ORBIT is a four-layer quantitative trading platform designed for physical alpha in ERCOT energy markets, augmented by a macro-aware oil overlay. Traditional HFT firms exploit symmetric-information markets through latency. Energy markets contain structurally asymmetric physical information — weather physics, grid topology, generation constraints, and policy-regime shifts — that speed alone cannot capture.
The platform fuses:
- 57ns execution core — competitive with top-tier HFT, targeting SCED bid timing in ERCOT RTC+B
- NLP intelligence layer — semantic extraction from market notices, weather narrative cross-checking, confidence-scored trade signals
- Macro regime engine — rolling signal decomposition, Bayesian regime classification, Jacobian sensitivity bridge from slow macro signals to fast HFT parameters
- VPP dispatch optimization — physical battery degradation curves, ORDC scarcity pricing, AS allocation weighting
All figures from rigorous walk-forward validation. In-sample period used for signal development only; out-of-sample period (2019–2026) was never touched during construction.
| Metric | Value |
|---|---|
| Full-sample Sharpe (2006–2026) | 1.55 |
| Full-sample 90% CI | [1.04, 2.04] |
| Full-sample ann. return (active periods) | +35.6% |
| In-sample Sharpe (2006–2018) | 1.93 |
| Out-of-sample Sharpe (2019–2026) | 0.71 |
| Out-of-sample Information Coefficient | +0.314 |
| IC decay in → OOS | −1.6% (near-zero — signal is live) |
| Max drawdown (full sample) | 9.9% |
| Rebalance frequency | 10-day (regime-conditional) |
| Capital deployment | ~20% of time (regime-gated) |
| All-weather blended return (est.) | ~10–11% annualized |
OOS context: The 2019–2026 holdout contained two unprecedented oil volatility events (COVID demand shock to −$37/bbl, Russia-Ukraine spike to $130). OOS annualized vol was 74% vs 14% in-sample. The signal maintained its predictive IC (+0.314, essentially unchanged from +0.309 in-sample), confirming the edge is real; the Sharpe penalty is variance-of-the-underlying, not signal degradation. A vol-targeting wrapper is the natural next layer.
Signals are derived from publicly available macro data (FRED, EIA) using rolling-window correlation decomposition. No proprietary data sources are required for the macro overlay.
Key relationships modeled:
- Momentum drift proxy (s2): 30-day WTI log-return normalized to [0,1]
- Regime conflict detector (s7): rolling 90-day Pearson ρ(DXY, WTI) — the petrodollar inverse relationship; a positive reading signals supply-stress regime
- Yield coupling (s8): 18-month ρ(WTI, 10Y Treasury) — inflation pass-through signal
Using the Grinold-Kahn fundamental law:
IR ≈ IC × √N
where IC is the Pearson correlation between signal value at time t and realized forward return at horizon h, and N is annual bet count. Regime-conditional ICs (easing-only) reach IC = 0.31–0.36 at the 10-day horizon, which is unusually high for a macro signal.
Size scales continuously with regime clarity:
regime_clarity = regime_confidence × (1 − e7_conflict_penalty)
position_scale = clamp((clarity − floor) / range, 0, 1)
Hard stops and yellow-zone caps prevent trading into regime conflict. When the regime conflict signal exceeds threshold, the system stands aside entirely.
The bridge solves:
ΔP = J(S) · Δs
where S is the slow macro signal vector and P is the HFT parameter vector. Partial derivatives map:
- Solar forecast error → bid-ask spread width
- Momentum drift → maximum position reduction
- ORDC scarcity → AS allocation weight shift
| Series | Provider | Endpoint |
|---|---|---|
| WTI Crude (daily) | FRED | api.stlouisfed.org — DCOILWTICO |
| Broad USD Index (DXY) | FRED | DTWEXBGS |
| 10-Year Treasury | FRED | DGS10 |
| Federal Funds Rate | FRED | FEDFUNDS |
| ERCOT Real-Time Prices | ERCOT MIS | mis.ercot.com |
| Weather / Solar | NOAA | api.weather.gov |
Register for a free FRED API key at fred.stlouisfed.org/docs/api/api_key.html.
Macro regime panel — real-time regime classification, rolling signal vector, Sharpe projection
ERCOT dispatch view — VPP position, DA/RT spread, ORDC scarcity signal
Physical oil & gas overlay — supply-stress regime detection, Jacobian parameter state
Policy state examples — HOLD (e7 hard-stop active) and ACTIVE with position scale
ORBIT Platform/
├── core/ HFT decision core, 57ns loop, memory allocator
├── execution/ Order routing, position management, kill-switch
├── intelligence/ NLP signal extraction, RAG cross-checking
├── strategy/ Macro context engine, policy engine, walk-forward backtest
├── bridge/ Jacobian calculator (slow→fast parameter bridge)
├── viz/dashboard/ React + Vite terminal dashboard
├── data/ Macro context, policy state, backtest results
├── backtest/ Historic nodal replay configs
└── configs/ ERCOT fee parameterization, tuning profiles
- Macro regime engine (FRED live data, rolling signals)
- Policy engine with regime-gated position sizing
- Jacobian bridge (macro → HFT parameter vector)
- Walk-forward backtest (2006–2018 in-sample / 2019–2026 holdout)
- React terminal dashboard with live policy state
- ERCOT historical SPP bulk ingest (e3 signal will be gathered over time)
- Vol-targeting wrapper on active periods
- Vercel deployment
ORBIT Platform — Physical Alpha through Regime-Aware Quantitative Intelligence