Credit Scores – could they turn toxic for insurers?

UK insurers enjoy pretty unrestricted use of credit scores in insurance pricing. Yet that may not last for many years more. Consumer groups think that some form of change is needed, while even within the sector, an increasing number think some form of change is inevitable. To paraphrase, there’s writing on the wall. The problem thought is ‘what exactly does it say’?

Working out what that ‘writing on the wall’ might be is tricky. People seem to take positions that offer little chance of converging. Yet are credit scores in insurance likely to continue unhindered? That seems very unlikely. Are they likely to be banned altogether in the UK? From my reading of the regulator, the answer is another no.

How will this play out then? Where might some common ground be found? And how will these divergent standpoints start to come together? The sector could of course just take a ‘wait and see’ position, sticking to its guns until the regulator ‘writes on the wall’ that some form of ban or restriction is being prepared. That I would caution against, for it could easily descend into a repeat of that ‘loyalty penalty’ challenge. Better to see what common ground can be found and build upon it.

Research Worth Reading

I’m going to draw for this article on some interesting research from the United States, by Professor Barbara Kiviat of Stanford University. She’s examined the positions taken in credit score debates by insurers and politicians and found patterns that I think will help UK insurers in three ways:

• to understand how the dialogue on credit scores in insurance might ‘erupt’ into a more public debate;
• what to look for in their own use of credit data;
• how they might best adjust their use of credit data in response to concerns

Let’s take a quick look at what Professor Kiviat was researching. She did a close analysis of policy debates in the US about the use of credit scores in auto insurance. The documentation she drew on included…

“…transcripts and recordings of hearings held by regulators and legislators, policy reports, actuarial and other studies, minutes from regulatory meetings, and texts of state laws. Hearings and meeting minutes are particularly valuable, because they reveal how actors work to convince each other that their way of thinking is the right one."

The US policy debate on credit scores in insurance is a long running and well documented one. Professor Kiviat analysed more than 6,600 pages of documents and 28 hours of recordings. What she was looking for was the moral basis upon which the two sides (insurers and policymakers) based their reasoning. In other words, why insurers saw their position as fair, and how policymakers saw certain outcomes as unfair. In Professor Kiviat’s words, her research…

“…examines the moral understandings justifying the use of personal data and algorithmic prediction in deciding who gets what, and the conditions under which stakeholders are likely to challenge those justifications.”

Constructive Engagement

So what, some of you may ask. Well, if you have two sides arguing about data and what is means, you’re only going to make that engagement constructive if you can show both sides how they overlap and how they differ. This allows them to move close towards an agreed solution, as opposed to just ‘talking at each other and getting no-where’. If that happens, any way forward is likely to be imposed by the side who can leverage the most power.

The US debate on the use of credit scores in auto insurance has played out against a widely accepted view that “…that it is fair to use data for pricing if the data actuarially relate to loss insurers expect to incur”. And by and large, policymakers have gone along with that, seeing it in general terms as sitting comfortable with their sense of ‘moral deservingness’.

So when insurers show that credit scores are a good predictor of losses, policyholders on the one hand see this, but on the other hand, wonder why that should be so. Insurers answer that question by pointing to the strength of the mathematical correlations. The cause (or to put it another way, the ‘why’) is something that insurers aren’t interested in. This is something the market has institutionalised as an undisputable fact, going by the name of actuarial fairness. And that was a fairness the market say as being not only in their own interest, but in the interest of consumers as well.

Perennially Questioned

Yet it is a fact (the correlation between credit score and losses) that is perennially questioned by US policymakers. This was not so much in relations to the maths, but into why credit scores predicted loss. This was in order to “…determine if any links in the causal chain held the wrong people to account, and as a consequence gave them prices they did not deserve.” And the reason for policymakers interest in that was the regulatory requirement across all US states for rates not to be ‘unfairly discriminatory’ – an absolutely key term in US insurance regulation.

So why were policymakers pushing insurers on this? One example brought up was low income families. Compared with a high income family with the same risk rating, the low income family was less able to afford to self-fund the cost of an accident. The conclusion - that the correlation of the credit score was not so much with risk but with making a claim.

Policymakers therefore saw income, not the risk inherent in personal responsibility, as being a driver of lower credit scores and more insurance claims. And so while policymakers saw it as fair that personal responsibility should influence premiums, they saw it as unfair that consumers should pay more because of their low earning power.

What this told policymakers was that unpicking the ‘why’ behind the use of credit scores in auto insurance raised issues around how well insurers were complying with that ‘not unfairly discriminatory’ requirement. This led them on to wonder why people had low credit scores in the first place. Five aspects of the subsequent discussions stand out.

Why Low Credit Scores Happen

One relates to factors that can impact credit quality, such as relationship breakdowns, job losses and health issues. Were these ‘life circumstances’ a matter of personal responsibility? In Professor Kiviat’s words…

“Individuals who fail to pay credit card bills because they lose a job and those who fail to pay their bills because they do not take financial obligations seriously are interchangeable to a credit scoring algorithm, but policymakers did not see them as morally equivalent.”

The issue here was context. So long as credit scores were determined largely on the basis of correlation, and with little interest on the causation, then the context that is lost causes insurers to miss out on information that can influence that ‘fair or unfair’ judgement. What this highlighted therefore was that what data you collected, and what you did with it, carried moral judgements within it. Actuarial fairness was not as fair as the market was making out.

