Two Sigma Technical Round 3 Interview: Stats & Math

The Two Sigma Technical Round 3 interview is stage 4 of 6 in the Two Sigma interview process, after the Technical Round 2. Format: 45 min · whiteboard. Two Sigma's signature round, heavy on regression theory: derive the OLS estimator, set up a maximum-likelihood problem, walk through bias and variance, compare OLS with weighted least squares, and reason about asymmetries such as regressing Y on X versus X on Y and omitted-variable bias.

What the Technical Round 3 tests

Expect follow-ups that change an assumption and ask what breaks.

QuantVault maps 143 practice problems to this round. By topic, the biggest groups are expectation (36), regression (32), statistics (26), probability (21); by difficulty, 7 easy, 74 medium and 62 hard. 142 of them were reported by candidates who interviewed at Two Sigma, most recently in Jun 2026.

How to prepare for the Technical Round 3

• Re-derive OLS β̂ on a napkin until reflexive. Same for $\text{Var}(\hat\beta) = \sigma^2 (X^\top X)^{-1}$.
• Know the asymmetry. OLS slope of $Y$ on $X$ is $\hat\beta_{Y\sim X} = \text{Cov}(X,Y)/\text{Var}(X)$; slope of $X$ on $Y$ is $\text{Cov}(X,Y)/\text{Var}(Y)$. Their product equals $R^2$.
• Three core MLE setups they reuse: Bernoulli, Normal-mean known-variance, and Uniform. If those are reflex, Pareto-MLE / AR(1)-MLE / GBM-MLE follow the same pattern.
• WLS in one line: OLS minimizes $\sum (y_i - x_i^\top\beta)^2$; WLS minimizes $\sum w_i (y_i - x_i^\top\beta)^2$. When $w_i = 1/\sigma_i^2$ (inverse-variance), WLS is BLUE under heteroskedasticity. Be ready to derive when WLS = OLS and why.
• Ridge $\lambda$ derivation: $\hat\beta_{\text{ridge}} = (X^\top X + \lambda I)^{-1} X^\top Y$. Optimal $\lambda$ falls out of bias-variance MSE minimization OR cross-validation. Know both arguments.
• Omitted-variable bias formula: if true model is $Y = \beta_1 X_1 + \beta_2 X_2 + \epsilon$ but you fit $Y = \gamma X_1$, then $\mathbb{E}[\hat\gamma] = \beta_1 + \beta_2 \cdot \text{Cov}(X_1, X_2)/\text{Var}(X_1)$. The DIRECTION of bias matters.
• Cold-start framework for unfamiliar territory. Identify the four primitives — variables, constraints, objective, data — then commit to exploring ONE branch deeply rather than gesturing at the whole problem. Senior interviewers care that you can navigate ambiguity, not that you know the domain.
• Most senior interviewers respect this much more than half-confident hand-waving.

Further reading: Behavioral questions at trading firms · How quant superdays work · Market-making game interview guide.

Question types in this round

The 5 largest question types in the Two Sigma Technical Round 3 practice set; across them the most common sub-topics are expected value, variance, conditional probability, independence, matrix operations, discrete probability. The questions themselves, with worked solutions, are in the round practice.

Prepare with QuantVault

The same round at similar firms

Compare with similar firms