Affordability versus Risk

Inherent in traditional credit scoring is the assumption that past behaviour predicts future behaviour. And it does, most of the time. When that link is broken, it is usually because of (i) the influence of some random outside force or because of (ii) a significant change in debt burden.

The first of these is an unavoidable part of forecasting, which is why even super prime consumers default from time to time. People get sick, people get divorced, people get hit by a natural disaster, and there’s not much lenders can do about it outside of insurance.

But when it comes to the significant change in debt burden, there is.

I spent eight years working in and around credit bureau scores, including helping to roll out the first-ever national scores in Thailand, Malaysia, and the Philippines. And there’s a delicate balance in a credit score, which essentially says these are the very accurate odds that a consumer will continue to repay all their existing obligations, but very often lenders are looking at that credit score just before they add on one new obligation.

So, one question that has to be asked at the same time, is ‘will this new loan upset the balance?’, ‘will this new loan make the total debt burden untenable?’, ‘can this consumer, all good intentions aside, actually afford to take on this new loan?’. This is why we should be taking steps to consider affordability separately to risk in our models.

A good credit history and money management skills speak to a consumer’s willingness to repay, affordability checks speak to their ability to repay.  And, at least in theory, calculating a consumer’s ability to repay a loan should be the simpler of the two tasks, being that ‘ability’ is all about the numbers while ‘willingness’ requires us to get inside the borrower’s head to some extent.  Of course, it is not so easy in the real world, but simply put, a consumer is able to repay a loan when they have more money available to meet their debt obligations than they need to keep those obligations up-to-date.

A lack of affordability can therefore originate from two events – a decrease in available cash or an increase in the cost of debt obligations.

Often, a decrease in available cash is due to one of those random events: COVID-19 knocked a lot of incomes, for example, and businesses can close for a number of reasons, but a fair bit of income risk can be modelled nonetheless. Economic data has long been used to forecast general rises and falls in incomes, but these days we can even take it down to an individual level. In the UK, we actually have bank account inflows reported to the credit bureau at a headline level, from which its possible to accurately proxy income and income trends – consumers with a history of income instability are likely to have a future of income instability, too, but with more complicate maths.

The UK is something of an outlier on that front, but globally now, Open Banking is enabling that ability in markets around the world. And even in markets like the UK, where income is available for modelling, the extra level of detail is a step-change, especially when it comes to net incomes. But that's a topic for another episode, for now, let’s just say that we do have ways to estimate a likely range of available cash.

That range of available cash then needs to be seen in the context of the cost of the total debt obligations, or the level of affordability required. The first source of change, of course, is the loan currently under review, which no matter how low-risk it is cannot be so big as to be unaffordable. The single largest factor in this is always going to be the loan amount, but the loan term and the loan price are important factors, too. This is also why I always recommend running the credit risk model first, so a risk-based price can be set and used in the affordability calculation - risk sets price, affordability sets size and term.

Albeit, that price can often change, especially in something like a mortgage where prices are often set relative to a movable base rate. This is the second source of debt burden rises, but again, these can be forecasted with some confidence.

The two sides of the affordability equation should both be considered so that the lender is comfortable that a normal range of income changes and a normal range of cost changes could both occur without leading to financial disaster.

It’s typically easier to prove that a consumer cannot afford a loan than it is to prove that a consumer can afford one, a certainty gap that closes as data quality increases. As such, the relative availability of data and systems will dictate the type of affordability decisions that can be made – if an organisation only has access to basic, customer-supplied data it will be impossible to rule out manipulation so a bigger margin of error will have to be needed where Open Banking or bureau-level incomes are available.

But still, some margin of error will always be needed and the affordability decision will almost always be negatively framed, and almost always be used as a means of identifying otherwise good accounts to be declined, not as a means of identifying otherwise risky accounts to be approved.

It’s typically easier to prove that a consumer cannot afford a loan than it is to prove that a consumer can afford one, a certainty gap that closes as data quality increases. As such, the relative availability of data and systems will dictate the type of affordability decisions that can be made – if an organisation only has access to basic, customer-supplied data it will be impossible to rule out manipulation so a bigger margin of error will have to be needed where Open Banking or bureau-level incomes are available.

But still, some margin of error will always be needed and the affordability decision will always be negatively framed – as a means of identifying otherwise good accounts that should be declined, never as a means of identifying otherwise risky accounts that should be approved.

There’s one more angle worth pursuing, however, and that’s after the fact, when a loan is going bad. I’ve written before about the can’t pay/ won’t pay conundrum in collections, and adjusting the mechanics of affordability, where possible, is a great way to help ‘can’t pay’ customers. If we’re looking for silver-linings in the COVID chaos, one of them might be that both lenders and borrowers will leave the crisis with a far better understanding of what can be achieved with payment accommodations and especially payment freezes. All of these can be best designed within the same context – what has changed to cause the consumer to be no longer able to pay their debt, and what can we change within the product terms to help? Is it a blunt instrument like a few months with no payments while the consumer looks for new work, or something more subtle like a term extension that reduces monthly repayments to a manageable amount?