Two approaches can be used to assess and monitor the level of “crowding” in different quantitative strategies. The authors introduce these measures and present anecdotal evidence suggesting that many quantitative strategies were crowded in the period leading up to the global financial crisis.
What’s Inside?
The authors present two measures to assess the level of crowdedness of common factors used in quantitative investment processes. The first measure—the difference in utilization of stock available for borrowing (for shorting purposes) between stocks with positive characteristics and those with negative characteristics—is best suited for assessing crowdedness for long–short funds. The second measure—a measure of the comovement of returns in a basket of stocks with desirable factor attributes compared with the comovement of returns in a control basket—can be used by long-only fund managers. The authors use these measures to understand the events of the quant crash of August 2007.
How Is This Research Useful to Practitioners?
Quantitative investment managers have long considered the concentration risk of strategies—thus, the constant search for innovations to refine and differentiate investment processes. The simple measures proposed by the authors allow quantitative managers to monitor the crowdedness of their processes over time. Therefore, these measures should be included in the tests performed as part of the research process. The ideas may also be useful for investment consultants because they assess the diversification provided by different investment managers.
The authors use these measures to provide insight into the infamous August 2007 quant crash, when quantitative funds around the world suffered a short period of large losses before quickly rebounding. This crash has been blamed on the liquidation of a large quantitative fund, which, because of crowding, affected a large number of other funds. The authors suggest that, consistent with a liquidation event, there was unwinding of crowded trades at this time before crowding levels quickly rebounded.
How Did the Authors Conduct This Research?
The measures proposed by the authors are relatively simple and can be constructed from readily available data.
The first measure involves calculating the difference in utilization of stocks available for borrowing (for shorting purposes) between funds with positive characteristics and those with negative characteristics. The authors’ hypothesis is that when a lot of long–short investors want to pursue a particular factor, stocks with unfavorable characteristics will be more in demand than stocks with favorable characteristics. Utilization, rather than absolute short interest levels, is used because some stocks are more difficult to short sell than others. The data used to calculate utilization are from Data Explorers and are only available since 2006. The authors use a regression framework, and such explanatory variables as volatility, size, and recent returns are included to enable the attribution of utilization differences to the stock characteristic under investigation rather than to other common factors. Interestingly, a variable to strip out sector effects is not included in the regression framework.
The second measure uses intraday trading data to assess the degree of return comovement in a basket of stocks exhibiting positive characteristics (for a particular quantitative factor) compared with the return comovement exhibited by a control basket.
Confidence intervals are calculated for both measures, and the results are compared to confirm that they are producing broadly similar outcomes.
Abstractor’s Viewpoint
The authors present two relatively simple measures that allow investment professionals to assess the degree of crowding across different investment themes. Although designed with quantitative managers in mind, the measures could be used by other managers and consultants to improve their understanding of current market dynamics.
Given the rapid growth of funds managed under quantitative strategies during the 2000s, it would have been useful to explore whether there is a connection between this growth and crowdedness measures. Because utilization data are not available before 2006, this extension would have used only the second measure. It would have been useful to assess crowding with the stock comovement measure over a longer time period, given that these data are readily available.
With respect to the security lending measure, it would have been useful to gain a sense of the absolute level of utilization associated with crowdedness. For example, the level of utilization for crowded measures reaching levels of 90% or higher might result in different outcomes from the ones where the “crowded” utilization reached only 40%. In other words, is there a tipping point that utilization levels must reach before crowding occurs?