Elicitation of Subjective Risk Perceptions
Before risk can be managed it has to be defined. And most of the risks that individuals, households, corporations and societies deal with involve subjective risks. This might seem counter-intuitive in some classical fields of financial risk management, but even actuaries rely on subjective judgments about future trends. And before we can evaluate if someone is making good or bad decisions towards risk, we need to know what risk they perceive. It could be that it is the perception of risk that causes bad decisions, and not attitudes towards risk. Or it could be that risk perceptions are not correctly updated as new information arrives, leading to bad decisions over time.
The first task is to develop tools to elicit subjective risks. In the case of eliciting subjective probabilities for binary events, such as the chance that there will be a positive return to the stock market next year, empirical methods are relatively well advanced. In the case of eliciting subjective belief distributions for continuous events, such as the return on the stock market next year, things are not so advanced. CEAR-sponsored research has led to the characterization of methods for eliciting these distributions, with incentives for respondents to take the task seriously. Those theoretical results can be tested and evaluated in controlled laboratory experiments before being applied.
The second task is to apply these tools. One application is to elicit subjective belief distributions of Chief Risk Officers over major international financial risks for the next year. This application, in conjunction with Bloomberg, considers economic and financial risks. Each month a panel of CROs are contacted on the web to update their beliefs about these risks, with incentives in the form of contributions to charity for better predictions. These beliefs are then collated and compared to those from a conventional econometric forecasting model to allow risk managers to see if they agree or disagree for each risk. If they disagree, then that is a signal that risk managers need to evaluate the statistical forecasts more carefully: is there something that these “canaries in the cave” can sense, from their close exposure and study of these risks, that the mechanical statistical forecasts cannot? If the CROs are in considerable disagreement with each other, is that also a signal that there are some background risk factors that need to be given more careful attention?
A second application is to provide a richer characterization of financial and statistical literacy. We can think of literacy as reflecting someone’s knowledge of a proposition, and that knowledge need not be “all or nothing.” That is, someone might have relative precise knowledge of some proposition, such as the implication of a compound interest calculation. Or they might have very imprecise knowledge. Or they might have biased knowledge, whatever the level of precision. One feature of this approach is to provide a more informative measure of the extent of literacy that an individual has in different domains: someone might be literate in terms of financial choices, but quite illiterate in terms of health risks, and of course both will be important for making decisions about retirement. By eliciting their subjective belief over the true answer one can immediately. CEAR has used these ideas to evaluate financial and statistical literacy of individuals, and extensions to consider health literacy, household literacy, and even social literacy are planned.