What Short Sellers Know

February 9th, 2012 by

The year 2011 was a fairly volatile one for markets – they went up, they went down, and then ended up right about where they started. But that didn’t make it impossible to make money. Indeed, nearly all StarMine quantitative models performed quite robustly in 2011 in the US, particularly those based on long-short strategies. Given the relative volatility experienced in 2011, we were not surprised to see investors flock towards the relative safety of stocks offering high earnings quality. More surprising was our realization that momentum signals performed solidly in 2011, since momentum indicators generally struggle during periods of market volatility. Most value signals also generated positive returns in a long-short framework in 2011, although not as strongly as momentum or quality signals. Although short interest models tend to be mildly correlated with valuation models, we were curious how such a signal would perform during the volatile year that just ended. As it turns out, it performed very well.

Short Interest models take advantage of the broadly accepted and empirically verified intuition that high levels of short interest reflect negative sentiment on the part of sophisticated investors and are associated with negative excess future returns. The StarMine Short Interest model ranks US stocks based on the hypothesis that stocks with a high (low) number of shares shorted will under (out) perform. The StarMine Short Interest model improves upon a basic short interest model by accounting for well-known arbitrage strategies (merger and dividend) that could affect levels of short interest. The model also incorporates institutional ownership as a supply factor that measures the number of shares available to be lent to short sellers and therefore serves as a gauge for how costly it is to short a given stock. We view high demand, in the form of a high number of shares shorted, in the presence of tight supply (and therefore high cost), as a sign of conviction on the part of short sellers. To compare the performance of the StarMine Short Interest model to that of a benchmark signal, we created a basic short interest signal by dividing shares short by shares outstanding for each security. We used the equal-weighted Russell 3000 as our universe in this study and rebalanced portfolios monthly, although we evaluate and plot performance on a daily basis.


Figure 1: Returns for the top decile, bottom decile, and decile spread of the StarMine Short Interest model. The performance of the equal-weighted Russell 3000 index is also included. Portfolios are rebalanced monthly. Source: QA Studio, StarMine

Figure 1, above, shows the performance of the top decile (the most lightly-shorted 10% of companies after applying our adjustments), bottom decile, and decile spread for the StarMine Short Interest model. We also plotted the Russell 3000 index for reference. The results indicate that the Short Interest model performed very well during 2011. The signal generated particularly strong performance on the short side, indicating that investors would likely get value simply by eliminating the lowest ranked stocks from their long portfolios. Adding a short interest filter to a long-only strategy would help avoid an investment in the riskiest stocks which declined significantly throughout the year. Furthermore, the model is valuable as a long-only signal, since the absence of short interest shows a lack of negative sentiment on the part of the short sellers. Figure 2, below, compares the 2011 decile spread for the StarMine Short Interest model and the benchmark model; it shows clearly that the StarMine model significantly outperformed the baseline signal throughout the year. The enhancements we make by accounting for the cost of shorting and also discounting arbitrage-related shorted shares drove the outperformance of the StarMine Short Interest model. These enhancements also help reduce the correlation with other quant factors, such as momentum, valuation, and quality, ensuring that the StarMine Short Interest model truly provides orthogonal alpha.


Figure 2: Returns for the decile spread of the StarMine Short Interest model and the baseline short interest model which is calculated as shares short divided by shares outstanding. Portfolios are rebalanced monthly. Source: QA Studio, StarMine

Despite the market volatility witnessed in 2011, a model based on short seller behavior performed solidly. Given its strong performance and low correlation with other quant signals, investors would be wise to consider including the StarMine Short Interest model in their investment process.

Learn more about how QA Studio and StarMine Quantitative Models can help you discover alpha. Request a demo today.

  • Rpinventor

    Very interesting way to control for downside risk.  Two quick questions: 
    1. In creating your short signal by “… 
    dividing shares short by shares outstanding for each security.” is there a reason your used TSO and not free float shares (which is what virtually all benchmark vendors use for the cap weighted indices?  While this may not be a big deal for mega caps, as you move down the capitalization scale, it becomes more significant, especially when looking at the constituents of the R3000.
    2. What’s the daily variation (variance) of daily short interest versus monthly?