Nine quarters is the regulator's horizon, not yours

Almost every stress-testing template in circulation inherits its shape from the supervisory exercise, and the most-copied piece of that shape is the horizon: nine quarters of projection under a prescribed severely adverse scenario. It is copied because it is the visible standard, and because a horizon someone else chose is easier to defend in a meeting than one you had to justify.

It is also, for most community bank books, the wrong length.

Why nine quarters is right for the exercise it belongs to

The supervisory test asks a specific question about large institutions: does regulatory capital stay above minimums through the trough of a prescribed macro path. That question is about the depth of the dip and the capital buffer standing against it. Nine quarters covers the trough comfortably, the scenario is handed to you, and the comparability across institutions is a feature: the point is partly that everyone is run through the same path.

None of those conditions describe a community bank stress testing its own concentration. The scenario is not handed to you. Comparability across institutions is not the objective. And the question is not whether you clear a minimum at the trough; it is how much of this book is going to go bad, and when, and whether the earnings and capital you have can absorb it as it arrives.

Credit losses arrive late

This is the part that makes the borrowed horizon actively misleading. Commercial credit losses do not peak with the macro downturn. They peak well after it.

The sequence is mechanical. The economy turns. Occupancy softens or revenue falls, but the borrower has reserves and a lease term, and keeps paying. Payments slow. The loan is restructured or extended, often sensibly. Eventually the property is reappraised into a market that has already repriced, or the borrower exhausts liquidity, and only then does the loss crystallise. In commercial real estate that chain has repeatedly taken two to four years from the start of the downturn to the worst charge-off year.

Where the losses actually landed in the last commercial real estate cycle

Percent of CRE loans, all commercial banks, annual average of quarterly readings

Trace the dates against a nine-quarter window and the problem is obvious. The recession began in December 2007; nine quarters later is the end of 2009. Charge-offs peaked the following year and were still running above their 2008 level through 2011. A projection that stops where the supervisory exercise stops would have closed its file with the ratios deteriorating and the losses it was meant to measure still a year away.

A nine-quarter projection stops inside that chain. It captures the deterioration in the ratios, delinquency rising and the watch list growing, and closes the file before the losses those ratios were predicting actually land. Run it against a CRE-concentrated book and it will show you a manageable result, correctly computed, that answers a question you did not ask.

A model can be arithmetically flawless and still be measuring the wrong window. Nothing inside it will tell you.

Five years, and say why

For a concentrated commercial book the defensible horizon runs roughly five years, because that is long enough to contain the peak loss year rather than stopping short of it. The horizon should be chosen from the loss emergence profile of the portfolio you actually have. A consumer book behaves very differently, with losses arriving fast and resolving fast, and does not need the same window.

The honest objection to the longer horizon is that the further out you project, the more of the result is assumption rather than measurement. That objection is correct, and the answer is not to shorten the horizon back to where the arithmetic feels safer. It is to widen the band. Report the range across scenarios rather than a point estimate, be explicit about which assumptions dominate the tail years, and show what the answer looks like if those assumptions are wrong. A wide, honest five-year range is far more useful to a board than a narrow, confident two-year number that stops before the problem.

A horizon is a claim about timing

The horizon looks like a technical parameter and is not. It is an implicit statement about when the losses you are worried about arrive, and borrowing it from another exercise means borrowing that exercise's answer to a question about your own book.

The supervisory number is not wrong. It answers a capital-adequacy question about large institutions under a prescribed path, and for that question it is well chosen. What travels badly is the assumption that a window fitted to one loss profile fits another.

Which leaves the harder version of the problem in plain sight: any horizon long enough to contain the peak is also long enough that much of the result is assumption rather than measurement. There is no length that escapes both, and choosing between them is a judgement about which kind of error the institution would rather make.