The advent of big data, fuelled by declining data-storage and processing costs as well as increasing connectivity, fundamentally changes the environment in which insurance operates.
Big data, however, also raises a number of concerns. These do not only involve ethical and societal concerns about privacy, data protection and “unfair” discrimination, but also questions about the welfare consequences.
This article uses standard economic welfare analysis to consider the implications of enhanced risk understanding versus banning the use of risk information in the underwriting process.
For the purpose of this analysis, we uniquely focus on the welfare implications of risk classification and do not consider broader concerns about privacy and data protection. The motivation of this analysis is to develop an analytical framework to evaluate the efficiency implications of different policy choices.
Life in Eden: The value of premium risk insurance
We refer to “Eden” as a situation in which no one (neither insureds nor insurers) knows about individual risk characteristics. Take life insurance before the invention of genetic testing as an example. A holder of a specific gene may have a high risk for developing a certain illness, but neither he nor the insurer knows about this. Hence, all individuals pay the same premium rate, irrespective of whether they are holders of a specific genetic predisposition or not.
Now suppose that we bite into the apple of knowledge, and both insurers as well as the insured learn about individual risk characteristics. Insurers will use this knowledge to differentiate premiums between high- and low-risk individuals.
Low-risk individuals clearly benefit from this scenario, as they now pay a lower premium, while individuals with a specific genetic predisposition will have to pay a higher rate. However, ex ante (ie before information about risk types is available), individuals may prefer to stay in Eden for fear of being revealed as a high risk. The absence of individual risk information thus acts as a protection against the risk of being revealed as high risk and having to pay a high insurance premium. We will refer to this in the following as “premium risk insurance”.
Risk-averse individuals will typically value this insurance. As a result, for risk-adverse individuals, the expulsion from Eden would lead to an overall loss of welfare. But also insurers may benefit in Eden: they not only sell an additional insurance (the insurance against premium risk) and earn additional premiums, they also save costs for risk assessment. The value of premium risk insurance to individuals will increase with the degree of risk aversion and the uneven distribution of risk types.
In reality, however, we hardly ever observe any “Eden” insurance markets. In many cases, individuals have at least a sense of their risk type. Moreover, both insureds and insurers may have an incentive to bite into the apple of knowledge to get an information advantage. This information advantage would enable them to engage in adverse selection behaviour.
Escaping Eden: The impact of adverse selection
As has been shown by Dionne and Rothschild (2014) and others, as long as the insurer cannot use risk-sensitive information and/or is not allowed to differentiate insurance premiums, asymmetry of information can cause adverse selection, with high social costs.
Suppose individuals now know their risk types, and insurers are banned from using risk information. As high- and low-risk individuals would pay the same premiums, this may give rise to the classical adverse selection scenario: the premium rate set by insurers is unattractive to low-risk types and they will stop buying insurance coverage. As a result, the average losses of the remaining risk pool will increase, and so will the premium, which will trigger additional low-risk individuals to drop out.
In equilibrium, only high risks will buy insurance, and for low risks, there will be no insurance coverage. From a welfare perspective, this situation is clearly inferior to risk-adjusted premiums. Adverse selection represents a social cost that must be taken into account in the design of regulatory policies.
Knowledge as the architect of fortune
While biting into the apple of knowledge may reduce the value of premium risk insurance, there may also be important benefits. In particular, knowledge about different risk levels and premium rates that reflect those risk levels may provide an incentive to invest in mitigation. In order to reduce their risk, individuals can build their property in a different location, install sprinklers, change their lifestyle or have preventive medical treatments.
In a world without insurance, they would be rewarded by the reduced level of risk and expected losses. Hence, they would want to invest in risk mitigation as long as the utility gain from the reduced risk outweighed the costs.
However, if individuals were insured, they would only invest in risk mitigation if the measure led to a premium risk reduction that outweighed the costs. This can only be the case if premiums are risk sensitive. Then risk mitigation leads to an increase in individual and aggregated utility. If the net gain from risk mitigation is large enough, the aggregated utility can even exceed aggregated utility in Eden.
The welfare effect from banning the use of risk information by insurers depends on the net effect between the gain from premium risk insurance, the cost of adverse selection, and the loss of risk mitigation incentives. These elements can differ significantly between different kinds of insurance products. Hence, from an economic welfare perspective, there is no general answer to the question whether the use of risk information should be banned
In the following, we apply our framework to a number of different cases for illustrative purposes. These case studies represent only a first cursory assessment. More in-depth quantitative research would be necessary to come to a definitive conclusion of these cases:
• Banning the use of gender information as a risk indicator in motor insurance: In most countries, motor insurance is mandatory and, therefore, adverse selection is eliminated. It is also unlikely that many men will change their gender in order to pay less on their motor insurance and that premium-rate differentiation between genders leads to any kind of risk mitigation. So it is unlikely that banning the use of gender as a risk indicator for motor insurance leads to a significant welfare loss.
• Banning the use of genetic information for life protection products (mortality/disability): Under the current circumstances, banning the use of genetic information by insurers provides a high benefit from premium risk insurance. At the same time, as most individuals do not possess genetic information, costs from adverse selection are currently low.
It is expected that genetic tests will become affordable for a significant part of society and thus create the basis for adverse selection. Under these circumstances, banning the use of genetic information could lead to a reduction in welfare.
Whether the use of genetic information incentivises risk mitigation has to be analysed case by case. Genetic information could allow for preventive actions to reduce the likelihood of the outbreak of a specific illness. In these cases insurers should offer a uniform insurance product which also covers the costs for risk mitigation.
• Banning the use of geolocation as risk indicator in property insurance: The location of a property has a significant effect on its exposure to natural disasters. Even though it may be difficult for individuals to correctly assess the risk exposure of their location, they usually know whether the risk is relatively high or low. Therefore, banning the use of geolocation as a risk indicator offers the possibility for adverse selection and banning the use of geolocation information in property insurance is very likely to reduce welfare.
• Banning the use of telematics and health trackers to incentivise prudent behaviour: New technologies have the potential to support and incentivise prudent behaviour by policyholders, thus reducing the overall level of risk. Telematics, for example, has the potential to positively affect prudent driving behaviour. A ban on the use of telematics in motor insurance would therefore likely reduce welfare.
As we have seen, whether a ban on the use of risk information in underwriting enhances or reduces welfare needs to be assessed on a case-by-case basis.
Figure 1 summarises these results. In the bottom left corner, both the likelihood that a ban on using risk information creates adverse selection costs that outweigh the loss of premium insurance and the likelihood that a ban reduces risk mitigation incentives are low. Hence, a ban on using risk information is likely to be welfare enhancing. As we move towards the upper right corner, the likelihood that a ban enhances welfare decreases.
Dr Benno Keller is Head, Research and Policy Development, Zurich Insurance Company Ltd and Dr Christian Hott is Senior Economic Advisor, Zurich Insurance Company Ltd. The paper reflects the personal view of the authors and not necessarily that of Zurich Insurance Group. The authors would like to thank the participants at the 2015 meeting of the Geneva Association’s Annual Circle of Chief Economists and David Swaden for helpful comments and suggestions.