I could fairly accurately define a mortgage as a loan that represents the very largest amount of money a person can possibly borrow and still expect to pay back. Though they are thought of as the safest kind of lending, when put in this light you can see that there is the potential for risk: the amounts are large and the timescales long.
If I were to try to create some way of predicting who would be a good credit risk, I would probably look at the household budget: how much money comes in, how much do they spend on what, and from there try to figure out how much that household could afford to put towards their prospective house purchase. Then, I’d have a few thousand potential scenarios with a number of important parameters — periods of unemployment, new children, interest rate changes, stock market returns, inflation in other spending categories, house price returns, etc., etc. — and run some Monte Carlo projections to see how likely this borrower would be to pay me back even if stressed.
That would probably give me a pretty good idea of who I could lend money to. But it is an absolutely ludicrous practice to think about for the real world. Firstly, there is going to be a lot of uncertainty: what if the borrower wanted to defraud me into giving them more money (or “soft fraud” so they could buy more house sooner) and under-stated how much they spent on food, vacations, and Halloween decorations so it looked like they could handle more mortgage than they really could? How would I ever collect and verify all the information they gave me to go into my model? Secondly, I will likely not be the one out vetting prospective borrowers and collecting their information, it will likely be someone who has a high school education (or a liberal arts bachelor degree) who has never heard of a Monte Carlo simulation working in the customer service part of my bank1. My “mortgage specialists” likely couldn’t figure out their own detailed household budget, let alone audit a potential customer’s.
So I want something else that is going to be a pretty good predictor of ability to pay that will also be easy to verify and easy to apply by minimally skilled staff. And when you boil it down there are basically two things that are rather good predictors: how much money you’re making (income) and how much money you already have (down payment).
A larger down payment will mean that you’ll need to borrow less, but it also demonstrates that you have the ability to stick to a budget and save (or relatives who at some point did and liked you enough to give it to you). It gives you “skin in the game” — you’re much less likely to walk away from your house after a 20% downturn in prices if you put 20% of your own money into it. 20% is the cut-off, by the way, below which you must have mortgage insurance for a bank to lend to you
More income means of course that you have more money to pay the monthly bills to service your debt as well as keep the house properly maintained, etc. The question though is what criteria do I set for “more income”? There are many options, so we’ll have to think about what aspects of the problem to consider.
We could simply set a threshold of price-to-income: say you can get a mortgage for 5X your income. So if you had a $50,000 income you could only qualify for a $250,000 mortgage. But a mortgage is so long-lived that the interest rate that applies greatly affects your ability to afford it. At 10%, the mortgage alone would be over half your income, while at 5% it would be just over a third. At 20% it would take everything you had. Interest rates are one aspect to somehow build into our metric then.
So instead of comparing to the total mortgage amount, we could compare to something that more closely mirrored income: the debt service costs — which, if you’ve read ahead, you know is what banks currently do. There are some problems with using debt service costs: because they are interest rate sensitive, you might give out too much money during times of low rates. This was only recently (and only partly2) fixed by introducing the concept of the qualifying rate: your loan is tested against an interest rate of 5.4%, even if you’re able to borrow at 3%.
If we’re using the monthly cost, then we just need to come up with a rough average household budget, figure out how much money can go to the mortgage versus everything else, and we can set up a fairly simple multiple test for our front-line bank staff. But there are some wrinkles to iron out: there’s a big difference between how households split their budget. When I was in grad school, almost 80% of my income went to rent. Food, rent, utilities, and transportation were pretty much the only expenses I had — I was dependent on gifts for clothes and entertainment. Clearly, the average household can’t afford to allocate 80% of their income to housing expenses.
If we look at some hierarchy of needs, shelter is up there, and if you don’t pay your mortgage you could lose your shelter. But it can take 120+ days for the bank to evict you if you skip your mortgage payment, and you’re hungry now. So food expenses, the transportation costs to get to work, and certain addictive vices (smoking, drugs, World of Warcraft) may get paid first. So if we say it takes about 15% of the household budget to cover that, then we can say that debt service costs can occupy the next tier of the budget. But not the entire rest of the budget — we want a household experiencing some stress (perhaps one wage earner out of the work force) to still be able to make those basic needs and service their debt — and an un-stressed household will still have other uses for money (vacations, saving, other discretionary spending), so there has to be headroom there.
Exactly how big that tier should be is a good question and I can’t really answer it from first principles. But the banks and CMHC have come up with 32% for debt servicing and that looks to be roughly in the right range.
Then the next question is what do we use to measure income? After-tax income makes a lot of sense: that is, after all, what you have to actually spend. However, it gets back to our initial problem of trying to make the model too good complex: how do we verify after-tax income? If you’re about to buy a house you might be putting as much as possible into your RRSP, with the intention of using the Home Buyer’s Plan. Those contributions make more of your income tax-free, making your after-tax income look bigger. But will that rate of RRSP contributions continue once you buy the house? What of all the other tax credits and deductions?
Gross income, on the other hand, is fairly easy to verify: an employment contract and/or T4 slip will do it. No fist-fights break out in the bank offices over what should be considered an eligible, sustained deduction, no shoe-boxes full of receipts to dump on the desk. Self-employed people will have trouble with either measure, so forget them for now.
Another advantage of using gross income rather than after-tax income is how it scales: people with higher incomes are able to put more of their budget towards a mortgage if they so choose, because the very basics (food, etc.) are covered by such a small part of what they make, so more of their salary is truly discretionary. They’re also taxed more. If the criteria is 32% of gross income, then someone making $40k and paying an average tax rate of 15% would be able to take on $12.8k/yr of debt servicing costs, and have an after-tax income of $34k to pay that — it works out to 37.6% of their after-tax income. Someone making $200k might have to pay 40% in tax, so their $64k in allowable debt servicing would be 53% of their after-tax income. A fixed ratio of gross income leads to a sliding scale for after-tax income.
Ease of verification, ease of calculation, some prediction of credit risk, and some modest scaling with income actually make the gross debt service ratio a pretty good choice for qualifying people to borrow.
There are still some weaknesses with it, in particular with how interest-rate sensitive it can be. Using a conservative qualifying rate helps mitigate that to a large extent, so I don’t know why there’s a loophole for the most popular kind of mortgage around (5-year fixed). Down payment is a little more straightforward. 20% is, roughly speaking, enough to weather a mild downturn and have enough capital to cover transaction costs if you need to bail. But providing mortgage insurance complicates things: being able to buy with no capital is an accelerating factor for bubbles, and is strongly predictive of default risk[3].
I’ve suggested some simple, easy to apply rules to modulate mortgage insurance: cutting back on the maximum price, scaling back on maximum loan-to-value or increasing the cost of insurance depending on price history, price-to-rent, or affordability metrics could help cool a bubble before it gets dangerously inflated4. However, that would require different rules for different cities depending on local economic conditions, and that doesn’t seem to be politically tenable for the CMHC, even though the rules could be made simple and transparent enough that it would be clear that no particular region was being targeted.
I don’t have a time machine or internal CMHC documents to explain it, but that’s my best guess for why GDS is used as the qualifying metric.
1. In this thought experiment I have a bank. I don’t really have a bank; I don’t even have bank stock.
2. A gaping loophole in the qualifying rate is that you get to side-step it if you lock into a 5 year mortgage (or longer), as most Canadians do. Then you get to use your contracted rate. These loans should also be tested against the qualifying rate for conservatism’s sake.
3. In all the research I’ve done, down payment/equity/LTV appears to be the second-strongest predictor of default after credit score.
4. And provide automatic, measured stimulus in a correction. Lean into the wind to introduce some negative feedback (negative is good).