Is There Alpha in the Wisdom of the Crowd? What Retail Investor Sentiment Tells Us...

If you’ve noticed some unusual behavior in your long-short portfolios, it’s possible that market moving retail flows may be the culprit.

Retail investors are piling into the market in record numbers and taking advice from non-traditional sources, such as Reddit’s WallStreetBets and PennyStocks.com. It’s quite possible that the investment behavior from these retail investors, which is typically speculative and not rooted in traditional investment theses, is deeply hurting traditional long-short strategies that are based on fundamentals.

We saw earlier this week in an article published by Bloomberg that roughly 40,000 accounts on the popular retail investing app, Robinhood, bought into Tesla (TSLA) on Monday, July 13. Despite the stock dropping over 15% from it’s intraday peak by the end of the trading day on Monday, the number of new accounts buying into TSLA almost topped 50,000 for the day, according to Robintrack.net, which tracks popularity data from Robinhood to show how many users hold a particular stock.

With the sheer mass of trading volume coming from these retail investors, it may be that we’re seeing an emergence of a new signal in the market coming from the ‘wisdom’ (or lack thereof) of the crowd.

To test this theory, we loaded the popularity data from Robintrack into Omega Point as a proxy to evaluate a ‘Retail Investor Sentiment’ signal. We’re looking to understand if the performance impact from the behavior of retail investors can be explained by traditional factors or if there is in fact a new signal that can be derived from this crowding effect. To start, we loaded the data into Omega Point as a long portfolio weighted by popularity (i.e. the number of customers holding a stock).

Methodology for Creating a Popularity-Weighted Portfolio

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 at the end of each trading day. To create the popularity-weighted portfolio, we started by evaluating the distribution of the Robintrack dataset. The data, as one might expect, is heavily skewed. Below are box and whisker plots at various points in time for the number of customers holding a certain stock.

To control for outliers and skewness in the distribution, the data was winsorized by capping the top 10% of the dataset. Though this is an aggressive winsorization, this helped dampen the possibility that an individual stock or a handful of stocks might carry all of the performance of the resulting portfolio. Here is a look at the resulting box and whisker plots of the dataset after cleansing.

Once the data was cleansed, we used the number of customers holding a stock as the economic exposure for that stock to create the popularity-based weighting scheme for the portfolio.

Another way to think about this portfolio is as an index tracking the Retail Investor Sentiment in the market. When considering the portfolio in this way, it’s interesting to note the number of positions.

The number of stocks covered goes from ~4000 at the inception of the dataset in May 2018 to almost 5500 names in July 2020. Though the portfolio does include some international names, it is largely a US domiciled portfolio and with 5500 names, covers more securities than the largest standard US market index. Given this, the size of the portfolio is large enough to operate as an indicator of the market in the same way as other standard and factor-based market indices. We’re off to a good start in terms of finding a broad-based signal that is representative of the market.

Diving Into the “Robintrack” Portfolio

We’ll evaluate the long, popularity-weighted “Robintrack” portfolio relative to the Russell 2000 since this benchmark was the best aligned on sector & market cap exposures. The Robintrack portfolio massively outperformed the Russell 2000 and was up by over 28% in the YTD period ending July 7. Though factor contribution was positive over this period, the alpha contribution really carried the performance and there was essentially no alpha degradation even during the March market turmoil.

Looking into the factor performance using the Axioma US 4 Medium Horizon model, we can see that recent usual suspects in terms of market moving factors (Volatility and Market Sensitivity) are definitely at play here. Overexposure to Volatility hurt performance during the market downturn in March, but this same overexposure allowed the portfolio to ride the rebound and ended the period with over 1% of performance contribution (red line). The portfolio was initially underexposed to Market Sensitivity but this exposure turned very positive in March and gave the portfolio 3% of performance contribution (blue line).

The unusual behavior of these factors over the past several months meant that overexposure to these factors paid off.
(For additional reference on recent behavior of the Market Sensitivity factor, please see our previous Factor Spotlight posts: “Innoculating Against Runway Beta” and “Diving Deeper into Runaway Beta”.)

