What our own data says about what comes next.

FlowDesk checks every reading it stores against what each stock did over the following 5 trading days, compared with the rest of the market. Only readings that held up in both halves of the history count as an edge, and the projection is used only after it has beaten chance on days it never saw.

In plain words

None of the readings FlowDesk stores has yet shown a reliable edge over the next 5 trading days.

No measurable edge: volatility (20 days), call/put volume tilt, tape composite, composite change (3 days).

The projection model is not trusted yet: Only 15 test days; out-of-sample edge not yet significant (t 0.14); top projected stocks did not beat the bottom ones. It is shown for transparency and does not change trades.

862

stocks studied

trading days

2026-08-18 → 2026-10-04

18,463

checked readings

Which readings predict

Reading	IC	t	Top − bottom fifth	Both halves	Verdict
Quality higher is better · 48 days	+0.038	+1.7	+0.47 pts	agree	weak
Confidence higher is better · 48 days	+0.040	+1.6	+0.66 pts	agree	weak
Price momentum (5 days) lower is better · 43 days	−0.048	−1.3	−0.51 pts	agree	weak
Last day's move lower is better · 44 days	−0.042	−1.1	−0.41 pts	agree	weak
Volatility (20 days) higher is better · 38 days	+0.037	+0.6	+1.04 pts	agree	noise
Call/put volume tilt lower is better · 45 days	−0.004	−0.2	−0.24 pts	disagree	noise
Tape composite higher is better · 48 days	+0.000	+0.0	−0.08 pts	disagree	noise
Composite change (3 days) lower is better · 45 days	−0.000	−0.0	−0.02 pts	disagree	noise

IC: average daily rank correlation between the reading and the next 5 days’ return over the market. t counts overlapping windows once (days ÷ 5). An edge needs |t| ≥ 2 and both halves of the history pointing the same way.

The projection

Not trusted yet. Only 15 test days; out-of-sample edge not yet significant (t 0.14); top projected stocks did not beat the bottom ones. It is shown for transparency and does not change any trade.

+0.005

out-of-sample IC

t 0.1

−0.01 pts

top − bottom fifth

50.4%

right side of the market

unseen test days

2026-09-20 → 2026-10-04

#	Stock	Projected vs market	Range	Main reasons
1	ARM	+1.17 pts	−3.6 … +6.0	Quality, Price momentum (5 days), Volatility (20 days)
2	APLD	+1.15 pts	−3.7 … +5.9	Quality, Price momentum (5 days), Volatility (20 days)
3	BB	+1.14 pts	−3.7 … +5.9	Call/put volume tilt, Price momentum (5 days), Volatility (20 days)
4	INTC	+1.10 pts	−3.7 … +5.9	Confidence, Price momentum (5 days), Volatility (20 days)
5	IONQ	+1.07 pts	−3.7 … +5.9	Call/put volume tilt, Price momentum (5 days), Volatility (20 days)

Points of return over the next 5 trading days relative to the average stock, with a range of one typical out-of-sample error. A projection, not a promise.

How it is measured

One reading per stock per day (the last of the day) from the tape history. The target is each stock’s return over the next 5 trading days minus the average of every stock that day, so a market-wide rally does not look like skill. Windows that cross a split or a bad price print (a daily move over 40%) are dropped. The projection is a ridge regression on each day’s rank-ordered readings. It is tested walk-forward: every 5 days it is refitted on all earlier days (leaving a 5-day gap) and scored on the next 5 days it has never seen, so most of the history is used as a test. It is rebuilt about once a day.