Relative morning IV across names predicts relative rest-of-day realized volatility

The pooled morning implied-volatility level already predicts the rest of the session (the lab's calibration result). This asks a different question: once the common market-wide component that drove that finding is removed, is there anything name-specific left? Put plainly, does a name whose implied volatility is elevated relative to its peers today go on to realize relatively more volatility this afternoon, beyond its own relative morning realized volatility? A second, independent dimension of information would be a genuine forecast input rather than a restatement of the first.

Tests whether the CROSS-SECTIONAL structure of IV (who is implied-expensive relative to peers) carries incremental information about relative realized volatility. Not a term-structure or skew claim; those representations are not instrumented (no IV-surface endpoint onboarded); this is the relative-IV structure available from current instrumentation. A null does not falsify term/skew claims once surfaces are onboarded.

Seven mega-cap names over seven months of sessions. Index ETFs are excluded on purpose: relative structure is only meaningful among comparable single names. Within each complete session the signal, the control and the outcome are each demeaned across the seven names, so the market-wide volatility level differences out entirely and only the cross-section remains. The signal is a name's log mean implied volatility over the first hour (09:30–10:30 ET) relative to its peers that session; the control is its relative log realized volatility over the same window; the outcome is its relative log realized volatility from 10:31 to the close. OLS with the relative morning realized-volatility control, session-clustered errors, permutation inference with 999 draws, and a preregistered positive direction.

Yes, and cleanly: beta 0.30, t = 3.94, permutation p = 0.002 over 136 sessions and seven names (952 name-sessions), with the relative morning realized-volatility control carrying its own strong effect alongside it. Relative implied volatility is not simply the market level wearing a different label. Who is implied-expensive against comparable peers this morning is a separate fact about the afternoon, and the cross-sectional structure survives the removal of the very component that produced the pooled result. This became a confirmed input to the Fair Volatility forecast, and the first evidence that the program had more than one dimension to assemble.

Formation windows close roughly a minute before outcome windows open, and measured feed publication latency consumes most of that slack. Live availability of each signal at the formation boundary is therefore tight rather than generous.

Sessions with incomplete implied-volatility inputs drop from the sample rather than appearing with imputed values. Reported session counts reflect those drops; nothing is filled in or interpolated.

This is the relative-implied-volatility structure available from current instrumentation, not a term-structure or skew claim: no implied-volatility surface is onboarded, so those representations were not tested and are neither supported nor ruled out here. Cross-sectional demeaning also means the result is silent about the market level, which is a separate question answered separately.

A discovery-substrate result: preregistered, but produced on data the research process has already seen. Queued for prospective confirmation on naturally accruing unseen data before anything graduates. This is a volatility forecasting claim only; it says nothing about direction.