S&P 500 cross-sectional strategy

Walk-forward LGBM model, 14 momentum/volatility/skew features, autonomously trading paper since 2026-05-15.

Dashboards

Pick a ticker + index; compare buy-and-hold of the ticker, the index, and the full strategy on a shared Y axis.

Today's top 10 longs and shorts from the autonomous daily-predict cron; live IC tracker vs backtest 0.10 reference.

Self-contained simulator's equity curve, current long/short positions, trade log. 20-day rebalance, 5 bps slippage.

Run a fresh walk-forward backtest with your own choice of model, horizon, quantile, cost, and training window.

Quick history of any S&P 500 name overlaid with V_alt model decile markers and predicted-vs-realized scatter.

Data: ~822 S&P 500 tickers (95% survivorship-clean via yfinance + Tiingo). Refreshed daily by daily_predict.py cron.
Live model: LGBM regression on 14 cross-sectionally ranked features (V_alt). Predicts after each market close, logs ranked picks to Postgres.
Paper trading: 200% gross long/short, 20-day rebalance, $100k starting equity, 5 bps one-way slippage. Started 2026-05-15.
Headline backtest: Sharpe ~3.46 net of 10 bps, IC ~0.10 (panel-observed; realistic real-world ceiling ~Sharpe 1.5–2.0 after additional frictions).

Built with Streamlit, LightGBM, Grafana, Postgres, Tiingo. Updated continuously.