Jane Street Data Analyst Case Interview: Market Microstructure, Execution Quality, and Experiment Design

This case simulates partnering with a trader/quant to diagnose and improve execution quality across multiple venues—very aligned with Jane Street’s collaborative, teaching-first culture. You’ll reason from raw trade/order-book style data, quantify edge, and design principled experiments while narrating assumptions and trade‑offs. What you’ll work on: 1) Quick calibrations (5–10 min): mental math and Fermi estimates (e.g., compute slippage per fill in bps, estimate variance reduction from batching). Expect to explain your shortcut arithmetic and error bounds. 2) Microstructure deep‑dive (25–30 min): given a small, messy dataset (e.g., timestamp, venue, side, fill_px, size, prevailing bid/ask, queue position/latency proxies), you’ll: - Define and justify execution metrics (arrival-price slippage, implementation shortfall, adverse selection after t milliseconds). - Identify confounders (tick size, volatility regime, hidden liquidity, self‑selection bias) and propose controls (time‑bucketing, venue fixed effects, robust stats like median/MAD). - Reason about conditional expectations and Bayes-style updates (e.g., post-fill price move probability conditioned on spread state and queue depth). 3) Experiment design (10–15 min): propose an A/B or switchback to test a new routing heuristic (e.g., fade quotes after stale signals). Cover unit of randomization, sample size intuition, power drivers under heavy‑tailed returns, guardrails (max drawdown, tail risk), and how to stop early without p‑hacking. 4) Communication & iteration (5–10 min): walk the interviewer through results, tradeoffs, and a next-step plan. Expect pushback; they’re looking for humility, clarity, and willingness to change your mind. What interviewers look for at Jane Street: - Precise thinking under uncertainty; ability to quantify assumptions and propagate error. - Comfort with probability, conditional reasoning, and back-of-the-envelope math tied to market intuition. - Practical data habits: define clean denominators, pick robust estimators, check for leakage, and design metrics that align with PnL. - Collaborative style: ask clarifying questions, surface constraints (latency, inventory limits), and explain reasoning as if teaching a colleague. Tools/skills that may come up (lightweight, discussion-first): - SQL-style aggregations (windowing by time/venue, joins to quotes), Python/pseudocode for feature definitions. - Diagnostic plots you’d want (spread state vs. slippage, venue FE, pre/post fill drift), even if you only sketch them. Deliverable by the end: - A clear definition of execution-quality metrics, a ranked list of likely drivers with rationale, and a test plan that Jane Street could implement safely in live trading. Evaluation rubric (implicit): - Rigor (correctness of definitions, handling of confounders), signal-vs-noise judgment, communication, and cultural fit (humble, kind, team-first).