KS
← Back to projects
CompletedJun 15, 2024 — updated Feb 10, 2025

X_QUANT: INSTITUTIONAL-GRADE QUANTITATIVE TRADING PLATFORM

90K+ line Python platform for systematic quantitative trading with ML pipeline, live IBKR paper trading, and CPCV/PBO validation achieving Sharpe 1.236.

lines 90K+sharpe 1.236strategies 18
pythonmachine-learningquantitative-financexgboostlightgbmibkrfastapi
View on GitHub →

Overview

X_Quant is a production-grade quantitative trading platform designed to compete with institutional-level hedge fund systems. Built from the ground up in Python, it implements a complete algorithmic trading pipeline from data ingestion through execution to risk management.

The system achieves an out-of-sample Sharpe Ratio of 1.236 with a CAGR of 16.97% net of costs — validated through institutional-grade statistical methods (CPCV, PBO, DSR) to guard against overfitting.

Currently running 24/7 as a live paper trading daemon connected to Interactive Brokers.

Architecture

┌─────────────────────────────────────────────────────────┐
│                    DATA INGESTION                       │
│   Market Data API → Normalization → Feature Store       │
└────────────────────────┬────────────────────────────────┘
                         ▼
┌─────────────────────────────────────────────────────────┐
│              ALPHA GENERATION (30 signals)               │
│   Technical · Microstructure · Fundamental · Cross-Asset │
│   Universal shift(1) anti-lookahead guard               │
└────────────────────────┬────────────────────────────────┘
                         ▼
┌─────────────────────────────────────────────────────────┐
│           ML PIPELINE v4.3 (6-model ensemble)           │
│   137 raw → 411 normalized → IC selection → ensemble    │
│   XGBoost · LightGBM · CatBoost · Ridge · Lasso · EN   │
└────────────────────────┬────────────────────────────────┘
                         ▼
┌─────────────────────────────────────────────────────────┐
│              SIGNAL GENERATION (MetaEnsemble)            │
│   7 base strategies → 6 sub-ensembles → IR-boost        │
└────────────────────────┬────────────────────────────────┘
                         ▼
┌──────────────────────┬──────────────────────────────────┐
│   EXECUTION ENGINE   │        RISK MANAGEMENT           │
│   Almgren-Chriss     │   VaR/CVaR (4 methods)           │
│   TWAP / VWAP / SOR  │   5-level drawdown controls      │
│   IBKR Live Gateway  │   Circuit breakers + regime       │
└──────────────────────┴──────────────────────────────────┘
                         ▼
┌─────────────────────────────────────────────────────────┐
│              MONITORING (24/7 daemon)                    │
│   Prometheus metrics → Grafana dashboards               │
│   Position tracking · PnL · Exposure · Alerts           │
└─────────────────────────────────────────────────────────┘

The platform follows a modular monolith architecture with 32 core modules.

Strategies

18 strategies across multiple timeframes and styles:

  • Trend-following — Moving average crossovers, breakout systems, channel strategies (daily/weekly)
  • Mean-reversion — Bollinger band, RSI-based, pairs trading (intraday/daily)
  • Momentum — Cross-sectional momentum, sector rotation (daily/weekly)
  • Statistical arbitrage — Cointegration-based, spread trading (intraday)
  • ML-driven — Ensemble predictions from the 6-model pipeline (daily)

11 of 18 strategies pass all 12 institutional validation checks.

Validation

The system uses institutional-grade statistical validation:

  • CPCV — Combinatorial Purged Cross-Validation (12 folds, 5-day purge)
  • PBO — Probability of Backtest Overfitting: 0.01 (target < 0.30)
  • DSR — Deflated Sharpe Ratio: 1.189 (target > 1.0)
  • Hansen's SPA Test — Superior Predictive Ability
  • Walk-Forward Efficiency — 71.18% (target > 70%)

Key Results

  • Out-of-Sample Sharpe Ratio: 1.236
  • CAGR (net of costs): 16.97%
  • Maximum Drawdown: 13.05%
  • Win Rate: 68.7%

Production Notes

  • Monitoring — Prometheus metrics exported for Grafana dashboards: PnL, exposure, drawdown, signal quality
  • Testing — 450+ pytest tests, CI via GitHub Actions on every push
  • Deployment — Docker containerized, runs as systemd daemon on dedicated server
  • Live trading — 24/7 paper trading on IBKR with automated position reconciliation
  • Alerting — Slack notifications for circuit breaker triggers, drawdown warnings, system errors

Tech Stack

Python, NumPy, Pandas, XGBoost, LightGBM, CatBoost, scikit-learn, Optuna, asyncio, multiprocessing, Interactive Brokers API, PostgreSQL, Docker, Prometheus, Grafana, pytest