A self-driving quantitative trading lab — paper money, real statistics. live validation phase · portfolios →
A cross-sectional machine-learning system that ranks ~1,400 US stocks every afternoon by expected 5-day relative return, trades the top decile on a paper account, and — the actual point — measures whether its own edge is real. Three gradient-boosted models (1, 5, and 21-day horizons) vote on rank-normalized price, volume, and sector-relative features. Out-of-sample rank-IC: +0.0340 (t = 4.12) across 718 walk-forward days. That's a small, statistically real edge — the kind professional quant runs on — held to honest accounting: purged cross-validation, Newey-West statistics, transaction-cost brackets, and a research kill log where failed ideas stay buried.
| # | Symbol | Score | Price | Sector |
|---|---|---|---|---|
| 1 | STX | 0.998 | $775.70 | Information Technology |
| 2 | TER | 0.996 | $393.25 | Information Technology |
| 3 | INTC | 0.994 | $106.61 | Information Technology |
| 4 | MU | 0.992 | $1,035.40 | Information Technology |
| 5 | FCX | 0.990 | $70.83 | Materials |
| 6 | FIX | 0.989 | $1,697.23 | Industrials |
| 7 | MRVL | 0.986 | $271.31 | Unknown |
| 8 | LRCX | 0.986 | $317.56 | Information Technology |
| 9 | JBL | 0.985 | $297.19 | Information Technology |
| 10 | KEYS | 0.985 | $373.90 | Information Technology |
Live rank-IC over 79 scored sessions: +0.0005 (backtest OOS: +0.0340). A meaningful comparison needs ~60 sessions.
| Date | Rank-IC | Names |
|---|---|---|
| 2026-09-18 | +0.2634 | 1468 |
| 2026-09-21 | +0.2072 | 1471 |
| 2026-09-22 | +0.1935 | 1471 |
| 2026-09-23 | +0.1509 | 1468 |
| 2026-09-24 | +0.1588 | 1470 |
| 2026-09-25 | +0.2318 | 1470 |
| 2026-09-28 | +0.1689 | 1469 |
| 2026-09-29 | +0.0917 | 1466 |
| 2026-09-30 | -0.0100 | 1466 |
| 2026-10-01 | -0.2297 | 1465 |
Mean +2.8 bps per side over 500 fills (11 sessions). Backtest survival brackets: Sharpe 1.06 @5bps · 0.71 @10bps · ~0 @20bps. Paper fills exclude market impact — treat as a lower bound.
| Date | Account | Equity | Positions |
|---|---|---|---|
| 2026-10-02 | A | $26,923 | 0 |
| 2026-10-02 | B | $27,027 | 202 |
| 2026-10-05 | A | $26,923 | 0 |
| 2026-10-05 | B | $27,050 | 202 |
| 2026-10-06 | A | $26,923 | 0 |
| 2026-10-06 | B | $27,224 | 202 |
| 2026-10-07 | A | $26,923 | 0 |
| 2026-10-07 | B | $26,838 | 202 |
| 2026-10-08 | A | $26,923 | 0 |
| 2026-10-08 | B | $26,430 | 202 |
Paper trading only. The backtest universe is current index members (survivorship bias inflates results). Free IEX data. The 2025 backtest year was weak (IC +0.016). None of this is investment advice — it's an experiment in doing retail quant with professional statistical hygiene.