ramseshk
7517163142
Track 1 — Grid param sweep in vbt_runner:
- _generate_signals accepts params dict: grid_levels, spacing_bps, rebalance_every
- run_strategy passes params through to signals
- _strategy_params reflects actual runtime params
- Grid param sweep results: spacing is critical, levels don't matter
Tight spacing (1-2bps) = 1 trade, positive EV
Wide spacing (20bps+) = many trades, negative EV
Candle simulation can't model grid MM fills accurately
Track 6 — quant/optimizer.py:
- ParanOptimizer: automated IS/OOS parameter walk-forward
- add_param() to define parameter grid
- Composite score: Sharpe × sqrt(trades) for robustness
- IS optimization per window, OOS testing per window
- WFParamWindow + OptimizerReport with consistency + stable params
Grid MM walk-forward results (3 windows):
W0: IS S=-3.75 → OOS S=+2.23 (+12.7%, 1t)
W1: IS S=+2.52 → OOS S=-3.49 (-19.0%, 11t)
W2: IS S=-3.30 → OOS S=0.00 (0t)
Consistency: 33.3%, Stable params: {levels=5, spacing=1bps, rebalance=5}
Verdict: Candle-based grid MM is fundamentally unreliable.
Real fills require queue simulation with L2 data.
FTDT Quant Lab
Production multi-strategy quant trading system running on Hyperliquid.
Live testnet node, paper trading simulator, historical backtesting, and real-time dashboard.
Live: https://ftdt.io/cv
Stack
| Layer | Technology |
|---|---|
| Runtime | Python 3.13 (async trading) |
| API Client | nautilus_trader (Hyperliquid SDK, Rust bindings) |
| Dashboard | Next.js 16 (static export) + shadcn/ui + Framer Motion |
| Design System | Hallmark Cobalt — Ubuntu font, hairline borders, cool paper palette |
| Reverse Proxy | Caddy → auto HTTPS |
| WebSocket | FastAPI (live/paper streaming) |
| Data | PostgreSQL 17 (ftdt_quant), JSON metrics files |
| Backtesting | Custom dollar-bar engine + numpy |
| Infra | OVH VPS (4 vCPU, 8GB RAM, Debian 13), 2GB swap |
~5,300 lines of Python + TypeScript. 67 commits since July 2026.
Repository Structure
ftdt-quant-lab/
├── live/
│ ├── node.py # Live trading node — testnet, 9 strategies
│ └── paper_trader.py # Paper trading — mainnet data, 10 strategies
├── strategies/
│ ├── orderbook_imbalance.py # L2 bid/ask volume skew (OBI)
│ ├── iceberg_detection.py # Whale TWAP accumulation detection
│ ├── funding_arb.py # Delta-neutral carry — spot/perp funding
│ ├── pairs_trading.py # BTC/ETH ratio Z-score (1.5σ)
│ ├── avellaneda_stoikov.py # Dual-sided stochastic control MM
│ ├── kalman_pairs/ # Kalman-filter adaptive hedge ratio
│ ├── hawkes_ofi.py # Hawkes process order flow
│ ├── deep_lob.py # Deep LOB CNN feature extraction
│ ├── queue_imbalance.py # Weighted queue dynamics
│ ├── hurst_vpin.py # Hurst exponent + VPIN directional
│ ├── hurst_vpin_live.py # Lightweight Hurst/VPIN for live tick stream
│ └── quant_report.py # QF-Lib style quant analytics
├── dashboard/
│ ├── server.py # FastAPI backend — WS, REST, static files
│ └── next/
│ └── src/
│ ├── app/ # Main page + layout
│ ├── components/ # QuantReport, StrategyCard, L2Terminal
│ └── lib/ # Types, API client
├── backtests/
│ ├── run.py # Backtest runner
│ └── results/
│ └── historical/ # JSON backtest snapshots (32 entries)
├── common/ # Shared utilities
│ ├── risk.py, risk_manager.py
│ ├── hyperliquid_api.py
│ └── portfolio.py, metrics.py
├── config/
│ └── fee_tiers.py # Perp/spot fee schedules
└── infrastructure/
├── Caddyfile # Reverse proxy config
└── systemd/ # Service units (pending)
Strategies — Current State
Live Node (Hyperliquid Testnet — 9 strategies, $100 each)
| # | Strategy | Type | Asset | Size | PnL | Trades | Win |
|---|---|---|---|---|---|---|---|
| 1 | Order Book Imbalance | reversal | BTC | 0.000200 | $0.00 | 0 | — |
| 2 | Iceberg Detection | momentum | BTC | 0.000210 | $0.00 | 2 | 0% |
| 3 | Funding Rate Arb | carry | BTC | 0.000220 | $0.00 | 0 | — |
| 4 | Pairs Trading | stat_arb | ETH | 0.006000 | +$0.74 | 9 | 67% |
| 5 | Avellaneda-Stoikov | market_making | BTC | 0.000230 | -$1.35 | 32 | 0% |
| 6 | Momentum Breakout | momentum | ETH | 0.000500 | $0.00 | 0 | — |
| 7 | Mean Reversion | reversal | ETH | 0.000500 | $0.00 | 0 | — |
| 8 | Kalman Pairs | stat_arb | ETH | 0.005000 | $0.00 | 0 | — |
| 9 | Hurst VPIN | momentum | BTC | 0.000240 | $0.00 | 0 | — |
Execution: GTC POST-ONLY limit orders. Signals every 5 ticks (5s), dual-sided for A-S.
