Undertow
NQ ● holds overnight ● live testing 24/7Short-term mean reversion on the Nasdaq-100. When the index gets oversold on a short-period RSI, Undertow buys the dip at the close and exits the next close — harvesting the bounce. Long-only, in the market only ~10% of days, built from many small wins (~60% win rate). Validated lookahead-free over seven years: Sharpe ~1.2, with a maximum drawdown about a quarter of buy-and-hold's, and positive every single year — including the 2022 bear, so it isn't just long beta. Uncorrelated to the trend, breakout and gamma sleeves (+0.05) — it's the book's mean-reversion counterweight, cushioning their drawdowns. Live since June 2026 — the seven-year backtest is shown alongside the growing live track record.
- Short-period RSI (oversold)
- Mean-reversion dip-buy
- Exit next close
- Long-only NQ
Signals
$100/mo
- Daily bias + exact execution plan via API
- Live event feed (entries, scale-outs, exits)
- Dashboard view + one URL for Claude/LLM use
- Cancel anytime
Full source code
$9,999 one-time
- Complete strategy source (Rust) incl. the model
- Backtest harness + parameters
- Own it, run it, modify it — no lock-in
- Delivered within 24h to your account email
Track record
Backtest
Backtest 2019-03-06 → 2026-02-05 · 6.9y — lookahead-free, fills on real trade prints only, 1 contract incl. commission + slippageLive paper-trading
Live paper-trading on tickstream data since 2026-06-29| Date | Side | Entry | Exit | P&L | Reason |
|---|---|---|---|---|---|
| 2026-08-06 | FLAT | — | — | +$0 | eod |
| 2026-08-05 | FLAT | — | — | +$0 | eod |
| 2026-08-04 | FLAT | — | — | +$0 | eod |
| 2026-08-03 | FLAT | — | — | +$0 | eod |
| 2026-07-31 | FLAT | — | — | +$0 | eod |
| 2026-07-30 | LONG | — | — | +$17,838 | eod |
| 2026-07-28 | LONG | — | — | −$5,475 | eod |
| 2026-07-27 | FLAT | — | — | −$7 | eod |
| 2026-07-26 | LONG | — | — | +$453 | eod |
| 2026-07-24 | FLAT | — | — | −$7 | eod |
| 2026-07-23 | FLAT | — | — | +$0 | eod |
| 2026-07-22 | FLAT | — | — | +$0 | eod |
| 2026-07-21 | FLAT | — | — | +$0 | eod |
| 2026-07-20 | LONG | — | — | +$338 | eod |
| 2026-07-19 | LONG | — | — | +$70 | eod |
| 2026-07-17 | FLAT | — | — | −$7 | eod |
| 2026-07-16 | FLAT | — | — | +$0 | eod |
| 2026-07-15 | FLAT | — | — | +$0 | eod |
| 2026-07-14 | FLAT | — | — | +$0 | eod |
| 2026-07-13 | FLAT | — | — | +$0 | eod |
Live results are paper-traded on real tickstream market data, one NQ contract, with each sleeve's own commission and slippage charged on every fill — between $9 and $24.50 round turn depending on how it exits. Every record runs from 1 June 2026. Rows dated before the sleeve itself shipped are a backfill and carry a bf tag wherever they surface in the trade table; everything from the launch date on was traded forward.
Rebuilt 31 July 2026. The whole record was re-derived from our own archived tick path: every session the archive covers end to end was replayed through the agent's unmodified production code — the backfilled portion included, so it is no longer backtest output but the same code run over recorded prints. That corrected fills which had been booked at the trigger level instead of at the price that actually printed; on gap opens the difference reached 470 points. Sessions the archive cannot cover end to end keep their rows as recorded, because deleting a real trade over a hole in our recorder would flatter the record. On the sessions used as the control, the replay reproduces the live rows to the cent.
The backtest is lookahead-free: weekly CFTC data is anchored on its actual publication day and fills are simulated on real trade prints only — the live agent runs the identical model. Past performance, simulated or live, does not guarantee future results. Not financial advice.