The strategy in one paragraph
Each weekday after the US market closes we scan ~520 large-cap US stocks for three specific patterns that have documented academic edge: post-event price drift on gap-up volume spikes, post-earnings-announcement drift, and 52-week-high breakouts. Each pattern triggers a "setup" that historically continues for 60-90 days. We publish today's setups, suggest position sizing for any capital, and forward-track the strategies live in paper portfolios so you can see whether they keep working.
Three strategies, ranked by evidence
⚡ Gap-up drift
When a stock closes up 5%+ on heavy volume (1.5x normal+) in a liquid name, it tends to continue drifting up over the next ~60 days. This is the post-event drift effect (Bernard-Thomas 1989, repeatedly validated). Most gap-ups are earnings beats; the rest are M&A, FDA approvals, or other material catalysts. Either way, the drift persists.
Honest update (2026-05-24): on the expanded R1k universe with survivorship correction and 10bps costs, the aggregate 4-year alpha is +27pp (+15%/yr). BUT year-by-year is variable: drift beat SPY in only 2 of 5 yearly periods. The aggregate is real but driven by 2 standout years (2022 defensive, 2024 huge). Per-trade bootstrap confidence intervals show +0.88pp alpha is statistically distinguishable from zero with 53% win rate, so the edge IS real, but the per-trade edge is small and the compounding produces high variance. This is a long-horizon strategy. Expect multi-quarter drawdowns.
Parameter update 2026-05-24: ran 144-combo sweep optimizing for yearly consistency. Best variant: hold 45 days (was 60). Lifts aggregate alpha from +27pp to +35.9pp, increases years beating SPY from 2 of 5 to 3 of 5, lower variance. Production drift detector now uses 45d hold for new picks. Numbers below reflect the 45d variant.
| Year | Drift return | SPY | Alpha | Events |
|---|---|---|---|---|
| 2022 | -15.9% | -19.9% | +4.0pp | 1,757 |
| 2023 | +11.1% | +24.8% | -13.7pp | 1,454 |
| 2024 | +43.3% | +24.0% | +19.3pp | 1,598 |
| 2025 | +22.6% | +16.6% | +5.9pp | 2,265 |
| 2026 YTD | +4.0% | +9.1% | -5.2pp | 712 |
| Aggregate | +92.0% | +56.1% | +35.9pp | 10,171 |
↗ 52-week high breakout
Stocks making new 52-week highs on confirming volume above their 200-day moving average tend to keep trending up (Jegadeesh-Titman momentum effect). Hold 90 days. Less explosive than gap-up drift.
Honest update (2026-05-23): earlier backtest claimed +11pp alpha. After applying the survivorship-bias correction we use for drift (including delisted tickers), the alpha shrinks to ~+1pp/yr after 10bps costs. The original number reflected only winners that survived to today. Demoted from secondary to supporting. Kept in the system as a confirming signal for stacked conviction, not as a standalone trade.
📊 Post-earnings drift (PEAD)
Stocks that report YoY EPS growth AND gap up on the SEC filing day tend to drift further. Bernard-Thomas (1989) used YoY growth as the surprise proxy, which lets us backtest across our full 4-year fundamentals history. Hold 90 days.
Honest update (2026-05-23): 4-year aggregate looks great (+33pp alpha) but year-by-year is inconsistent — 2022: +9.5pp, 2023: -20.9pp, 2024: -3.0pp, 2025: +2.2pp. The aggregate is carried by two strong years. Should NOT be relied on as standalone signal. Kept as a stacking signal for top-conviction picks (when PEAD agrees with drift, conviction goes up).
Composite ranking
Originally the main product. 19 factors weighted into a 0-100 score per stock. The 4-year walk-forward backtest showed it returns almost exactly the same as SPY (+0.5pp alpha over 4 years). Still useful as a screening filter, but should not be used as the primary buy signal.
Caveats (read these before risking real money)
This is not investment advice. cloudtrades publishes algorithmic setup output. You decide if and how to act.
Backtest biases stripped where possible. Survivorship bias accounted for via delisted-ticker correction (+2pp/yr drag). Transaction costs modeled (5-30bps). What remains: in-sample tuning risk (the criteria were chosen partly by looking at recent data), single 4-year sample (insufficient regime coverage), no slippage on entry fills.
Live forward test is young. We started tracking in May 2026; nothing has closed yet. Until we have 60+ closed trades, the backtest is the only meaningful evidence.
Drawdowns are real. The drift strategy had a -22.4% drawdown in the 2022 bear-market backtest. You will see down months. The strategy works on average over time, not every month.
Position sizing matters more than picking. 2% per position across 80 concurrent positions diversifies tail risk. Concentrating into the "highest conviction" pick is the most common way retail traders blow up. Don't.
The pipeline (every weekday after US close)
- Ingest fresh closing prices for ~520 stocks (Yahoo via curl_cffi Chrome impersonation)
- Scan for gap-up events, 52w breakouts, post-earnings drift setups
- Publish today's setups to the picks page with rationale and target/horizon
- Mark-to-market the three live paper portfolios (drift / PEAD / composite)
- Close positions whose 60-day horizon reached or target hit
- Re-bake landing page with today's data
Every dataset in the backend is browsable from the data lab with a time-machine filter — pick a date, see only what was knowable on that date. Verifies no look-ahead bias.