In 1921 the economist Frank Knight drew a distinction that most of finance has spent a century forgetting. Risk is measurable. The outcomes and their probabilities are knowable, at least in principle. Think of a roulette wheel, a fair die, a large actuarial pool where the law of large numbers makes the total predictable even though each case isn't. Uncertainty, now often called Knightian uncertainty, is when you can't put reliable probabilities on the outcomes at all. The event is novel, or the system can change its own rules, or the relevant history is too short or too non-stationary to estimate from. A technological disruption, a war, a regime change in policy and a first-of-its-kind product all qualify. And the difference matters in practice. For risk, you can compute an expected value and size to match. For uncertainty, the true probability is unknown, and pretending otherwise with a precise-looking number is how smart people blow up.
Advertisement
Why do markets mix up risk and uncertainty?
Markets and models mix the two up all the time. The mix-up is seductive because uncertainty is uncomfortable and a number feels like control. The classic disasters all come from treating uncertainty as if it were tame, measurable risk. In 1998 Long-Term Capital Management sized positions off volatility estimates that assumed a stable, estimable distribution. A Russian default that its models rated a multi-standard-deviation impossibility destroyed it. The 2008 crisis was built on mortgage models fed a decade of data in which house prices had never fallen nationally. So the models put the "probability" of that event at essentially zero, right up until it happened.
Value-at-Risk numbers, credit ratings on structured products and any talk of a "5-sigma event" share the same flaw. They take a domain of real uncertainty and dress it in the mathematics of risk. That manufactures false precision and, worse, false confidence and oversizing. Nassim Taleb's whole body of work, The Black Swan included, is an extended sermon on this error. Damodaran gives the practitioner's version: when you cannot estimate the distribution, admit it, and stop hiding behind decimals.
How do you invest when you cannot estimate the odds?
The right response isn't paralysis. Match your method to the regime you're in. Expected value and Kelly are the right tools where outcomes can really be estimated. Think of a portfolio of many independent, well-understood positions, a diversified insurance book, a recurring-revenue business with years of cohort data. Real uncertainty is the other regime. It covers a single novel technology and a business whose future depends on an unprecedented event. It also covers macro forecasts, and Phase 7 will end by insisting most of macro is Knightian. There you need different tools. The first is margin of safety (per Graham/Klarman). Buy so cheaply that you profit across a wide range of outcomes without needing to know the odds. The second is robustness, a portfolio that survives many futures instead of one optimised for your forecast. The third is small size or avoidance. Don't bet big on things whose odds you can't honestly estimate.
The most useful habit here is one question, asked of every position: am I dealing with risk I can quantify, or uncertainty I'm pretending to quantify? Then widen your margin of safety, and shrink your size, in proportion to how much of the second you've got.