Original title: Velocity- and Regime-Aware Detection of Intraday Options Market Manipulation, with Explainable Attribution
Alex Chen, Maria Hybinette · August 2026
Catching cheaters who manipulate financial markets during the trading day is tough. Their tracks are brief, lost in millions of quotes, and look like normal price swings. If standard tools flag them, they trigger far too many false alarms. To fix this, researchers tracked the speed of market changes instead of just price levels. They looked for a specific pump-and-crash pattern in how quickly things shift minute by minute. In an Indian index-options market, their autoencoder computer model successfully caught 10 of 10 manipulation days flagged by regulators. They also tested thinly traded U.S. stocks and found the exact same pattern shape. In fact, many extra alerts the tool raised shared a nearly identical profile with the real illegal days. This suggests those extra flags might not be system errors, but rather real manipulation that regulators simply missed. Overall, the pattern itself transfers cleanly across different markets.
Imagine listening to a song and trying to tell if the singer made a mistake. You focus on the pitch, but sometimes the singer intentionally slides between notes, making it hard to separate error from art. Here, the song is the market’s price movements, and the mistake is manipulation—hard to spot because it hides among normal volatility. The researchers found that mistakes show up as a sudden acceleration in the singer's speed, not a change in pitch.
Regulators drown in false alerts because normal volatility looks like manipulation. This work reveals a real signature—a pump-and-crash pattern visible in how fast the market changes—that appears across index options and thinly traded stocks. Although precision stays modest, the signature itself is robust and can help prioritize which alerts to investigate.
The method achieves only about 25% precision under the closed-world assumption, meaning most alerts are false positives. Conditioning detection on market regimes actually reduces recall without improving precision. The velocity magnitude does not transfer across markets, only the shape. High similarity between confirmed and unconfirmed alert attributions (cosine similarity 0.99) suggests many unlabeled days may also contain manipulation, making true performance hard to assess.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Velocity- and Regime-Aware Detection of Intraday Options Market Manipulation, with Explainable Attribution — Alex Chen, Maria Hybinette
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.