Aggregate Short Interest as a Stock Market Indicator
Posted in Short Selling
August 14, 2009
Does aggregate short interest serve as an intermediate-term stock market indicator based on either momentum (with an increase/decrease in aggregate short interest signaling a continuing stock market decline/advance) or reversion (with an increase/decrease in aggregate short interest signaling a stock market advance/decline)? To investigate, we compare the behavior of NYSE aggregate short interest with the behavior of the the NYSE Composite Index). The NYSE has measured aggregate short interest monthly (about mid-month) through August 2007 and approximately biweekly (mid-month and end of month) since. Using monthly/biweekly short interest data culled from NYSE news releases and contemporaneous NYSE Composite Index level/volume data for the period January 2002 through July 2009 (69 monthly followed by 35 biweekly observations), we find that:
First, we focus on monthly data. The following chart compares monthly levels of the NYSE Composite Index (approximately mid-month, as of the short interest measurement closing dates) and NYSE short interest over the entire sample period (91 months). Visual inspection reveals no clear and consistent relationship between the two series, although short interest does rise dramatically before the 2008 plunge in stock prices.
For a closer look at the relationship, we relate monthly changes.
The following scatter plot relates the monthly change in the NYSE Composite Index to the same-month change in NYSE aggregate short interest over the entire sample period. The Pearson correlation for these two series is 0.00 and the R-squared statistic is 0.00, indicating that the monthly change in aggregate short interest explains nothing about contemporaneous stock returns.
This result is counterintuitive. One might reasonably expect that an increase/decrease in aggregate short interest would accompany a general decline/advance in the stock market. If we exclude the outlier in the lower left of the plot (October 15, 2008), the best-fit line does slope down from left to right per this intuition, with an R-squared statistic of 0.14. This one outlier thoroughly disrupts the expected relationship (and perhaps disrupted many portfolios).
Might the ratio of aggregate short interest to aggregate trading volume be more predictive?
The next chart compares monthly levels of the NYSE Composite Index and the ratio of NYSE aggregate short interest to NYSE Composite Index trading volume over the entire sample period. The trading volume is the average index daily trading volume over the 21 trading days preceding (and including) the final trading date for measurement of short interest. Visual inspection reveals no obvious relationship between the two series.
For a closer look, we again relate monthly changes.
The next scatter plot relates the monthly change in the NYSE Composite Index to the same-month change in ratio of NYSE aggregate short interest to trading volume over the entire sample period. The Pearson correlation for these two series is 0.18 and the R-squared statistic is 0.03, indicating a modest positive (and counterintuitive) relationship. This time, if we exclude the outlier in the lower left of the plot (again October 15, 2008), the best-fit line flattens and the R-squared statistic decreases to 0.00.
It appears that the simple change in the aggregate short interest is potentially more informative than the change in the ratio of short interest to trading volume.
Do changes in aggregate short interest or stock index level reliably lead the other over intervals up to six months?
The next chart plots correlations between monthly changes in the NYSE Composite Index and monthly changes in NYSE aggregate short interest for various lead-lag scenarios ranging from stocks lead short interest by six months (-6) to short interest leads stocks by six months (6) over the entire sample period, with and without the one disruptive outlier.
With or without the outlier, there is hardly any support for a belief that either series reliably leads or lags the other. Without the outlier, there is support for the intuition that the stock market advances/declines as aggregate short interest decreases/increases. A coincident relationship is not tradable.
Might there be a reliable non-linear relationship?
The sample size is fairly small for a ranking test. Segmenting the 90 monthly changes in aggregate short interest into ranked thirds yields unsystematic results: the average NYSE Composite Index return during the month after the lowest / middle / highest monthly changes in aggregate short interest is -0.10% / 0.87% / -0.49%. Excluding the October 15, 2008 outlier increases the first number from -0.10% to 0.76%.
Is the more recent biweekly data (September 2007 through July 2009) more informative?
The next chart relates the biweekly change in the NYSE Composite Index to the same-interval change in NYSE aggregate short interest during September 2007 through July 2009 (only 45 observations). As for the monthly data, the Pearson correlation for these two series is 0.00 and the R-squared statistic is 0.00, indicating that the biweekly change in aggregate short interest explains nothing about contemporaneous stock returns. Again, the result is counterintuitive, and the October 15, 2008 outlier (lower left of the plot) is to blame. Excluding this outlier makes the best-fit line slope down from left to right with an R-squared statistic of 0.06.
Does biweekly data exhibit any lead-lag relationship between stock market returns and changes in aggregate short interest?
The final chart plots the correlations between biweekly changes in the NYSE Composite Index and biweekly changes in NYSE aggregate short interest during September 2007 through July 2009 for various lead-lag scenarios, ranging from stocks lead short interest by three biweeks (-3) to short interest leads stocks by three biweeks (3), with and without the one disruptive outlier.
With the outlier, there is some indication that change in short interest is contrarian (an increase/decrease in short interest precedes a market advance/decline). Excluding the outlier eliminates the indication.
In summary, evidence from simple tests on a limited set of data offer little support a belief that changes in aggregate short interest reliably predict future stock market behavior. There is an example in the sample period of a materially disruptive single observation (Black Swan?).
