← All publications

From the literature

Risk or uncertainty: the distinction that changes how a project is read

We often use the words "risk" and "uncertainty" as synonyms. Yet Frank Knight devotes a large part of his book to explaining that this confusion prevents us from understanding profit, enterprise and decision.

In everyday language, a "risky" project is simply one whose future is uncertain. For Frank H. Knight, this definition is far too broad. In Risk, Uncertainty, and Profit, published in 1921 from a thesis defended at Cornell five years earlier, he gradually reserves the word risk for situations in which the distribution of outcomes can be known or sufficiently approximated. Uncertainty, by contrast, refers to situations that cannot be reduced to an objectively measurable probability.

The difference looks terminological. It is in fact economic. When a risk is measurable, one can group a large number of cases, observe frequencies, calculate an average loss, pool the outcomes and, in many cases, insure the event. Risk then ceases to be a mystery: it becomes a cost line that the organisation builds into its economic calculation, in the same way as rent or a raw material.

Three ways of saying "probable"

Chapter VII provides a clarification that summaries often leave out, even though it is the heart of the argument. Knight distinguishes not two but three types of probability.

The first is a priori probability: the kind deduced from the very structure of the set-up. A fair die, a roulette wheel, a perfectly defined draw. Here, probability is a known property of the object, prior to any observation.

The second is statistical probability: the kind inferred from a frequency observed over a large number of sufficiently similar cases. No one knows the "structure" that produces warehouse fires, but an insurer with a large and homogeneous portfolio can derive a solid estimate from it.

The third category is that of estimates. There is neither a known structure nor a class of comparable cases: the decision-maker forms a judgement about a largely unique situation. And Knight adds the decisive remark: in real economic life, it is this third category that dominates. The first two are the comfortable exceptions.

Why the boundary shifts

Knightian uncertainty arises when the situation is too singular to form a reliable statistical class. What is the probability that a new technology will become a standard? That an inexperienced team will flawlessly execute an unprecedented strategy? That a regulatory change will transform a market that does not yet exist? One can produce figures, of course. But Knight asks us not to confuse producing a number with knowing a probability.

Nor is the boundary fixed. It depends on the ability to form a class of comparable cases, and therefore on the information available, the size of the organisation and the time one is prepared to invest. A hazard that a craftsman cannot measure can become a risk that can be priced for a player handling ten thousand files of the same type. It is precisely this shift of the boundary that explains part of economic organisation.

The use in project finance

This distinction is particularly useful in financial analysis. Many models treat all uncertainty as if it could be described by a distribution. Sometimes that is an excellent approximation. Sometimes it is a very elegant way of hiding the fact that we do not know. The problem is not using scenarios or subjective probabilities: Knight himself acknowledges that we have to estimate. The problem is forgetting the nature of the estimate.

A Knightian reading of a project therefore begins with a question of almost embarrassing simplicity: what kind of unknown are we talking about? If the data make it possible to build a robust frequency, we are on the side of risk. If the situation is unique, if the comparables are weak and if most things depend on judgement, we are in uncertainty.

In practice, a single application mixes all three. The volatility of a commodity price often has a usable history. The loss rate of an industrial asset belongs to insurance statistics. The pace of adoption of a technology that does not yet have a market is, for its part, a matter of pure judgement. Putting these three items in the same "risks" column of a table is convenient. Treating them with the same tools is an analytical error.

Knight's merit is that he does not turn this uncertainty into a failure of method. It is a fact of economic life, not an admission of intellectual laziness. The aim is not to make it disappear by decree, but to know when we are calculating and when we are judging. It is less spectacular than a Monte Carlo simulation. But often far more honest.

Frank H. Knight, Risk, Uncertainty, and Profit (1921), chapters VII and VIII. Editorial adaptation, not a quotation.