Skip to main content
The Investor’s MindLesson 2 of 6 · The classic traps

Judgment & behaviour

Base rates and the outside view

Before asking what makes this company special, ask what usually happens to companies like it. The outside view is the cheapest correction available to an investor.

7 min readBeginnerLast checked against the product on

Two ways to forecast

The inside view looks at the case in front of you: this management, this product, this market, this quarter. It feels like analysis because it uses all the specific information you have.

The outside view ignores nearly all of that and asks what usually happens to cases of this type. What fraction of companies growing revenue at 40% sustain it for five years? What happens to margins after a competitor enters? How often does a turnaround turn?

The consistent finding across decades of forecasting research is that the outside view is more accurate, and that people abandon it the moment they have details — which is always.

What a base rate is

A base rate is the frequency of an outcome in a reference class: the set of cases yours belongs to.

If 20% of companies that grew earnings at 30% for three years were still doing it five years later, then 20% is your starting point for the next one — before you have read a single word about its management. Everything specific you learn should move you off 20%, and most of it should move you a little, not a lot.

The mistake is not using the details. The mistake is starting from them.

Why it is so hard

The details are vivid and the base rate is boring. A compelling story about a new product produces a confident forecast; "most companies like this one revert to the mean" produces no story at all.

There is also a supply problem: base rates require you to define a reference class and go and find the numbers, and nobody hands you either. The inside view is free.

That asymmetry — one view free and thrilling, the other costly and dull — is enough to explain most of the gap.

How this connects to Pythia

Sector-neutral scoring is a base rate made mechanical. PAS inputs are z-scored within a GICS sector and country — each company judged against its own reference class rather than against the market. A 40% gross margin does not get credit for being large; it gets credit for being large among companies like this one.

The reference class widens when it has to. A bucket with fewer than 20 local peers falls back to the whole GICS sector pooled across countries, because a base rate computed from eight companies is worse than a broader one. Below even that there is no class, and no score.

Percentile ranks say the same thing more directly: this company sits at the Nth percentile of its peers on this metric. That is a base rate you did not have to assemble.

And the valuation model refuses when the case leaves the reference class its worked examples came from — a company earning 98% on equity is outside the regime the equity-compounding lens was built for, so the lens declines rather than extrapolating.

The limits of this idea

A reference class is a choice, and the choice does the work. Is the class "software companies", "companies growing at 40%", or "companies whose founder still runs them"? Different classes give different base rates, and someone determined to reach a conclusion can usually find a class that supplies it. State the class you used, so the argument is about something checkable.

Base rates also describe populations, and you are buying one company. If your specific information is genuinely strong — and occasionally it is — the right move is to depart from the base rate deliberately and by a stated amount. The discipline is not "never leave the base rate". It is "know what you left, and by how much".

Check yourself

4 questions. Nothing is recorded unless you are signed in, and nothing here affects anything else.

1. An analyst studies a company in depth — management, product, market — and forecasts five more years of 40% growth without asking how often growth like that persists across companies. Which error is this?
2. Suppose 20% of companies that grew earnings at 30% for three years were still doing it five years later. You are analysing one such company and its management genuinely impresses you. What should that impression do?
3. A company sits in a scoring bucket with only eight local sector peers. What does the platform do, and why?
4. Two analysts reach different base rates for the same founder-led software company — one from the class of software companies, the other from companies growing at 40%. What is the discipline the lesson prescribes?

4 questions left. An unanswered question counts as a miss, so the check waits for all of them.

Sources