Price Discrimination In Insurance

In previous posts I have defended the practice of “price discrimination,” charging different prices for the same service to different customers based on their willingness to pay. Without it many mutually beneficial transactions, perfectly satisfactory to the customer and profitable to the producer, would not happen. Some industries would run at much higher costs, innovation would happen more slowly, and conceivably some business models or even industries wouldn’t exist at all. It makes perfect sense for businesses to offer a high-ball “sticker price,” then try to attract the marginal customer with various sneaky discounts for the same product.

Have you heard about our senior discount?

The software suite is $150. Oh, you’re a poor college student, you say? I mean it’s only $25.

That’ll be $425 for the frames plus the lenses. Of course that’s with the anti-reflective coating. Without the coating, it’s only $250. Oops, they went ahead and manufactured them with the coating anyway. We’ll give you that for free, then.

Examples abound.

The insurance industry is finally getting into this game, within the past decade or so. Insurance companies know a lot of details about their customers. They have big databases of every single customer including dozens or even hundreds of attributes. They make frequent changes to their rates, and they can observe when people leave their company for another insurer. They can use this kind of data to create predictive models that estimate price elasticity. In other words, “What is the probability that Bob will renew his policy if I increase his rates 5%? 10%? 15%?” By doing this on a large portfolio of insurance customers, they can figure out what the effect of a rate change will be on customer retention. In fact, they can use this kind of information to decide how big of a rate change to take. Or (much more likely) for a given rate change, how should this rate be spread around to their various customers? If I know I need $1 million in additional premium, this kind of elasticity measurement can help me minimize the disruption by giving price-insensitive customers the biggest rate increases and price-sensitive customers the smallest increases. It sounds almost sinister, until you realize that every other industry does this in some way. Still, within the industry many consider this a big no-no. It’s called “price optimization,” and it’s sort of a political hot-button right now.

The barriers to implementing price optimization in the insurance industry are both cultural and political. Actuaries, the people who set insurance rates, all study a common syllabus. We all get our certificates from one of two organizations: the Society of Actuaries and the Casualty Actuarial Society. (That’s in the U.S.; other countries have their own accrediting bodies.) We practice our profession based on a set of “actuarial standards of practice,” one of which defines an “actuarially sound rate.” These standards of practice and the pricing equations that we study all tend to assume that the premium charged for an insurance policy has to be based on actual cost. Suppose two people have different risk attributes (age, credit, prior claims, etc.), but my predictive model says they both have an expected claims cost of $650. I have to charge them both the same rate, at least according to various actuarial equations and standards of practice. But actually, maybe they have the same expected claims costs, but one is likely to be a customer for a single year and then leave while another is likely to renew for the next ten years. The longer-retaining business actually saves me on expenses. I have to underwrite (inspect) both policies when I first write them, and this incurs an up-front business cost. But in one case I amortize those costs over one year and in the other I do so over ten years. The per-year expense is smaller for the policy with longer expected retention, so I can justify a discount based on expense differences.

Suppose now I have two people with the same expected claims costs and the same expected expenses. What I can’t now do is say, “His premium is $1,000, but his premium is $1,100 because he is less price sensitive.” Or maybe you can. It’s kind of a grey are. People disagree about whether the actuarial standards of practice explicitly forbid this kind of price differentiation. States vary as to whether they allow it or not. Some have specifically passed statutes banning price discrimination, so case closed. In other states, there is no specific statute outlawing price optimization. But there are various catch-all statutes that incorporate the definition of an actuarially sound rate. A rate has to cover the expected cost of a risk transfer, it cannot be inadequate, excessive, or unfairly discriminatory, etc. (More here.) A state with such a statute can object to price optimization because it runs afoul of the “unfairly discriminatory” clause (which really means some regulator just doesn’t like it), or because the rate isn’t explicitly based on the cost of that business. So anyone playing in this area potentially faces the threat of professional and legal sanctions. Objection letters and filing forms often ask perfunctory questions like "Are you practicing price optimization?" or "Do your models predict price elasticity? Do those estimates impact your rating factors?"

Some very confused regulators mistake cost-based pricing for price optimization. In my example above, I explained how someone who retains longer saves us money. That justifies a cost-based discount that has nothing whatsoever to do with that customer's price elasticity. I expect you to retain 1 year longer, thus saving me (say) 5% on expenses and justifying a premium reduction. These are called "lifetime value" discounts, based on expected retention differentials. Price optimization is different, and it often yields results that run in the opposite direction. Price optimization says, "I'm predicting an additional year of expected retention if I give you a 5% rate increase instead of the cost-based 10% increase. So I'll give you the 5%." (Except this is optimized across and entire book of business, not done one-by-one for individual customers.) In the first case you're not trying to affect behavior. You're simply offering a discount based on the calculated cost-savings from a higher retention. In the second case, you're effectively offering a discount to someone who is unlikely to retain unless you give them the discount. This is true price discrimination: tweaking the price in order to gain the marginal customer. These are very different concepts: lifetime value versus price optimization. But I had a recent back-and-forth with a regulator who repeatedly accused us of the latter, despite our numerous clarifications. (He finally gave up and approved our filing.)

Here is how I envision price optimization being used in actual practice. Insurers run predictive models all the time in order to change their rating plans. The rate differentials based on age, credit, gender, mileage, etc. are always changing, and new variables get added all the time. Adding "annual mileage driven" to your rate plan is going to change things like the male/female differential or the married/unmarried differential, because presumably these things correlate with mileage. So you run a new predictive model and everyone's rate changes a bit. Except some people's rates change a lot. Whenever we re-run these models, there is a bell-shaped curve of indicated rate changes. Some people are out on the tails getting big increases and big decreases. This is where price optimization comes in. Perhaps we don't immediately implement our new rating factors and cause all of that disruption to our book of business. Instead, we run a "price elasticity" model and do some complicated math to figure out how to spread these changes to the rating plans over several years. Instead of ramping my "male" factor from 1.05 to 1.20 like my loss model says to do, I'll raise it first to 1.14, then to 1.20. Or maybe I'll make this adjustment over three or more years. A sophisticated enough process can optimize what this transition looks like. That's the value of price optimization.

What I believe will not happen is the following: an insurer identifies a segment of the population that is particularly price insensitive, then permanently charges them a higher rate. That overcharged segment of the population will eventually leave that insurer for one offering a true cost-based premium. Insurers know this and want to retain that business at a profitable price. This price-insensitive segment may see a slower movement toward the cost-based premium, and this may mean temporarily higher-than-cost-indicated premiums. But it will probably not be permanently overcharged. I think this is what concerned regulators and consumer "watchdog" groups have in mind when they disapprove of price optimization, and it is unlikely to happen in an industry as competitive as personal-lines insurance.

Price discrimination is economically efficient. It's a way to maximize consumer and producer surplus. It means charging more to people who don't mind paying, and less to people who are do mind. It doesn't imply overall higher prices, as naive observers tend to assume. The "discounts" and "surcharges" tend to cancel overall. I've seen textbook models that describe price discrimination as "the producer captures much more of the consumer's surplus." This is a faulty model. I think it's more accurate to think of price discrimination as a set of discounts and surcharges that net out to zero, but that increase the overall surplus by making otherwise unpalatable transactions possible.