a statistical-physics spoofing & layering detector
A 2023 paper treats a limit order book like a particle system: every resting order is "mass" sitting at a price, and each placement or cancellation is a flux of that mass toward or away from the mid-price. From this they build a momentum-like statistic over order-book flow and show it catches spoofing (big fake orders parked away from the touch, then cancelled before they can execute) and layering (stacked fake orders on one side) that a standard rolling Z-score anomaly detector lags on or misses outright — validated on the LUNA flash crash and Bitcoin markets. Below is a live book with both detectors running on the exact same order stream. Inject manipulation and watch the physics metric fire where the Z-score baseline stays asleep. arXiv:2308.08683
Based on: "Detecting Financial Market Manipulation with Statistical Physics Tools" — arXiv:2308.08683 (q-fin). Toy simulation; not the authors' data.
Π = Σlevels distance × (Δbid − Δask) —
summed over the whole book and weighted by lever-arm distance from the mid. A far-parked spoof is precisely
large mass × large distance appearing and vanishing, so it produces a sharp Π spike on placement
and an opposite spike on cancellation, well above its own quiet baseline. Same data, same threshold —
the difference is flow-vs-stock and the distance weighting. Everything above is computed from the real signals
crossing the real threshold; nothing is a scripted "alert."