Score Momentum — How Far Has the Score Fallen?

Score momentum measures how much the Alphameter composite score has deteriorated from the best level it reached since the current episode began. A large drop from the episode peak historically increases the probability of a regime flip.

Why Episode Peak — Not a Fixed Lookback

A simpler approach would compare today's score to where it was 14 days ago. The problem is that during shock events the score can crash and partially recover within the same window — making the move look much smaller than it really was.

April 2026 example

Score peaked at +75. The tariff shock in early April dropped it to near zero, before recovering to +31. A 14-day window measured after the recovery would show only a modest change. The episode peak approach records the full 44-point drop from +75, regardless of when within the episode it occurred.

By measuring from the episode peak, the signal captures the cumulative damage — and stays elevated until the score recovers to a new episode high.

Historical Flip Probability by Drop Size

Validated across 30 years (1996–2026) on days where 2 or more indicators were diverging.

Drop from peakN (days)P(flip 20d)Median days to flip
1–9 pts5634%23d
10–19 pts26642%21d
20–29 pts13147%19d
30–39 pts42856%15d
40–49 pts29947%20d
50–59 pts54666%11d
60–74 pts40558%15d
≥75 pts77162%10d

Larger drops generally correlate with faster flips — median time falls from 23 days at a small drop to 10 days when the score has fallen 75+ points. The dip at 40–49 pts reflects episodes that stall in neutral territory without fully committing to a reversal.

P(flip within 20 days) by drop from episode peak — div≥2, 1996–2026

Validated Against Fixed Lookback

Fixed 14-day lookback

Precision63%
Recall89%
Sample size1,050

Episode peak (chosen)

Precision59%
Recall96%
Sample size2,580

The episode peak approach trades a small amount of precision for significantly higher recall and more than twice the sample size. For a caution indicator, missing a genuine flip is more costly than the occasional false alarm.

For educational and informational purposes only. Not financial or investment advice. Historical probabilities are not guarantees of future results.