2.1Pooling as an old practice
The arithmetic of pooling is older than its theory. Hacking (1975) traces the concept of probability to around 1660 and finds it with two faces from the start: the frequency of chance events, and the degree of belief that evidence warrants. Daston (1988) follows probability through the eighteenth century as a model of reasonable judgment, including its use in insurance and annuities. Bernstein (1996) offers a popular history of the idea of risk, from the arithmetic of chance to modern finance. None of the three notes dates the first commercial use of probability, and nothing here turns on that date. What matters is narrower. When Knight turns to insurance in Chapter VIII, he describes an institution already at work in many branches, from life and fire cover to ships, crops and credit, not one he proposes.
Knight's chapter, read against these histories, is a survey of devices built on pooling: insurance, the corporation, the spreading of shareholdings, hedging, speculation, specialised promotion. Each is presented as a way of gathering cases so that their errors cancel.
2.2The first account: the tail outside the class
Taleb (2007) is concerned with rare events of extreme impact and with the way they are explained after the fact, as though they could have been foreseen. The note in the parent study records nothing more about his argument, and this article takes nothing more from it. That much is a claim about time: the explanation comes after the event and presents it as predictable. The further step, that the rare loss lies outside any class counted in advance, is the first account's step, not Taleb's.
Kay and King (2020) widen the claim. On their argument, many decisions face a radical uncertainty that probabilities cannot capture. They recommend asking what is actually happening in the situation at hand, reasoning with narratives, and choosing strategies robust and resilient to what cannot be foreseen, rather than optimising on invented numbers. Their recommendation, then, concerns the choice of strategy: robust and resilient arrangements rather than optimisation on invented numbers.
Put together here, the two works locate the trouble in knowledge: the average fails at the tail because nothing counted in advance contained the tail. Neither note states that conclusion. Knight's remark in Chapter VIII on a sixth factor of uncertainty, occurrences so revolutionary and so unforeseen by anyone that they hardly count as errors of judgment at all, belongs on this side.
2.3The second account: the tail inside the pool
Daníelsson (2002) argues that risk models such as value-at-risk perform worst in the crises they are meant for, and that when many firms use them under regulation their use changes the market: risk is endogenous, and the models can deepen the crises they measure. From that note the second account draws a contrast the note does not state: the first account's trouble is missing knowledge, the second's is shared knowledge, since when many pool by the same rule their cases are no longer independent.
Bookstaber (2007) describes markets made complex and tightly coupled by their own innovations, in which liquidity vanishes when it is most needed. The coupling is the market's own work. Perrow (1984), writing about technological systems rather than markets, treats interactive complexity and tight coupling as two independent dimensions, and holds that in systems high on both, accidents are normal: to be expected from the structure, not from a single error. This article reads Bookstaber's markets as a financial instance of the structure Perrow describes, though Bookstaber's note does not draw that link.
Rajan (2005) argues that managers paid on returns measured against their peers have reason to take risks that pay off most of the time and lose heavily only rarely (tail risk), and to herd, so that financial development can raise the risk of a severe crisis even as it spreads risk. The second account takes herding as one more route to dependence: managers choosing separately may end up holding the same tail. The note speaks of herding and tail risk; the link to Daníelsson's endogenous risk is made here.
Minsky (1986) states the financial instability hypothesis: in good times finance moves from hedge to speculative to Ponzi positions, and stability breeds instability. The note says nothing about strategies or tails. A calm period, on the second account, is not by itself evidence that danger has passed.
2.4The two accounts side by side
| First account, tail outside the class (built from Taleb 2007; Kay and King 2020) | Second account, tail inside the pool (built from Daníelsson 2002; Bookstaber 2007; Rajan 2005) | |
|---|---|---|
| Where the rare loss comes from | Events no prior count contained | Coupling, shared models and incentives within the pooled system |
| Why the average misleads | It was built on cases that did not include the rare event | It treats as independent cases that the system has made dependent |
| What a better average would do | Little; the gap is in knowledge | Something, if dependence under stress is measured separately |
| What the decider should prefer | Robust and resilient strategies over optimisation (Kay and King) | Not stated in the notes as a single remedy |
| Knight's nearest remark | Revolutionary occurrences that are hardly errors of judgment | Fire losses that spread and come in groups |
The table should not be read as two camps in open dispute; the works were written for different purposes, and the rows state what each account holds, not what any one work says. Its use is to show that the two accounts recommend different things to a person reading a record. On the first account, a record that asks for better measurement of the tail asks for something that cannot be had. On the second, a record that separates calm-period statistics from stress-period statistics asks for exactly the measurement that matters.
2.5The managerial reply
Power (2007) describes risk management as a way organisations turn uncertainty into objects they can manage and account for. In an earlier book, Power (2004) argues that risk management spreads into all organisational activity and turns towards secondary risks, the reputational and legal risks to the organisation and its managers, so that defensive documentation can take the place of judgment. Clarke (1999) describes plans written for catastrophes that organisations cannot control, plans that could not work as written but serve to persuade others, and the organisations themselves, that the risk is under control.
These works do not take a side on where the tail comes from. They describe what happens after the tail has been named: it becomes a budget line, a limit, a test, a table. Whether such an object captures the tail or only certifies that someone looked at it is the question they leave open, and it is the question the two records' final lines raise.
2.6The older dispute, briefly
The dispute between LeRoy and Singell (1987) and Langlois and Cosgel (1993) concerns where Knight's grouping stops. LeRoy and Singell read Knight's line as a line about insurance markets: uncertainty is where moral hazard and adverse selection keep a risk from being insured. Langlois and Cosgel answer that for Knight uncertainty arises where instances cannot be sorted into classes of like cases, so no frequency can be had, and judgment must take over. A companion article drawn from the same parent study reads that dispute at length. It is mentioned here because Knight's account of grouping, read for its conditions rather than its limit, supplies a third consideration neither reading puts first: a class can exist, and an insurance market can exist, and pooling can still fail if the cases in the class fail together.