Detecting market manipulation at machine speed
Market manipulation, stock market manipulation, price manipulation and pump-and-dump schemes appear on about 20 recent papers, split between the AI and finance journals. Spoofing, wash trading and coordinated pumps now happen in milliseconds and across venues, including crypto. Can detectors trained on known cases find new ones, and what do regulators miss?
What prompted this
Since October 2025, "market manipulation" appears on 10 papers, "stock market manipulation" on 5, "price manipulation" on 5, "pump-and-dump schemes" on 2 and "wash trading" on 1, split between AI journals (which build detectors) and finance journals (which study episodes). High-frequency trading and limit order books appear on 34 papers between them.
Questions
- Trained on prosecuted spoofing and pump-and-dump cases, how well do detectors flag new cases, and how many false alarms do they raise per true case at market scale?
- How much wash trading is there on crypto exchanges, by exchange, and does it respond to enforcement?
- Do coordinated pumps organized on social media leave signatures in order flow before prices move?
Data that could help
Enforcement case records with dates and instruments, order book data (LOBSTER, exchange feeds), crypto trade data and social media archives.
Next step
Train a detector on the order book patterns of prosecuted spoofing cases and measure its alarm rate on a month of ordinary trading in the same stocks.