Risk

Risk versus Uncertainty: Deciding When the Probabilities Do not Exist

Tell the difference between a decision you can price and one you cannot, and switch tools when you cross the line, so you stop attaching invented probabilities to genuinely unknowable outcomes and start buying reversibility instead.

  • Advanced
  • 12 min total
  • 14 chapters

What decision this helps you make: Which regime a given decision sits in, and therefore whether to optimise an expected value, buy information first, or design for the case where the model is simply wrong.

What this topic is

Risk is uncertainty you can put numbers on: the outcomes are listable and the odds are knowable or estimable, which is why insurance exists at all. Uncertainty in the strict sense (the distinction Frank Knight drew in 1921) is the situation where you cannot list the outcomes, cannot assign meaningful probabilities, and have no data-generating process to sample from. Both get called "risk" in ordinary speech, and the tools that work brilliantly in the first regime quietly mislead in the second.

Why it matters

Almost every decision-support technique a business uses (expected value, discounted cash flow, simulation, sensitivity analysis) assumes you are in the first regime. Applied to the second, they do not fail loudly; they produce a confident number with a decimal point on it. That number then anchors the discussion, and the fact that its inputs were invented disappears somewhere between the model and the meeting.

Who should learn it

Anyone entering a market that does not exist yet, pricing a product nobody has sold, planning against a regulatory regime not yet written, or being asked for a probability they have no basis to supply.

What you will understand

  • The three regimes, and the specific test that tells you which one you are in
  • Why expected value stops being the right objective when the distribution is unknown
  • What replaces it: robustness, reversibility, staged commitment, and buying information
  • The honest counter-argument that all uncertainty can be expressed as subjective probability

Prerequisites

Common misconception

"If we do not know the probability, we should estimate one: a rough number beats no number." Sometimes, and it depends entirely on where the number comes from. An estimate anchored in something real (a base rate, an adjacent market, your own history) is genuinely better than nothing. A number produced because a cell in the model needed filling is worse than nothing, because it converts an acknowledged unknown into a stated input, and every calculation downstream inherits a confidence that nobody actually has. The failure is not imprecision. It is that the output no longer carries any signal that its foundation was invented.