[ AHQ : LAB — MICROSTRUCTURE RESEARCH ]
Research that trades and detects
AHQ Lab publishes original quantitative research — starting with market manipulation detection on Level-2 order book data. The same microstructure machinery that finds spoofing also sharpens execution.
PUBLISHED · SSRN
Market manipulation detection: spoofing & layering in Level-2 order book data
An ensemble approach combining Hawkes processes, XGBoost and autoencoders to detect spoofing, layering and quote stuffing in limit order book message data — validated on LOBSTER Level-2 feeds and synthetic adversarial data, with separate calibrations for Indian (NSE/SEBI) and US (SEC/CFTC) market structure.
- ▸Hawkes process modelling of order-flow self-excitation
- ▸Feature engineering on order placement / cancellation asymmetry
- ▸XGBoost + autoencoder ensemble for anomaly scoring
- ▸Tested on real AAPL LOBSTER message data + synthetic spoofing injections
Detection pipeline
01
Ingest
Level-2 message data — every order placement, modification and cancellation, microsecond-stamped.
02
Feature
Order-book imbalance, cancellation ratios, queue position dynamics and burst intensity per window.
03
Score
Ensemble model flags windows where placement behaviour is statistically inconsistent with genuine intent.
04
Verdict
Ranked alerts with the exact message sequences that triggered them — auditable, not black-box.
More research is coming
VRP harvesting, regime detection and execution studies are in the pipeline. Waitlist members read first.