Base rate neglect in stock picking

In 1973, Daniel Kahneman and Amos Tversky published "On the Psychology of Prediction" in Psychological Review. They showed that when subjects were given a vivid personality description of a graduate student and asked which field the student was studying, they ignored the known proportions of students across fields and answered based purely on the description's resemblance to a stereotype. The base rate — the prior probability before any individual evidence — was treated as if it did not exist.

The bias generalizes. Kahneman, in "Thinking, Fast and Slow" (2011), called it the most consequential failure of intuitive judgment in domains with a meaningful base rate. The condition for the bias to bite is exactly the condition that holds in equity selection: a vivid case-specific narrative, a knowable but uncomfortable base rate, and a forecast horizon long enough for the base rate to dominate.

Michael Mauboussin, Dan Callahan, and Darius Majd formalized the investment-specific version in "The Base Rate Book" (Credit Suisse, September 2016). They compiled distributions of historical sales growth, gross margin, operating margin, and return on invested capital across the Russell 3000 by sector and starting decile. The data show that sustaining high rates of sales growth across long horizons sits in the tail of the historical distribution — yet analyst long-term-growth estimates routinely cluster well above the base rate for any company whose recent narrative supports it.

Werner De Bondt and Richard Thaler, in "Does the Stock Market Overreact?" (The Journal of Finance, 1985) and the follow-up "Further Evidence on Investor Overreaction and Stock Market Seasonality" (1987), documented the asset-price consequence. Stocks that had produced the most extreme three-to-five-year returns subsequently reverted toward the base rate of their cohort, producing the long-horizon mean reversion that underlies the value premium. Investors had extrapolated the recent narrative and ignored the prior.

The practical failure mode for a thesis is consistent: the analyst constructs a forward model around a company-specific story ("this is the next platform shift," "this management team is exceptional," "this margin structure is defensible"), and the model's terminal assumptions sit in the tail of the historical distribution for comparable companies. The story is internally coherent. The base rate says the story almost never plays out.

The structural defense is to require, at the point of thesis construction, an explicit comparison between the thesis's load-bearing forecast and the base rate of that outcome for the relevant reference class. If the thesis depends on a top-decile outcome, the thesis has to name the specific reason this company is in the top decile — and that reason becomes a falsifiable condition, not an assumption.

Tensile surfaces the base-rate distribution for each load-bearing forecast in a thesis and requires the analyst to name the specific differentiator that justifies a forecast in the tail of that distribution.