MARKET STRUCTURE VALIDATION
Does Contessa Describe Real Intraday Market Behavior?
This report evaluates structural accuracy rather than strategy PnL. The goal is to test whether Contessa regimes identify larger moves, volatility separation, reversal pressure, and meaningful zero-gamma or wall-bias states.
Top 1% absolute 5-minute moves occurred in negative-gamma regimes.
Negative gamma represented this share of all intervals.
Negative-gamma 5-minute absolute move size versus positive gamma.
Top-decile versus bottom-decile next-interval absolute move size.
CURRENT SAMPLE
13,104 Complete-Session Snapshots Across 168 Trading Days
The sample spans Aug 11, 2025 - May 20, 2026, uses 12,761 valid same-day next-interval observations, and excludes partial sessions. This makes the structural read cleaner than using every raw file indiscriminately.
- The strongest validation is regime classification, not raw up/down prediction.
- Negative-gamma and amplifying states captured nearly nine out of ten of the largest intraday SPY moves while representing about 61% of intervals.
- Zero-gamma distance and wall-bias strength both showed monotonic relationships with next-interval realized movement.
- GEX bias alone is weaker as a structural metric in this sample and should not be the headline proof point.
LARGE-MOVE CAPTURE
Negative Gamma Captured The Majority Of Tail Intraday Moves
Negative gamma represented about 61% of intervals, but captured 87.3% of top-5% absolute 5-minute moves and 89.1% of top-1% absolute moves. That is the cleanest product-validation metric currently available.
REGIME SEPARATION
Realized Volatility By Market State
If the regime labels are structurally useful, realized movement should separate across states. In the current sample, negative-gamma intervals had materially larger forward moves than positive-gamma intervals.
| Regime | Intervals | Share | Avg abs 5m | Avg abs 15m | Avg abs 30m | Top 1% move share | Lift |
|---|---|---|---|---|---|---|---|
| Negative gamma | 7,763 | 60.8% | 5.56 bps | 9.33 bps | 13.01 bps | 89.1% | 1.46x |
| Positive gamma | 4,982 | 39.0% | 3.50 bps | 5.95 bps | 8.36 bps | Baseline compare | n/a |
| Amplifying dealer behavior | 7,769 | 60.9% | 5.56 bps | 9.33 bps | 13.00 bps | 89.1% | 1.46x |
MONOTONICITY
Signal Strength Should Map To Market Behavior
Monotonicity is harder to fake than a single threshold. Zero-gamma distance and wall-bias strength both showed larger next-interval absolute moves in their top deciles than in their bottom deciles.
| Signal | Bottom decile move | Top decile move | Move lift | Bottom reversal | Top reversal | Read |
|---|---|---|---|---|---|---|
| Zero-gamma distance | 3.82 bps | 8.08 bps | 2.12x | 48.0% | 51.6% | Strong positive relationship with realized movement. |
| Wall-bias strength | 4.09 bps | 7.93 bps | 1.94x | 46.4% | 51.9% | Higher wall strength maps to larger moves and more reversal. |
| Chain GEX magnitude | 4.05 bps | 7.33 bps | 1.81x | 48.2% | 51.2% | Useful for volatility/reversal context, weaker than zero-gamma and wall. |
| GEX bias magnitude | 5.23 bps | 4.07 bps | 0.78x | 47.1% | 48.8% | Not a headline structural metric in the current sample. |
EXTREME STATES
Top-Decile Signal Behavior
| Signal | Hit | Mean move |
|---|---|---|
| Zero-gamma distance | 51.7% | 8.09 bps |
| Wall bias | 51.9% | 7.93 bps |
| Chain GEX | 51.2% | 7.33 bps |
| GEX bias | 48.8% | 4.08 bps |
CRITICAL READ
What This Does Not Prove Yet
- The run covers the generated complete-session subset available now, not the final full-year sample.
- Dealer behavior currently appears mostly amplifying in this sample, so stabilizing-regime comparisons need more complete data before they are reliable.
- Structural validation measures market behavior conditional on Contessa states; it is not the same as a live execution backtest.
- The strongest claim is large-move and volatility separation, while raw directional accuracy remains close to coin flip.
BEST CUSTOMER-FACING CLAIM
Contessa identifies intraday options regimes where realized movement and tail-risk concentration are materially different.
This is a stronger data-product claim than quoting strategy Sharpe. It says the signal layer has measurable structural relationship to the market, which makes it useful for research, risk monitoring, and systematic strategy development.