Concept/Decision Theory/No. 0864
Risk vs. Uncertainty
Risk vs. uncertainty is the distinction between outcomes with usable probabilities and situations where probabilities or possible outcomes are unclear. Economist Frank H. Knight set out the distinction in 1921; it is foundational in decision theory and is not a measure of danger.
- Evidence
- Well established
- Read
- 6 min
- Links
- 13 connections
01You've seen this when…
- in life
You compare mortgage payments using interest-rate scenarios. Then you consider moving into a new line of work, where you don’t yet know what employers will pay you.
- at work
Your team can estimate routine equipment failures from years of maintenance records. The forecast for a product nobody has sold before rests on three interviews and a hopeful spreadsheet.
- out in the world
A city sizes drains using decades of rainfall records. Engineers disagree about how much those records tell them about storms over the next fifty years.
02The idea
A forecast can look equally precise whether it rests on thousands of observations or one person’s guess. Risk vs. uncertainty asks you to inspect what supports the numbers before using them.
In Frank Knight’s distinction, risk means you have a usable probability model. You don’t know which outcome will happen, but you can describe the possibilities and assign reasonably defensible chances to them. A fair die is the clean example. Insurance claims across a large, well-understood population are a practical one.
Uncertainty means that those probabilities are poorly specified. You may lack relevant observations, disagree about how the situation works, or not know all the outcomes to consider. Launching into an unfamiliar market can involve all three.
The distinction isn’t between safe and dangerous. A risky activity can have a well-understood chance of disaster. An uncertain activity can turn out harmless. You can still use numbers under uncertainty, though their authority depends more heavily on assumptions and judgment.
The practical question is: Are these probabilities reliable enough for the decision we’re making? A rough estimate might suffice for a small trial but not for a commitment that could bankrupt you.
03Why it matters
Different gaps in knowledge call for different decisions. Treating everything as calculable risk can lead you to optimize confidently around a fragile forecast. Treating everything as unknowable causes you to miss useful evidence.
- Use probabilities when you can defend them. Compare costs and benefits while accounting for the range of possible losses. Expected value can help. Assess whether you can survive a bad outcome separately from the average.
- Test several futures when the probabilities are weak. Ask which choice remains acceptable across different assumptions. Scenario planning explores those futures; robust decision-making looks for choices that hold up across them.
- Buy information before making the large commitment. A pilot, customer test or inspection can replace some guesses with observations. Use Bayesian updating to revise your beliefs, while remembering that a small test may not represent full-scale conditions.
- Preserve ways to change course. Short contracts, staged spending and reversible designs let you respond as the situation becomes clearer. Their value is captured by real options.
The difference changes what you buy: under measurable risk, perhaps insurance; under uncertainty, perhaps flexibility, a smaller initial commitment, or a plan that doesn’t depend on one forecast being right.
04A worked example
Consider a fictional bakery choosing whether to sign a five-year lease beside a commuter station. At its existing shops, ordinary weekday demand is well described by three levels. The estimated probabilities are 20% each for demand of 80 or 120 loaves and 60% for demand of 100 loaves. That gives average demand of 100 loaves.
The new site looks similar. But the rail operator is considering a timetable change, and a nearby office building may lose its largest tenant. The bakery has no reliable basis for assigning probabilities to either development.
What it looks like One forecasting problem. The owner uses the average from a slightly adjusted copy of the existing shops’ demand model to justify the lease.
What’s actually going on There are two different problems. Day-to-day variation at established shops is reasonably modeled as risk. Whether the new location will attract the same customer base is uncertain. The historical probabilities describe fluctuations around an established pattern. Whether that pattern will transfer remains uncertain, even after averaging the forecast.
What would have helped Separate ordinary demand variation from assumptions about the location. Check whether the shop can survive much lower commuter traffic while treating the exact probability of that traffic drop as unknown. Seek a break clause or run a temporary stall before signing. The trial can reveal whether passing commuters actually buy bread. The railway’s plans remain a separate uncertainty.
