Does 0DTE trading raise index volatility?
Three published answers, one tape: a day-level panel over 1,083 sessions
Whether 0DTE options destabilise the index is a live academic dispute with three camps. Brogaard, Han & Won (SSRN 4426358): a one-standard-deviation increase in the 0DTE volume share raises underlying volatility by about 9% of its mean, even after controlling for market-maker gamma. Dim, Eraker & Vilkov (SSRN 4692190): 0DTE positioning does not propagate volatility. Adams, Dim, Eraker, Fontaine, Ornthanalai & Vilkov (SSRN 5641974): 0DTE intermediation lowers volatility by 60–90 annualised basis points. All three built their samples before or during the regime's adolescence; this panel is the mature regime, measured on the same per-print SPXW tape the terminal runs on.
The panel
One row per session: realised volatility (the standard deviation of the 390 one-minute log returns, annualised — mean 10.9%), and the 0DTE share — contracts expiring that day divided by the near book's total (mean 66.2%, sd 7.2 points). Two caveats up front. The denominator is the near book of roughly 21 expiries, not the whole chain — it carries the large majority of SPXW volume, but the share is a near-book share. And the share swings mechanically with the weekday, so every regression carries day-of-week and year fixed effects plus the previous session's log volatility.
Test 1: the raises-volatility claim
The Brogaard object is contemporaneous: same-day volatility on the standardised 0DTE share, with the controls above.
| sample | days | one sd of 0DTE share ↔ log rv | t |
|---|---|---|---|
| 2022 (from Apr 14) | 174 | +0.004 | +0.1 |
| 2023 | 250 | −0.011 | −0.3 |
| 2024 | 252 | +0.046 | +1.4 |
| 2025 | 249 | −0.040 | −0.9 |
| 2026 (to Aug 20) | 158 | +0.098 | +1.5 |
| all | 1,083 | +0.006 ± 0.017 | +0.3 |
In the units of the published claim: one standard deviation of 0DTE share corresponds to +0.6% of mean volatility, against a published +9%. Signs alternate year to year; no year is significant. Whatever their 2016–2022 sample carried, the association does not exist in 2022–2026 SPX at the daily level.
Test 2: does the share at least lead volatility?
If 0DTE flow caused volatility, today's share should predict tomorrow's volatility given today's. It does not: −0.015 ± 0.017 (t −0.9) pooled. One year prints positive (2024, t +3.2), three print negative — the pooled zero is alternation, not absence of data.
The one robust fact runs the other way
Reverse the arrow — today's volatility against tomorrow's share, given today's share — and the panel finally produces a stable number: −0.14 sd of share per log-unit of volatility, t −4.3. After a volatile session the 0DTE share falls. A ratio can fall for two reasons, so the decomposition matters:
| log rv today → | elasticity | t |
|---|---|---|
| next-day 0DTE volume | −0.047 | −3.6 |
| next-day total near-book volume | −0.026 | −2.2 |
Both legs mean-revert after a volatility shock (given the day's own volume level), but the 0DTE leg retreats roughly twice as hard as the rest of the book. The folklore says volatility spikes attract 0DTE speculation; the tape says 0DTE flow steps away the next day, disproportionately.
What this says about the three papers
The raises-volatility association is absent, which is the Dim–Eraker–Vilkov side of the dispute. The Adams dampening claim is not testable here, and the honest reason is identification: their evidence comes from the staggered Tuesday/Thursday SPXW listings of April–May 2022 as a natural experiment, and this archive begins 2022-04-14 — days before the first new listing, so there is no pre-period to difference against. This page refutes the plus sign; it neither confirms nor refutes the minus.
What it is not
Day-level associations with named controls — not a causal design. The share is endogenous: volatile markets change who trades everything, and no fixed effect fully unwinds that. Nothing here is investment advice; see the Terms.
Reproduce it
The volatility half is free: every session's minute series is in its public JSON at
https://gex.live/snapshots/YYYY-MM-DD.json (fields minutes,
spot; dates at /sessions; ~7 MB per day, fetch
sequentially) — realised vol is one line of numpy over spot. The share half
needs per-expiry SPXW volume, which is not in the public files; the source used here is
the ThetaData per-print tape (thetadata.net, a paid
subscription), and any per-expiry daily volume source — OCC series-level data included —
reproduces the same ratio. Regressions are ordinary least squares with year and
day-of-week dummies; nothing is fit.
Part of gex.live research. Measured on the free session archive; every session is free to replay.