Every dealer-positioning story assumes the same mechanism: customers force option delta onto dealers, dealers hedge it in the underlying, the hedge moves the index. The premise is stated everywhere and measured almost nowhere — the canonical footprint study (Hu 2014) is equities, pre-0DTE, daily grain. So we measured it on SPX 0DTE directly: when a customer imbalance lands this minute, how many minutes until the index has absorbed the hedge?
Setup. 1,088 SPX sessions (2022–2026) at one-minute grain. The dealer's option-delta inventory is a published series on our terminal; its minute change, negated, is the delta dealers must hedge that minute. Regress forward index returns at 1–30 minute horizons on that imbalance, with day and minute-of-day fixed effects, reversal controls, and the imbalance's absolute size as a magnitude control. Day-clustered errors; last 40 sessions held out.
Answer: two to three minutes.
| horizon | coef, bp per 1σ | t |
|---|---|---|
| 1 min | +0.038 | +4.4 |
| 2 min | +0.050 | +4.2 |
| 3 min | +0.048 | +3.5 |
| 5 min | +0.021 | +1.2 |
The footprint is front-loaded exactly as a hedging story requires — strongest in the first two minutes, statistically gone by the fifth. On the held-out 40 sessions the sign is right at all seven horizons tested.
The placebo is the point. Rerun everything with one change: every minute's imbalance keeps its exact magnitude and gets a coin-flip sign. The placebo predicts nothing, at any horizon, in either sample. The information is not in how much traded — it is in which way the signing engine says it traded. Random signs kill the result; the tape-signed book carries it.
What this is not: 0.04bp per standard deviation is not a trading edge after any spread. It is a measurement of the hedging machine's latency — and, as far as we can find, the first published figure of its kind for SPX.
Full study with the reproduce recipe (it runs entirely off our free public session files): How fast do dealers hedge?
