Loss aversion is a tendency that behavioral finance has studied for decades. It’s also a reality that investment managers deal with every day. We as humans struggle with facing losses as opposed to gains. We irrationally value things we own versus their economic complements. Pioneering research by Daniel Kahneman, Jack Knetsch, and Richard Thaler brought this to light in the early 1990s with simple experiments. A recent paper by Klakow Akepanidaworn, Rick Di Mascio, Alex Imas, and Lawrence Schmidt (“AMIS”) echoes their findings across the arena of institutional asset management.

These researchers studied nearly 800 institutional portfolios, evaluating trade-level position-level data across 16 years. At Novus, because we passionately believe in the value of analyzing position-level data, we found the research naturally attractive. The average portfolio in the study held $573 million in assets.  This is important, as prior research on trade-level behavior collected brokerage account data from individuals, which behave differently, and perhaps less skillfully than professional investors. The average manager across AMIS’ sample study generated nearly 300 basis points of annualized alpha against carefully selected benchmarks. These are no slouches! What they found confirmed those seminal experiments of behavioral finance: selling decisions are far less effective than buying.

How much you may ask? To the tune of a 200 bps per year. If managers acted upon this, their annualized alpha could increase from 300 bps to 500 bps; an almost 70% increase. Figure 1 demonstrates the composite drag analyzed across nearly 4.4 million trades over preceding intervals of time. It is calculated versus a hypothetical alternate (“counterfactual”) trading decision (more on that below):

‍

‍

This is a dramatic trend that demonstrates efficacy in decision making for buys and significant value destruction for sells. But why is this? The researchers sought to unpack this a few ways:

1. Could trading decisions be a function of risk? If so, would risk-adjusting the counterfactual trade narrow this gap?
2. Could this be a function of portfolio importance? If so, can we analyze core trades versus minor trades?
3. Could this be a function of attention? If so, would earnings-related trades demonstrate a different picture?

‍

Regarding risk-adjustments (1), AMIS’ core findings assume an alternate trade to measure the impact of both buys and sells. They assume that the portfolio manager would simply buy or sell a random security in their portfolio instead of the actual trade made.  This is the counterfactual for which the chart shows a spread in return outcome. Well, what if the sell decision was made to reduce risk exposure to a given factor, for example? To control for this, AMIS tested a factor-neutral counterfactual, and the results were directionally the same; perhaps diminishing the positive impact of buys, but not improving the drag of sells.

Regarding major trades versus minor trades (2), the researchers split trades into two groups: major (greater than 50% of an existing position either bought or sold) versus minor (less than 50%).  The idea being that maybe for meaningful trades, there’s a difference in future outcomes.  Instead, they found amplified results for major trades, where the annual spread between buys and sells balloons to ~300 bps (the chart above for all trades shows ~200 bps of spread after 365-days between counterfactual buys and sells).

Finally, on earnings related trading (3), we find something truly peculiar: buy decisions around earnings releases were effectively noisy, but sell decisions were undoubtedly profitable! That is, the future return of the counterfactual portfolio underperformed the sell-decisions of the portfolio around earnings-releases at every future time interval (28 days – 365 days in the future).  This flies in the face of the core findings regarding all sell decisions.  The obvious question is, why?

The researchers conclude that attention is a limiting factor to sale efficacy, and they point to earnings releases as a method of focusing the decision making around selling.  Earnings focuses a selling decision on the facts at hand.  Otherwise, selling decisions are too-often driven by emotions (or drawdowns) rather than process, going back to the biases exposed by Kahneman. That behavioral bias is undoubtedly a meaningful drag on performance.

What are we to do about it? Force our PMs and risk committees to pay better attention? That sounds unlikely. Perhaps there are a few things we as investors should do to improve our odds:

1. Collect your data in a way that allows you to shed transparency on these trends. Without knowing, we navigate in the dark in a game whose odds are tilted against us.
2. Leverage your data to build structure around trading process. Without structure, we lose that vice-grip that forces focus on process. This means revisiting priors and structuring sell decisions the same way one structures buy decisions. Review these decisions intermittently and improve. Rinse, repeat.
3. We must become learning machines. Like analyzing a golf swing, learn more about your own behavioral tendencies around trading decisions.

‍

A Tool for Timing Analysis

At Novus, we help portfolio managers gain insight into their tendencies, as evidenced through their position-level data. You can see if the findings of the paper apply to yourself using Novus Event Analysis, which allows you to measure the presence (and impact) of various behavioral biases, including loss aversion. The screencap below depicts an example where trade traces (prices trajectories before and after a sale decision) can be examined to construct statistical tests around the presence of behavioral biases.

‍

Ultimately, this research provides a glimpse into how talented managers, managing significant sums of capital, consistently destroy value through selling decisions across a statistically significant time-period and sample set. The fact that earnings-related sells drastically outperform all other sells points to the opportunity for improvement to this pervasive problem. The data is the record of facts. How you choose to use it depends on your willingness to improve.