05Where people trip up
- A simulation’s assumptions need separate validation. A Monte Carlo simulation calculates consequences of the probabilities you supply. Thousands of simulated futures can still omit the development that matters most. That’s a form of model risk.
- Data volume and relevance are separate questions. Ten years of sales under one business model may tell you little about demand after a new regulation or competitor arrives. Check whether the process generating the observations still applies.
- Unknown probabilities leave relative likelihoods open. Assigning equal weights to scenarios may be a useful exploratory convention. Evidence is needed to establish that those futures have equal chances.
- Risk aversion and uncertainty avoidance describe different preferences. Risk aversion concerns preferences over gambles with specified probabilities. Ambiguity aversion concerns reluctance to choose options whose probabilities are unclear. Someone can accept large known risks while avoiding small ambiguous ones.
- The labels leave the decision to you. Uncertainty calls for examining your exposure and assumptions and identifying escape routes before deciding whether to stop. Waiting can also be costly, especially when a modest experiment would teach you something.
06When it isn’t a clean split
Most real decisions contain both. You may estimate how often a component fails while remaining unsure whether the whole system faces an overlooked failure mode. Probabilities can also be usable without being exact: weather forecasts are estimates, yet often informative enough to act on.
Some approaches to decision theory represent uncertainty through subjective probabilities. That is legitimate if you make the judgments explicit. An opinion expressed as a percentage still rests on judgment, while an observed frequency rests on recorded outcomes.
Knight’s distinction differs from epistemic vs. aleatory uncertainty, the distinction between gaps in knowledge and variability in a process. You can lack knowledge about the bias of a coin while still having a useful model for repeated tosses. The two distinctions ask different questions: what causes the not-knowing, and how well can you represent it probabilistically?
07Roots
Frank Knight developed the argument in his doctoral work at Cornell, later published as Risk, Uncertainty and Profit in 1921. His puzzle was profit. If competition prices labor, capital and predictable costs, what explains the money left over for an entrepreneur?
An owner pays wages and orders materials before knowing what customers will buy. Knight argued that some possible losses can be grouped, estimated and treated as business costs, much as insurers pool claims. But judgment about a new enterprise cannot always be reduced to that kind of calculation. Bearing this unmeasurable uncertainty helped explain entrepreneurial profit in his theory, along with the possibility of loss.
The distinction traveled beyond economics into how people choose. Daniel Ellsberg’s 1961 paper used urns containing colored balls to contrast known proportions with unspecified ones. His examples highlighted a preference for known probabilities that standard accounts of choice struggled to explain.
Later planning methods made the distinction operational. Analysts could test policies across many plausible futures even when they disagreed about the probability forecast. The aim shifted from finding the best answer for a trusted model to finding an answer that survives doubts about the model itself.
08How solid is this?
The distinction is foundational in decision theory, not an empirical rule that cleanly divides every situation. Experiments on ambiguity show that people respond differently to known and unspecified probabilities, but do not establish a single best response to uncertainty.
09Connections
- Often confused with Asymmetric Information, Epistemic vs. Aleatory Uncertainty
- Countered by Scenario Planning
- Can lead toAmbiguity Aversion (Ambiguity Effect)
- Includes Risk of Ruin
- See also Probability Neglect, Monte Carlo Simulation, Robust Decision-Making, Expected Value, Model Risk, Risk Aversion, Real Options, Bayesian Updating
+ 3 more in the list
10Origin and sources
Frank H. Knight drew the influential economic distinction in Risk, Uncertainty and Profit (1921), which grew out of his doctoral work at Cornell.
- [1]Knight, F. H. (1921). Risk, Uncertainty and Profit. Houghton Mifflin Company.
- [2]Ellsberg, D. (1961). Risk, Ambiguity, and the Savage Axioms. The Quarterly Journal of Economics, 75(4), 643–669.
- [3]Lempert, R. J., Popper, S. W., & Bankes, S. C. (2003). Shaping the Next One Hundred Years: New Methods for Quantitative, Long-Term Policy Analysis. RAND Corporation.
Suggest an edit· Updated 2026-10-02