QuantTrader — ML-Powered Systematic Trading Platform

Production-grade quant trading dashboard with regime detection, multi-factor signal generation, Kelly criterion position sizing, and Monte Carlo risk management. Includes a paper trading simulator for model validation before deploying real capital.

Architecture

quanttrader/
├── api/
│   ├── quant/
│   │   ├── engine.py       # Core quant engine (FeatureEngineer, RegimeDetector, 4 signal generators, ML ensemble)
│   │   ├── simulator.py    # Paper trading simulator with walk-forward backtest
│   │   └── data.py         # yfinance data layer with caching + sanity checks
│   ├── analyze.py          # Vercel Python serverless: /api/analyze
│   ├── backtest.py         # Vercel Python serverless: /api/backtest
│   └── watchlist.py        # Vercel Python serverless: /api/watchlist
└── src/
    ├── app/
    │   ├── page.tsx         # Main dashboard
    │   └── api/             # Next.js API routes (local dev via Python subprocess)
    ├── components/
    │   ├── panels/          # SignalPanel, IndicatorsPanel, SimulatorPanel, WatchlistPanel
    │   ├── charts/          # EquityChart, MonteCarloChart, IndicatorGauge
    │   └── ui/              # Card, Badge, Spinner
    ├── store/trader.ts      # Zustand state
    ├── lib/api.ts           # API client
    └── types/quant.ts       # TypeScript types

Quant Methodology (based on QuantBasics.txt — Proof Trading, Feb 2026)

Features (§5: Data Normalisation)

• 14 ADV-normalised features: RSI, MACD histogram, Bollinger %B, triple EMA spreads
• Hurst exponent via R/S analysis (rolling 40-bar window)
• 12-1 month momentum (Jegadeesh-Titman), ADV volume ratio
• ATR-normalised volatility, HL range relative to ADV

Regime Detection (§9.3 clustering)

Regime	Condition	Strategy Weight
Trending Up	Hurst > 0.6, EMA spread > 0	Trend x2, Momentum x1.5
Trending Down	Hurst > 0.6, EMA spread < 0	Trend x2, Momentum x1.5
Mean-Reverting	Hurst < 0.4	MR x2.5, Trend x0.3
Volatile	ATR > 85th pctile	All x0.3, ML x1.0
Quiet	Otherwise	Balanced

Signal Generators (§10.5 decision trees, §10.6 neural nets)

• MeanReversionSignal: Bollinger %B < -0.85 + RSI < 35
• TrendFollowSignal: EMA 8/21 golden/death cross + MACD confirmation
• MomentumSignal: 12-1 month momentum with ADV volume confirmation
• MLSignal: GBM ensemble (150 trees, walk-forward CV, no look-ahead per §7.1)

Position Sizing — Half-Kelly Criterion (§3)

f* = (b*p - q) / b  *  0.5  (half-Kelly, capped at 25%)

CVaR gate: if 5th-percentile CVaR < -8%, position size is halved.

Monte Carlo Risk (§7.2)

• 500 bootstrap paths x 21-day horizon
• Bootstrap from empirical return distribution (no Gaussian assumption — §4.4)
• Reports VaR, CVaR, median/worst drawdown, P(positive)

Paper Simulator

• Walk-forward backtest with 60-bar warm-up (prevents look-ahead)
• ADV-normalised slippage: slip_bps = max(0.5, participation_rate * 500)
• Commission: $0.005/share, min $1 (IB-style)

Validated Results (2026-06-11)

Symbol	Period	Return	Sharpe	Win Rate	Max DD
NVDA	1y, $50k	+13.1%	2.08	78.9%	2.9%
AAPL	2y, $100k	+2.2%	0.42	60.5%	2.9%

Local Development

# Install Node deps
npm install

# Install Python deps
pip install -r requirements.txt

# Run dev server
npm run dev -- --port 3001

# Open dashboard
open http://localhost:3001

Deploy to Vercel

# Install Vercel CLI
npm i -g vercel

# Deploy (Python serverless functions handle /api/*)
vercel --prod

The vercel.json wires Python serverless functions for the quant API endpoints. Next.js API routes (src/app/api/) are used for local development via subprocess.

Disclaimer

This software is for educational and research purposes. Past performance of backtests does not guarantee future results. Always validate with paper trading before deploying real capital.