Risk-based pricing
There are two routes an organisation can take when setting a price for its product: they can calculate the cost of producing that product and add a margin, hoping the market will pay the resulting price; or they can start by taking their price from the market and then find ways to produce the product for a cost low enough to generate a sufficient profit.
While the former approach may initially sound like the orthodoxy, in reality, any business operating in a competitive environment must follow the latter route to some extent. This is particularly true for lenders whose product, money, is many ways a commodity.
However, calling lent money a commodity is also something of an oversimplification. In a customer’s hands, $10,000 is $10,000 but we don’t price loans based on the lump sum paid out today, we price them based on the series of incoming payments we expect to return to us in the future — and that is where risk comes into the equation. If you know with great certainty that you will receive all 36 of your expected payments, you need only add a small margin on top of your costs to make it profitable. If you are worried that some of the later payments might never actually arrive, you need to add a larger margin so that the payments you do receive will compensate for the ones you don’t.
Risk-based pricing strategies, then, seek to assign a fair price to each customer – or, more accurately, to each group of customers.
In its simplest form, the logic is that every customer is equally expensive to manage, it is only the cost of risk that fluctuates.
This is not strictly true. Typically it is cheaper to acquire high-risk customers, for example, but since it is also typically more expensive to service customers that frequently end up in collections we can get away with the simplified logic for now. So when issuing loans, unless risk-based pricing is implemented lower-risk customers will be subsidising the cost of higher-risk ones— where the combined costs of the loan exceed the price.
This situation is doubly problematic as it doesn’t only lead to sub-optimal profit but also to a sub-optimal portfolio structure. The lowest risk customers (who are being over-charged) can be tempted to leave for cheaper competitor offers while the higher risk customers (who are being under-charged) will gravitate towards your product; leading to a riskier portfolio on average. Risk-based pricing addresses this by lowering the rates charged to low-risk customers and raising the rates charged to high-risk customers.
The key to successful risk-based pricing is, not surprisingly, a good understanding of risk so scorecards are once again at the heart of any strategy. But now we’re not thinking in terms of a binary approve/ decline decision. Instead, we’re thinking on a near-continuous spectrum where we accept everyone at the right price (only declining when the ‘right price’ is too high to be legally, ethically, or otherwise acceptable).
Ideally, all costs should be added together to get an accurate ‘right’ price for each loan but in most cases, it is sufficient to use a relative pricing approach. This means that the average risk group is assigned a baseline price – the sort of price everyone would be charged before risk-based pricing was implemented. All other risk groups are then assigned a price premium or discount based on their risk relative to that of the ‘average’.
In a simplified example, if the risk of the full portfolio is such that a 10% interest rate is required, then the more sophisticated the models in our arsenal are, the more individual groups of differently-risky customers we can identify. We’d then find the smaller group whose risk is exactly average and charge them 10%, and adjust each other groups price up or down based on how much more or less risky they are.
When implementing risk-based pricing for the first time, though, it is worth following a phased approach starting by charging premiums to higher risk customers first. High-risk customers that are being undercharged are already unprofitable and so even if they leave due to the higher prices, they increase the lender’s profitability while doing so.
It is more complicated when we implement a price discount, though. For a discount to increase the overall profitability it needs to cause a compensatory increase in demand from the market. If all that happens is that the customers who used to apply before the discount was advertised are the same ones that are applying again now, or if the increase in demand is relatively small, then portfolio profitability can be erased.
As such, some understanding of price sensitivity needs to be established. And the prices offered need to make sense within the context of the market’s wider offerings. Mapping prices in the market is relatively easy with a bit of research, but estimating price sensitivity can be difficult. The best way to try is via a series of test-and-learn or champion-challenger experiments.
The goal of these experiments would be to see how small adjustments to the size or form of a discount move campaign response rates in different customer segments. You might start by identifying the score segment that you wish to attract with a discounted price, and then divide them randomly into one of three test groups. The first of those groups is our control. The consumers randomly assigned to this group will receive an offer to apply for a loan at the non-risk-based price, so 10% in our earlier example. The purpose of this group is to establish a baseline level of demand for our loans.
Then come the more interesting groups. Each of these should receive an offer to apply with all the product details left exactly the same, except now the offered price is adjusted by a smaller and a larger amount: so maybe test group one is offered a loan with 9.9% interest rate, while test group two is offered a loan with 9.75% interest rate.
