Short Selling as an ex ante alpha signal

Efficient stock market pricing requires all market participants to feed their individual expectations regarding future and value-relevant information into the process. If a market participant considers the market price to be too high, the decision will be taken against purchasing the share. The higher the confidence of the presumed overvaluation and/or the higher the overvaluation itself, the more likely are professional investors in particular to be willing to enter into a short sale. For this purpose, interested parties borrow the desired securities from the custodians against collateral and a fee and sell them on the market. If the share price declines as expected, the investor buys back the securities for a lower price and receives the difference as a profit.

Investment hypothesis

Short selling is an operationally highly regulated, complex and economically very risky process, as the loss exposure for the short seller is unlimited. It is fair to assume that especially well-informed institutional investors with above-average analytical capabilities are active in this market. The assumption that this group of investors has valuable, share-specific information, paired with the observation of an increase in short selling, signals a negative sentiment for the respective stock. Activities that suggest a directional opinion are particularly interesting for return forecasting. Consequently, the short selling of securities around the dividend date purely motivated by tax reasons leads to distortions – such trades have no information value, which is why we adjust the signals for this effect.

Basis of data

The information on how many short-selling contracts have been concluded is available with a time lag from the exchanges themselves –or very promptly from specialised data providers. In addition to the short-selling positions entered into, these data providers also supply information on the potential volume of securities available for lending and thus also cover the supply side. For our global sample, we thus obtain information on the number of securities available for lending, the amount and number of short sale contracts concluded and the fees incurred for the loan. This data has been available to us since January 2007.

Modelling

In order to complement the alpha model by the short-selling information, we create a composite signal from multiple dimensions of the short-selling market –in line with the already existing factor groups. The elements that have turned out to be particularly valuable in factor research cover:

• the volume of outstanding short-selling contracts (volume);
• the breadth of short-selling activities (breadth);
• the capacity utilisation of securities available for lending (capacity utilisation).

The benefits of this signal in the context of our multi-factor approach will be discussed in more detail in the following section.

Factor performance

To assess the benefits of the short-selling composite signal, we create monthly quintiles sorted in ascending order according to our signal. The annual outperformance of each quintile against the broad market for the investment markets – Emerging Markets (EM), Europe (EU) and North America (NA)–is shown in Figure 1. The positive correlation between composite signal and outperformance is readily apparent and stable for all regions presented. The observation that the short-selling factor works both in the first quintile and in the fifth quintile is remarkable. Naturally, one would expect stocks with high short-selling activity to perform particularly poorly. As long-only investors, however, the absence of short selling is a particularly interesting aspect for us, as these stocks also outperform on the buy side.

The annual excess return of about 2% for the fifth quintile may look low at first glance. However, the less the new signal correlates with the signals already used in the forecast model, the more valuable it becomes. To demonstrate this, we calculate the correlation of the short-selling composite with known investment factors already considered in the model.

It becomes clear that a high or low level of short-selling activity occurs practically independently from Value, Sentiment, Size and Beta–the factor groups already in use. Against this background, the short-selling composite signal creates a positive benefit for our alpha model.