← Citadel Interview Insights

Citadel·Data Scientist·Technical Phone Screen·Senior

Senior
Jun 2026

Summary

Citadel data science interview that went deep into stochastic processes and trading strategy math. Four-part problem covering random walks, profit distributions, and Brownian motion limits. Felt like a quant interview wearing a data science costume.

Questions Asked (4)

Q1

For a simple symmetric random walk, consider a strategy where you open a long or short position at each step based on the direction of that step, hold all positions until the final time T, and liquidate at the terminal price. Derive the expected total profit and its variance.

Algorithms & Data StructuresTechnical Trade-offs
Author's notes

The expectation part wasn't bad once I wrote out what each position contributes: a long at time t earns S_T minus S_t, a short earns the negative.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Model the random walk as a sequence of independent steps, define the position at each step, and express the total profit as a sum of products of step signs and final position. Use linearity of expectation and properties of symmetric random walks to compute the mean and variance, possibly leveraging martingale theory or combinatorial identities.

Pro tip: Emphasize that the strategy's profit is a quadratic function of the final position, and use the fact that for a simple symmetric random walk, the expected value of the fourth power of the final position is known. This shows deep understanding and avoids lengthy combinatorial sums.

1. Define the random walk and strategy

Let S_t be the position after t steps, with S_0=0 and increments X_i = ±1 with equal probability. At each step i, take a position of +1 if X_i=+1 (long) and -1 if X_i=-1 (short), i.e., position = X_i. Hold until time T and liquidate at S_T.

2. Express total profit

The profit from the position taken at step i is X_i * (S_T - S_{i-1})? Wait, careful: The position is opened at step i based on X_i, and held until T. The profit is X_i * (S_T - S_i) because you enter at price S_i? Actually, if you open at step i after observing X_i, the entry price is S_i. Then profit = X_i * (S_T - S_i). Sum over i=1 to T. Alternatively, if you open at step i based on X_i, you might enter at S_{i-1}? The problem says 'at each step based on the direction of that step', so likely you observe X_i and then enter at S_i. So profit = sum_{i=1}^T X_i (S_T - S_i).

3. Simplify the profit expression

Note that S_T - S_i = sum_{j=i+1}^T X_j. So total profit = sum_{i=1}^T X_i sum_{j=i+1}^T X_j = sum_{1<=i<j<=T} X_i X_j. This is a sum over all pairs of steps.

4. Compute expected profit

By linearity of expectation, E[profit] = sum_{i<j} E[X_i X_j]. Since steps are independent and mean zero, E[X_i X_j] = E[X_i]E[X_j] = 0. Thus expected profit is 0.

5. Compute variance

Var(profit) = E[profit^2] since mean is 0. profit = sum_{i<j} X_i X_j. Square and take expectation. Use independence: E[X_i X_j X_k X_l] is 1 if each index appears an even number of times, else 0. Count the number of ways to get non-zero contributions. Alternatively, note that profit = (S_T^2 - T)/2? Actually, sum_{i<j} X_i X_j = (S_T^2 - sum X_i^2)/2 = (S_T^2 - T)/2. So profit = (S_T^2 - T)/2. Then E[profit] = (E[S_T^2] - T)/2 = (T - T)/2 = 0. Var(profit) = Var((S_T^2 - T)/2) = (1/4) Var(S_T^2). For simple symmetric random walk, S_T^2 has mean T and variance? Var(S_T^2) = E[S_T^4] - (E[S_T^2])^2. E[S_T^4] = 3T^2 - 2T for T>=2? Actually, for simple random walk, E[S_T^4] = 3T^2 - 2T. So Var(S_T^2) = (3T^2 - 2T) - T^2 = 2T^2 - 2T = 2T(T-1). Thus Var(profit) = (1/4)*2T(T-1) = T(T-1)/2.

Key Points to Mention

  • Linearity of expectation to show expected profit is zero.
  • Independence of steps and mean-zero property.
  • Expression of profit as sum over pairs of steps.
  • Simplification to (S_T^2 - T)/2.
  • Known moments of simple symmetric random walk: E[S_T^2]=T, E[S_T^4]=3T^2-2T.
  • Variance calculation leading to T(T-1)/2.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

Now modify the strategy so that each position is closed after just one step instead of held until the end. What are the expectation and variance of a single one-step profit, and what happens to the cumulative profit after T steps?

Algorithms & Data StructuresProduct Analytics & Metrics
Author's notes

Much cleaner than part one.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, clarify the trading strategy and the distribution of one-step returns. Then compute the expectation and variance of a single one-step profit, and finally analyze the cumulative profit after T steps by considering the sum of independent one-step profits.

Pro tip: Emphasize that if one-step profits are independent and identically distributed, the variance of the cumulative profit grows linearly with T, which is crucial for risk assessment. Also, mention that if the expected one-step profit is zero, the cumulative profit is a martingale.

1. Clarify the Strategy

Restate the modified strategy: each position is opened and closed within one step, so profit is realized immediately. Define the one-step profit random variable and its distribution.

