← Citadel Interview Insights

Citadel·Data Scientist·Technical Phone Screen·Senior

Senior
Jun 2026

Summary

Citadel Data Scientist interview with a pretty intense stochastic processes question. The kind of thing that feels manageable until you're actually in the room trying to derive variance on the fly.

Questions Asked (1)

Q1

A stock price starts at 0 and follows a simple symmetric random walk (steps of +1 or -1). At each time step T, you buy one share if the price went up and short one share if it went down. Given that positions accumulate across time, derive the variance of your total cumulative profit at time T. Follow-ups: what's the expected profit and why, and how does this extend to continuous-time Brownian motion?

Algorithms & Data StructuresTechnical Trade-offs
Author's notes

Spent the first couple minutes just making sure I understood the position accumulation mechanic correctly, which I think was the right call because it's easy to misread.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by formalizing the random walk and the trading strategy, then express the cumulative profit as a sum of products of past and current steps. Compute the variance by analyzing the covariance structure of these products, and finally discuss the expected profit and continuous-time extension.

Pro tip: Emphasize that the profit is a quadratic function of the random walk, and use the fact that the steps are independent to simplify the variance calculation. This shows you can handle complex stochastic processes efficiently.

1. Define the random walk and trading strategy

Let S_t be the price at time t, with S_0=0 and S_t = sum_{i=1}^t X_i, where X_i are i.i.d. with P(X_i=1)=P(X_i=-1)=1/2. At each time t, you buy 1 share if X_t=1 and short 1 share if X_t=-1, so your position after time t is X_t.

2. Express cumulative profit

The profit from the position taken at time t is X_t * (S_T - S_t) for t < T, and the position at time T yields no profit yet. So total profit at time T is P_T = sum_{t=1}^{T-1} X_t (S_T - S_t) = sum_{t=1}^{T-1} X_t sum_{i=t+1}^T X_i.

3. Simplify profit expression

Swap sums to get P_T = sum_{i=2}^T X_i sum_{t=1}^{i-1} X_t = sum_{i=2}^T X_i S_{i-1}. Alternatively, note that P_T = (S_T^2 - T)/2, since S_T^2 = sum X_i^2 + 2 sum_{i<j} X_i X_j = T + 2 sum_{i<j} X_i X_j, and P_T = sum_{i<j} X_i X_j.

4. Compute variance

Using P_T = (S_T^2 - T)/2, and S_T^2 ~ chi-squared with 1 degree of freedom scaled? Actually S_T is sum of T i.i.d. ±1, so S_T^2 has mean T and variance 2T? Wait, Var(S_T^2) = E[S_T^4] - (E[S_T^2])^2. For simple random walk, E[S_T^4] = 3T^2 - 2T? Let's derive: S_T = 2B - T where B~Bin(T,1/2). Then S_T^2 = 4B^2 -4BT + T^2. E[B]=T/2, E[B^2]=T/4 + T^2/4, E[B^3]= T/8 + 3T^2/8 + T^3/8? Actually use known: E[S_T^4] = 3T^2 - 2T for T>=1? Check T=1: E[S_1^4]=1, 3-2=1 ok. T=2: S_2^4: values 0,4,0,4? Actually S_2 can be -2,0,0,2 with probs 1/4,1/4,1/4,1/4. E[S_2^4]= (16+0+0+16)/4=8. 3*4-2*2=12-4=8 ok. So Var(S_T^2)= (3T^2-2T) - T^2 = 2T^2 - 2T = 2T(T-1). Then Var(P_T) = Var((S_T^2 - T)/2) = (1/4) Var(S_T^2) = (1/4)*2T(T-1) = T(T-1)/2.

5. Address follow-ups

Expected profit: E[P_T] = E[(S_T^2 - T)/2] = (T - T)/2 = 0. This makes sense because each trade has zero expected profit due to martingale property. Continuous-time: replace random walk with Brownian motion B_t, strategy: buy if dB_t>0? Actually in continuous time, the strategy is to hold sign of dB_t? But dB_t is infinitesimal, so more appropriate: hold position equal to sign of previous increment? The discrete analogue leads to P_T = (B_T^2 - T)/2, and variance T^2/2? Actually for Brownian motion, if we define profit as integral of sign of past increments? The continuous limit of the discrete strategy yields P_T = (B_T^2 - T)/2, which has variance T^2/2? Let's check: Var(B_T^2) = 2T^2, so Var((B_T^2 - T)/2) = T^2/2. Expected profit 0.

Key Points to Mention

  • Random walk properties: independent increments, symmetric steps, variance T.
  • Profit expression as sum of products of past and current steps.
  • Simplification to (S_T^2 - T)/2 using algebraic identity.
  • Variance calculation using moments of S_T (E[S_T^4] = 3T^2 - 2T).
  • Expected profit is zero due to martingale property.
  • Continuous-time extension: Brownian motion, profit = (B_T^2 - T)/2, variance T^2/2.

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