Abstract

Chapter VIII of Frank Knight's Risk, Uncertainty and Profit (1921) is usually read for the names it gives to two kinds of uncertainty. This article reads it instead for its account of grouping: the method by which insurers, investors, speculators and promoters pool many cases so that their gains and losses cancel. Knight makes a strong claim for that method. He holds that pooling works even across cases that share no visible kinship, provided enough of them are thrown together and competently judged. He also names, in passing, the conditions under which it fails: losses that arrive together, as when a fire spreads from one insured building to the next, and offsetting that shows itself only after the fact, over a span that may outlast the person who must wait for it. A later literature can be arranged, as this article does, into two accounts of the rare large loss. The first draws on Taleb (2007), on rare events of extreme impact that are explained afterwards as if they had been predictable, and on Kay and King (2020), on a radical uncertainty that probabilities cannot capture; it places the loss outside any class counted in advance. The second draws on Daníelsson (2002), on risk models whose use changes the market so that risk becomes endogenous, Bookstaber (2007), on tightly coupled markets whose liquidity vanishes when most needed, and Rajan (2005), on managers who take tail risks and herd; it places the loss inside the pooled system. The article reads Knight's grouping beside two synthetic decision records published for study by WSTiger, a studio of the RISCON family: a strategy with good average returns and a rare loss that damages the account, and a portfolio whose many named strategies lose together on days of sharp moves. Both records are modern summaries written for the platform and report no real firm or event. Read with Knight, they show the two failure conditions of grouping at work in a single decision each. They cannot show which side of the later debate is right.