Arrowstreet Capital Interview Process & Prep

Candidate-reported (several 2025-26 accounts): CodePair, 'interactive but really just doing problems': conceptual Python questions plus two LeetCode easy/medium problems (the one remembered: 'count the words in a list that have exactly k distinct letters').

Slicing (negative indices, out-of-range bounds, views vs copies, list/NumPy/pandas indexing); @staticmethod vs @classmethod (what each receives, instance/class calls, inheritance with cls). Pandas: merge (join types, keys, duplicated keys), groupby (agg/transform/apply), rolling (window semantics, min_periods, centered), resample (up/down-sampling, label, closed). The must-do set below rebuilds all of it on small frames with verified outputs, plus both coding problems on the judge.

Candidate-reported (pooled from four 2025-26 accounts): the onsite opens with a paper test of statistics and asset pricing, closed-form derivations not computation. The five must-do problems below are original worked versions, numbers verified.

Topics: mean-variance (tangency portfolio, capital market line). Risk parity / ERC (w_i (Sigma w)_i = w_j (Sigma w)_j, two-asset closed form, not inverse-vol unless correlations are equal). X on Y vs Y on X (slopes differ, product is R-squared). Autocorrelation (N_eff about N(1-rho)/(1+rho) for AR(1), t-stats, Bartlett / Newey-West). Log utility (certainty equivalent, growth-optimal sizing). Pricing a payoff under a given utility / SDF.