Citadel / Citadel Securities interview questions

How Citadel / Citadel Securities's process runs, as candidates and the firm describe it, and the questions candidates report, each with a worked solution. Free, no sign-up.

Practise these in the app Mock interview

How Citadel / Citadel Securities interviews

Sources: Citadel: Investment & Trading Intern, Europe (official) · Citadel: What we do (official) · Glassdoor: Citadel quant trader

Questions (16)

Try each one before opening the solution. Interviewers change the numbers, so learn the method.

Quantitative Trader · Interview · Expected value · reported by a candidate

A fair die is rolled and you receive the cube of the result.

  1. What is the expected value?
    Answer and worked solution

    Answer: 73.5

    Key idea. Average the six payouts directly.

    1. Cubes: 1, 8, 27, 64, 125, 216. Sum = 441.
    2. E = 441/6 = 73.5.

    Check. Shortcut: 1³ + … + n³ = (1 + … + n)² = 21² = 441.

  2. Follow-up (ours): what is (E[X])³, and why is it smaller?
    Answer and worked solution

    Answer: 42.875

    Key idea. Jensen's inequality: for a convex function f, E[f(X)] ≥ f(E[X]).

    1. (E[X])³ = 3.5³ = 3.5 × 12.25 = 42.875.
    2. x³ is convex for x > 0, so E[X³] (73.5) exceeds it.

    Check. The same idea is why options gain value with volatility: their payoff is convex.

  3. Follow-up (ours; another Citadel report mentions rerolls): you may reroll once and are paid the cube of the final roll. What is the game worth?
    Answer and worked solution

    Answer: 635/6 ≈ 105.83

    Key idea. Keep a first roll whose cube beats a fresh roll's value, 73.5.

    1. 4³ = 64 < 73.5, so reroll 1–4; keep 5 (125) and 6 (216).
    2. Value = (125 + 216)/6 + (4/6) × 73.5 = 56.83 + 49 = 105.83.

    Check. The threshold is higher than with a linear payout (keep 4+), because the payoff is convex.

    Trap. Rerolling only 1–3 because "4 is above average": compare cubes, not faces.

Interviewers may push further: What is the variance of the payout?

Source: Glassdoor · Try it in the app →

Quant Trader Intern · Interview · Expected value · reported by a candidate

You roll a fair die until every face has appeared at least once.

Also reported at Da Vinci.

  1. What is the expected number of rolls?
    Answer and worked solution

    Answer: 14.7

    Key idea. Coupon collector: one stage per new face.

    1. With k faces seen, a roll is new with probability (6 − k)/6: mean wait 6/(6 − k).
    2. Sum: 6(1 + ½ + ⅓ + ¼ + ⅕ + ⅙) = 6 × 2.45.
    3. = 14.7.

    Check. For n faces it grows like n ln n.

Interviewers may push further: Roughly how does this grow for an n-sided die? (n ln n.)

Source: Glassdoor · Try it in the app →

Quant Trader Intern · Interview · Statistics · reported by a candidate

Which is more likely: 40 to 60 heads in 100 fair flips, or 400 to 600 heads in 1000?

  1. What is the probability of 40–60 heads in 100 flips (to two decimal places)?
    Answer and worked solution

    Answer: 0.9648

    Key idea. Measure each range in standard deviations.

    1. 100 flips: σ = 5, so 40–60 is ±2σ. P ≈ 0.95; with the endpoints included, exactly 0.965.
    2. 1000 flips: σ ≈ 15.8, so 400–600 is about ±6.3σ: essentially certain.
    3. So the 1000-flip range is far more likely.

    Check. The same ±10% window covers more standard deviations as n grows (σ grows like √n).

Source: Glassdoor · Try it in the app →

Quantitative Trader · Interview · Expected value · reported by a candidate

The numbers 1 to 5 are placed around a circle in random order.

  1. What is the expected number of numbers smaller than both their neighbours?
    Answer and worked solution

    Answer: 5/3 ≈ 1.6667

    Key idea. Linearity of expectation over the 5 positions.

    1. A number and its two neighbours are three distinct values in random order.
    2. P(it's the smallest) = 1/3.
    3. E = 5 × 1/3 = 5/3.

Source: Glassdoor · Try it in the app →

Semi-Systematic Trader · Interview · Market making · reported by a candidate

Make a market on the square root of 30.

  1. First, your best estimate of √30 to two decimal places.
    Answer and worked solution

    Answer: 5.4772

    Key idea. Start from a nearby square and take one Newton step.

