Daniel Kahneman died on March 27, 2024, at the age of 90.
He spent the last fifty years of his life proving that human beings — including very smart human beings — are systematically wrong about how confident they should be. His final book, written with Olivier Sibony and Cass Sunstein, was about a quiet form of error he called "noise."
The lesson at the center of it is simple, and it is the entire reason prediction markets exist.
Confidence is not accuracy
In a famous study, Kahneman and his colleagues asked U.S. Army officers to forecast the performance of cadets in officer training. The officers were extremely confident in their predictions. They were wrong about as often as random chance.
The result reproduced almost exactly when the same study was run on doctors predicting patient outcomes, judges predicting recidivism, recruiters predicting job performance, and financial analysts predicting earnings.
The pattern is universal. The more confident the expert, the larger the gap between their stated certainty and the actual hit rate.
Kahneman called this overconfidence. The behavioral fix, he argued, was to never trust a single confident forecast — and to systematically check whether your past predictions had matched reality.
The forecasting tournament that proved him right
In the 2010s, Philip Tetlock at Wharton ran what is now the most rigorous forecasting tournament ever conducted. Over 2,800 volunteers and several teams of intelligence analysts predicted thousands of geopolitical and economic outcomes over four years.
The most accurate forecasters had two traits in common, and only two:
First, they assigned probabilities. Not "likely." Not "probably." A specific number, like 64%.
Second, they updated those numbers constantly. The best forecasters revised each prediction about 12 times on average. The worst revised maybe 3 times.
Confidence didn't matter. Education didn't matter. Access to classified information didn't matter. What mattered was the willingness to commit to a number and then change it when the world changed.
Why a number is so much better than a word
Imagine two analysts evaluating the same Thai macro question — say, whether the Bank of Thailand will cut rates before the end of Q3.
Analyst A says: "I think the BoT will probably hold."
Analyst B says: "I price the probability of a cut before October 1 at 32%."
A year from now, you cannot evaluate Analyst A. "Probably" is unfalsifiable. He was right whether the BoT held or cut.
You can evaluate Analyst B. If the BoT cuts, his 32% was on the lower end of correct. If the BoT holds, his 32% was on the higher end of correct. You can score him over hundreds of forecasts and learn whether he is calibrated.
Kahneman's insight, distilled: anyone who refuses to give you a number is also refusing to be accountable.
What prediction markets do mechanically
A prediction market is a system that forces every participant to commit a number — and to put real capital behind it. The price of a contract is, by definition, the probability the market assigns to the outcome.
If "BoT cuts before Q3" is trading at 32¢, that's a 32% probability. Not "probably." Not "likely." 32%.
And because every participant has capital at risk, the price is the most calibrated estimate available — better than any single analyst's view, better than any survey of analysts, better than the consensus quoted in research.
That isn't an opinion about prediction markets. It's the empirical finding from Tetlock's research, validated against 25 years of forecasting data.
How to use this in daily life
You don't need a prediction market account to start thinking like Kahneman wanted you to. Try this for one week:
Read a news headline. Before you read the article, write down your own probability for one outcome. "Probability that the Fed cuts in June: 41%." "Probability that Trump signs the trade deal: 73%."
Then, when the actual outcome arrives, check your number against reality. Over a hundred forecasts, you will see whether you are calibrated — whether your 70%s actually happen 70% of the time.
Most people who try this discover something uncomfortable: their 90%s happen about 60% of the time. Their gut is overconfident, exactly as Kahneman predicted.
The good news is that calibration is trainable. The act of writing down a number and checking it later, repeated, fixes the bug.
The line worth remembering
Kahneman's last book closes with a thought he repeated often in interviews: "We are confident when we are wrong."
The asymmetry Kahneman described — markets admitting uncertainty while experts rarely do — is the exact design premise behind Juno's prediction market platform, which aggregates crowd forecasts on Thai economic and political outcomes rather than relying on any single analyst.
It is the most useful sentence ever written about forecasting. The market admits this. The expert almost never does. That asymmetry, in a single sentence, is why prediction markets work.