Risk vs. uncertainty: Knight, Kahneman, and a Mauritian sugar export

Frank Knight, a Chicago economist, drew a line in 1921 that most finance education still ignores.

Risk is when you know the probability distribution. Flipping a coin. Rolling dice. The bond market for AAA-rated treasuries. You can compute expected value, variance, drawdown, all the textbook numbers. They mean something.

Uncertainty is when you don't know the probability distribution. The 2026 Mauritian sugar harvest. Whether the Fed will pivot on rates by Q4. Whether your high-conviction stock pick will be the founder's biggest break or his biggest mistake. The textbook risk numbers still exist as calculations, but they don't mean what the textbook implies they mean.

Almost every interesting trading decision is uncertainty, not risk.

Why this distinction matters for you

Most introductory finance courses teach risk management as if all trading decisions are like dice rolls. Calculate the standard deviation, compute value-at-risk, size your position accordingly. That works fine for systematic strategies with statistical edge over thousands of trades. It works terribly for the kind of trading most retail investors and discretionary funds actually do — making 5-20 high-conviction decisions a year about specific companies in specific market regimes.

When you're trading uncertainty:

• Volatility is a poor proxy for "how wrong could I be." A stock can have 15% historical volatility and still go to zero. (Lehman did.) Volatility measures the past frequency of small moves. The risk that matters is the unmeasured probability of large unprecedented moves.
• Position sizing must reflect ignorance, not just variance. When you genuinely don't know the odds, sizing should be smaller than the formula suggests. The Kelly criterion assumes you know your edge precisely; most discretionary traders overestimate their edge by 2-3× when they self-assess.
• The shape of the distribution matters more than the mean. A trade with 80% chance of +10% and 20% chance of −60% has positive expected value, but it's a terrible trade to size large because the loss outcome is catastrophic. Conventional risk metrics will tell you it's fine.

Kahneman's contribution

Daniel Kahneman and Amos Tversky spent four decades showing that humans are bad at estimating probabilities under uncertainty in systematic, predictable ways:

• Anchoring: the first number you see colors all subsequent estimates
• Representativeness: you mistake what's vivid for what's likely
• Availability: you overweight recent and personally-experienced events
• Overconfidence: when 1,000 experts say "I'm 90% sure," they're right closer to 70% of the time

The trader who has internalized these biases — not just read about them — has a structural advantage over the trader who hasn't. Most trading mistakes aren't bad models. They're bad probability estimates from biased intuitions.

Concrete example: 2024 Mauritian sugar export decision

Suppose you're advising a Mauritian sugar producer in early 2024 on whether to hedge their EU sugar export prices via futures. You have:

• Risk numbers: futures price = €450/ton. Historical 1-year volatility = 22%. Standard hedging cost = 8% of notional.
• Uncertainty factors: EU sugar import quota review pending in Brussels; Brazilian crop forecasts uncertain due to La Niña; Indian export restrictions could lift or tighten; emerging Asian demand from rising middle class is harder to forecast than ever.

The risk-only analysis says: "your historical volatility implies a 95% confidence interval of €350-550, hedge the downside if your operating margin can't tolerate prices below €400."

The risk + uncertainty analysis says: "the historical distribution is unreliable here. The base rate of a 30%+ year-over-year move in sugar is higher than vol implies because of the policy-tail risk. Hedge more aggressively than the model suggests, AND build optionality into your physical contracts so you can switch buyers if the EU quota tightens."

Same data, very different recommendation.

The trap of false precision

A trading desk that reports "we have 15.2% expected return at 8.4% volatility" looks rigorous. The numbers are wrong in two distinct ways:

1. The expected return is itself uncertain. The +/− on that 15.2% is probably +/− 8%.
2. The volatility number is computed from a sample that may not represent the future regime.

Treating these numbers as if they're as solid as the price of a treasury bill is exactly the failure mode that gets risk managers fired in regime changes (2008, 2020, every emerging-market crisis).

What you should do differently

1. For every trade you place in the simulator, write down two probability ranges: your best estimate, and the range of possible-but-not-crazy values. If those ranges are very different, you're trading uncertainty, not risk. Size accordingly (smaller).
2. Keep a journal of your probability estimates and check them against reality. Over 6 months you'll find specific biases — most students discover they systematically overestimate their conviction on tech stocks and underestimate it on cyclicals.
3. For long-tail decisions, ignore the volatility number entirely. Use a thought experiment: "what's the worst that could plausibly happen, and can I tolerate it?" If you can't tolerate it, the position is too big regardless of what the model says.

Reading

• Frank Knight, Risk, Uncertainty and Profit (1921, free in public domain) — the canonical statement, dense but rewarding
• Daniel Kahneman, Thinking, Fast and Slow (2011) — the popular account
• Nassim Taleb, The Black Swan (2007) — strident but useful, especially the parts about Mauritius and risk-management failure modes

Next lesson: Reading an income statement under time pressure — what 60 seconds with a 10-K should actually tell you.