I did a study a month or so ago on using data on the number of Robinhood accounts that hold a certain security to predict cross-sectional equity returns. Here are some of the results. Please feel free to comment, ask questions (no guarantee I can answer them), and/or point out mistakes I made.
Shoutout to SMB Capital as they posted a video showing the work one of their traders has done to use Robinhood Data along with other datapoints to develop trading strategies. This encouraged me to share some of the work I have done on the topic.
Summary of Key Points (more discussion in attached notebook)
- Robinhood is a fast growing FINRA broker dealer whose clients are relatively inexperienced, yet trade actively. It has a popular
trading app and was a pioneer in commission free trading.
- Data from Robintrack.net tracks the number of Robinhood accounts that hold a given stock.
- A 5-day, long-short account growth factor (i.e. the percent change in number of accounts holding the stock) appears to show positive alpha over the next 1 to 5 trading days. However, performance does deteriorate in the out-of-sample period, but exceptional market volatility makes it difficult to determine whether this deterioration was simply a result of a difficult market regime, or if it was due to overfitting and/or signal decay.
- There is evidence that the source of alpha could be a result of a herding mentality where Robinhood users visually see which stocks are
“trending” in the app, and then pile on, leading to a momentum effect.
- The factor does show some exposure to the short term mean reversion factor, suggesting Robinhood traders may have a tendency to buy
beaten down names.
- Results are noisy, and should be looked at skeptically, especially given the relatively short sample period (2 years). The factor alone
is not likely to be robust on its own. Combination with other complimentary factors is likely necessary.