In parallel there have been rising concerns about data privacy and hidden implications of using “black-box” price predictions.
An obvious concern when using black-box models for pricing is that of implicit discrimination. As we will see below, this is always a potential issue, regardless of the model, and this is regulated in EU-law, see [4]. A related question is that of (algorithmic) fairness, and below it will be seen that these two concepts will often fail to agree.
Further, in order to adjust for implicit discrimination, discriminatory information needs to be collected (more on this below), which is a privacy concern in itself. More generally, this relates to the question of which covariates that are suitable to use for pricing. This in turn connects to discussions about covariates’ causal effects and risk factors. However, below it will be seen that this is not essential for avoiding implicit discrimination.
Lees dit artikel verder onder Download.
Download
- Insurance pricing: discrimination, causality, and fairness .pdf • 0,08 MB