INVESTED: In Markets, in the Future, and in Society.

Freaking Out

The disposition effect is a well-studied behavioural finance anomaly which shows that individuals dislike losses more than they enjoy gains. In financial markets, this translates into investors selling their “winners” too early and holding onto their “losers” too long, in the hopes that they will recover and avoid taking a loss on the position. In a new paper released last month, titled When Do Investors Freak Out? Machine Learning Predictions of Panic Selling, the authors find that investors panic sell during sharp market downturns, and these so-called “freak-outs” are not only predictable but also fundamentally different from the disposition effect and other previously studied behavioural finance anomalies. One of the most interesting conclusions from the paper is that not only does the median investor earn a zero to negative return after they engage in panic selling behavior, 30.9 percent of those investors never return to reinvest in risky assets (i.e., stocks). The meat of the paper is centered around the ability of machine learning models to predict when investors may panic sell, and unfortunately the question around causation isn’t addressed. Given that a large proportion of individuals who panic sell never return to the market or only return to the market after a subsequent rally, thus realizing a large opportunity cost to do so, this is an important topic for investors – especially because certain demographic profiles are found to be more prone to freak-outs than others.
 
The authors go on to opine that the negative connotations around panic selling may not be warranted, because although “panic selling in normal market conditions is indeed harmful to the median retail investor, freaking out in environments of sustained market decline prevents further losses and protects one’s capital.” However, predicting market structure as an investor is no easy feat. For example, the authors found that “a person who liquidated [their portfolio] at the start of the [Great Financial Crisis of 2008] and left the market for 15 months… saved [themselves] from losing another 17 percent.” In contrast, the COVID-19 drawdown was a much different beast than the financial crisis of 2008; the S&P 500 took only four and half months to recover from the 34 percent drop experience over February and March. Considering that the research suggests only about half of the individuals who panic sell and later re-enter the market do so within six months, the net effect of panic selling will greatly depend on what type of market structure is prevalent at the time. While not a topic of the paper, a natural extension could be analyzing how a systematic, rules-based trend and momentum ensemble could help influence behaviour and protect investors from not only the disposition effect, but panic selling freak-outs. By codifying a decision-making framework to dictate when to reduce exposure to risky assets and when to increase exposure to risky assets, investors can work toward removing some of the behavioural biases that cause sub-optimal decision-making in the heat of the moment.