WP 2013-08 Subjective Beliefs and Statistical Forecasts of Financial Risks: The Chief Risk Officer Project
Published in 2014 as a book chapter in Contemporary Challenges in Risk Management: published by Palgrave Macmillan UK, and edited by Torben Juul Andersen & Stanley Mayes.
AUTHORS: Glenn W. Harrison and Richard D. Phillips
ABSTRACT: Information about financial risks comes from many sources. We formally consider how one can elicit and use information from two important sources when making forecasts. One source is a traditional statistical forecast, using familiar econometric methods for extrapolating from the past to the future. The other source is the elicited subjective belief distributions of “experts” in this domain: Chief Risk Officers of major international corporations. We demonstrate how these beliefs can be elicited in a formal, structured and incentivized manner, and critically contain information on the precision of the individual’s belief for each risk. We characterize the manner in which these two sources tell different stories about these risks, arguing that any distributional differences, or similarities, between the two sources are informative for risk managers. Finally, we characterize the degree of consistency among our experts: are they “on the same page” in their beliefs? We argue, again, that consistency or inconsistency of subjective beliefs is in itself informative for risk managers.
We first explain the basic idea of bringing together these two sources of information on future risks, and what we seek to learn from this exercise. Then we review the subjective belief elicitation procedures we developed, and document the statistical model we developed to provide the traditional forecasts for comparison. The subsequent section offers some initial findings from our experts, and discusses how we characterize the consistency of beliefs: aggregate subjective beliefs and the statistical forecasts on the one hand, and the heterogeneous subjective beliefs themselves on the other hand. Finally, we conclude and discuss the possible extensions of our approach.