← Citadel Interview Insights

Citadel·Software Engineer·Technical Phone Screen·Senior

SeniorPrefer not to say
Jun 2026

Summary

Citadel quant researcher interview with a pretty gnarly probability/statistics question that required you to actually derive an answer from scratch rather than just pattern-match to something you'd seen before. The expected answer was around 16 days, which at least gave me something to sanity-check against.

Questions Asked (1)

Q1

You run an LLM-based sentiment analysis pipeline and each day it produces a vector of scores over the same dataset. The pairwise correlation between any two days' outputs is about 0.95. If you average together n days of outputs into a single vector, how many days n do you need so that two independently-averaged n-day windows have a correlation above 0.99 with each other?

A/B Testing & ExperimentationTechnical Trade-offs
Author's notes

This one took me a minute to even set up correctly.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Model each day's output as a signal plus independent noise, where the correlation of 0.95 implies a signal-to-noise ratio. Then derive how averaging n days reduces noise variance by a factor of n, and solve for the n that yields a correlation of 0.99 between two independent n-day averages.

Pro tip: Explicitly state the assumptions (e.g., equal variance, independence of noise across days) and note that the result is an approximation; this shows rigor and awareness of real-world complexities.

1. Model the data

Assume each day's output vector is the sum of a fixed true signal and independent noise with equal variance. The correlation between two days is then signal variance divided by total variance.

2. Relate correlation to variance components

Given correlation ρ = 0.95, express the signal variance as ρ times the total variance, and the noise variance as (1-ρ) times the total variance.

3. Determine the effect of averaging

Averaging n independent days reduces the noise variance by a factor of n while the signal variance remains unchanged. Thus, the correlation between two independent n-day averages becomes ρ_n = signal variance / (signal variance + noise variance / n).

4. Solve for n

Set ρ_n = 0.99 and solve for n. Using ρ = 0.95, the equation is 0.95 / (0.95 + 0.05/n) = 0.99, which simplifies to n = 19.

5. Interpret and discuss

State that approximately 19 days are needed. Mention that this assumes independence and equal variances, and that in practice the required n may differ due to correlations or non-stationarity.

Key Points to Mention

  • Signal-plus-noise model for daily outputs
  • Correlation as a measure of signal-to-noise ratio
  • Variance reduction by averaging independent samples
  • Formula for correlation after averaging: ρ_n = ρ / (ρ + (1-ρ)/n)
  • Solving for n yields n = ρ(1-ρ_target) / (ρ_target(1-ρ))
  • Assumptions: independence, equal variance, stationarity

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