SPXW Hybrid Switch Note
A walk-forward SPXW switch study where direct option-surface filters stayed more useful than added realized-volatility estimators.
Why this note matters
This note sits next to the 0DTE SPXW iron condor case study for a reason. The case study shows what happens when a promising rule set is promoted only after walk-forward review. This note shows something slightly different: even when more sophisticated volatility features are available, the best live decision rule can still come from the option surface itself.
Benchmark ladder
The initial comparison set was simple enough to read quickly:
| benchmark | total pnl |
|---|---|
| standard full | 3,655.48 |
| narrow full | 5,717.01 |
| standard skip-vix | 5,032.98 |
| alpha skip-vix | 9,873.47 |
Those numbers were useful, but they were not enough on their own. The real question was which rule set could stay stable when the promotion standard moved from full-sample ranking to walk-forward behavior.
Stable candidate
The promoted candidate used a regime-first switch with direct option-surface conditions:
- surface richness threshold at
1.3 - minimum extra credit threshold at
5.0 - no additional realized-volatility gate required for promotion
Out of sample, that candidate produced:
+$9,108.02total test pnl85.71%positive fold rate389test trades
That result finished +$2,453.58 ahead of the standard skip-vix benchmark while preserving fold stability.
What the test rejected
The point of including realized-volatility estimators was not to prove they never matter. It was to see whether they added enough value to justify being part of the live decision rule. In this run, they did not.
That is a useful finding. It keeps the promoted rule closer to the surface features the desk can observe directly instead of adding extra complexity for weak practical gain.
What this changes on the desk
The note reinforces a working bias: use regime context early, but prefer direct option-surface evidence when it carries the real decision edge. A feature can be statistically interesting and still fail the test for operational usefulness.
That distinction matters more than squeezing one more attractive column into a research notebook.