Detect and quantify wash trading

Quick Overview

This question evaluates a candidate's skills in fraud detection analytics, feature engineering from order and trade logs, graph-based identity inference, model calibration and backtesting for market-manipulation detection.

Detect and quantify wash trading

Company: Coinbase

Role: Data Scientist

Category: Analytics & Experimentation

Difficulty: hard

Interview Round: Technical Screen

Define an analytics approach to detect and quantify wash trading on BTC‑USD and thin‑liquidity altcoin pairs on a centralized exchange. Detail: a) features from order/trade logs (self‑matches, rapid round‑trips, size/price mirroring across linked accounts, order‑book position churn), b) graph heuristics to infer common control (shared devices/IPs, on‑chain funding links) with privacy‑preserving hashing and strict access controls, c) thresholds that separate legitimate market making from manipulation (include precision/recall trade‑offs), d) backtesting using synthetic injected wash trades plus any available enforcement ground truth, e) a daily risk score with confidence intervals and calibration checks, and f) how to avoid penalizing bona fide liquidity providers and how you would surface cases to Compliance for review.

Quick Answer: This question evaluates a candidate's skills in fraud detection analytics, feature engineering from order and trade logs, graph-based identity inference, model calibration and backtesting for market-manipulation detection.

|Home/Analytics & Experimentation/Coinbase
Coinbase logo
Coinbase
Oct 13, 2025, 9:49 PM
hardData ScientistTechnical ScreenAnalytics & Experimentation
9
0

Detecting and Quantifying Wash Trading on a Centralized Exchange

Context

You are designing an analytics approach for a centralized exchange to detect and quantify wash trading on BTC‑USD (deep liquidity) and thin‑liquidity altcoin pairs. Assume access to:

  • Full order/trade logs (orders, cancels, fills, order book snapshots), per account.
  • Device/IP/session metadata and permissioned KYC signals.
  • On‑chain deposit/withdrawal addresses linked to accounts.
  • Maker/taker fees, rebates, and any incentive program data.
  • Historical enforcement outcomes (if available).

State any minimal additional assumptions you need.

Task

Define a comprehensive approach that addresses:

a) Features from order/trade logs to flag potential wash trading, including:

  • Self‑matches, rapid round‑trips, size/price mirroring across linked accounts, and order‑book position churn.

b) Graph heuristics to infer common control among accounts (e.g., shared devices/IPs, on‑chain funding links), implemented with privacy‑preserving hashing and strict access controls.

c) Thresholds that separate legitimate market making from manipulation, including discussion of precision/recall trade‑offs and differences between BTC‑USD and thin‑liquidity pairs.

d) Backtesting methodology using synthetic injected wash trades plus any available enforcement ground truth.

e) A daily risk score with confidence intervals and calibration checks.

f) Safeguards to avoid penalizing bona fide liquidity providers and how to surface cases to Compliance for review.

Loading comments...