Fundamentals

Knightian uncertainty: deciding when risk cannot be calculated

By Daniel Sardá · Published on October 13, 2026

6 min read1,228 words

Knightian uncertainty arises when a decision lacks a sufficient basis for assigning defensible probabilities. Understanding it helps distinguish what can be estimated from what requires judgment.

Not every difficult decision involves the same kind of unknown. Sometimes there is enough data to estimate the likelihood of different outcomes. At other times, the situation is so new or distinctive that assigning precise percentages would create an appearance of knowledge we do not actually possess.

Knightian uncertainty names this second situation. Frank H. Knight set out the distinction in Risk, Uncertainty, and Profit, published in 1921: risk permits some meaningful probabilistic treatment; uncertainty, by contrast, provides no defensible basis for constructing a probability distribution.

The distinction does not divide rational calculation from improvisation. It separates two problems that call for different tools.

Key idea: There is risk when we can estimate probabilities on a relevant basis. There is Knightian uncertainty when that basis is missing, even though we can still reason and act.

Risk and uncertainty are not synonyms

In everyday language, “risk” and “uncertainty” are often used interchangeably. For Knight, the distinction is more exact.

There is risk when possible outcomes can be grouped into a sufficiently comparable class and we have a basis for estimating their frequencies. Prediction need not be perfect. It is enough that the probability rests on relevant grounds and can guide a decision.

An insurer, for example, does not know whether a particular home will catch fire next year. Yet it can assemble many cases with comparable characteristics, observe frequencies, refine its estimates, and spread expected costs across many policyholders. The individual outcome remains open, but the group can be treated statistically.

There is uncertainty, in the Knightian sense, when that procedure cannot be defended. The decision may be unique, past cases may not be comparable, or relevant changes may prevent the construction of a stable class. People can form opinions and estimates, but these are not the same as measurable probabilities.

The difference can be summarized this way:

• Under risk, data and a reference class make useful probability estimates possible.
• Under uncertainty, the case’s uniqueness or the lack of comparable cases makes that basis inadequate.
• In both cases, knowledge is incomplete; what changes is how a decision can be justified.

A product without close precedents

Imagine a firm considering the launch of a genuinely new product. It may estimate some elements: manufacturing costs, plant capacity, or distribution time. It can also study similar products and ask potential customers for their views.

But the decisive questions may not have a sound probabilistic answer. Will an unforeseen use emerge? Will public tastes change? Will a competitor arrive with a better offering? Is comparison with earlier markets valid?

Assigning a 60 percent probability of success does not solve the problem if there is no sufficiently comparable class of product launches from which that percentage can be drawn. The number may express a disciplined opinion, but that does not give it the same foundation as a frequency calculated from many similar losses.

Both domains can coexist within a single decision. Routine machine failures may be insurable, while the product’s future reception remains uncertain. Knight’s categories are analytical distinctions, not pure boxes into which every real-world decision fits neatly.

Key idea: A percentage does not, by itself, turn uncertainty into risk. Before relying on it, ask where it comes from and whether the cases used as a reference are genuinely comparable.

When no formula is sufficient, judgment enters

Acting under uncertainty does not mean acting blindly. It means deciding with partial knowledge: experience, signals from customers, understanding of the sector, the team’s capabilities, and arguments about possible futures. None removes the unknown, but all can improve judgment.

Models remain useful as well. Scenario analysis can show what would happen if demand were high, medium, or low. Sensitivity analysis can reveal which assumption has the greatest effect on the result. What a model cannot do is create a known distribution where there is no basis for one.

That is why judgment is not a license to disregard evidence. It consists in interpreting incomplete evidence, comparing explanations, and committing resources without a rule that determines the correct answer. The decision can still be reasonable, open to criticism, and revisable, but it cannot be reduced to applying a calculation.

This is a central dimension of entrepreneurial judgment. The decision-maker must discover which information deserves attention, which analogies mislead, and when an opportunity warrants taking on commitments. Experience can then make learning possible, though it can never guarantee that the next situation belongs to the same class.

Profit and loss: the outcome can go either way

Knight connected uncertainty to business organization and to the emergence of profit and loss. Before actual demand is known, someone must hire, produce, set prices, or invest. Those commitments rest on fallible anticipations.

If the outcome exceeds the costs and commitments undertaken, a positive residual return may remain. If the anticipation proves wrong, a loss appears. In this account, profit is not an automatic reward for exposure to the unknown, nor can every gain be explained by uncertainty alone. It is one possible result of deciding before the outcome is known.

This relationship helps explain the entrepreneurial function: different people try plans, correct errors, and reallocate resources. From a classical liberal perspective, the process matters because economic coordination does not begin with complete information. Yet uncertainty alone does not establish the superiority of any institution. Comparing institutional arrangements still requires examining incentives, responsibility, rights, and the capacity to learn.

Key idea: Under uncertainty, profits are not assured. The same judgment that opens an opportunity can produce a loss, and both signals can prompt plans to be reconsidered.

Concepts worth keeping distinct

Knightian uncertainty does not mean absolute ignorance. A firm may know a great deal about its technology and customers while lacking a defensible distribution for the outcome that matters most.

Nor is it synonymous with volatility. A price may fluctuate sharply and still have a history that is useful for certain models. Conversely, a decision may have only a few visible outcomes and remain deeply uncertain if comparable precedents do not exist.

Later concepts such as ambiguity, the Ellsberg paradox, or “black swans” engage with related problems, but they should not be treated as equivalents or attributed retrospectively to Knight. The same applies to dispersed knowledge: the connection is useful for considering the limits of economic knowledge, but it belongs to a different line of development.

A practical question before calculating

The distinction becomes useful before opening a spreadsheet: what justifies the probabilities we are using? If there are many relevant and reasonably comparable cases, risk analysis can organize a decision, even while retaining margins of error. If the case is singular and comparisons are fragile, it is wiser to acknowledge the limits of calculation.

That does not require inaction. It requires combining scenarios, small experiments, learning, and responsibility with a modest claim: some consequences can be estimated; others call for judgment. Knightian uncertainty matters because it prevents us from confusing numerical precision with knowledge and reminds us that deciding also means acting reasonably when the future cannot yet be turned into a table of probabilities.

Sources

• Frank H. Knight, Risk, Uncertainty, and Profit (1921), especially chapters VII and IX.
• G. Townsend, R. Hunt, and J. Rady, “Chance, probability, and uncertainty at the edge of human reasoning” (2024).
• Ross B. Emmett, “Uncertainty and the social organization of economic activity” (2021).