While factors overall gave the portfolio 8.3% in performance, this pales in comparison to the alpha, which contributed over 20% to active performance.

‍So far, all signs point to Retail Investor Sentiment as an alpha signal emerging in the market. But before we rush to allocate all of our capital into this signal, we should investigate the ‘health’ of this alpha by looking at the alpha concentration. Given the nature of the data, one might expect for there to be heavy alpha concentration in a handful of names, likely to be FAANG stocks, TSLA, and other (in)famous names sweeping public newsfeeds. However, we see quite the opposite. The alpha is actually very well diversified, with no single standout name carrying or detracting from the performance and the top 10 contributors driving about 2.25% of performance. Some of the potential alpha concentration was likely dampened from the winsorization applied during the data cleansing process, but nonetheless we see what appears to be a relatively healthy alpha signal.

By analyzing this popularity-weighted portfolio, we’ve seen that traditional factors are unable to describe the strong performance. More analysis is yet to be done, but this is compelling evidence pointing to an emergence of Retail Investor Sentiment as a new alpha factor outside of traditional signals rooted in fundamentals or technical analysis. Perhaps there is something to this ‘wisdom of the crowd’ idea?

Next week, we’ll continue our investigation of the Retail Investor Sentiment signal by overlaying the Robintrack data onto portfolios as an exposure.

US & Global Market Summary

US Market: 7/13/20 - 7/17/20

• The market saw continued strength despite major concerns about the economy as coronavirus cases surged across the country, with the US hitting a record of new infections on Thursday (77,499).
• As a response, California announced lockdown measures and 21 other states have either increased restrictions or slowed down their reopening roadmaps.
• The early reading of the July U Michigan consumer sentiment survey showed a decrease from 78.1 last month to 73.2. Final results will be released at the end of the month.
• Initial weekly jobless claims came in at 1.3 million for the week ending July 11, vs. consensus estimates of 1.25 million.
• Earnings season kicked off with some of the major banks reporting. JP Morgan and Citigroup beat consensus estimates on the strength of their diverse business lines, while Wells Fargo suffered its first quarterly loss since the GFC. We'll see more companies report next week, with investors getting the popcorn ready for TSLA on Wednesday.

‍Factor Update: Axioma US Equity Risk Model (AXUS4-MH)

• Size saw ongoing strength as it climbed to the threshold of Overbought space at +0.95 standard deviations above the mean. Recall that this factor was -3 SD below the mean at the beginning of June.
• The rally in Momentum continued, albeit at a slower pace, as it now sits solidly in Overbought territory.
• Profitability is now inches away from being a perfectly Neutral factor on a normalized basis.
• Value continued its decline on a normalized basis, heading further into negative normalized space.
• The bleeding slowed down for Market Sensitivity, but it remains Oversold at -1.49 SD below the mean.
• Volatility was the week’s biggest loser, falling by a little over half a standard deviation to end up in Oversold territory.
• US Total Risk (using the Russell 3000 as proxy) declined by 7bps and remains elevated relative to Global risk.

Factor Update: Axioma Worldwide Equity Risk Model (AXWW4-MH)

• Earnings Yield vaulted up into Overbought territory, moving +0.68 SD in the past week.
• The rally in Momentum continued after a recent nadir of -3.13 SD below the mean on 6/10, now sitting deep in Overbought territory.
• Size calcified its Overbought label as the rally in this factor continued since hitting trough in late May.
• Market Sensitivity descended further into Oversold space, but at a slower rate.
• Volatility also continued its decline on a normalized basis, earning an Oversold label.
• Exchange Rate Sensitivity took a sizable tumble from its recent peak of +2.62 SD above the mean on 7/10, and is now poised to shed its Extremely Overbought designation.
• Growth was the biggest loser globally, falling by 0.58 SD in the past week.
• Global Risk (using the ACWI as proxy) declined by 7bps, exactly in line with US risk.‍

Regards,
Alyx