Evaluating smoothed secondary counts with a chi-square test

Secondary offerings as a weekly tape statistic

A secondary-offering is a large-block sale placed through an underwriter so the position can be distributed with less obvious pressure on the open-market price. The archive describes that routing as a way to reduce the chance of a sharp decline while the position is placed.

The weekly count of those offerings is treated as an informed-selling-proxy. The working premise is that sizable holders are better informed, so a cluster of secondaries is read as a cautionary tape statistic rather than as ordinary liquidity. A high count is mapped to a bearish reading and a low count is mapped to a bullish reading.

From raw count to a smoothed forecast object

The working series is a 40 percent exponential-smoothing of the ordered weekly count. That decaying-weight average is presented as roughly equivalent to a four-week moving average, and the smoothing constant is described as roughly comparable to a four-week lookback.

A precommitted normal-band

In the stated 10-year sample the mean weekly count was 2.7 and the standard deviation was 2.2. A normal-band was set two-thirds of a standard deviation above and below that mean, a corridor used to separate quiet readings from signal swings. Swings outside the band were treated as directional signals.

Scoring later direction with a chi-square test

Those signals were scored against later market direction over a six-month horizon and over a one-year horizon. Agreement between signal and later direction was assessed with a chi-square-test: a one-degree-of-freedom test with the Yates correction, used to judge whether later market direction lined up with those swings more often than chance would suggest. The statistic was reported above 32 for each horizon.

Chart notes in the sample

Chart notes contrast an elevated-count episode in 1986 with elevated counts from May to August 1987 and with subdued counts after the crash.