Research (see here, here and here)has shown that investor sentiment (emotions) plays a significant role in international market volatility and generates return predictability of a form consistent with the correction of investor overreaction; and total sentiment is a contrarian predictor of country-level market returns, as high investor sentiment predicts low future returns and vice versa. In particular, the research shows that younger, smaller, more volatile, unprofitable, non-dividend paying, distressed stocks are likely to be the most sensitive to speculative demands and more affected by shifts in investor sentiment. Conversely, “bond-like” stocks are less driven by sentiment.
Domonkos Vamossy and Rolf Skog contribute to the literature with their December 2021 study, EmTract: Investor Emotions and MarketBehavior, in which they employed a unique dataset combined with advances in text processing to examine the connection between firm-specific investor emotions and asset price movements. Specifically, they explored the following research questions:
Do the results of controlled laboratory experiments relating investor emotions to trading behaviour replicate observational data?
Do investor emotions forecast daily price movements?
Whose emotions matter and for what type of firms?
Do investor emotions help explain first-day IPO returns and subsequent long-run underperformance?
To answer these questions, Vamossy and Skog used data from StockTwits, a social networking platform for investors to share stock opinions, that covered 63 million messages linked to particular stocks spanning the period 2010-May 2021. They developed a tool that quantifies investor emotions from financial social media text data (i.e., informal text containing less than 30 words).
The authors explained: “Our models are powered by deep learning and a large, novel dataset of investor messages. In particular, our tool takes social media text as inputs, and for each message it constructs emotion variables corresponding to seven emotional states: neutral, happy, sad, anger, disgust, surprise, fear.” They then tested whether firm-specific investor emotions before the market opens predict a firm’s daily price movements. Following is a summary of their findings:
Neutral messages constituted about 42 percent, followed by happy posts of about 31 percent. Fear and surprise were the third and fourth most frequent emotions, followed by sad, anger and disgust.
Investors are more likely to share their enthusiasm than their pessimism on social media.
Investor emotions extracted from social media data behave similarly to those in
controlled laboratory experiments, providing validity for previous lab experiments (for example, see here and here).
Most posting activity on the platform happened when the markets were open — consistent with investors updating their beliefs in real time as financial events unfold.
Social media investors are more interested in discussing firms with higher dollar trading volume, volatility, larger market cap, higher short interest and lower institutional ownership.
Within-firm investor emotions can predict the company’s daily price movements — variation in investor enthusiasm is linked with marginally higher daily returns. The result was driven both by messages conveying original information and by those disseminating existing ones.
When considering messages that convey information directly related to earnings, firm fundamentals or stock trading relative to those messages that consist of other information, the latter has a slightly larger impact on daily returns.
The impacts of emotions are larger when volatility or short interest are higher, and when institutional ownership or liquidity are lower.
StockTwits users tend to discuss stocks that have gone up or are currently going up in value: The average past monthly return was 6 percentage points, the close-open return was 0.5 percentage point, and the one-day lag open-close return was 0.1 percentage point higher than in the CRSP sample. These stocks ended up with a 0.4 percentage point lower open-close return, suggesting mean reversion.
Emotions before the market opened explained a small fraction of the variation in daily returns — a standard deviation increase in non-market hour happiness (before the market opened) was associated with a 0.7 percent standard deviation increase in daily stock returns. The effects were smaller for larger-cap stocks. There were stronger effects for stocks with larger user engagement (at least 100 messages) — a standard deviation increase in happiness before the market opened was associated with a 3.1 percent standard deviation increase in daily open-close stock returns.
When it came to messages disseminating existing information, only the level of fear and happiness was associated with statistically significant differences from the baseline (neutral) level.
The predictive power of investor emotions diminishes over a few days.
Investor enthusiasm is a predictor for first-day IPO returns and subsequent long-run underperformance — the set of IPOs with low investor enthusiasm prior to the IPO had first-day returns of 16.5 percent on average, while the set of IPOs with high investor enthusiasm had a much higher first-day return of 30.9 percent on average.
IPOs with large first-day returns driven by investor enthusiasm underperformed average firms in the same industry over the long run. In contrast, IPOs experiencing large first-day returns without high investor enthusiasm prior to IPO did not experience long-run reversal — neither investor enthusiasm nor first-day return alone predicts long-run IPO underperformance, though the interaction between investor enthusiasm and first-day return does.
Their findings led Vamossy and Skog to conclude: “Investor emotions extracted from StockTwits provide information relevant to stock valuation not accounted for by unobservable time-invariant stock characteristics, by time patterns, or by recent price movements. ”They added: “These impacts are larger when volatility or short interest are higher, and when institutional ownership and liquidity are lower.” And finally, their findings showed that “investor emotions can help rationalise two stylised facts about IPO returns”.
Evidence From Robinhood investors
Behavioural finance professors Brad Barber and Terrance Odean have done extensive research on the performance and habits of individual investors. Among their findings is that, on average, individual investors lose money from trading—and not all the losses can be explained by trading costs. In their 2008 study, All that Glitters: The Effect of Attention and News on the Buying Behavior of Individual and Institutional Investors, they made the case that limited attention prevents retail investors from considering all available information and possible stock choices. Instead, many retail investors choose stocks to buy from the subset of stocks that catch their attention. Because most investors own only a few stocks and do not sell short, limited attention plays a smaller role in their sales decision.
In their November 2020 study, Attention Induced Trading and Returns: Evidence from Robinhood Users, Barber and Odean, with co-authors Xing Huang and Chris Schwarz, examined the behaviour of Robinhood users and found that they are even more subject to attention biases and more likely to chase stocks with extreme performance and volume than other retail investors. Herding by Robinhood investors can be forecasted by attention measures, such as lagged absolute returns and lagged abnormal volume, previously shown to affect the buy-sell imbalances of retail investors. Most importantly, they found that Robinhood herding episodes are followed by abnormal negative returns. Of particular interest is their finding that sophisticated investors were exploiting the patterns created by Robinhood investors by shorting stocks or buying puts in response to Robinhood herding events. They found a marked increase in short selling for stocks involved in Robinhood herding events — for the stocks with the top 25 returns for the period, the average change in short interest was three times greater. They concluded that their results “suggest strongly that market participants examined Robinhood ownership data, knew about the subsequent poor performance caused by Robinhood herding, and traded against Robinhood order flow.”
Investor takeaways
The research demonstrates that investor sentiment (emotions) not only plays a significant role in market volatility and generates return predictability of a form consistent with the correction of investor overreaction but also is a contrarian predictor of future returns. This is particularly true for stocks of companies that are younger, smaller, more volatile, unprofitable, non-dividend paying and distressed. Thus, investors are best served by having a well-though-out investment plan, including an asset allocation table that is adhered to (rebalancing along the way). Having and adhering to such a plan provides the greatest chance of not allowing emotions to impact decisions. Forewarned is forearmed.
Another takeaway is that while the effects of information on fundamentals can be identified with well-established techniques in finance, studying the emotional component requires new tools, such as the artificial intelligence tool used by Vamossy and Skog (which they made available here).
© The Evidence-Based Investor MMXXIV. All rights reserved. Unauthorised use and/ or duplication of this material without express and written permission is strictly prohibited.