OBEN
The widespread digitisation of the economy has created a wealth of data on risk exposures for insurance companies. But how insurers exploit what has become known as Big Data has prompted widespread discussion around data ethics, especially in regard to transparency and fairness where consumer data is used to market and to price risk.
Insurers are already familiar with the concept of adverse selection. Now they are getting to grips with the new concept of inverse selection that arises with Big Data.
It’s useful to look at a basic risk spectrum to explore how Big Data can fundamentally change the information balance of insurers and insureds, and the latter’s perception of risk.
Usually, the consumer has private information about where s/he is located on the risk spectrum, exposing the insurer to the risk of adverse selection. Using Big Data combined with machine learning, the insurer might be able to uncover the location and eliminate the information asymmetry. Where the consumer’s risk perception is poorly calibrated, the insurer might be able to reverse the information asymmetry in its own favour by having a more precise read of the risk than the consumer has, and even determine the consumer’s perceived location.
With this reversal in information asymmetry in mind, how should the insurer price risk – on the basis of actual risk or the consumer’s perception of risk? And should the insurer make Big Data available to the consumer or to society?
Consider the basic consumer risk spectrum (see below). The insurer has a read of the average risk on this spectrum but doesn’t know the locations of the individual consumers. The consumers on the other hand know their locations.
Knowing only the average risk, the insurer prices the policy to the centre of the spectrum. All higher risk consumers will purchase insurance and all lower risk consumers will not. In consequence, the insurer won’t break even. It’s a simple representation of adverse selection.1
When the insurer is equipped with Big Data, this situation changes, however. The insurer now has granular consumer information, potentially observed at high frequency, and can gauge the consumers’ true locations on the risk spectrum – and differentiate premiums accordingly.
In this case, every consumer will purchase insurance. The lower risk consumers now have access to insurance at premiums commensurate with their actual risks. The higher risk consumers will purchase insurance also, albeit at higher premiums, no longer collecting an information rent. And the insurer will break even. Clearly, a socially desirable outcome.
What happens if the consumer’s perception of risk is not well calibrated? Let’s look at consumers 4 and 8 (see below). Consumer 4 erroneously believes him/herself to be lower risk, whereas consumer 8 errs in the opposite direction. If the insurer prices to the actual location on the risk spectrum, then consumer 4 will not purchase a policy; consumer 8 on the other hand will still purchase a policy, at a premium that s/he perceives as a bargain.
Inverse selection arises if the insurer sets the premium to the maximum of the actual risk and the consumer’s perceived risk. Then the insurer will collect an information rent on consumer 8.2
If consumer 4 knew that s/he is higher risk, then s/he would purchase insurance at the quoted premium and s/he would be better off. This situation of the consumer’s misperception of risk raises the question of whether the insurer should make its information available to the consumer.
Objectively, Big Data has the potential to broaden access to insurance by removing information asymmetry – the elimination of the consumer’s information rent comes at no loss to society.
However, there is the potential for shifting the information balance in favour of the insurer, which would allow the insurer to earn an information rent – the broader availability of insurance remains.
Consumers with poorly calibrated risk perception would benefit from having as equally a precise read of their risks as the insurer. That’s why some have called for making the insurers’ Big Data available to the consumer or, more generally, to society.3
