Bufetico
Two days ago my own screen showed Amazon at a price 437% above what my model said the business supported. Roughly every other valuation I could find said the opposite: undervalued, by something like half.
I don’t believe ten strangers are all wrong at once and I’m the one who’s right. So I spent a day finding out which of us was broken.
It was me.
Where I was starting from
To value a company you have to start from a number, and I was starting from the cash left over after the business had paid for everything it needed, including everything it was building.
For Amazon, that figure in the last closed year was 7.7 billion dollars, in a year when the company was spending heavily on new data centers. Grow 7.7 billion forward and you get something small. Put the price next to it and the price looks absurd.
The mistake, in one line
Spending money is not the same as losing money, and I was treating them the same way.
A company putting cash into new warehouses and a company bleeding cash produce the same figure at the bottom of the page, and only one of them ends up owning something. Subtract every dollar of investment and the two become impossible to tell apart, which is exactly what my model was doing.
What I should have been measuring is the cash a business throws off after the investment it needs to stay as it is. What it spends on top of that, trying to get bigger, is a different thing.
I now separate those two. How I draw the line is my own business and not very interesting; what matters is that a company expanding and a company shrinking stopped looking identical.
What changed on the screen
It happened in two steps, and things looked worse before they looked better.
I added a rule so a projection can’t start from a bad year: if the recent cash has negative years in it, there’s nothing to anchor on and the square goes gray. Amazon’s recent cash had negative years in it, so the model refused to give a number at all. Better than 437%, and still useless.
After that I stopped subtracting the growth investment as if it were a loss, and the chart gave a number again, on the same side as everyone else’s.
The list of companies that get through all six layers didn’t change. Thirty-six before, thirty-six after. That’s what keeps me from feeling clever about it.
The repair changed what the screen says about a company when you go and look it up, and that’s all it changed. Most repairs are like that: they improve the explanation and leave the answer alone.
Where this breaks
The correction has a failure mode, and it’s easy to state. A company that’s under-investing now looks better than it deserves. Let the factories age, spend less than you should, and my estimate of what maintenance costs follows you down. The number improves while the business rots, and the arithmetic doesn’t complain.
Which is why this adjustment touches what I think a company is worth and nothing else. The cash square, the layers, and every test on the shape of the cash series still run on the plain, unadjusted figure. Who passes the filter gets decided without it.
Adjusted earnings are how a bad business gets to look like a good one. Keeping the estimate out of the filter is what stops that.
I spotted this because my screen disagreed with everyone else and that bothered me enough to go looking for my own mistake instead of theirs. That discomfort is the whole skill.