@quantpad: Order book imbalance tested on 566 million Level 2 depth updates on E-mini S&P 500 futures, measuring whether resting size at the bid and offer predicts the next few minutes of price. Order flow traders are right, and this is the part nobody shows you. Imbalance is a genuine predictor: the information coefficient is +0.072 one second out, t-stat 15.5, and it is positive in 99% of 70 sessions. That is about as statistically solid as anything in intraday futures gets. Then it decays. IC falls from 0.072 at one second to 0.039 at five seconds, 0.015 at thirty, and 0.0019 at five minutes, where it is statistically indistinguishable from zero. The book predicts the next few seconds, not the next few minutes. The economics are what kill it. The edge is worth about 0.13 ticks per trade, stable across both halves of the window at 0.1269 and 0.1212. The round trip costs 1 tick, and we measured that rather than assumed it: the median spread was exactly 1.000 tick in 70 of 70 sessions. We swept 252 versions across every threshold and holding time. One beat a 1-tick cost. None beat 2 ticks. Pick that one winner using only the first half of the data and it makes −$9.66 a trade on the untouched second half. We also tested the version order flow traders actually defend, not a strawman. Absorption, where large resting size keeps getting hit and does not disappear, is real: imbalance is 2.7 to 3.1 times more informative when it fires. It still adds 0.00008 to R squared. And depth weighting across all ten levels made things worse, not better, because the information lives at the touch. Buy and hold made $9,638 on one contract over the same window with two decisions. The 10-level order book history is included with QuantPad. There is no free version of this dataset. #QuantPad #orderflow #marketmicrostructure #quanttrading #marketdata

QuantPad

Open In TikTok:

Region: US

Thursday 20 August 2026 23:53:47 GMT