Auditing advance-decline confirmation for scale, horizon, and composition

Read confirmation as a three-part audit

A standard advance-decline line is a running sum of each session's advancing-minus-declining issue count, so its step size is the raw net breadth of that session. Market breadth is that session-level tally of how many listed issues participate in an advance versus a decline, used to test whether an index high is widely supported.

On-balance volume is a running accumulation that adds or subtracts session volume according to the direction of price. It is a participation check that is conceptually parallel to counting advancing versus declining issues.

Editorial view: run the same cross-examination whether the vote is an issue count or a volume tally. First ask whether the vote is still scaled for a changing listing universe. Then ask how many sessions the price-breadth split has actually lasted. Then ask whether stripping interest-sensitive names removes noise or deletes the liquidity information the signal was built on.

Ask whether the vote is still scaled

A net reading of 100 issues was treated as more meaningful when about 600 names traded than after more than 3,400 names were listed, because the same raw difference becomes a smaller share of the tape. That distortion is listing-universe bias: the error that appears when a raw net-issue difference is compared across decades in which the number of traded names has grown.

A ratio-adjusted series uses advances minus declines, divided by advances plus declines, then multiplied by a constant such as 1000 that must stay fixed across the sample. That advance-decline ratio is a scaled breadth increment that keeps long samples comparable. Over spans of up to a couple of years that series can look much like the raw cumulative line.

On a multi-decade comparison the raw cumulative line had already moved above its 1958 high while the ratio-adjusted line had not, even though the ratio line stood above its 1998 peak. A multi-year failure to recapture the 1958 ratio high was not treated as proof that the 1980s and 1990s advances were only a failed rally.

Editorial view: a confirmation that still uses the raw step has not yet passed the first audit. The historical workflow treated the fixed-multiplier ratio as the scale check, not as a verdict that later advances were only a failed rally.

Then count how long the split has lasted

A price index making a higher high while the advance-decline series makes a lower high is treated as a breadth divergence that questions whether the advance is broadly supported. Splits of only a few days occur often and can mislead. Splits lasting weeks to months are used to judge the next upward impulse after an intermediate-cycle low.

In the 1926-1930 sample the NYSE advance-decline series peaked on 14 May 1928, more than a year before the 1929 industrial-average peak. In July 1998 the industrial average made a higher high while advance-decline did not, and after the late-1998 low the average rose more forcefully than the breadth series. Across data from 1926 onward, tops without a breadth divergence were followed by milder corrections than tops that had one.

Editorial view: duration is the second audit, not a decoration. A split that has lasted only a few sessions is a common chart condition. A split that has lasted weeks to months is the horizon the historical workflow used when judging the next impulse after an intermediate-cycle low.

Then inspect what the filter removes

A small-cap-to-large-cap relative-strength ratio moved closely with the cumulative advance-decline ratio, consistent with smaller names contributing heavily to daily breadth and with weaker liquidity showing up first in those names.

Common-only breadth is an exchange-filtered advance-decline count that drops symbols longer than three letters and therefore excludes most preferreds, rights, and warrants while still including many funds. Common-only exchange counts leave closed-end and country funds, and that series usually ran a bit weaker than the composite. Subtracting closed-end-fund votes from the common-only counts produced a nearly identical path. After a shared August 1989 top, a later joint top showed a higher high in the unfiltered composite and lower highs in the filtered versions, which did not recapture their 1989 highs until June 1997.

Uncommon breadth is the residual advance-decline vote obtained by subtracting common-only counts from the composite, isolating the more interest-sensitive names the filter removes. On the compared sample, that uncommon line standing well above its 200-day moving average coincided with firm corporate-bond prices, and a retreat toward or below that average coincided with weaker bond periods. In 2003 the uncommon line was rising through that average while the corporate-bond index was also rising.

Policy rate increases arriving while advance-decline was already weak, including 1974-75 amid an oil shock and 1999-2000 after a late-cycle tightening, were presented as coinciding with worse output outcomes than tightenings that arrived while breadth was strong.

Editorial view: the third audit is whether the filter is cleaning the tape or deleting the interest-sensitive residual. The historical workflow kept the uncommon line and read it against a 200-day moving average beside corporate-bond prices. That is a composition check, not a license to treat the filtered common-only path as the only legitimate vote.