    1. 5.5² = 30.25, slightly high.
    2. Newton: x − (x² − 30)/(2x) = 5.5 − 0.25/11 ≈ 5.477.
    3. √30 ≈ 5.48. Quote tightly around it, e.g. 5.47 @ 5.49.

    Check. 5.477² ≈ 29.998.

    Trap. Quoting very wide on something you can compute: it signals you don't trust your own arithmetic.

Source: Glassdoor · Try it in the app →

Trader · Interview · Market making · reported by a candidate

"I flip a fair coin 100 times. Make me a market on the number of heads." Then: "Now make it 10,000 flips."

  1. (a) What standard deviation do you size your 100-flip market from?
    Answer and worked solution

    Answer: 5

    Key idea. Size the market from the standard deviation.

    1. Mean = 100 × ½ = 50.
    2. Variance = 100 × ¼ = 25, so SD = 5.
    3. A first market: 48 @ 52 or so (within one SD), tightening if they trade with you.

    Check. The 95% range 40–60 is ±2 SD.

    Trap. Quoting 45 @ 55 "because it's a coin": wide enough that nobody will trade.

  2. (b) By what factor should the width of your market grow for 10,000 flips?
    Answer and worked solution

    Answer: 10

    Key idea. Width scales with √n, not n.

    1. For 10,000 flips the SD is √(10,000/4) = 50.
    2. 50/5 = 10: the market should be about 10 times as wide.
    3. Fair value went up 100 times, so relative to the price the market is 10 times tighter.

    Check. This is the law of large numbers in one line: absolute uncertainty grows, relative uncertainty shrinks.

    Trap. Scaling the width by 100 along with the fair value.

Interviewers may push further: As a fraction of the fair value, is the 10,000-flip market wider or narrower? (Narrower: 50/5000 against 5/50.)

Source: Glassdoor: Citadel trader interviews · Try it in the app →

Quantitative Trader · Interview · Expected value · reported by a candidate

You draw independent Uniform(0, 1) numbers and keep a running total.

  1. What is the expected number of draws until the total first exceeds 2? (2 dp)
    Answer and worked solution

    Answer: 4.6708

    Key idea. Let m(x) = expected draws until the total exceeds x, and condition on the first draw.

    1. For 0 ≤ x ≤ 1: m(x) = ex (the classic gives m(1) = e).
    2. For 1 ≤ x ≤ 2: m′(x) = m(x) − m(x − 1) = m(x) − ex − 1.
    3. Solving with m(1) = e: m(x) = ex − (x − 1)ex − 1.
    4. m(2) = e² − e ≈ 4.67.

    Check. Each draw adds ½ on average, so about 2/½ = 4 draws plus the overshoot: 4.67 is about right.

    Trap. Answering 2e (doubling the "exceed 1" answer). The overshoot past 1 carries into the second unit.

Interviewers may push further: And to exceed 1? (e ≈ 2.72.) Roughly how many draws to exceed a large x? (About 2x + 2/3.)

Source: Glassdoor: Citadel quantitative trader interviews · Try it in the app →

Quantitative Trader · Interview · Probability · reported by a candidate

Using two independent Uniform(0, 1) numbers U and V, produce a point uniformly distributed over a disk of radius 1.

The report asks how; the parts below check the reasoning with numbers.

  1. (a) Take angle 2πV and radius √U. What is the probability the point lands within distance ½ of the centre?
    Answer and worked solution

    Answer: 0.25

    Key idea. Uniform over area means P(R ≤ r) = r², the share of the area within r.

    1. With R = √U, P(R ≤ r) = P(U ≤ r²) = r²: correct.
    2. P(R ≤ ½) = (½)² = 1/4.

    Check. The inner disk of radius ½ has a quarter of the area.

  2. (b) If you used radius U instead, what would that probability be?
    Answer and worked solution

    Answer: 0.5

    Key idea. A uniform radius ignores that outer rings have more area.

    1. P(U ≤ ½) = 1/2.
    2. That's twice the true share: points bunch up near the centre.

    Trap. Using r = U looks natural and is the mistake the question is testing.

  3. (c) For a uniform point on the disk, what is its expected distance from the centre?
    Answer and worked solution

    Answer: 2/3 ≈ 0.66667

    Key idea. The density of R is the derivative of r²: 2r.

    1. E[R] = ∫₀¹ r · 2r dr = 2/3.

    Check. More than ½, because most of the area is near the edge.

