Risk Versus Uncertainty
Frank Knight (1921) drew the foundational line: risk describes situations where outcomes are unknown but their probabilities are measurable, so a loss distribution can be written down, whereas uncertainty describes situations where the probabilities themselves cannot be reliably measured. A separate organizing device, the known/unknown 2x2 matrix (popularized by Donald Rumsfeld in 2002), sorts our awareness into four quadrants: known knowns, known unknowns, unknown knowns, and unknown unknowns. The two ideas align but are not the same: known unknowns behave like Knightian risk and unknown unknowns like Knightian uncertainty, while the unknown-knowns quadrant (blind spots in our own organization) has no Knightian analogue. Tools such as VaR, hedging, and stress testing live in the measurable-risk world; they are structurally blind to genuine uncertainty.
Why it matters
Knight asks one question: can you put trustworthy numbers on the odds? A roulette wheel is risk, you know the probabilities. A brand-new pandemic or a financial innovation nobody has lived through is uncertainty, you can guess but the odds are not really knowable. The 2x2 matrix is a different lens about awareness: some things you know you know, some you know you do not know, some you have but have not surfaced (unknown knowns, the dangerous blind spots), and some you cannot even imagine (unknown unknowns, the black swans). Keep the two frames side by side: VaR is excellent in Knight's risk world and useless against true uncertainty.
Formulas
Worked examples
Sort each into Knightian risk or uncertainty, then place it in the 2x2 matrix: (a) the payoff of a fair coin flip; (b) tomorrow's loss on a diversified equity book under normal markets; (c) the systemic fallout of a never-before-seen financial instrument.
(a) and (b) are Knightian RISK: probabilities are measurable, so they sit in the known-unknowns quadrant (we know what we do not know and can model it with VaR or stress tests). (c) is Knightian UNCERTAINTY and belongs in unknown unknowns, the black-swan quadrant, where no reliable distribution exists and VaR is blind. A fourth case, a flawed model assumption the desk relies on without questioning, is an unknown known, a blind spot that maps to neither Knight category but is a real source of failure.
Why can VaR give a precise 1-day 99% figure yet miss the 2008 crisis?
VaR estimates a quantile of an assumed loss distribution, a Knightian-risk tool. The 2008 collapse involved structural shifts and instruments whose joint behaviour was effectively unknown unknowns, Knightian uncertainty. A model can be internally precise about the wrong, too-thin distribution, so a confident VaR number coexists with total blindness to a regime the model never contemplated.
Was the March 2020 COVID-19 market crash an "unknown unknown"? Use it to see why the label is contested.
Many risk teams treated the pandemic as a black swan, an unknown unknown that no model anticipated, and to a desk calibrated only on recent data the speed of the March 2020 drawdown was Knightian uncertainty. But Nassim Taleb, who coined the term, argued the opposite: a pandemic was a foreseeable, repeatedly warned-about hazard, so he called COVID-19 a "white swan," a known unknown that planning should have covered. The lesson is that the quadrant depends on the observer's information set. A genuine unknown unknown is unforeseeable to everyone; an unknown known is a blind spot where the warning existed but the organization ignored it.
Common mistakes
- ✗Risk and uncertainty are the same thing. Knight's entire point is that they differ: risk has measurable probabilities and a writable loss distribution, uncertainty does not.
- ✗The known/unknown 2x2 and Knight's risk-vs-uncertainty are one idea. They align but are distinct frames; in particular the unknown-knowns quadrant (organizational blind spots) has no Knightian counterpart, so the matrix should not simply be relabelled as Knight.
- ✗VaR and stress tests can quantify true uncertainty. These tools require an estimable distribution, so they operate in the risk world and are structurally blind to genuine, unmeasurable uncertainty.
- ✗Unknown unknowns can be eliminated with enough data. By definition they are outcomes not yet conceived; more history sharpens known risks but cannot reveal a scenario the model has never imagined.
Revision bullets
- •Knight (1921): risk = measurable probabilities; uncertainty = probabilities not reliably measurable
- •Known/unknown 2x2 is a separate awareness frame, not a relabel of Knight
- •Known unknowns ~ Knightian risk; unknown unknowns ~ Knightian uncertainty
- •Unknown knowns (blind spots / risk culture) have NO Knight analogue
- •VaR, hedging, stress tests work on risk; they are blind to true uncertainty
Quick check
In Frank Knight's (1921) distinction, the defining feature that separates "risk" from "uncertainty" is whether
Which quadrant of the known/unknown 2x2 matrix has NO clean Knightian analogue?
A bank reports a precise 1-day 99% VaR yet is devastated by a crisis it never modelled. This is best explained as
Connected topics
Sources
- Knight, F. H. Risk, Uncertainty and Profit. Houghton Mifflin, 1921.Original source of the risk (measurable probabilities) versus uncertainty (unmeasurable) distinction.
- Diebold, Doherty & Herring (2010)Diebold, F. X., Doherty, N. A., & Herring, R. J. (eds.). The Known, the Unknown, and the Unknowable in Financial Risk Management. Princeton University Press, 2010.Develops the known/unknown/unknowable framework for FRM and its limits for VaR-style measurement.
- Jorion (2007), Ch. 1Jorion, P. Value at Risk: The New Benchmark for Managing Financial Risk. 3rd ed. McGraw-Hill, 2007.Notes that VaR measures quantifiable market risk and does not capture model risk or unforeseen regime shifts.