6.1The claim
Chapter VIII of Knight's book gives two reasons why the typical business risk cannot be insured, and puts them in one passage: such risks group only in retrospect, and that grouping cannot be done by an outside party because the decisive factors lie inside the person who decides. The first reason is the classing reading's; the second is close to the market reading's. This case reads that passage, then the revisit lines and next checks of both records, which say when the class would be available and who writes it. It adds to the debate the observation that, in these records, the two reasons arrive in order: the class is missing first, and the threat to it appears when someone tries to build it.
6.2Insurance and its limits
Chapter VIII's survey of insurance runs from the best case to the worst. Insurance needs a case to be assimilated to a group of similar cases, so that the share of the group with each outcome can be expected. That assimilation can in principle be precise and the shares calculated, but in business they are always counted. The insurer must also be able to argue plausibly that each insured pays in proportion to that insured's own chance of loss, a fair share.
Life insurance is Knight's nearest approach to the calculated case. With a medical examination and attention to age, sex, residence, occupation and habits, insurers can pick lives whose chances of death are about as equal, and as truly undetermined, as anything in nature. Insurance against sickness and accident is the opposite: cases cannot be described and classed objectively, and the business is costly and limited. Fire insurance lies between. Only recently, Knight says, have insurers formed classes of buildings homogeneous enough to approach a real probability for a given case, and even with a great many classes defined, room remains for judgment. Two further limits follow. Because of moral hazard, the amount insured is held to the direct loss or less, so that the indirect losses from interrupted business are not covered. And fire losses are not independent, since one fire spreads to others.
At the far end lie the policies of underwriters who insure that a coronation will happen on the scheduled day, or guarantee the weather in a place with no records. Even there, Knight says, a rough grouping by judgment must lie behind the premium. And he draws a striking conclusion. Premiums on the most singular contingencies can still cover the losses, if the estimates are careful and competent, because the very exercise of judgment supplies a ground for grouping: the judgments of a competent underwriter across unlike cases tend to steady when taken together.
6.3Knight's two reasons in one passage
From this Knight infers that uniqueness alone does not limit insurance of business risks, and he makes two observations in anticipation of the next chapter. The first is that the typical business risk that cannot be insured is the risk of error in the business person's own judgment. Knight describes it as uninsurable because unmeasurable, and unmeasurable because unclassifiable. The second is that such judgments tend to settle into groups in which errors cancel, and so to approach constancy, but only after the fact, and, given how short a working life is, they give only a modest footing for prediction. He then adds a further limit. Grouping of this kind can be carried out by an agency outside the person only within narrow bounds, because a moral hazard is attached to these risks in a peculiarly stubborn way. The factors that decide the outcome lie so far inside the one who decides that the cases cannot be described objectively or controlled from outside.
Both readings of Knight have their sentence here. The classing reading has the chain from uninsurable to unmeasurable to unclassifiable. The market reading has the outside agency defeated by moral hazard. The passage does not ask the reader to choose. It says the risk is unclassifiable at the time of decision; that a class forms only afterwards, from the record of outcomes; and that even that later class cannot be built by an outsider, because the outsider cannot see or control what the decider did.
Knight goes on to say that these difficulties fall away when the grouping is done inside the person's own scale of operations, or inside an organisation where responsibility is centralised and interests are unified. He traces from this the growth of firms, the move from individual enterprise to partnership and from partnership to corporation, and he argues that limited liability by itself only shifts losses to creditors, while the real gain is a double consolidation of risks, with greater publicity. Specialisation, he adds, is the insurance idea applied again where moral hazard blocks insurance from outside: the speculator who hedges a miller's grain risk enters the market many times where the miller enters once, and the speculator's errors cancel. Unlike an insurer, who takes on a risk that is no smaller for being taken on, the specialist substitutes the specialist's own judgment for that of the person who passes the risk on. The uncertainty problem thus turns, at the end of Chapter VIII, into the problem of management. Where ignorance is partial, Knight writes, cases cannot be classed objectively, and any change in the conditions under which an opinion is formed tends to change the worth of the opinion.
That last remark is the hinge for the records. If the conditions under which an opinion is formed change its worth, then a rule that pays for the opinion changes the opinion.