Interviewers may push further: Another way with no square root? (Rejection: draw from the square [−1, 1]², keep it if it lands in the disk; you keep π/4 of them.)

Source: Glassdoor: Citadel quantitative trader interviews · Try it in the app →

Quant Trader Intern · Superday · Conditional probability · reported by a candidate

A bag holds four stones, each black or white; you don't know how many of each. You draw two without replacement and both are black.

The report doesn't say how the bag was filled. We assume every number of black stones from 0 to 4 is equally likely, then compare with a different assumption in (c).

  1. (a) What is the probability the next stone you draw is black?
    Answer and worked solution

    Answer: 0.75

    Key idea. Bayes over the unknown number of black stones k, each value 0–4 equally likely a priori.

    1. P(two black | k) = C(k, 2)/6: k = 2 → 1/6, k = 3 → 3/6, k = 4 → 6/6.
    2. Posterior weights: 1 : 3 : 6, i.e. 0.1, 0.3, 0.6.
    3. P(third black | k) = (k − 2)/2: 0, ½, 1.
    4. P = 0.3 × ½ + 0.6 × 1 = 3/4.

    Check. Laplace's rule of succession gives the same: (2 + 1)/(2 + 2) = 3/4.

    Trap. Answering ½: with an unknown mix, black draws are evidence that the bag is mostly black.

  2. (b) You're offered an even-money $1 bet that the next stone is black. What is your expected profit?
    Answer and worked solution

    Answer: 0.5

    Key idea. EV of an even-money bet = 2p − 1.

    1. 2 × ¾ − 1 = $0.50 per $1 staked.

    Check. A big edge: this is the moment to size up, within your bankroll (Kelly: bet 2p − 1 = 50% at even money).

  3. (c) If instead each stone had been black with probability ½, independently, what would the probability in (a) be?
    Answer and worked solution

    Answer: 0.5

    Key idea. With independent stones, past draws say nothing about the rest.

    1. The third stone is black with probability 1/2, whatever came before.

    Check. The model of how the bag was filled is everything. Ask the interviewer before you compute.

Interviewers may push further: How much would you pay to see the third stone before betting? Would you change your bet size?

Source: Wall Street Oasis: Citadel Securities interview reports · Try it in the app →

Quantitative Researcher Intern · Interview · Probability · reported by a candidate

Two people each arrive at a station at a uniformly random time between 4pm and 5pm, and each waits 10 minutes for the other.

  1. What is the probability they meet?
    Answer and worked solution

    Answer: 11/36 ≈ 0.30556

    Key idea. Unit square of arrival times; they meet when |x − y| ≤ 1/6.

    1. They miss in two corner triangles with legs 5/6: total area (5/6)² = 25/36.
    2. P(meet) = 1 − 25/36 = 11/36 ≈ 0.31.

    Check. With a 15-minute wait it's 7/16 and with 20 minutes 5/9: always 1 − (1 − w)².

Source: Glassdoor: Citadel quantitative research intern interviews · Try it in the app →

Quantitative Researcher · Interview · Conditional probability · reported by a candidate

X and Y are independent standard normal random variables, and you're told X + Y > 0.

The report's wording is garbled ("E[X>0|X+Y>0]"), so both natural readings are asked.

  1. (a) What is P(X > 0 | X + Y > 0)?
    Answer and worked solution

    Answer: 0.75

    Key idea. Rotational symmetry: probabilities are angles.

    1. X > 0 and X + Y > 0 overlap in a 135° wedge.
    2. P(both) = 135/360 = 3/8; P(X + Y > 0) = ½.
    3. P(X > 0 | X + Y > 0) = (3/8)/(1/2) = 3/4.

    Check. Above ½, as it should be: a positive sum makes a positive X more likely.

  2. (b) What is E[X | X + Y > 0]? (3 dp)
    Answer and worked solution

    Answer: 0.56419

    Key idea. Condition through the sum S = X + Y ~ N(0, 2).

    1. By symmetry, E[X | S] = S/2.
    2. E[S | S > 0] = √2 × √(2/π) = 2/√π.
    3. E[X | S > 0] = (2/√π)/2 = 1/√π ≈ 0.564.

    Check. Half of E[S | S > 0], since X and Y share the sum equally on average.

    Trap. Using √(2/π) ≈ 0.80 directly: that's E[X | X > 0], a stronger condition.

Source: Glassdoor: Citadel quantitative researcher interviews · Try it in the app →

Quantitative Researcher · Interview · Probability · reported by a candidate

X and Y are independent Uniform(0, 1). Let S = X + Y.

