of it. The reservation price and the inventory-adjusted spread are the two pieces you must be able to articulate and code live. Even a toy version with four parameters is sufficient if you can explain how each term moves.
  3. Cointegration is the statistical-arbitrage vocabulary. Not "correlation." Jane Street specifically tests whether you can articulate why correlation is insufficient and cointegration is the right statistical construct for pairs trading. Be able to explain the Engle-Granger procedure.
  4. Drill the classic probability pool hard. Three points on a circle, Penney's game, biased coin with conditional probability, waiting-time paradox, and birthdays in a circle are all in the Jane Street pool. You will get at least one. Have the closed-form answer and the derivation ready.
  5. Treat the behavioral chats as equal weight. Four separate behavioral conversations is unusual. Jane Street uses each to cross-check your answers. Keep your story consistent across all four. Contradicting yourself on motivation or timeline between chat one and chat four is a rejection signal.
  6. When asked about market microstructure, name price-time priority versus pro-rata. Saying "the book matches orders" is vague. Saying "the book uses price-time priority on most equity exchanges but pro-rata on some treasury futures venues" signals domain depth. Jane Street graders listen for vocabulary.

Jane Street's loop is a filter for research-style thinking under pressure. If you can derive the right formula from first principles under a stopwatch, defend it with follow-ups, and stay calm across five back-to-back technical rounds, you will do well. If any of those pieces is shaky, it will show.