Costly Shorts
Hard-to-borrow stocks tend to have negative returns
In recent years, the investment world has been shaken by several major short squeezes, with the AMC and GameStop fiasco of 2021 standing out as a prime example. These incidents served as a stark reminder to hedge funds and institutional investors of the risks associated with short selling, particularly when it involves stocks that are difficult or expensive to borrow.
The cost of borrowing shares for short selling varies significantly, often influenced by the stock's characteristics. Large, stable, and liquid stocks generally have lower borrowing costs, whereas smaller, volatile, and illiquid stocks can be much more expensive.
Short-borrow fees for individual companies are subject to change quite significantly through time. In Jan 2019, GameStop had an average short-borrow fee of 1%. By January of 2021, around the onset of the short squeeze, GameStop’s short-borrow fee had jumped to a whopping 34% (via IHS).
This week, we delve into the relationship between high short-borrow fees and future stock returns, highlighting the strategic value of recognizing hard-to-borrow (HTB) stocks. Through the development and application of a novel HTB classification model, we analyze how HTB stocks have fared over time relative to general collateral (GC). We hope to emphasize the value of HTB data as both a predictive signal and a risk-management tool.
While real-time borrow cost data is often available through brokers, extensive, retrospective datasets are either of poor quality or are expensive to obtain from third-party vendors. To address this problem, we developed an HTB classification model. Using a limited dataset from IHS that includes weekly borrow cost data from 2019 to 2023 across a broad range of equities (approximately 15,000) and integrating security-level data and factor exposures from our own database, we trained a borrow-cost classifier. We believe the key predictors of whether a stock would be hard to borrow are size, trading activity, and volatility, in addition to what exchange the stock trades on. This model proficiently differentiates between hard-to-borrow stocks and general collateral with a high degree of accuracy.
Our model achieves a precision and recall for identifying both hard-to-borrow stocks and general collateral at around 95-97% and an overall accuracy of 96%. Using this classifier, we generated daily short borrow classifications for approximately 30,000 stocks from 2005 to 2023. Beyond enhancing backtest realism, these classifications offer some compelling insights.
Our analysis shows that, in the US, hard-to-borrow stocks typically show negative returns over 3-, 6-, and 12-month horizons, whereas general collateral stocks tend to yield positive returns over the same periods (Figure 1a).
Figure 1: Forward Returns by Borrow Class
a. Median 3M, 6M, 12M FWD Returns by Short Borrow Class
b. Median 12M FWD Returns by Size Category & Short Borrow Class
Source: Verdad analysis. US Equities 2005-2023.
Short-sellers in the US appear to be fairly good at identifying companies to short, and this holds true independent of size (Figure 1b). It is worth noting, however, that this trend did not appear to hold globally. In Japan and Europe, hard-to-borrow stocks do not appear to systematically underperform as they do in US markets.
A follow-up analysis examined forward three-month returns based on short-borrow cost decile. Importantly, the underperformance occurs at the most extreme decile (most expensive to borrow).
Figure 2: Average 3M FWD Return by Borrow Cost Decile
Source: IHS, Verdad Analysis. US Equities 2019-2023.
High borrow fees can be a powerful signal for forecasting future returns. This is particularly evident in cases where large, previously liquid stocks become hard to borrow, often signaling impending negative returns. Although this points to a potentially lucrative investment opportunity, the practical challenges of short selling such stocks are non-trivial. After all, by definition, hard-to-borrow stocks are … wait for it … hard to borrow.
Thus, we believe in most instances the most pragmatic application of hard-to-borrow data is to steer clear of taking long positions in these stocks, serving as a strategic compass in the often-tumultuous seas of market speculation.