Virtual Summit Shorts: The Discriminating (Pricing) Actuary

Insurance has always relied on finding a balance on the appropriate level of discrimination. Here, we mean discrimination in the neutral sense – how we differentiate treatment (like an insurance price) between different people. However, the ‘right’ answers continue to change over time as the context of social expectations and economic incentives change.

A timely reminder of some of these principles was provided at the 2021 All-Actuaries Virtual Summit. Fei Huang presented some work (done jointly with Jed Frees) which was motivated by the increasing need to incorporate ethical considerations into the technical teaching in actuarial education, particularly in an era of big data. Some of this material is also being incorporated into Australian university coursework too. The paper is available at SSRN.

One of the key ideas in the presentation is a framework for thinking about why certain rating factors are acceptable or not.

Some of these reasons are social reasons (like favouring variables that are controllable by the customer or have a causal connection to claim risk) while others are economic (the insurer should recognise an important factor or risk adverse selection and unsustainable pricing pressures).

In making these points, the research takes an international view of what practices are permitted (or not!) in different jurisdictions. America sees tight regulation and often allows rating only on a small number of permitted attributes. Price optimisation is also commonly not allowed. The EU prevents gender-based differential pricing for many insurance classes.

Some contemporary examples were also presented. For example, Wisconsin state passed a law preventing a COVID-19 diagnosis (or suspicion of a diagnosis) be factored into health insurance offers. Such a rule conforms to social norms (collectivism in the face of the pandemic), even if there are strong economic reasons why a diagnosis is relevant to future health costs.

The research then explores the issue of proxy variables; what should happen when a variable is predicted (e.g. not allowed to price on gender), but other variables correlate (e.g. smaller engine sizes might be more common for females)? In such cases, there are a variety of treatments possible, although there may be multiple interpretations of what is ‘fair’ in such cases. Such issues are heightened in a world of big data; a greater number of variables are available, and the risk of proxies increases. Some variables may even be outputs of machine learning algorithms,

The wide variety of approaches internationally means that there is much to learn about what works and what does not. Huang & Frees have done an impressive job collating this information across countries and classes of business to give an informative take on the use of price discrimination in insurance.

CPD: Actuaries Institute Members can claim two CPD points for every hour of reading articles on Actuaries Digital.