5 | Case Studies: Challenges for insurers Life Insurance Pty Ltd (LIPL) is constructing a Generalised Linear Model (GLM) to price their term life insurance policies. A GLM is a common statistical model used in insurance pricing, allowing prediction of expected claims costs from historic claims and policy data. In conducting this analysis, LIPL observes that: • • The inclusion of the occupation category data results in a material change to the fitted model’s estimated effect of sex on risk. This means that including ‘occupation categories’ such as professional or blue-collar categories as an item of data in the analysis impacts the measured effect of a customer’s sex on the expected mortality risks in term life insurance, and ultimately, the price proposed by the model. Similarly, a decision to split a particular occupation type into two subtypes further changes the sex relativities produced by the model. LIPL is debating whether to include occupation categories within the model. LIPL notes that if it chooses not to include occupation or occupation categories this will cause the premiums for lower-risk males to be significantly greater than if occupation is included, as males tend to perform a higher proportion of riskier occupations than females, such as working in blue-collar occupations. Should an insurer include all available data? An insurer would not be expected to include all data it holds that may be correlated to risk in its model. Almost all information may have some impact on the model if it is included, and clearly it is unreasonable to require all data to be used or analysed merely because it may be available. The insurer must use their commercial judgement to decide what information to collect and include in their model. 36 If the model results in discriminatory outcomes, provided such discrimination is based on data on which it is reasonable to rely and objectively reasonable, then the data exemption under the SDA should apply. By including a particular occupation category, LIPL may be able to deliver outcomes for its customers that might be considered fairer. The premium that people pay will be adjusted in accordance with their occupation category and the expected risks it entails for the insurance cover. However, occupation category data should be used carefully if it is correlated with other protected attributes. LIPL should be confident that the use of occupation category data is reasonable, having regard to all other protected factors. Does the data exemption apply? The availability (or not) of the data exemption under the SDA in this case study is unlikely to turn on whether LIPL includes occupations and occupation categories in its analysis or not. Either practice may be defensible. It is assumed that statistical or actuarial data shows that men, all other things being equal, have a higher mortality rate than women (for whatever potential reason) and it is reasonable to rely on such data. The data exemption will apply if it is reasonable for LIPL to discriminate against men by charging them higher premiums for the higher expected risks. In determining whether this is reasonable, relevant factors may include the practice of others in the insurance industry, and the commercial judgement of LIPL behind this decision.