Memory and Probability

Behavioral economists connect the way our memory works to how we make probability judgments
 

By: Dan Rasmussen

In the first few months of the COVID pandemic, a group of behavioral economists surveyed Americans about their beliefs about the lethality of the pandemic. The survey found that the young dramatically overestimated their own risk, the elderly underestimated their own risk, and health problems were associated with higher levels of pessimism. And, perhaps most interestingly, people who overestimated the percentage of Americans with red hair were among the most pessimistic about COVID’s lethality.

These results don’t seem to fit into any standard behavioral models. But the economists who ran the survey believe these responses can be explained by the way human memory works and the role our memories have in shaping the way we understand the present and forecast the future.

The economists argue that older people have had more experience and, because they evaluate new experiences relative to a large database of memories, are less reactive. They remember surviving many other non-COVID adversities, so they underestimate the risk posed by this novel disease. The economists call this phenomenon “interference.”

The young, in contrast, respond more intensely to frightening media stories. This reaction pattern is similar to other behavioral economics studies that find that people will overpay to insure against specific, salient events (e.g., buying insurance right after a disaster hits). Young people don’t have a large set of memories of other adversities interfering with their thinking, and they are therefore more likely to be overly triggered by a novel, specific threat made salient on the news or social media.

People who overestimate the share of people with red hair are more likely to make generalizations based on their own experience and thus overestimate rare events. Overestimation of COVID lethality was correlated with a general inability to make good probability assessments.

The lead author on this COVID study, Pedro Bordalo, is now out with a major new paper he co-authored with Andrei Shleifer and Nicola Gennaioli called “Memory and Probability.” Their paper seeks to generalize these findings on memory and probability to offer a new model that can reconcile seemingly contradictory findings from behavioral finance around over and under reaction.

According to their model of memory, people have a database of experiences on which they draw when making probability judgments, and they draw selectively from this database of experiences based on similarity. “Similarity helps retrieve relevant experiences but also invites interference from experiences inconsistent with the hypothesis at hand (but similar to it),” they write. Probability judgments are shaped by content, not just objective frequency.

According to their model, our memory is triggered by similarity, and we tend to overestimate the frequency of things that are easy to recall. People go out and buy flood insurance right after major floods or invest in tail risk hedge funds right after a market crash. On the other hand, there’s a large body of work that shows that people under save for retirement because they don’t plan for “bills outside what we normally would expect: the garage door spring and cable that snapped and had to be replaced; the family member who asked for financial help; the x-rays and dentists’ fees for a sudden toothache," etc. Contractors also famously underestimate the time it will take to complete a project because the reasons for delays are always various and unpredictable—even though, in aggregate, every job experiences at least some number of these “death by a thousand cuts” small problems.

Their research builds on an interesting set of papers by Demis Hassabis, now CEO of DeepMind, that first hypothesized the connection between recall and forecasting in the brain. Hassabis’s paper, “The Future of Memory: Remembering, Imagining, and the Brain," argued that there were “striking similarities between remembering the past and imagining or simulating the future, including the finding that a common brain network underlies both memory and imagination.”

Bordalo, Gennaioli, and Shleifer’s paper could have much broader implications, building on Kahneman and Tversky’s work to develop a better model of how human cognition works and how humans make decisions. Bordalo et al.’s work also points to where quantitative investors might be able to find alpha in the market. Fundamental investors will likely rely on their memories to make probability judgments, matching prospective investments to similar ones that worked in the past. And they are more likely to project the recent returns and correlations of different asset classes to remain constant. Quantitative investors, in contrast, will draw on a fuller probability distribution and, ideally, be able to make more accurate probability assessments that are not as biased by recent experience and human memory.