Figure 1 — Equity Curve (2021-2025, 44-ticker Top-5 RVOL) — Shingan 5m/2R/1.5 Sharpe 1.21 · Paper §6.2
Anything added dilutes everything else. — tw93
Shingan is the hybrid that survives where hourly fade dies. Hourly sweep (2011-2026, 5m) loses 24% to 99% every hour, Sharpe -0.29 to -2.39 — archived to models/legacy/. See INCIDENT_REPORT.md.
| Model | 2021-2025 · 44 Top5 | 2024 | Sharpe | PF | MaxDD |
|---|---|---|---|---|---|
| Shingan 5m · RVOL≥1.5 φ[0.6,2.0] 2R | 1262 · +108.7% | 321 · +53.0% | 1.21 | 1.25 | 16.8% |
| Valkyrie 15m same filters | 2252 · +101.9% | 257 · +40.6% | 0.80 | 1.17 | 15.8% |
| Zarattini naive 5m | 6275 · +220% | 1260 · +70% | 0.37 | 1.04 | 60.1% |
| SPY naive 15m | -95% | — | -2.14 | — | — |
No RVOL → Sharpe 0.37 vs 1.21, DD 60% vs 16.8% (SHINGAN_Hybrid_ORB_Paper.md:186). Full WFO in SHINGAN_Hybrid_ORB_Paper.md.
| Figure 3 — Heatmap: OR × TargetR × RVOL (2024, φ[0.6,2.0]) — 5m/2R/1.5 peak 2.69 | Figure 2 — Drawdown: Hybrid vs Parents (2021-2025) |
|---|---|
 |
 |
| RVOL 1.5 is the knee — Paper §6.1 | Hybrid 16.8% vs Valkyrie 15.8% vs Naive 60.1% — Paper §6.2 |
pip install -r requirements.txt
python3 scripts/run_model_backtest.py --or-minutes 5 --target-r 2.0 --min-rvol 1.5 --top-k 5 --start_year 2024 --end_year 2024
# 321 trades · 53.03% · PF 1.69 · Sharpe 2.59 · DD 9.87%
# 2021-2025: --start_year 2021 --end_year 2025 → 1262 trades +108.7% Sharpe 1.21
# heatmap: --heatmap · tests: python3 -m unittest discover tests -v
Figure 4 — Monthly Returns — 1 trade/day across 44 names, β 0.05 to SPY
| Layer | Rule |
|---|---|
| Selection | RVOL = vol_OR / mean(vol_OR 14d) ≥ 1.5 and φ = spread/ATR14 ∈ [0.6,2.0] — rank 44, take Top5 |
| Signal | 5m C≥O → LONG · 15m low_first+green → LONG else REJECT trap — cuts 34% false breakouts |
| Bracket | high+0.01 / low-0.01 stop, SL opposite, TP entry ± 2R, BE at +1R, 15:00 +1% lock, 15:55 flat — 100% cash |
models/hybrid_orb.py:5 · swarm_runner.run_hybrid_portfolio · 150L, ponytail ultra 293→150L · 5 fade models in models/legacy/ kept for audit.
Why not hourly fade: $0.30 SL → 1,600 shares on $500 risk → $75 fee (18% of risk) → 2,700 trades → -$240k. Only Backtester.run() counts. See INCIDENT_REPORT.md.
shingan/
├── engine/ data_loader · backtester · metrics · swarm_runner
├── models/ hybrid_orb 150L · legacy/ 5 archived
├── scripts/ run_model_backtest.py — one CLI
└── tests/ 6 tests
- Zarattini et al. 2024 SSRN 4824172 · 4416622 — 5m OR, 1,000 stocks, Sharpe 2.40
- Valkyrie 2026 — STRATEGY.md · Live Executor
- Shingan 2026 — Paper · PDF · ArXiv
Shingan — Clear-Eyed Lens.