2. Compute Expectation and Variance

Calculate the expected value and variance of the one-step profit using the given distribution or assumptions. If not specified, denote them as μ and σ².

3. Analyze Cumulative Profit

Express the cumulative profit after T steps as the sum of T independent one-step profits. Use properties of expectation and variance for sums of independent random variables.

4. Discuss Limiting Behavior

If T is large, discuss the asymptotic behavior: by the Central Limit Theorem, the cumulative profit is approximately normal with mean Tμ and variance Tσ². Mention implications for risk and scaling.

Key Points to Mention

  • Independence and identical distribution (i.i.d.) assumption for one-step profits
  • Linearity of expectation: E[sum] = sum of E[each]
  • Variance of sum of independent variables: Var(sum) = sum of Var(each)
  • Cumulative profit after T steps: mean = Tμ, variance = Tσ²
  • Central Limit Theorem for large T
  • Martingale property if μ = 0

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

Generalize the original hold-to-end strategy to a biased random walk where the probability of an up step is p, not necessarily one-half. How do the expected profit and variance change?

Algorithms & Data StructuresTechnical Trade-offs
Author's notes

The expectation is no longer zero when p is not a half, which makes sense intuitively since the walk drifts.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, restate the hold-to-end strategy and define the biased random walk with up probability p. Then derive the expected profit and variance using the binomial distribution and properties of random walks, highlighting how they depend on p.

Pro tip: Emphasize the practical implications: when p ≠ 0.5, the expected profit becomes non-zero, and the variance changes, affecting risk management. This shows you understand both theory and application.

1. Define the problem

Clarify the hold-to-end strategy: you hold the position until the end of N steps, then profit is the final position. Define the biased random walk: each step is +1 with probability p, -1 with probability 1-p.

2. Compute expected profit

Let S_N be the final position. E[S_N] = N(2p-1). So expected profit is N(2p-1), which is zero only when p=0.5.

3. Compute variance

Var(S_N) = N * 4p(1-p). This follows from the variance of a single step: Var(X_i) = 1 - (2p-1)^2 = 4p(1-p).

4. Compare to symmetric case

When p=0.5, expected profit is 0 and variance is N. For p≠0.5, expected profit is non-zero and variance is maximized at p=0.5, decreasing as p moves away from 0.5.

5. Discuss implications

Mention that a biased walk introduces a drift, making the strategy profitable if p>0.5 (or unprofitable if p<0.5). Variance decreases with bias, meaning less uncertainty but also potentially lower risk.

Key Points to Mention

  • Expected profit formula: N(2p-1)
  • Variance formula: 4Np(1-p)
  • Drift term and its effect on profitability
  • Variance is maximized at p=0.5 and decreases with bias
  • Connection to binomial distribution and random walk properties
  • Practical implications for trading strategies and risk assessment

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q4

Replace the discrete random walk with standard Brownian motion sampled at equally spaced intervals. Apply the same trading rule using the sampled increments, then discuss what happens to the expected profit and variance as the sampling interval shrinks to zero. Are there additional assumptions needed, and does the limit exist?

Technical Trade-offsSystem DesignAlgorithms & Data Structures
Author's notes

This is where I started to sweat.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, define the trading rule precisely for the discrete random walk and then adapt it to Brownian motion sampled at intervals Δt. Analyze the expected profit and variance of the strategy as Δt → 0, considering the impact of transaction costs and other frictions. Discuss whether the limit exists and what additional assumptions (e.g., no transaction costs, continuous trading) are needed for the limit to be well-defined.

Pro tip: Emphasize that in the limit, the strategy's profit may diverge or converge to a finite value depending on the rule and costs; highlighting the need for realistic assumptions like transaction costs shows practical insight.

1. Define the trading rule

Clearly state the trading rule for the discrete random walk (e.g., buy if the last increment is positive, sell if negative) and how it translates to Brownian motion increments.

2. Compute expected profit and variance for fixed Δt

For a given sampling interval Δt, calculate the expected profit and variance of the strategy using properties of Brownian motion increments (independent, normal with mean 0 and variance Δt).

3. Analyze the limit as Δt → 0

Take the limit of the expected profit and variance as Δt approaches zero. Determine whether they converge, diverge, or depend on scaling (e.g., profit per unit time).

4. Discuss additional assumptions and frictions

Address the role of transaction costs, bid-ask spread, and the feasibility of continuous trading. Explain how these affect the existence of the limit and the strategy's profitability.

5. Conclude on the limit's existence

Summarize whether the limit exists under idealized assumptions and how realistic frictions alter the conclusion, tying back to practical trading.

Key Points to Mention

  • Brownian motion increments are independent and normally distributed with mean 0 and variance Δt.
  • The expected profit of a simple momentum strategy may scale with the number of trades, potentially diverging as Δt → 0.
  • Variance of profit may also diverge or converge depending on the strategy and scaling.
  • Transaction costs and market impact are crucial; without them, the limit may be infinite or undefined.
  • The concept of quadratic variation and its role in continuous-time trading.
  • The need for a risk-neutral measure or real-world drift when evaluating expected profit.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.