← Bank of America Interview Insights

Bank of America·Software Engineer·Technical Phone Screen·Senior

SeniorPrefer not to say
Jun 2026

Summary

Quant researcher interview with a pretty gnarly statistics question about estimating volatility from a discrete Brownian motion path. One question, no fluff, just math.

Questions Asked (1)

Q1

You observe a Brownian motion at integer times t=1 through 100, giving values X1 to X100. Define H as the running maximum, L as the running minimum, and C as the terminal value X100. After appropriate rescaling or bias correction, which of these is the best estimator of sigma: (i) |C|, (ii) |H-L| minus |C|/2, or (iii) (H-L)^2 + C^2?

Technical Trade-offsData ModelingProduct Analytics & Metrics
Author's notes

This wrecked me a little.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, clarify that the question is about estimating the volatility parameter sigma from discrete observations of a Brownian motion. Then, analyze each estimator's statistical properties, focusing on bias and consistency, and determine which one is the best after appropriate rescaling or bias correction. Conclude that (ii) is the best estimator because it effectively estimates the range-based volatility, which is more efficient than using only the terminal value.

Pro tip: In interviews, emphasize that while |C| is a simple estimator, it is very noisy; the range-based estimator (H-L) uses more information from the path and, after bias correction, provides a much more efficient estimate of sigma. This shows you understand the trade-off between simplicity and statistical efficiency.

1. Understand the problem and definitions

Restate that X_t is a Brownian motion with volatility sigma, observed at integer times 1 to 100. H and L are the running maximum and minimum, and C is the terminal value X_100. The goal is to estimate sigma.

2. Evaluate estimator (i): |C|

Note that C ~ N(0, sigma^2 * 100), so |C| has expectation sigma * sqrt(100) * sqrt(2/pi). After rescaling, it is an unbiased estimator of sigma, but it has high variance because it ignores the path information.

3. Evaluate estimator (ii): |H-L| - |C|/2

The range R = H - L of a Brownian motion over [0,T] has expectation sigma * sqrt(8T/pi). Thus, R is an estimator of sigma up to a known constant. The term |C|/2 is a bias correction to make the estimator unbiased for sigma. This estimator uses the full range and is more efficient.

4. Evaluate estimator (iii): (H-L)^2 + C^2

This estimator is not directly proportional to sigma; it involves squared terms, so its expectation is proportional to sigma^2, not sigma. After taking square root and rescaling, it might be an estimator, but it is not as straightforward and likely less efficient.

5. Compare and conclude

Compare the estimators in terms of bias, variance, and efficiency. The range-based estimator (ii) is known to be highly efficient for estimating volatility from discrete data. Therefore, (ii) is the best estimator after appropriate rescaling or bias correction.

Key Points to Mention

  • Brownian motion properties: C ~ N(0, sigma^2 T), range expectation E[H-L] = sigma * sqrt(8T/pi).
  • Bias correction: The term |C|/2 in (ii) corrects for the bias in the range estimator.
  • Efficiency: Range-based estimators use more information from the path and have lower variance than terminal-value-based estimators.
  • Estimator (iii) involves squared terms, so it estimates sigma^2, not sigma, and requires additional transformation.
  • Rescaling: All estimators need appropriate scaling by sqrt(T) or constants to estimate sigma.
  • Practical consideration: In real applications, range-based volatility estimators (e.g., Parkinson) are preferred for their efficiency.

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