Testing price-volume agreement after percent reversal filters

What the design is asking

The archive workflow builds observations first, then tests association on those observations alone. Editorial reading: TradersWeek treats the price-percent-filter as the sampling design, not a chart overlay. The filter chooses which swings become the trend-reversal-sample. A four-cell test then asks whether volume changed in the same direction as those filtered price swings.

How turning points are selected

A price-percent-filter records a new date only when the percent change from the latest peak or trough meets a chosen threshold and runs opposite the previous filtered change. The sequence of those filter dates is the trend-reversal-sample.

Each kept event stores the date, the industrial-average close, and exchange total volume so later signed changes can be computed from those tabulated points.

Signed pairs and the four-cell test

After each tabulation, the sign of the price change is paired with the sign of the change in cumulative volume. That signed-price-volume-pair is assigned to one of four cells.

Association is summarized with a price-volume-cross-correlation, the four-cell coefficient that records whether filtered price changes and volume changes tend to share the same sign. The chi-square-from-correlation equals that coefficient squared times the number of pairs.

The same test across filter sizes

The same statistics were produced for every integer filter size from 1 percent through 10 percent across four windows spanning the start of 1897 through the start of 1987.

In the window from 13 June 1949 through 2 January 1987, the 10 percent filter had the largest positive signed-change coefficient among the ten sizes, followed by the 9 percent filter. In that same later window, nine of the ten filter sizes had a chi-square confidence above 99.5 percent. The 8 percent size was the exception and still exceeded 99 percent.

Chart overlays of the kept dates

Charted overlays used a 10 percent filter on price and volume from the start of 1897 through 12 August 1982, then compared 1 percent and 10 percent filters from 12 August 1982 through 12 December 1986.

Editorial note: TradersWeek reads those overlays as pictures of the same trend-reversal-sample used in the four-cell counts, not as a separate trading rule.