Bayes research could deliver fairer insurance deals for customers
Academics from Bayes Business School and other universities have been researching 'proxy discrimination' in the insurance sector and some other financial services for several years. They may have identified a cure.
Insurance companies could use a new research-based tool to reduce ‘proxy discrimination’ in the pricing models that shape premiums in the sector.
Proxy discrimination happens when an algorithmic pricing model indirectly infers characteristics – such as ethnicity and sex – from other information provided by potential customers. For example, occupation can act as a proxy for sex, and postcode for ethnicity.
Researchers at Bayes Business School, which is part of City St George's, University of London, developed a framework that measures the extent of such bias. The tool also identifies the variables that contribute most to proxy discrimination, along with those which sometimes actually reduce it.
Direct discrimination in pricing based on protected characteristics is illegal in most circumstances.
Co-author Professor Andreas Tsanakas, Professor of Risk Management at Bayes Business School, says the sector could adopt the framework, which has been published in the European Journal of Operational Research, as it is applicable to most types of insurance cover. It could also be used to identify proxy discrimination in other financial services – such as credit scoring.
He said: “If they choose to, insurance companies could reduce indirect discrimination by using this framework. It could also be a useful diagnostic tool for auditors and regulators. It’s up to regulators to set out clear principles and incentives for insurers to act on the issue.”
Identifying the presence of proxy discrimination requires data on the policyholders’ protected characteristics, only some of which are collected. While potential customers are often asked their sex, for example, data about individual policyholders’ ethnicity are generally not collected. Earlier work by Professor Tsanakas and his co-authors has gone some way to addressing those challenges.
The paper also revealed that some variables can actually reduce proxy discrimination – suggesting, Professor Tsanakas said, that the interplay between pricing factors and fairness is even more complex than previously recognised.
A minor impact of proxy discrimination at a portfolio level, the paper found, can mask significant impact on specific demographic groups. For example, when the researchers introduced policyholder-specific measures to a real-world motor insurance pricing model, they found young drivers from one ethnic group were systematically quoted higher premia. That variation was partially attributed to proxy effects.
Such unfairness, Professor Tsanakas said, can drive financial exclusion for particular demographics.
The study is the latest in the discrimination-free insurance pricing research programme in which Professor Tsanakas is working with academics in Sweden and Switzerland. That programme involves the co-authors of this paper: Mathias Lindholm, Associate Professor at Stockholm University, insureAI CEO Ronald Richman, and Professor Mario Wüthrich from ETH Zurich.
Picture credit: JPSS imagery: CSU/CIRA & NOAA/NESDIS, Public domain, via Wikimedia Commons
.png){kind=link}