How Banks Test Borrower Forecasts

A bank does not take a forecast at face value, and it does not argue with it first either. It reads the forecaster, rebuilds the forecast, lends against its own version of it, and keeps checking after the money is out. The sections below take each step in turn.

Before a single assumption is examined, an experienced desk looks for what this management projected two and three years ago, and what actually arrived. Forecast accuracy is a property of the forecaster before it is a property of the spreadsheet, and it is the one input that cannot be produced for the occasion.

The reading is about direction more than size. A business that missed a year because a customer failed has a story. A business that has been optimistic by a similar margin in the same direction every year has a habit, and the habit is the information — every individual explanation can be reasonable while the pattern remains the most reliable predictor in the file. This is the same evidence the desk uses to read management quality, which is not a coincidence.

For a first-time borrower with no track record, the substitute is corroboration outside the forecast: contracts, an order book, capacity that already exists, terms already agreed. A number with a document behind it survives; a number with a rationale behind it is discounted.

The reading below is not a search for errors. It is a search for the assumptions that were made without being noticed, which is where forecasts usually fail.

The fourth is where most forecasts break, and it is the one borrowers least expect. Growth absorbs cash arithmetically: every extra dollar of revenue carries its share of the cash cycle. A forecast that grows sales while holding the working capital lines flat has quietly assumed that customers will pay faster or suppliers will wait longer, and has not said so.

A profit forecast on its own cannot be tested, because it cannot show whether the money exists in the month the payment falls due. What a desk wants is the projection reconciled across all three statements, month by month for at least the first year: the profit and loss, the balance sheet it implies, and the cash flow that connects them.

Monthly matters more than it sounds. An annual forecast can be comfortably serviceable and still contain a month where the facility is exhausted — a seasonal build, a tax payment, an insurance renewal landing in the same fortnight as a large stock purchase. Facilities are sized against peaks, not averages, and the peak is invisible in an annual view.

This is also the point where a forecast meets the question of whether the earnings behind it were real. A forecast that projects forward a profit which never converted to cash in the past is projecting the conversion problem forward with it. Issue 07 is the published file where that gap decides the answer.

The conventional next step is a downside case: revenue falls, margin compresses, collection slows, and the desk checks whether the facility is still serviced and the covenants still hold. It is worth doing, and it has one weakness — it requires choosing the right shock in advance, which is the thing nobody can do.

The more useful question runs in reverse. Rather than asking what happens if sales fall by some chosen amount, ask how far they would have to fall before this facility cannot be serviced. The answer is a single figure the borrower can hold in their head, and it converts an argument about whose forecast is right into a shared understanding of how much room there is if neither is.

It also changes what covenant headroom means. Headroom is not a negotiating concession; it is the distance between the forecast and the point at which the bank has to act. A borrower who understands that number negotiates for the right thing.

A forecast that survives this reading has four things attached. The prior forecast next to what actually happened, including the misses, with a sentence on each. The revenue line separated into volume and price. Working capital assumptions stated as days, so the reader can see they move with the sales. And a monthly view for the first year, so the peak is visible rather than discovered.

Sending those four is not extra work for the bank's benefit. It is the difference between a forecast that is questioned and one that is used, and it usually shortens the approval more than any argument about the assumptions would. Cash flow analysis for business lending sets out the historical half of the same reading.

Everything above is the desk reading the borrower's forecast. What it lends against is usually its own. The forecast as sent is the management case; the desk keeps its structure and replaces each assumption it could not corroborate with the one the evidence supports — the growth this business has actually delivered, the margin it has actually earned, the days its customers actually take to pay. That is the bank case, and the facility is sized and structured on it, not on the plan.

On the bank case the desk reads three things: debt service cover (DSCR) — the cash available for debt service against the interest and principal due — or fixed-charge cover where leases and other fixed payments weigh as much as the loan; leverage, with the requested facility included; and the lowest point of liquidity in the monthly view. Then it pushes the case. One assumption at a time first — sales, margin, the interest rate, collection days — to learn which one the answer rests on. Then several together, because a bad year rarely arrives one line at a time. Then in reverse, for the break-even in section 04.

None of the adjustments is standard. How far the desk moves each assumption comes from this borrower's own record and its industry's volatility, which is why the forecaster is read first. In a credit paper the sequence usually shows as management case, bank case, downside, break-even, and a conclusion on repayment, covenants and liquidity. Underneath all of it is one question: not whether the forecast is right — no forecast is — but how wrong it can be before the bank stops being repaid.

Two tests look past the first year. The covenants are projected quarter by quarter across the forecast, on the bank case as well as the management case, because a borrower can meet every interest payment and still breach a leverage or cover covenant in the quarter the seasonal build peaks. And the maturity is read against the plan: if the forecast only works because a facility is refinanced in three years, the refinancing is an assumption like any other, and it rests on markets nobody can forecast.

The test also does not end at approval. Each reporting period the desk sets actual results beside the forecast the facility was approved on, and reads the variance the way section 01 read the history: direction before size. A variance the borrower explains before being asked is a data point; one the bank finds for itself starts a closer kind of watching, set out in How Banks Manage Credit Risk. The forecast a borrower sends this year becomes the first test of the next one.