Clustered true-range days as a regime label rather than a top forecast

What the volatility-percent count records

A true-range day, in this construction, is a session whose true range is at least 3% of that session's price. The default reading counts how many such 3% true-range days occur in the prior 50 sessions, which is the lookback window, and reports that share as a percentage. That share is the volatility-percent count.

The count is only a record of how much clustered range expansion sits inside a fixed window. It does not say why prices moved, and it does not, on its own, classify the next decline as sudden or orderly.

How often 3% true-range days appeared

From 29 October 1928 through 13 July 1997, 3% true-range days occurred on fewer than 7.2% of sessions. They arrived in clusters rather than at a regular interval.

After the sample was started in July 1935 and 1 October 1987 through 29 February 1988 was dropped, the same 3% threshold appeared on only 2.9% of sessions and mostly during declining markets.

The count rose during ordinary declines

In most ordinary declining markets in the sample, the count did not rise before the slide. It increased during the decline and peaked near the low. That path matches an orderly decline: a downward phase in which the volatility-percent count rose mainly after prices had already begun to fall.

New advances often began while the count was still high, because the prior decline's final expansion left the lookback window elevated.

Crash analogs stayed elevated for months

A crash analog, here, is a historical episode in which the count stayed elevated for months before a sudden break. In the 1929 and 1987 crash analogs, the 50-session count reached 20% to 30% about six months earlier and stood at 12% or higher immediately before those breaks.

The crash comparison rested on two historical episodes and was presented as limited rather than statistically conclusive.

The 1997 reading sat below the crash analogs

On 22 September 1997 the same 50-session count stood at 4, below the crash-window readings used as analogs. The count was described as uninformative for anticipating an orderly, non-crash decline.