How many consumers get assigned to each group is determined by the size of the population and the level of accuracy needed by our models, plus the level of conservatism we wish to apply. So if there is significant organisational resistance to the idea of a risk-based price we might assign half of the prospects to the control group and forty per cent to test group one with its small discount so that only a tiny fraction of loans are issued at the most discounted price.
Over the following weeks, we will then measure response rates, to see if the uplift in demand from the control group to the test group and from test group one to test group two is sufficient to justify the foregone interest. And we’d do that again and again, adjusting the discount, maybe trying out a sign-on bonus instead of an explicit discount, always improving our understanding of the impact that price has on demand.
In an ideal world, this would allow us to turn the two-dimensional matrix above into a three-dimensional cube where risk determines the minimum price charged but price sensitivity determines how close to this minimum the actual offered price should be.
To be honest, though, while the knowledge of prices impact on demand remains vital to any lender, the degree to which price sensitivity can be used to optimise discounts upwards is arguably less significant in a world with online price comparison services.
When doing a traditional variable price campaign, we would always have used direct mail because we didn’t want a customer in one test group to know of a bigger or smaller discount lest it influences the response. I remember some campaigns that we tried to run in markets where customers often received mail at their work address, think miners or soldiers who live on site. These campaigns tended to be noisy with response rates affected by one prospect becoming aware of another’s offered price.
In many ways, price comparison services create the same scenario. Now, instead of a customer in one test group coming into contact with a customer in another, you have one loan campaign coming into contact with any number of other loan campaigns and it is hard to see how a lender would risk anything but their best price in such a situation.
In some research I did in Hong Kong a few years ago I was able to show that when consumers were considering a new loan offer, they evaluated the price of that offer by comparing it to the price of their previous loan and not the fair market rate – so if their last loan was 11%, then 10% might be considered good even if the market was offering 9%.
I can only believe that this was true because, in order to determine the fair market rate, they would have had to apply for multiple loans. So obtaining price awareness at the level of the individual used to be a tiresome process, but as price comparison services become ubiquitous, and as they incorporate individualised loan offers rather than generalised price offers, I’d expect to see less of this sort of price variability.
Which probably brings up one last question: is risk-based pricing fair? Well, as I introduced the topic, when that the price of a loan reflects the risk of that loan should not be controversial, and so the general answer to that question is ‘yes’. However, I do think there are some situations where the answer may be a little more nuanced.
The first such situation is the one described above, where a loan may be priced at a rate above the bare minimum just because the customer is willing to pay. This is the sort of problem that capitalist theory tells out the market will correct, and as I suggest probably at an ever-faster rate as innovation increases market transparency.
The second trickier situation is one that arises when we look more closely at the source of risk. Risk is another way of saying ‘uncertainty’, with a riskier loan being one where we are less certain that we will be repaid in the future.
The most common source of that uncertainty is the consumer’s past behaviour – if a consumer has a history of missing payments in the past, we have to doubt their willingness and/ or ability to make every payment in the future. Hence a loan to them is riskier. Hence a loan agreed, would be agreed at a higher price. It is easy to call that situation fair, with the consumer in charge of their own destiny.
However, uncertainty can also just be uncertainty. Sometimes we simply don’t know how a consumer will behave – think about consumers who are new to credit, for example, where we simply have no data on which to form an opinion one way or another.
In a developed market where the majority of economically active adults are also credit active and that credit activity is recorded in a centralised location, the new to credit population is small and usually temporary. Someone in their first job may have to ask their parents to stand surety for their first loan or maybe they have to pay a slightly higher price initially, but soon their credit history will establish itself, bringing them ‘inside the walls’. In developing markets, though, where the credit activities of the majority of adults may be invisible this is a far more pervasive problem.
In such markets, borrowers could spend all their working lives reliably paying informal lenders with no positive credit history being built. So here, a lender has nothing to reassure them, no data to put into a model to determine a more specific price, and so the borrower either has to go without credit or accept the price of the population’s average risk. I find it hard to call that fair. This is why I love hearing stories about innovation in risk scoring in developing markets – I have this discussion with Oscar Koster of Big Data Scoring in episode 8 of How to Lend Money to Strangers.