What market breadth can and cannot tell you
A headline index and the typical constituent can tell different stories. Breadth measures how widely a move is shared. It is descriptive information, not proof that a crash is imminent. This revision removes the earlier claim that a single warning signal explained the dot-com peak; the site's dataset does not cover that episode and cannot validate that comparison.
A reproducible two-stock example
Suppose stock A begins at 90% of a two-stock index and B at 10%. Over one interval A rises 10% and B falls 10%. The starting-weight return is 0.9×10% + 0.1×(−10%) = 8%. The equal-weight return is 0.5×10% + 0.5×(−10%) = 0%. Half the stocks declined even though the weighted index rose strongly. This is a hypothetical illustration; no real index has only these two holdings.
Four separate measurements
Our Bubble Meter uses concentration, a trailing-return leadership gap, participation relative to a moving average, and leader earnings yield. Those measurements can disagree. Concentration is about weights, participation about how many stocks meet a trend rule, and earnings yield about a particular valuation ratio. A high weight does not by itself imply an expensive business.
How the score is constructed
The code keeps months where all four measurements exist, ranks each component against its full loaded history, and averages those four percentile ranks. The result is a score from 0 to 100. A score of 70 is not a 70% crash probability and is not necessarily the 70th percentile of the combined score. Adding later months can change the historical component ranks.
Why an attractive historical chart can mislead
The input universe uses current S&P membership and share counts applied backward. Stocks removed from the index are not restored to their historical positions. Annual fundamentals use reporting-lag approximations. These limitations matter when interpreting the meter as a record of what an investor could actually have known. The code does not establish that a high or falling score predicts subsequent returns.
A practical reading worksheet
Record the observation month, each raw component, the combined score, and the available historical window. Explain which components caused a change. Then look at your own holdings for overlap; do not use the dial as an automatic instruction to buy or sell. Read methodology for implementation details and compare endpoint growth with the independent drawdown study.