Fee model: Maker 0.02% (testnet).
Paper Trader (Hyperliquid Mainnet data — 10 strategies, $100 each)
Same set + Queue Imbalance. Real mainnet orderbook + funding data. Fee model: taker 0.05% / maker 0.02%. Trades simulated with 1bps slippage.
Historical Backtests
32 backtest snapshots across 8 strategies × 4 coins (BTC, ETH, HYPE, VVV).
Hurst/VPIN BTC: 46 trades, 96% win rate, +1.10% on synthetic trending data.
Priority Analysis
Strategies showing real signal
| Strategy | Signal | Status |
|---|---|---|
| Pairs Trading | ✅ | +$0.74, 67% win rate — only profitable live strategy |
| Avellaneda-Stoikov | ⚠️ | 32 trades but losing — spread capture not covering fees |
| Iceberg Detection | ⚠️ | 2 trades — rare signals, needs threshold tuning |
| Hurst VPIN | 🔬 | 96% win in backtest, 0 live trades — very selective |
| Mean Reversion | ⏳ | 0 trades — VWAP deviation not crossing 1.0σ |
| Momentum | ⏳ | 0 trades — Bollinger 1.2σ too tight for ETH |
Recommendation: focus investment here
-
Pairs Trading —
#1 priority. Only live winner. Extend to more pairs (SOL, ARB, OP). Add Kalman dynamic hedge ratio. This is the clearest path to sustained PnL. -
Hurst/VPIN —
#2 priority. Backtest shows strong edge (96% win). Needs real market data (not synthetic) and 3-day candle feed to trigger more signals. The selectivity IS the edge — don't dilute it. -
Avellaneda-Stoikov — Needs inventory control. 32 trades losing because adverse selection. Add skew-aware quoting (update reserve price based on queue imbalance).
-
Iceberg Detection — Lower detection threshold. Currently requires 7/10 consecutive ticks same direction — too strict.
-
Funding Rate Arb — Real Hyperliquid funding data already plumbed. Test threshold from 3% → 1% APR. Prefunding detection (predict next rate before announcement).
-
Backtest engine — Replace synthetic data with real Hyperliquid candles. Add walk-forward optimization. The
hurst_vpin.pyinfrastructure is ready.
Skip for now
- OBI / Mean Reversion / Momentum — 0 trades. Signal thresholds need fundamental redesign, not just tuning.
- Cartea-Jaimungal / Gueant MM — academic models, not adapted to crypto microstructure.
- DeepLOB / Hawkes OFI — dependency-heavy, no live integration.
Next Steps
# Clone and deploy
git clone https://git.ftdt.io/rams/ftdt-quant-lab.git
cd ftdt-quant-lab
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt # (pending — currently manual)
# Start services
python live/node.py & # Trading node
python live/paper_trader.py & # Paper simulator
python dashboard/server.py --port 9175 # Dashboard backend
Roadmap
- Docker Compose for reproducible deployment
- Walk-forward backtest on real Hyperliquid candle data
- Extend Pairs Trading to BTC/SOL, BTC/ARB
- Hurst/VPIN 3-day candle feed → real live signals
- Memory leak proofing — current guard at 512MB RSS
- systemd service unit files for auto-restart
- Grafana + Prometheus monitoring dashboard
Built with Hermes Agent · Hallmark Cobalt · Ubuntu fonts