BTC Multi-Timeframe Stacked Ensemble Arbitrage

The strategy combines predictions from XGBoost and LightGBM models trained on different BTC timeframes (5m, 15m, 1h) to identify mispricings in Polymarket's Up/Down markets. The ensemble approach aims to improve prediction accuracy by stacking heterogeneous model outputs. Entry occurs when the ensemble's probability estimate diverges significantly from the market-implied probability, creating an arbitrage opportunity. Exit is either at market resolution or when the edge normalizes. The repo includes no verified backtest results, but the approach is theoretically sound for capturing cross-timeframe information asymmetries. Risks include model overfitting, stale data, and rapid market moves that can eliminate the edge before execution. Implementation requires a robust data pipeline for real-time BTC price feeds and Polymarket order book data. The bot is active but unverified with no reported PnL.