The report says the questions were about convolution but gives no numbers; these are ours.

  1. (a) What is the density of S at s = 1.5?
    Answer and worked solution

    Answer: 0.5

    Key idea. Convolve the two uniform densities: f(s) = ∫ fX(x) fY(s − x) dx.

    1. For 1 ≤ s ≤ 2, x must lie in [s − 1, 1], an interval of length 2 − s.
    2. f(1.5) = 2 − 1.5 = 0.5.

    Check. The triangle has total area 1: base 2, height 1.

  2. (b) What is P(S > 1.5)?
    Answer and worked solution

    Answer: 0.125

    Key idea. Area of the tail triangle beyond 1.5.

    1. Base 0.5, height 0.5: area ½ × 0.5 × 0.5 = 1/8.

    Check. By symmetry P(S < 0.5) is also 1/8.

Interviewers may push further: What does the density of the sum of three uniforms look like? (Three quadratic pieces, peak 3/4 at 1.5.)

Source: Wall Street Oasis: Citadel Securities interview reports · Try it in the app →

Trader · Interview · Options · reported by a candidate

How would you build a position that is long vega and short gamma?

Model answer

Sell short-dated options and buy long-dated ones at the same strike (sell the front-month straddle, buy the back-month straddle).

  • Gamma is concentrated in short-dated options, so the short front leg dominates: short gamma.
  • Vega grows with time to expiry (roughly with √T), so the long back leg dominates: long vega.
  • Theta: positive. The front option decays faster, and you're short it.

You profit if implied volatility rises (especially further out) while the stock stays still. You lose on a big move now. Mention that you'd delta-hedge to stay near flat.

Source: Glassdoor: Citadel trader interviews · Try it in the app →

Quantitative Research Intern · Interview · Statistics · reported by a candidate

What are the assumptions of linear regression, and how would you check each one in practice? When would you use ridge rather than lasso?

Model answer
  • Linearity (the mean of y is linear in the features): plot residuals against fitted values and against each feature; look for curves.
  • Independent errors: in time series, check residual autocorrelation (ACF plot, Durbin–Watson).
  • Constant variance: residuals fanning out against fitted values; Breusch–Pagan test. Fix with transforms, weighted least squares or robust (HC) standard errors.
  • No perfect multicollinearity: correlated features inflate coefficient variances; check variance inflation factors.
  • Normal errors (only needed for small-sample inference): Q–Q plot of residuals.
  • Exogeneity (errors uncorrelated with features): the hardest to check; think about omitted variables and look-ahead.

Ridge vs lasso. Both shrink coefficients to cut variance. Ridge (L2) shrinks smoothly and keeps every feature: use it when many correlated features each carry a little signal. Lasso (L1) sets some coefficients to exactly zero: use it when you believe few features matter and want selection. With groups of correlated features lasso picks one arbitrarily; elastic net mixes the two. Choose the penalty by cross-validation, respecting time order for market data.

Source: Wall Street Oasis: Citadel Securities interview reports · Try it in the app →

Trader · Interview · Behavioural · reported by a candidate

Behavioural questions candidates report:

Answer one out loud, as you would in the interview.

Model answer

Use a short story: the situation in one sentence, what you did (not the team), the result with a number if you have one, and what you'd do differently. Aim for about two minutes.

Source: Glassdoor: Citadel trader interviews · Try it in the app →

Trader · Interview · Behavioural · reported by a candidate

Behavioural questions candidates report:

Answer one out loud, as you would in the interview.

Model answer

Three parts, about a minute each. Why trading: a specific moment you got hooked (a game, a market you followed, a decision under uncertainty), not "I like maths". Why this firm: something concrete about how it trades or trains people, ideally from talking to someone there or from its own material. Why you: evidence you already think in probabilities and react well to being wrong. Avoid anything about money or prestige.

Source: Glassdoor · Try it in the app →

Other firms

Jane Street · SIG (Susquehanna) · Optiver · IMC · Five Rings · Two Sigma · Flow Traders · Maven Securities · Wincent · Akuna Capital · Da Vinci Derivatives · DRW · Common formats · Hudson River Trading · Jump Trading · Eclipse Trading · Old Mission · Belvedere Trading · Squarepoint Capital · Valkyrie Trading · Virtu Financial · Tibra · Tower Research Capital · D. E. Shaw · Mako · Question library

Practise by subject instead: topic guides.