v1.0.0 — Initial Open Source Release

qlib-tw-trader v1.0.0

The first public release of qlib-tw-trader — an end-to-end quantitative trading system for Taiwan stocks.

Highlights

Model & Backtesting

• DoubleEnsemble (ICDM 2020): +55% IC and +71% Sharpe over single LightGBM
• 156-week Walk-Forward backtest: Sharpe 1.724, +23.9% annualized excess return
• ~300 alpha factors: Alpha158 (109) + Taiwan institutional flow (107) + interaction (50) + enhanced (37)
• Strict lookahead bias prevention: T-day trades use T-1 features only, 7-day embargo

Full-Stack Dashboard (9 pages)

• Training: Week calendar with batch training and Optuna hyperparameter search
• Evaluation: Rolling IC charts, cumulative return curves, factor importance, CSV/JSON export
• Backtest: Walk-Forward results with per-week IC and multi-strategy comparison
• Quality: IC stability monitoring, Jaccard similarity, ICIR tracking
• Positions: Current holdings, trade history, holdings timeline
• Datasets: Multi-source data coverage and sync status

Infrastructure

• Docker support (docker compose up --build)
• FastAPI backend with Swagger/ReDoc documentation
• React 18 + Vite + TailwindCSS frontend
• 28 passing tests

Getting Started

git clone https://github.com/Docat0209/qlib-tw-trader.git
cd qlib-tw-trader
cp .env.example .env
docker compose up --build

See README for full documentation.