Liminal / History
Market Memory
Thesis
Closest analogue: narrow-leadership advances running on flat, complacent vol
Three features define the current setup: index highs on narrow leadership, implied vol running below realized, and credit sitting at its tights. That combination has four close precedents in the trailing two decades. Three resolved through rotation into the laggards, and one resolved through a vol-led drawdown, a 75% split this module treats as a base rate worth noting, not a forecast worth trusting. Memory's job here is narrower than prediction: it keeps the analysis from treating today's setup as unprecedented, an error that makes both panic and complacency cheap.
Evidence
Four episodes match on all three selection features: leadership concentration, vol discount, and credit tights. In the three benign resolutions, laggard participation improved before the leaders cracked. In the drawdown case, credit spreads widened roughly three weeks before equities noticed.
At n=4, statistics has nothing to say about whether the 75% split is a genuine rotation tendency or simple luck. Treat the number as a base rate, not a distribution. What it does establish holds regardless: this setup has repeatedly resolved without a drawdown, and in the one case it didn't, credit announced the shift first.
Four episodes match on all three selection features: leadership concentration, vol discount, and credit tights. In the three benign resolutions, laggard participation improved before the leaders cracked. In the drawdown case, credit spreads widened roughly three weeks before equities noticed.
Market Read
None of the four precedents currently shows up as a factor risk desks are pricing for. Index skew carries no premium for a repeat of the drawdown case, and credit protection isn't bid anywhere near where it sat three weeks ahead of that one prior episode. Protection priced this cheap only makes sense if the market simply hasn't gotten around to the historical parallel yet, not if it weighed the analogy and consciously waved it off.
Four Is Not A Distribution
Every analogue engine oversamples the memorable: episodes get selected because they resemble today on the features someone happened to record, and nothing guarantees those were the features that actually mattered. This module's claim has an honest ceiling, and 'this has happened before' is it. Nothing about the four precedents licenses anything sharper.