INTERIM BACKTEST REPORT
SPY 5-Minute Options-Positioning Backtest
This report summarizes all complete-session generated data currently available. It is designed to show whether Contessa metrics contain exploitable structure, not to present a finished trading product.
WindowAug 11, 2025 - May 20, 2026
Snapshots13,104
Trading days168
Intervals12,936
Rules tested272
Cost model1 bp one-way
READ FIRST
What The Current Data Actually Says
- The broad/simple rules weakened as the sample grew; the raw GEX and wall direction rules are not enough by themselves.
- The strongest exploratory families are reversal-oriented: zero-gamma distance reversion and wall-bias fading in high-strength regimes.
- Most high-return rules still underperformed SPY during their active windows, so excess return remains the key validation gap.
- Sharpe ratios are reported for transparency but should be treated as fragile because they are annualized from 5-minute returns and selected after exploration.
TOP EXPLORATORY RULES
Ranked By Net Return After 1 bp Slippage
These are hypothesis-generation results. Notice that several high-return rules still show negative active-window excess return versus SPY, which is why they should not be sold as finished alpha.
| Rule | Trades | Net | Excess | Sharpe | Max DD | Hit rate | Holdout net |
|---|---|---|---|---|---|---|---|
| Zero-gamma reversion, top quartile gap, negative gamma | 2,555 | +1,004 bps | -141 bps | 5.16 | -394 bps | 51.8% | +467 bps |
| Wall-bias reversal, top quartile strength, amplifying regime | 2,554 | +912 bps | -153 bps | 4.78 | -350 bps | 51.8% | +383 bps |
| Wall-bias reversal, top 15 percent strength | 1,941 | +878 bps | -187 bps | 6.18 | -257 bps | 50.7% | +354 bps |
| Hybrid: fade extreme GEX, else zero-gamma reversion | 3,710 | +871 bps | -293 bps | 3.61 | -394 bps | 50.7% | +264 bps |
| Zero-gamma reversion, top quartile gap | 3,234 | +868 bps | -371 bps | 3.90 | -394 bps | 50.7% | +331 bps |
| Hybrid: fade extreme wall, else zero-gamma reversion | 3,305 | +856 bps | -393 bps | 3.81 | -394 bps | 50.1% | +331 bps |
| Wall-bias reversal, top 20 percent strength, amplifying regime | 2,163 | +856 bps | -120 bps | 5.28 | -299 bps | 51.6% | +269 bps |
| zero reversion direct top 15 | 1,941 | +815 bps | -144 bps | 5.54 | -315 bps | 51.3% | +249 bps |
METHODOLOGY
How The Backtest Was Scored
- Signals are built from generated Contessa metric JSON files, not from direct price-only indicators.
- Each strategy acts on one 5-minute interval and compares the position to the next same-day SPY spot move.
- Days with fewer than 70 generated snapshots are excluded from the report sample to avoid partial-session distortion.
- One-way slippage of 1 bp is charged on position changes. Slippage sensitivity is shown at 0, 0.5, 1, 2, and 5 bps.
- The chronological holdout uses the final 30 percent of intervals. It is useful, but it is not a replacement for a true future out-of-sample test.
LIMITATIONS
Why These Results Stay Preliminary
- This is exploratory research, not a production trading model.
- The strategy search creates data-mining risk; best rules should be treated as hypotheses to validate, not final alpha.
- High Sharpe ratios can be inflated by intraday annualization, return clustering, and selective active windows.
- Buy-and-hold comparisons are active-window comparisons, not portfolio-level capital allocation results.
- The strongest evidence so far is that market-structure states contain measurable reversal behavior, not that a finished strategy is ready to sell.