Which home loans will go bad, and what should a lender do about it?

How to read this: How much money is still owed on loans that are being repaid: the amount at risk. It grows as new loans are added each year and shrinks when borrowers pay off early, as in the 2020-2021 refinance boom.

How to read this: The share of the book paid off early each year, mostly because borrowers refinance when mortgage rates fall (2003, 2012, 2020-21). Early payoffs are not losses, but they end the interest income a lender or investor was counting on.

How to read this: The share of the book falling into default each year. It spiked in 2009-2010 as house prices fell and unemployment rose, the combination that turns missed payments into losses.

How to read this: The share of borrowers who were paying on time last month but missed a payment this month. It moves first when the economy weakens, months before defaults show up.

How to read this: Roll-forward rates show how often late borrowers fall further behind; the cure rate shows how often a borrower 30 days late gets back to current. Rising roll-forwards with falling cures, as in 2008-2009, mean losses are on the way.

How to read this: Each cell is a group of today's loans, by the borrower's credit score and the loan-to-value ratio (LTV: how much was borrowed compared with the home's value). Darker cells are expected to lose more over the next 12 months. The rows with credit scores below 680 are the darkest.

How to read this: The same expected-loss rate by state. Differences come from local unemployment, house-price trends and the mix of borrowers in each state. Every state is drawn the same size so small states are as visible as large ones.

How to read this: Credit score is the single strongest predictor of default in the model. Borrowers with scores under 620 are many times more likely to be behind on payments today than those above 760.

How to read this: Most simulated economies are calm, so most bars sit at small losses. The long tail to the right is the recessions. Risk teams and regulators focus on the tail: the 99th percentile is the loss exceeded in only 1 economy in 100, and the capital a lender holds has to cover it.

How to read this: The shaded band holds 90% of the simulated futures. The coloured lines are the named scenarios regulators ask for, from a baseline to a severe recession, plus a replay of what actually happened in 2008-2010.

How to read this: House prices matter because a borrower whose home is worth less than the loan (underwater) is far more likely to default, and the lender recovers less when the home is sold.

How to read this: Losses stay small until unemployment and house prices both move against borrowers. Job losses make people unable to pay; falling prices take away the option to sell the house and repay. Use the sliders to ask, for example, what the book loses when unemployment rises 4+ points and prices fall 20%+.

How to read this: Approving everyone is the zero line. Declining the very riskiest applicants raises profit, because their expected losses exceed the margin they would earn. Decline too many and profit falls again: you turn away good borrowers along with the bad. The best policy sits at the top of the curve. The recommended cutoff was chosen on earlier loans (2012-2013) so the result shown here is a fair, out-of-sample test.

How to read this: If the model were useless, declining 10% of applicants would avoid about 10% of defaults (the grey line). The further the blue curve sits above it, the better the model separates risky borrowers from safe ones.

How to read this: The share of automated checks that passed, by type of file. Checks on a rejected batch count as failures here even after the file was fixed and reloaded, so a figure just under 100% is the gate doing its job.