82.3K
VICan you predict when volatility spikes before it happens?
A new paper suggests you might not predict direction, but you can time volatility. By modeling order flow as a Markov chain and tracking its entropy in real time, low-entropy periods show about 2-3x larger absolute moves, with no directional edge (about 45%, basically chance). The decoupling is structural: entropy is invariant under buy/sell permutation, so it cannot carry directional information by construction.
A constrained rule built on this signal shows over 1,000 bps out-of-sample across 5 walk-forward folds, driven almost entirely by timing, not direction.
One important caveat: 36 days, one asset, results concentrated in specific periods, and idealized execution assumptions. This is early evidence, not a production edge.
The real question isn’t whether it works here. It’s whether it survives across assets, regimes, and time.
Further reading:
-> Singha - « Hidden Order in Trades Predicts the Size of Price Moves »
arXiv:2512.15720
#finance #quant #trading #algotrading #stoc
@vince.quant