6.4The first record's revisit line
Confidence score as size sets its condition for revisiting the judgment in one sentence.
If, after calibration, score inflation is not observed, or if a size cap excessively blocks returns, the matter is re-judged.
Two events would reopen the matter. The first is that calibration, once in place, shows no inflation. That depends on the per-agent record the example judgment asked for, which does not exist at the moment of writing and will exist only after enough orders have resolved. Knight's after-the-fact grouping is written into the condition: the judgment stands until the class of each agent's past calls has formed enough to show whether the scores were honest. The record gives no number of resolved orders, no threshold for what counts as inflation, and no date; the parent study's code found nothing countable in it. When the class will be ready is left open.
The second event is that a cap blocks too much return. Kelly's fraction (1956) is the stake that maximises long-run growth, so a cap protects against an overstated edge but, when the edge is real and the cap sits below that fraction, gives up growth. The record names the cost without measuring it.
The incentive the example judgment identifies, the agent's reason to inflate, is Knight's moral hazard placed inside an automated decider. The engine is an outside agency in Knight's sense. It allocates capital to decisions it cannot see into, on the strength of the decider's report about itself. Knight's ground for saying an outside party cannot do the grouping is that the decisive factors are inside the person. In the record, the agent's learned tendency to inflate is such a factor. Calibration is the engine's attempt to group from outside anyway, by keeping the one thing it can observe, the gap between report and result. Jensen and Meckling (1976) name the cost of that kind of watching as an agency cost, and name what is left when the watching fails as the residual loss. The size cap is the option that accepts the residual loss and bounds it.
Read in this light, the record shows both readings at once, but at different moments. At the time of allocation, the trouble is a missing frequency: no hit record exists for the agent. Once the engine tries to supply one, the trouble becomes the agent's incentive to shape what goes into it. The record does not describe an insurance market, and nobody in it buys cover. What it describes is one party that supplies capital and another whose conduct the first cannot see into. LeRoy and Singell (1987) place Knight's uncertainty where moral hazard keeps a risk from being insured, and the record shows moral hazard of that general kind without the insurance.
6.5The second record's revisit line
Timeframes in collision sets its condition in two labelled parts.
Change condition: once the conditions of the execution timeframe are met, entry is re-reviewed, and if the higher structure changes, the allowed scenario itself is re-determined. / Next check: record the analysis timeframe and the actual order timeframe separately in the trade log.
The change condition keeps the two classes apart in time as well as in function. The short-term class reopens the entry when its own conditions are met; the long-term class reopens only the allowed scenario, and only if the higher structure changes. Neither class is allowed to reopen the other's decision. Like the first record's condition, it contains no number, threshold or date the parent study's code could detect, and it names no one to do the re-reviewing.
The next check is the record's most Knightian line. It asks for the timeframe on which the analysis was done and the timeframe on which the order was placed to be written separately in the trade log. That is the act of fixing the class at the moment of decision, so that the trade can be counted later in the class it was actually taken in. Without it, a trade log would show only entries and exits, and the decider could assign each trade to whichever class flattered it. With it, the after-the-fact grouping Knight describes becomes possible, and becomes harder to bend.
Here the inside factor is not an incentive bought by an engine but the decider's own later account. The record treats that account as untrustworthy enough to require a written commitment beforehand, in the same way the first record treats the agent's confidence as untrustworthy enough to require a written gap afterwards.
6.6What the case adds
Chapter VIII of Knight's book joins the two readings: business risks are uninsurable because unclassifiable at the time; they become classifiable only after the fact; and no outsider can do the classing, because what decides the outcome lies inside the decider. The two records put that chain into practice. Each writes a class-forming instrument into the decision: a per-agent gap between report and outcome, and a separate entry for analysis and order timeframes. Each says when the class would be consulted, though neither says when it would be ready. And each treats the description supplied by the decider, an inflated score in one case and a convenient timeframe in the other, as the thing most likely to spoil the class. For the debate, the records suggest that the missing frequency and the missing market are not rival accounts of Knight's uncertainty but two stages of it, and that a written record is where the first is supplied and the second is resisted.