Periodic Reporting for period 4 - SkewPref (Skewness Preferences – Human attitudes toward rare, high-impact risks )
Período documentado: 2023-10-01 hasta 2024-02-29
The large share of decisions that humans make involves at least some risk. Among the most important risks we face are rare, high-impact risks (i.e. risks involving rare, high-impact events). For example, many useful medications come with rare but devastating side effects. The goal of this project is to provide a fundamental understanding of humans’ skewness preferences—our attitudes toward such rare, high-impact risks. Skewness preferences are much more influential on behavior than previously realized, and thus must take a central place in the economic analysis of risk. One reason is that skewness preferences have unexpected and far-reaching implications in dynamic decision situations, and thus their complex interaction with time is a major topic of this project. Besides offering fundamental results on the psychology of risk-taking, we explore applications, for example, in the domain of finance.
Part of the work focused on skewness preferences in static (i.e. one-shot) decision situations. We developed a general notion of what it means that skewness preferences are “strong,” that is, that humans are really sensitive to rare, high impact risks. We then studied this notion in various well-known theories of human risk preferences (for example, the psychologically motivated prospect theory of Kahneman and Tversky). We could show that leading theories make different predictions regarding the strength of skewness preference. While the leading “rational” theory, expected utility theory, cannot explain the strongest type of skewness preference we define, psychologically grounded (“behavioral”) theories can. We believe that this insight is important for economic modeling quite generally because most economic models require assumptions about how humans deal with risk and skewed risks in particular. Later in the project, we also proposed and analyzed a new notion of the strength of prudence, a preference trait that relates to precautionary saving.
Another subproject was concerned with the repeated risk-taking of skewed risks. Penny-picking refers to the often-observed phenomenon of repeatedly taking negatively skewed risks and seems directly at odds with the evidence on (positive-)skewness-seeking as observed in static settings. I show that penny-picking may not only occur despite skewness-seeking, but—seemingly paradoxically—because of skewness-seeking. With sufficient time available, risks with arbitrary negative skewness can be gambled in such a way that, overall, skewness is positive. Therefore, classical behavioral theories like prospect theory straightforwardly explain penny-picking. More generally, I show that the versatile dynamics of skewness reconcile apparent preference reversals concerning the avoidance and acceptance of (skewed and non-skewed) risks.
Another work stream focused on whether, and how, skewness preferences reflect in the prices of financial assets (e.g. stocks and options). Specifically, we studied the asset pricing implications of skewness preferences as implied by probability weighting (the idea that investors overweight rare, high impact events). We obtained and empirically validated several novel implications for asset prices, in particular on what experts call the option-implied premiums on variance and skewness. Moreover, our model allows us to disentangle pricing effects of variance and skewness. The latter are often confounded using standard modelling approaches. This subproject, but also some of the others, benefitted of an in-depth understanding of how skewness and correlation reflect and interact within simple two-outcome risks. In one subproject, we also collected some results on such risks together with economic applications.
In another, experimental subproject, we proposed a method to elicit stopping times—each subject’s complete contingent plan for taking a risk for up to five times—to study repeated risk-taking under precommitment. In addition to time- and outcome-contingent risk-taking, we allow some subjects to use path-dependent or randomized stopping times. Our experimental design thus allows for hundreds of different risk-taking plans. Using an unsupervised machine-learning algorithm, we find that individuals’ risk-taking strategies map well to stop-loss, take-profit, or buy-and-hold strategies. Most strategies are of a continue-when-winning and stop-when-losing type, with a profit-trailing stopping barrier. Path-dependence and randomization are used extensively, even if they are costly. We further analyze dynamic consistency in a sequential risk-taking task and find that subjects largely follow the unconstrained plans that we elicited.
The work as described above has resulted in six research papers thus far. The results were disseminated to researchers worldwide at various occasions, including conferences, workshops, invited seminars, and keynote addresses.
Another subproject was concerned with the repeated risk-taking of skewed risks. Penny-picking refers to the often-observed phenomenon of repeatedly taking negatively skewed risks and seems directly at odds with the evidence on (positive-)skewness-seeking as observed in static settings. I show that penny-picking may not only occur despite skewness-seeking, but—seemingly paradoxically—because of skewness-seeking. With sufficient time available, risks with arbitrary negative skewness can be gambled in such a way that, overall, skewness is positive. Therefore, classical behavioral theories like prospect theory straightforwardly explain penny-picking. More generally, I show that the versatile dynamics of skewness reconcile apparent preference reversals concerning the avoidance and acceptance of (skewed and non-skewed) risks.
Another work stream focused on whether, and how, skewness preferences reflect in the prices of financial assets (e.g. stocks and options). Specifically, we studied the asset pricing implications of skewness preferences as implied by probability weighting (the idea that investors overweight rare, high impact events). We obtained and empirically validated several novel implications for asset prices, in particular on what experts call the option-implied premiums on variance and skewness. Moreover, our model allows us to disentangle pricing effects of variance and skewness. The latter are often confounded using standard modelling approaches. This subproject, but also some of the others, benefitted of an in-depth understanding of how skewness and correlation reflect and interact within simple two-outcome risks. In one subproject, we also collected some results on such risks together with economic applications.
In another, experimental subproject, we proposed a method to elicit stopping times—each subject’s complete contingent plan for taking a risk for up to five times—to study repeated risk-taking under precommitment. In addition to time- and outcome-contingent risk-taking, we allow some subjects to use path-dependent or randomized stopping times. Our experimental design thus allows for hundreds of different risk-taking plans. Using an unsupervised machine-learning algorithm, we find that individuals’ risk-taking strategies map well to stop-loss, take-profit, or buy-and-hold strategies. Most strategies are of a continue-when-winning and stop-when-losing type, with a profit-trailing stopping barrier. Path-dependence and randomization are used extensively, even if they are costly. We further analyze dynamic consistency in a sequential risk-taking task and find that subjects largely follow the unconstrained plans that we elicited.
The work as described above has resulted in six research papers thus far. The results were disseminated to researchers worldwide at various occasions, including conferences, workshops, invited seminars, and keynote addresses.
The research has advanced the state of the art in the research field of decision analysis (and decision-making under uncertainty in particular) in various ways. We advanced the fundamental understanding of skewness preferences in static and dynamic decision situations. In particular, we succeeded in establishing a theory-free notion of skewness preference, the first to allow for a non-trivial analysis in leading theories of decision-making under risk. We gave a comprehensive theoretical analysis of the forces that are at work in determining the repeated risk-taking of skewed risks. And we derived and tested novel and important implications of skewness preferences for asset pricing, in particular concerning the so-called variance premium and the so-called skewness premium. The project has also resulted in a new method to elicit stopping times and tests for time-inconsistency.