A second aspect that policymakers in the US addressed was equality. Here’s what Professor Kiviat found in the policymakers’ position:

“If low scores stemmed from long-standing inequities, such as exclusion from mainstream credit markets, predatory lending practices, and residential segregation that made it tougher to find good-paying jobs and build housing wealth, then one could argue—and many did—that it was not fair to penalize minorities for their lower scores with higher prices. Minorities could not deserve higher prices by virtue of the discriminatory actions of others.”

What policymakers were doing was positioning insurers’ emphasis on actuarial fairness with their emphasis on equality. The problem for insurers was that actuarial fairness was a market practice, while equality was a legal obligation.

Weak Data

The third aspect was that credit scores can move not just because of what consumers did, but of what credit providing firms did. Professor Kiviat again…

“Too many inquiries (on a credit file), and a person’s credit score drops. A large number of inquiries can indicate someone is desperately applying for credit, but the same pattern can also mean a person is shopping around for the best rate on a loan.”

To not fail that key ‘unfairly discriminatory’ test, insurers needed to use credit scores with greater care and understanding. A correlation was not enough. Policymakers were particularly concerned about people having to pay more in premium just because they displayed financial responsibility.

And the fourth aspect concerned the level of credit information upon which the score was based. Sometimes credit information was missing because creditors failed to report complete information. Other times, a low credit score was because the person had chosen to borrow little to no money. Why should such people have to pay more for their insurance because of these two things? Why should they be disadvantaged for making morally justified choices? Policymakers couldn’t see any justification in it.

The outcome of many such debates between insurers and policymakers was that several states now ban the use of credit scores in insurance pricing, while others put various types of restrictions on how they’re used.  Insurers were in a sense paying the price for, in my opinion, clinging to a too simplistic application of actuarial fairness to commercially defined data sets. Maths, tradition and economics were now not enough.

Signals Insurers Are Picking Up

Is either a ban or restrictions likely here in the UK? Yes, but more likely the latter. There are clear links in the points raised above to the ethnicity penalty, the poverty premium and to digital poverty here in the UK. And those links are what I referred to earlier as the points on which the debate on credit scores in UK insurance will ‘erupt’.

So how might an insurer, not so much react, but prepare, for such a situation? I differentiate in that way because I think there’s time for them to prepare if they act now. If they delay, then they will have to rely more and more on ‘react’.

In summary, they need to review what credit related data they’re using, the makeup of that data and how they’re using it (which includes analytics). From this, they need to build a ‘use’ justification that is a more than just ‘well, the two are correlated’.

Sure, the insurer will have done a data minimisation assessment for compliance with data protection legislation, but that is rather a round peg for the square hole we’re talking about here. Just because your data minimisation says you can use data in that way, doesn’t mean you should. Remember that data protection is about privacy, not fairness or discrimination.

Why Insurers Need to Ask Why

As I said at the outset, the use of credit scores in insurance is contentious. The defence of actuarial fairness breaks down if not applied with care, particularly in relation to personal responsibility. That’s not happening enough at the moment and so challenges are being raised as a result. Politicians recognise the social justice implications of those challenges and are, at best asking questions of insurers, at worse challenging their use of credit data.

It's all a US thing, some of you may be thinking. I don’t think so. What is unique about the US is that the use of credit scores in insurance has been debated in several political forums, making the issues raised by both sides more comprehensive and more transparent. In other countries, the same (if not very similar) issues exist. The debate is just beneath the surface, waiting for challenges to erupt it onto policymaker’s agendas, as it is beginning to do here in the UK.

Asking these ‘why’ questions should help insurers to then take steps that make their use of credit scores more acceptable to the public and to politicians. It helps them remove points of contention from their credit data, before politicians and policymakers become tempted to ban their use of it altogether. The fate of lifetime value modelling should not be forgotten.

So, here are six types of question you should be asking of your credit related data. If you feel they are more than a bit challenging, ease your way into them by picking a particular issue and following it down into your data through a series of ‘why’ questions That issue could be gender, divorce, data poverty or the like. Follow threads and see where it takes you and what you find.

Six Things to Look for in Your Credit Data

You need to look at how the data that you have came to be packaged in the way that it did. What assumptions were used and why? What structural characteristics are built into it, and how do those characteristics affect the decisions you can draw from it?

In looking at the data you have, you also need to think about the data you don’t have, and ask ‘why’ questions about that too. The answers to those ‘why’ questions could have meaning for the ways in which you can use the data you do have.

What options do you have for adjusting the credit data you use so that it either doesn’t pick up certain things (such as life circumstances like divorce) or does pick up certain things (like more context)? In other words, does the credit data you use fit your needs? And how can that data be changed to reflect your evolving needs (such as more responsible and discerning use of certain pieces of data)?

As and when your data needs change, how can you replicate that in what your analytics have learnt? If your use of credit data was to change as your needs changed, are you able to learn that new change into your analytics, and unlearn now discontinued features out of your analytics?

While you have underwriting, claims and counter fraud needs for your data, what other  needs do you for your data? How have you framed these and embedded them into your digital strategy? That it is fair and non-discriminatory are two obviously ones, along with consent.

What responsibilities and obligations exist within your contractual arrangements with data brokers and software houses? Are you able to have certain pieces of data removed, or added, or treated differently? If your data is provided pretty much on an ‘oven ready’ basis, what options do you then have?

Language Matters

As Professor Kiviat demonstrates, unpicking the language of data is central to unpicking the different ways in which data is seen and then used. In particular, recognising that fairness is very much a multi-dimensional thing allows you to better take on board the alternative views of fairness that challenges are being based upon. What that results in then is this thing called the equality of fairness. In other words, the need to balance the different dimensions of fairness: fairness of need, fairness of access, fairness of time and the fairness of crowds, with of course the insurers’ favourite, the fairness of merit.