Bufetico
When a model projects a company’s cash years forward, most of the answer doesn’t come from the years you forecast. It comes from one number at the end: the multiple you put on the final year to stand in for everything after the forecast stops.
Get that number wrong and the rest of the arithmetic is decoration.
There’s a common way to pick it. Take what the market pays for the company today, its price divided by its free cash flow, and use that. It feels defensible. It’s specific to this company, it reflects how the market actually values the business, and it’s one lookup away. It’s also circular.
The price is on both sides
Price-to-free-cash-flow contains the price. That’s what the P stands for.
So when you use today’s ratio as the number at the end, you take today’s price, run it through the model, and compare the result with today’s price. You’re asking the market whether it agrees with itself.
Take two companies with the same cash, the same growth, the same debt. One trades at 40 times its free cash flow, the other at 12. Give each its own current ratio and the expensive one comes out worth more, because you handed it a bigger number to work with.
The model took today’s price, ran it through and gave it back with more decimals. The business never got measured.
And it fails worst exactly where it matters. A stock that has run up carries a high ratio, which inflates the number at the end, which makes it score well. A stock that has been beaten down carries a low ratio and scores badly. The method flatters whatever has already gone up.
What I use instead
The median of the company’s own price-to-free-cash-flow over a long stretch, long enough to cover a full cycle. The middle value across those years, and this week’s price is just one of them.
Why the median? For the same reason as with growth: one wild year shouldn’t set the level for the whole window. A company that traded at 90 times its cash in a single bubble year shouldn’t carry that number into every valuation afterward.
And why a long stretch? A three-year window inside a long bull run is barely different from using today’s number. You need the bad years in there too.
That’s a narrower question than what the company is worth. It asks: when the market has been calm about this business, what has it usually paid for a dollar of its cash? That’s a fact about the past, and it doesn’t move when the price moves this week.
Where this breaks
Three things. It’s wrong when a company genuinely deserves to be valued higher than it used to be. A business that moved from selling hardware every few years to selling software every month earned its new multiple. The long median drags the old, lower number along for years and calls the company expensive the whole time. That’s the one that costs real money.
It needs years of history. A company listed recently doesn’t have them, and that leaves out exactly the newer companies where the valuation gets argued about the most.
It can’t be worked out when a company reports in one currency and trades in another, because the ratio would compare a price in one currency with cash in another. For those, a fixed conservative number is used instead. It’s a patch. I don’t have anything better for those yet.
The ten-second test
Any input that contains the price can’t be used to judge the price. Written down, that sounds obvious. It gets broken all the time: in the number at the end of a model, in comparisons against a group of peers that have all moved up together, in any screen that ranks by a ratio with market value on top.
The test takes ten seconds. Would your answer change if the stock jumped 30% tomorrow with nothing happening to the business? If it would, the price is somewhere inside your model where it doesn’t belong.