← All articles · September 2026 · 3 min read
Why Experts Get Predictions Wrong — and Who Does Better
Two decades of forecasting research, what it really found about experts, and the habits of the forecasters who beat them.
"Why Did Nobody Notice It?"
In November 2008, visiting the London School of Economics, Queen Elizabeth II asked why nobody had seen the financial crisis coming. A group of economists convened by the British Academy later replied in a letter that it was, in large part, a failure of the collective imagination of many bright people to understand the risks to the system as a whole. The question is a good one to ask more generally: how good are expert predictions, and why do they miss?
What Tetlock Found
The most systematic answer comes from the psychologist Philip Tetlock. Starting in the 1980s, he collected tens of thousands of forecasts from 284 people who made their living commenting on political and economic trends, and scored them when the outcomes were known. The results, published in Expert Political Judgment (2005), were humbling: on average, the experts did little better than simple rules such as "assume the current trend continues", and on longer-range questions they struggled to beat chance.
Two details matter more than the headline. First, the experts were not all the same. Those Tetlock called "foxes" — who drew on many ideas and were comfortable with uncertainty — did clearly better than "hedgehogs", who explained everything through one big theory. Second, the experts who were more prominent in the media tended to be less accurate. Confidence and a compelling story make good television; they do not make good forecasts.
The Forecasters Who Did Better
Tetlock's next project, the Good Judgment Project, entered a forecasting tournament run by the US intelligence research agency IARPA from 2011. It recruited volunteers, scored thousands of their forecasts, and identified a small group — roughly the top 2% — who were consistently accurate. Tetlock and Dan Gardner describe these "superforecasters" in Superforecasting (2015). Most had no special access or credentials. What they shared were habits: starting from base rates, breaking questions into parts, using precise probabilities, updating often in small steps, actively looking for reasons they might be wrong, and — crucially — keeping score.
Why Confident Experts Miss
Overconfidence: decades of calibration research, including work by Sarah Lichtenstein and Baruch Fischhoff, show that people who say they are very sure are right less often than their confidence implies.
One big theory: a single explanatory framework makes the world feel predictable and makes contrary evidence easy to explain away.
Hindsight: after the outcome, experts and audiences alike remember forecasts as closer to the truth than they were, so the lesson is never learned.
Incentives: bold calls get attention; careful probabilities do not. A pundit who is memorably right once is remembered; the many misses are forgotten.
No scorekeeping: most public predictions are never written down precisely enough to be checked, so nobody — including the expert — learns their real accuracy.
How to Read Expert Predictions
Ask whether there is a track record: a list of past predictions, with dates and probabilities, that someone other than the forecaster can check. Prefer forecasts stated as probabilities over vague language. Be wary of a single explanation for everything. And notice whether the forecaster changes their mind when the facts change — that is a sign of good judgment, not weakness.
Building a Record Others Can Trust
The lesson for your own forecasting is the same: write predictions down precisely, with a probability and a deadline, score them honestly, and review what you got wrong. For the predictions you want others to trust, make the record impossible to edit afterwards. icalledit.io writes a SHA-256 hash of your written prediction into its own Bitcoin transaction and gives you a PDF certificate and a proof.json file that anyone can verify without us. It proves what you wrote and when; it does not score you or prove you were right — the outcome does that.
A timestamp proves that content existed at a point in time. It is not legal advice, a patent, a trademark or a copyright registration. Compare the options.