Understanding How Short Interest, Market Structure and Sentiment Impact Risk and Returns

Author:

Jeff Messina

Chris Kleparek

Sat Bhattacharya

November 23, 2022

Highlights

Using S3 data, we created three factors to evaluate how changes in short interest, financing rates and utilization can impact risk and returns in broader portfolio construction
From 2015 to today, when used in isolation, these three factors had a positive impact on risk-adjusted returns
When these three factors were combined over the same time period, results were even stronger

Introduction

S3’s data is derived from our Blackwire OMS technology, which is connected to every major bank and financing counterparty, 67 stock exchanges for filings, 58 regulators for disclosures, every significant financial twitter follower and integrated with major OMS providers. S3’s unique data lake provides investors with the ability to build uncorrelated investment strategies. In this research paper we provide a combined factor that generates consistent returns since 2016. There are significant opportunities to enhance the strategy by leveraging additional S3 data attributes and combining our data with external data sets (Px/Vol, news, options, fundamental data including estimate revisions, etc.).

S3 Partners has been the global industry leader in high frequency short interest data for over a decade. Our specialty is the financing space—securities lending, prime brokers and swaps—and this gives us a unique view into broader market sentiment for both long and short market positioning. Sentiment is one of the most important — yet opaque — factors that drives investment returns. At S3 we solve for sentiment in a systematic way through our data lake, which allows us to collect the data that make our products the gold standard for understanding long and short market positioning:

S3 delivers a consolidated, unbiased, data-driven view of short interest, market structure and sentiment across all buy-side profiles in the investment ecosystem, including hedge funds, mutual funds, retail and passive investors. The core products and analytics are built based on our proprietary data lake – a treasury management system that has existed for 20 years.

Our Core Products Include:

Hourly

Last Rate

Intraday, when applicable

Bid rate
Offer Rate
Indicative availability
Utilization

Daily

S3 Short Interest as a % of the S3 Float (1)
Utilization (Borrowing Availability)
Bid/Ask Spread
Mark-To-Market P&L

Predictive Analytics

Crowded Score
Squeeze Risk Score
Momentum Score

Data

History: Point-in-Time back to 2015
Data Lake: +120 buy-side institutions across all buy-side profiles
S3 Short Interest Coverage: +65,000 securities globally

Methodology

S3 has partnered with Quantision (2) – a leading data science and quantitative research firm – that has developed an institutional approach to extracting value from S3 data systematically. Leveraging some of the unique attributes of the dataset, several factors have been constructed as a research guideline. For this analysis, we considered an investable universe of US equities with at least a $500M market capitalization and $10M in average daily trading volume. To avoid overfitting, we started with sound market structure hypotheses and tested those with historical point-in-time data. To be consistent with industry research standards, the factors were tested using long/short quantile portfolios, taking into account transaction costs assumptions.

The following is a description of the factors’ methodology:

Short Term Changes in S3 Short Interest (% of S3 Float)

Short term changes in S3 Short Interest can be an indication of investor sentiment and future short selling (SI increasing) or covering (SI decreasing). We construct this factor by comparing the average S3 Short Interest percentage of S3 Float of the last 5 days with the previous 10 days moving average.

Factor Sign: Long (Decreasing short-term short interest) / Short (Increasing short-term short interest)

Changes in Financing Rates

Increasing rates can be an indication of future short covering while decreasing rates can lead to increased short selling as borrowing costs decline. The factor is constructed by comparing the last rate – mid rate spread with the historical spread for each security.

Factor Sign: Long (Increase in last rate) / Short (Decrease in last rate)

Crowding (% Utilization)

Securities that are close to max borrowing availability (as measured by short interest utilization) may see future, positive relative returns. This factor is primarily effective on the long side, as low levels of short interest are not necessarily an indication of future selling.

Factor Sign: Long (Approaching max borrowing availability) / Short (Retreating from max borrowing availability)

Combined Factor

The 3 factors above are combined into a single factor with the following method. The first two factors (SI Changes and Rates) are combined by equally weighting each factor's standardized values (Z-Scores). Given the nature of the crowding factor being primarily indicative of positive relative returns, it is applied as an additive factor of 1.0, if the normalized crowding values (% utilization) are included in the top 5% of crowded shorts for that period. While this ensures the factor is only applied on the long side, it also acts as a risk management input to reduce potential short positions for crowded short securities. Finally, the resulting combined factor is smoothed using an exponentially weighted moving average to reduce the overall turnover of the signal.

Results

To analyze the performance of each factor, we created sample portfolios by splitting each factor into deciles, and equally weighting securities in the top decile (long) and bottom decile (short). Each factor is also de-meaned by GICS sector to reduce any net exposures across sectors. Each resulting sample portfolio is dollar neutral with total gross exposure of 100% and performance metrics are computed net of estimated trading costs.

Performance and portfolio characteristics for each factor are presented in Figure 1. The table shows each underlying factor exhibits positive returns while being uncorrelated with each other as shown in Figure 2. Finally, the combined factor maintains consistent risk-adjusted returns and lower maximum drawdown while still exhibiting a reasonable level of portfolio turnover.