Research library

2024-01-02 to 2026-08-14

The independent SPX 0DTE edge search

Eight causal ideas, real SPXW prices, conservative fills—and no candidate earned access to confirmation.

VerdictNO EDGE FOUND

In simple words

What happened

Seven candidates lost money or failed stability rules. T1 showed a small positive average, but the typical trade still lost 50%, 2025 lost 50% per play, modest parameter changes turned negative, and its control side was never defined. It is un-runnable, not a hidden winner.

Who would pay?

Each hypothesis named a possible structural payer: traders overpaying for familiar levels, underpricing expansion after compression, or failing to confirm an SPX move in SPY/VIX. Those stories only matter if the real option population beats time-matched and opposite-side controls.

How it was tested

Exact next-minute SPXW opens, 10% adverse entry and exit execution, stop-first grading, one leader per candidate per day. Missing synchronized structure marks stayed unresolved.

What would prove it wrong?

Negative 2024 results, weak medians, poor Wilson intervals, control failure, missing-state rates, unstable sensitivities, or unavailable raw-event multiplicity were each enough to refuse promotion.

Leader-collapsed evidence

What each setup actually did

657 fixed sessions; discovery used 219 mod-3 strip-A sessions

SetupLeadersMeanMedianHit rateControlsDecision
V1Directional continuation after cheap implied move16 / 16 resolved-20.4%-50.0%25.0%No-feature -18.0%; opposite -10.9%REFUSED
V2Buy a compressed ATM straddle141 / 78 resolved-25.3%-35.0%7.7%No-feature -32.4%REFUSED
P1Continue after level acceptance103 / 101 resolved-22.7%-50.0%23.8%No-feature -17.6%; opposite -10.7%REFUSED
P2Reverse after level rejection13 / 13 resolved-20.9%-19.7%23.1%No-feature -17.4%; opposite -29.6%REFUSED
T1Lunch compression, then held break15 / 15 resolved+5.4%-50.0%40.0%Undefined by preregistrationUNRUNNABLE
X1SPX break confirmed by SPY and VIX48 / 46 resolved-23.4%-50.0%21.7%Partial confirmation -4.1%; opposite -22.9%REFUSED
X2Reverse an SPX break confirmed by neither market1 / 1 resolved-50.0%-50.0%0.0%Partial confirmation -18.1%REFUSED
R1Buy relatively cheap 10-point wings138 / 138 resolved-22.1%-35.0%15.2%ATM straddle -23.0%; no-feature -20.2%REFUSED

No modeled premiums

Real option-price audit

Observed debit is the real next-minute open. Stressed debit adds the binding 10% adverse fill.

2024-01-02V1
O:SPXW240102P04735000
Observed
$8.64
10% stress
$9.504
Outcome
-50.0% stop
2024-01-02V2
O:SPXW240102C04735000 + O:SPXW240102P04735000
Observed
$19.24
10% stress
$21.164
Outcome
-35.0% stop
2024-01-02P1
O:SPXW240102P04735000
Observed
$8.7
10% stress
$9.57
Outcome
-50.0% stop
2024-01-05P1
O:SPXW240105C04715000
Observed
$9.3
10% stress
$10.23
Outcome
-1.5% timed exit
2024-01-05R1
O:SPXW240105C04695000 + O:SPXW240105P04675000
Observed
$7.18
10% stress
$7.898
Outcome
+50.0% target
2024-01-24R1
O:SPXW240124C04905000 + O:SPXW240124P04885000
Observed
$4.17
10% stress
$4.587
Outcome
+50.0% target

What we keep

Learnings

  1. 01

    A plausible market story is not an edge. The price paid for the option dominated every directional narrative tested here.

  2. 02

    Buying both sides did not solve timing. The compressed ATM straddle lost 25.3% on resolved trades after conservative execution.

  3. 03

    Level acceptance and cross-market confirmation were worse than their controls, so the feature added no usable information.

  4. 04

    Missing data must enter the denominator. V, X, T, and R all had explicitly counted missing-state sessions.

  5. 05

    The classifiers did not retain raw event counts; raw event counts were not retained anywhere else. The report says unavailable and treats that as a promotion failure; it never copies leader counts into a fake raw column.