Following the terminal's big-print flow arrows in the stock, with and against the day's flow lean
Does buying or selling the stock when a big option print's arrow appears on the chart make money, and do the arrows that agree with the day's running flow lean do better?
No. Arrows that agree with the day's flow lean lost 1.5 bp a trade to the close after costs over 31,475 trades on 8 tickers from Nov 2025 to Nov 2026 (day-weighted -1.6 bp, day-clustered t -0.78), and the version with the study's regular-session floor rule lost 2.4 bp a trade (day-weighted -3.5 bp, t -1.66). Pooled, every arrow group lost after costs at 30 minutes, 60 minutes and the close. By ticker the agreeing arrows were slightly positive on three names, none near significance. Pooled, the arrows against the lean were the worst groups (-7.7 and -7.8 bp), but the ordering does not hold on every ticker. The lean on its own, read at 10:30 ET, made +2.0 bp (t 0.60): no edge either.
Backtested by QS Research
Pre-registration
Rules fixed before any multi-day result: a replay of the chart's arrows as they appeared (the 20 largest single-leg prints so far, grouped by 5-minute bar and direction), entry at the next 1-minute open in the arrow's direction, exits at +30 minutes, +60 minutes and the close, costs 2 bp for the index ETFs and 5 bp for stocks. The lean is bullish minus bearish premium so far. Judged: arrows agreeing with the lean, and the same for the floor-rule arrows (added later the same evening, still before any result), each needing a positive to-close mean with a day-clustered t of at least 2, positive in both halves, and beating all arrows. The rules were written into the study script; the pre-registration document was committed after the run.
Registered 2026-11-20Trials · 2 trials · 0 passed · judged at N = 2 trials; 2965 trials run in total
| Trial | Verdict | |
|---|---|---|
| B: arrows agreeing with the lean | FAIL | counted |
| FB: floor arrows agreeing with the lean | FAIL | counted |
| A: every arrow | RESULT | not counted |
| B: arrows agreeing with the lean | RESULT | not counted |
| C: arrows against the lean | RESULT | not counted |
| B': agreeing with the put/call lean (exploratory) | RESULT | not counted |
| F: every floor arrow | RESULT | not counted |
| FB: floor arrows agreeing with the lean | RESULT | not counted |
| FC: floor arrows against the lean | RESULT | not counted |
| E: the lean at 10:30 ET, held to the close (baseline) | RESULT | not counted |
| Pooled to-close mean > 0 | RESULT | not counted |
| Day-clustered t >= 2 | RESULT | not counted |
| Mean > 0 in the first half (to 2026-05-15) | RESULT | not counted |
| Mean > 0 in the second half | RESULT | not counted |
| Beats A (FB: beats F) | RESULT | not counted |
| Against E, the lean alone | RESULT | not counted |
| SPY | RESULT | not counted |
| QQQ | RESULT | not counted |
| IWM | RESULT | not counted |
| INTC | RESULT | not counted |
| NVDA | RESULT | not counted |
| TSLA | RESULT | not counted |
| AAPL | RESULT | not counted |
| AMD | RESULT | not counted |
Limitations
- The stock, not the option: options on the same arrows were not priced.
- Prints are the 1,000 largest per ticker per session, chosen over the full day, so the lean sums those, not every print; the arrows are unaffected.
- The floor-rule arrows approximate the chart's floor; under its exact rule the agreeing floor arrows lose about 3.0 bp (t about -1.9), still a fail.
- Eight hand-picked tickers over one year.
- A prototype run: no signed engine release, no trial ledger rows and no pinned data snapshot; the study script and its result file are not published, and inputs are recorded only by content hash.
- SPY's sessions from June to November 2026 overlap another study's sealed holdout; that study records the read.
- The t-statistics are of the day-weighted mean, clustered by day; the per-trade means are shown beside them.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
1 published version · current ea121565d086cfd7, published 2026-11-21