Bias, guess and expert judgement in actuarial work

Expert judgement is frequently used within general insurance. It tends to be a method of last resort and used where data is sparse, non-existent or non-applicable to the problem under consideration. Whilst such judgements can significantly influence the end results, their quality is highly variable. The use of the term “expert judgement” itself can lend a generous impression of credibility to what may be a little more than a guess. Despite the increased emphasis placed on the importance of robust expert judgements in regulation, actuarial research to date has focussed on the more technical or data-driven methods, with less emphasis on how to use and incorporate softer information or how best to elicit judgements from others in a way that reduces cognitive biases. This paper highlights the research that the Getting Better Judgement Working Party has conducted in this area. Specifically, it covers the variable quality of expert judgement, both within and outside the regulatory context, and presents methods that may be applied to improve its formation. The aim of this paper is to arm the insurance practitioner with tools to distinguish between low-quality and high-quality judgements and improve the robustness of judgements accordingly, particularly for highly material circumstances.