The Robintrack portfolio has seen stellar performance YTD (~30% relative to the Russell 2000) that traditional factor models are not able to explain. We also saw massive alpha in the portfolio and inferred that there may be a signal emerging rooted in the wisdom of the crowd. In this case, retail investors.
The rush into online brokerage trading continues, with trendy stocks like Pfizer, Tesla, Nikola, and Snap being added to tens of thousands of Robinhood user accounts over the last week (according to Robintrack.net), and E-Trade adding over 320,000 new accounts in each of the first two quarters of 2020. Recall that last week’s article focused on a long-only, popularity-weighted “Robintrack” portfolio using Robinhood’s popularity data, as tracked by Robintrack.net, which shows how many user accounts hold a particular stock. We found that the Robintrack portfolio has seen stellar performance YTD (~30% relative to the Russell 2000) that traditional factor models are not able to explain. We also saw massive alpha in the portfolio and inferred that there may be a signal emerging rooted in the wisdom of the crowd. In this case, retail investors.
This week, we’ll continue exploring this signal by transforming the data into a factor exposure and overlaying it on top of additional portfolios.
Methodology for Creating a Retail Investor Sentiment Exposure
As discussed last week, the Robintrack dataset goes back to May 2018 and is updated on an hourly basis throughout each trading day. For our purposes, we grabbed a snapshot of the number of customers holding a given stock at the end of each day and our resulting universe (post-cleansing) is ~5500 stocks as of July 22, 2020. Because the cross-sectional data for a given day follows an exponential distribution, we chose to transform the data by taking the natural log in order to make it more intuitive for exposures analysis.
The resulting distribution looked much closer to a normal distribution, so no winsorization was required (in contrast to last week when we used this same data to create a popularity-weighted portfolio). Below is the resulting box and whisker plot for the transformed Robintrack data.
After the transformation, we converted the daily data to a z-score, with the resulting value being our exposure. Below is a plot of the exposures for July 22, 2020 - the distribution goes from -4 to +4 and looks similar to what we might expect from a traditional risk model factor exposure.
Names like Tesla (TSLA), which have seen a major increase in popularity over the past several weeks, have a large, positive and increasing exposure.
Less popular names will have a low or even negative exposure - an example is BioNTech (BNTX). Prior to announcing a partnership with Pfizer (PFE) to co-develop a coronavirus vaccine on March 17, BNTX was on the lower end of the popularity spectrum, showing a negative exposure in January 2020. This exposure increased to just above 0 prior to the announcement and skyrocketed afterwards.
How Meaningful is the Retail Investor Sentiment Signal?
To level set the analysis, we’ll get a feel for how our Retail Investor Sentiment factor exposure behaves within the Robintrack portfolio that we evaluated last week. The expectation is there should be very strong exposure within the portfolio, given that the portfolio and the factor exposure were constructed from the same dataset.
As expected, there is a high Retail Investor Sentiment exposure to the portfolio, about 1.86 on average YTD, and this exposure has increased over time, with a spike in mid-March which aligns with the trends we’ve observed in the market. Given the Robintrack portfolio has very strong positive performance over this period, the increase in the factor exposure makes sense here.
How Does Retail Investor Sentiment Impact Shorting Strategies?
Considering the impetus of our analysis being the recent unusual behavior of traditional long-short equity strategies and our curiosity of the cause (other than the obvious ‘coronavirus’ effect), we decided to look at the intersection of a Short Interest signal, which can be a proxy of the aggregate short side of long-short portfolios, and our Retail Investor Sentiment signal. To do this, we loaded a portfolio from our partner Wolfe Research to represent Short Interest. This portfolio is constructed based on a signal that uses daily securities lending data aggregated from broker-dealers by Markit and specifically measures shorts outstanding relative to available inventory.
As our heuristics might tell us (and as long-short managers would hope, given this portfolio is comprised of many of their short bets), Short Interest is a factor that is known as an underperformer; however, on a YTD basis, we’ve seen a sharp reversal of this factor, with YTD return gaining back almost a third of the losses since 2007.
Perhaps Retail Investor Sentiment can be partially to blame for throwing the Short Interest factor onto a different trajectory? If we look at the portfolio representing the Short Interest factor from Wolfe and overlay our Retail Investor Sentiment exposure, we may be able to tease out a relationship.
As one might have guessed, there is a high exposure of Retail Investor Sentiment within the Short Interest portfolio. While we’re not able to specifically attribute the increasing performance of Short Interest to our Sentiment factor since it is exogenous to the risk model, we can certainly see that there’s a positive correlation between the two signals.
While it remains to be seen if Retail Investor Sentiment is here to stay or will become more of a transient, pandemic-induced signal, it’s certainly worth checking the Robintrack leaderboard to see if your next short idea might be foiled by the wisdom of the crowd.
This article, "Is Retail Investor Sentiment Leading to the Collapse of Our Short Ideas?," originally appeared on the Omega Point blog.