Philip Tetlock, the Wharton researcher who ran the most rigorous forecasting tournament ever conducted, found something that surprised even him.
The best forecasters in his study — the people who beat CIA analysts with classified intelligence by more than 30% — did not have higher IQs than the average participant. They did not have more education, more domain expertise, or more access to information.
What they had was a habit. The habit took about five minutes a day.
This is what the habit is, why it works, and exactly how to start it tomorrow morning.
The habit: predict, write, check
Every morning, the best forecasters in Tetlock's study did three things, in order:
One. They read the news, looking for questions whose answers would be known within a defined window. "Will the Fed cut rates at the next meeting?" "Will Apple's earnings beat consensus?" "Will the SET close above 1,500 by end of quarter?"
Two. For each question, they wrote down a specific probability. Not "likely." Not "probably." A number. "47%." "82%." "12%."
Three. When the answer arrived — days, weeks, or months later — they checked their prediction against reality, and they tracked whether their numbers matched outcomes over time.
That's it. That's the entire practice. The most decorated forecasters in the world were doing this five minutes a day, on the side, while holding regular jobs.
Why it works
The habit attacks the single most important weakness in human reasoning: poor calibration.
Calibration is the match between your stated confidence and your actual hit rate. If you say you're 80% confident in something, well-calibrated people are right about 80% of the time when they say that. Most people, when they say 80%, are right about 55% of the time.
The gap between stated confidence and actual accuracy is enormous, and almost nobody is aware of theirs. The only way to close the gap is to write down predictions, check them later, and see where you're systematically wrong.
Daniel Kahneman, in the work he produced before his death in 2024, described this as the single most useful exercise a thinking person can do. "We are confident when we are wrong," he wrote. The exercise of writing the number, then checking, slowly retrains the part of your brain that produces confidence.
What you need (almost nothing)
You need three things to start. None cost money.
A notebook, a Notes app, or a spreadsheet. Anything that lets you record a date, a question, a probability, and an outcome.
A list of questions that resolve in the next 1-30 days. The news produces these for free.
Five minutes per morning.
The template
For each entry, capture five fields:
- Date: when you made the forecast
- Question: phrased as YES/NO with a specific resolution date
- Your probability: a number between 0 and 100
- Resolution date: when the answer will be known
- Outcome: filled in when you check, marked YES or NO
That's the entire structure. After 30 days, you'll have ~30 entries. That's enough to start spotting patterns.
How to read your own track record
After 60 to 100 forecasts, group them by your stated probability. Take all your 70%-confident predictions. How many of them were right?
If they were right about 70% of the time, you are well-calibrated at the 70% level. Congratulations.
If they were right 55% of the time, you are overconfident — your "70%" should probably be "55%."
If they were right 80% of the time, you are underconfident — your "70%" was actually a "80%."
This is the entire skill. You are not trying to be more accurate. You are trying to make your confidence match your accuracy. That match is the difference between a useful forecaster and a confident-sounding one.
What to forecast
Pick questions that resolve cleanly. "Will Bitcoin close above $100K on July 31?" is good. "Will inflation be transitory?" is bad — there's no clean answer.
Mix domains. Markets, sports, weather, politics, your own work. The more diverse the questions, the more honest the calibration data.
Avoid questions where you have a strong emotional stake. Your beloved football team's playoff odds are a worse training set than anything else, because you'll either be too pessimistic (defensive) or too optimistic (loyal).
What changes after 90 days
Three things, reliably, based on what Tetlock and others have measured in calibration studies.
First, you stop trusting your own first instinct. The 90% confidence that used to feel like certainty starts to feel like a candidate for revision. You become more comfortable holding 60%s and 40%s — which is where most real-world questions actually live.
Second, you read the news differently. You stop consuming opinion pieces as truth claims and start consuming them as data — what does this analyst's track record say about how to weight their view today?
Third, when you encounter a prediction market, you can read it instantly. "47¢" stops being abstract. It's the same kind of number you've been producing yourself for three months.
The shortcut, if you want one
If five minutes a day sounds like a lot, here is the shortcut: pick one ongoing question. Macro is good. "Will the Fed cut by July?" Update your estimate weekly. Check at resolution.
That's not a full calibration exercise, but it's enough to feel the muscle. Most people who try the full version started with one question and got hooked.
The number you can train
The thing that makes this exercise different from most self-improvement habits is that there is a number at the end. You either are or are not better calibrated after 90 days. The data is testable, and it's yours.
Five minutes a day, for ninety days, costs you 7.5 hours total. That's the entire investment to be measurably better at thinking about the future than the rest of the population.
That 5-minute daily practice is exactly what Juno's prediction market platform is designed to support — real markets on Thai economic and political outcomes where your calibration score is tracked and improves with every forecast.
The best forecasters in the world figured this out. The exercise is sitting there for anyone who wants to copy it.