Persistence, Premium, and the Illusion of Residual Reversion in SPY Options
SVI smile residuals converge less than a scale-matched null. The tape's P&L was variance risk premium, not signal.
Observed one-step convergence runs 7–11 points below a no-reversion null. The tape returns line-3 Sharpe 0.34 [−0.18, 0.90]; a constant short side scores 1.75.
Smile residuals are widely believed to mean-revert fast enough to trade. Against a scale-matched null they converge less than chance, because they are persistent: AR(1) φ = 0.928 hourly, half-life about 1.3 trading days.
A faithful tape then fails. Leg attribution shows what the P&L actually was: variance risk premium, collected by the short side, not signal.
What’s in the repo
The scale-matched convergence null, the capture-vs-own-spread screen, the
faithful tape including its side_mode negative control, and per-year leg
attribution. Both pre-registrations ship verbatim, with deviations recorded
rather than edited away.
Reproducibility
Not runnable as shipped — requires licensed CBOE option chain data. The scripts are published for audit, not execution: they document exactly what was computed. Anyone with an equivalent SPY hourly chain panel can point the paths at it.