How often fast growth lasts

The inside view builds a forecast from the specifics: this company's products, its last five years, the management's plan. It is detailed, it feels well informed, and it is where almost every growth assumption comes from. The outside view ignores the specifics at first and asks what happened to a large reference class of similar cases. Its answer is blunt, and it is usually less exciting.

The working method is to do both, in that order. Start from the base rate: real sales growth of 5.4 percent a year at the median, with a standard deviation of 18.7 points around it, and lower and tighter for the biggest companies. Then move away from it only as far as the evidence about this business justifies, and write down why. A forecast of 15 percent real growth for a decade is not forbidden. It is a claim that this company will sit far out in the right tail, and it needs an argument to match.

Size is the first adjustment. The book finds that both the level of growth and its spread decline as companies get larger, so a very large company's reference class is narrower and slower than the full distribution in the figure.

When a distribution has a long right tail, the mean is dragged toward the tail by a small number of extreme values. The median is the middle company: half grew faster, half slower. Here the mean is 8.1 percent and the median 5.4 percent, a gap of 2.7 points a year, which compounds into a large difference over a ten-year first stage.

If you pick one company and want the growth rate most likely to describe it, the median is the better anchor, because most companies are nowhere near the tail. The mean is the right number for a different question, the average growth across a whole portfolio of companies, where the few fast growers really are in the pot. Using the mean as the default for one company quietly assumes it has a fair share of the tail's luck built in.

The spread tells you as much as the centre. A standard deviation of 18.7 points means two companies one standard deviation either side of the median differ by more than 37 points a year. That is the honest width of "a typical company", before anything specific is known.

A standard deviation of 34.6 points for actual three-year earnings growth means outcomes scatter far more than any single forecast suggests. Forecasts with a spread of 19.2 points are, on average, behaving as if the future were a little over half as uncertain as it turns out to be. Lean optimistic on top of that, and the typical forecast both sits too high and is held with too much confidence.

For a valuation, the practical consequence is about the range, not the point. In Analyst session 6 you built a range from the discount rate plus or minus one point and growth plus or minus two. If your growth input came from forecasts, those two points are almost certainly too few for earnings, because the forecasts you started from were narrower than the world. Widen the growth band, or at least write down that it is a floor on the uncertainty, not a measure of it.

The left tail is real too. The book finds 31 percent of companies had lower net income after three years, 29 percent after five, and 24 percent after ten. Roughly one company in four earned less a decade later. Any bear case that never lets earnings fall is not a bear case.

The book reports that sales growth is more persistent than earnings growth. Earnings sit at the bottom of the income statement, after costs that do not move one for one with sales, so a small change in sales or margins can produce a large swing in profit. That is why the earnings distribution in the figure is so much wider than the sales distribution in Part 1, and why a growth forecast anchored on sales is usually steadier than one anchored on last year's profit.

Growth also explains less of shareholder returns than a story might suggest. The book reports a correlation between sales growth and total shareholder return of 0.20 over one year, 0.25 over three years and 0.28 over five. Positive, rising with the horizon, and still modest: growth matters, but it is far from the whole explanation, which is one reason the price paid for that growth matters so much. The book also reports that median sales growth correlates with US GDP growth at 0.66.

For owner earnings, which sit even closer to the bottom line than net income because capital spending is subtracted too, expect less persistence still. The one-off years you met in Analyst session 2 are the same effect seen up close.

At the default five-year hold-out, mean absolute miss across the four filers: rung 4, the fade, missed by 47.7 percent; rung 3, the flat ten-year mean, by 50.2 percent; rung 2, the flat five-year mean, by 47.4 percent; last year's net income by 70.4 percent; and the anti-lever by 19,614.8 percent. Fading half the past growth helped a little over holding the ten-year mean flat, and the five-year mean still edged it.

The more important finding is the direction. Every honest rung undershot. At five years the bias equals minus the mean miss: all four companies' owner earnings rose sharply after 2020, and no rule built from the years before saw it coming. Exxon Mobil is the clearest case: from 2020, the fade forecast 3.74 dollars a share a year against an actual 8.87, a miss of 57.8 percent.

So the base rates in Parts 1 and 2 do not simply argue for lower growth. Over this window, lower would have been worse. They argue for wider ranges: outcomes scatter more than forecasts do, in both directions, and a range that cannot contain a boom is as wrong as one that cannot contain a bust. The fade is a sensible centre; it is not a forecast you should hold tightly.

Because rung 4 starts from rung 3's base, its value at the same sliders equals rung 3's in the bench; the fade changes the forecast of the hold-out years, which is what the ladder scores.

Our four filers grew their owner earnings per share at between 3.1 and 9.2 percent a year over their runs: Caterpillar 9.2, Exxon Mobil 8.1, Johnson & Johnson 7.3 and Procter & Gamble 3.1. Those are real numbers, computed in the build-along, and they are almost useless as a base rate.

First, all four are long-lived giants. They were chosen because their filings are long, and a long filing history exists only for a company that survived. Survivorship removes exactly the cases the Base Rate Book was careful to keep: the half of listed companies that disappear within ten years. Four survivors describe what happened to four survivors.

Second, the units differ. The filers' rates are nominal and include inflation; the book's are inflation-adjusted. They measure owner earnings per share over ten to fourteen years, not sales over three. A nominal 9.2 percent and a real 8.1 percent mean are not the same kind of number, and putting them on one axis invites a comparison that cannot be made.

What the four histories are good for is a sanity check on one company: does a forecast for Procter & Gamble far above its own 3.1 percent nominal history have an argument behind it? The base rate tells you how unusual a forecast is across companies; the company's own history tells you how unusual it is for that company. You want both, and you want to know which one you are reading.