Holding clocks
The same reversion premise behaves like four different strategies depending on how long you hold. This is why the worked example runs four models, not one.
A stretch that closes within an afternoon and a stretch that closes over a month are the same idea on paper and almost opposite strategies in practice. The holding clock — how long you expect the reversion to take — changes the mean you measure against, the size of move worth acting on, the cost that eats your edge, and the kind of risk that ends the trade. A single set of rules cannot serve all of them, which is the entire reason the worked example splits the premise across four models.
The shorter the clock, the smaller the move and the tighter the math
On a same-session clock, the mean is a recent intraday level and the stretches are small — fractions of a percent. The edge per trade is thin, so it has to be harvested often and protected fiercely from costs and slippage; a wide spread can erase a session reversion entirely. On a multi-week clock the mean is a longer average, the stretches are large enough that costs barely register, but you are exposed for days to the risk that the stretch was a real change in value rather than a temporary one. Same premise, completely different problem.
| Model (holding clock) | 2026 return | Win rate | Trades |
|---|---|---|---|
| Swing Reversion roughly one to four weeks of hold | +225% | 74.4% | 78 |
| Session Reversion opened and closed inside one trading day | +95% | 67.5% | 308 |
| Intraday Hold from half a session out to two sessions | +404% | 71.4% | 262 |
| Position Reversion a long, patient horizon | +502% | 73.8% | 42 |
Why the grade has to be set per clock
This is the practical payoff of the holding-clock idea. A 0.70% move is an exceptional session reversion and an unremarkable multi-week one. If you graded both against one absolute bar, every fast trade would look weak and every slow trade would look strong, and the grade would carry no information. So a measured model sets the grade-A bar against each model's own return spread:
| Model | Holding clock | What earns an A |
|---|---|---|
| Swing Reversion | roughly one to four weeks of hold | about 6.00% a trade |
| Session Reversion | opened and closed inside one trading day | about 0.70% a trade |
| Intraday Hold | from half a session out to two sessions | about 4.50% a trade |
| Position Reversion | a long, patient horizon | long-form, no single bar |
An A marks the top band of a model's own measured returns and a D the lowest still issued. Because the bar is fixed per holding clock, an A on a session trade and an A on a multi-week swing both read as “top band for this horizon” rather than a single absolute target stretched over very different holds. The scale stops at four bands: an E once existed but left the live product in 2026, so the A-to-D ladder reads cleanly.
Read the table as four translations of one sentence — “this trade sits in the top band for its clock.” That is what lets a reader compare an A on a session trade with an A on a swing trade without comparing raw percentages that were never meant to be compared. The mechanism behind it is on grades that are measured; the reason the grade can be trusted at all is on issued before the outcome.