Why Your Forecasts Are Probably Wrong & What to Do About It.
Most people don’t forecast. They guess, then defend the guess.
Ask someone what they think about a given outcome: an election, a rate decision, whether a company hits its earnings number and you’ll usually get a verdict, not a probability. Yes or no. Will or won’t. The confidence comes first; the reasoning gets built afterward to support it. This is how most humans process uncertainty, and it’s also why most humans are bad at it.
Bayesian thinking is the alternative. It’s not a trading strategy or a piece of software. It’s a discipline for updating what you believe when new information shows up and it’s the closest thing forecasting has to a foundational law.
The Core Idea, Without the Math
Thomas Bayes was an 18th-century minister who never published the theorem that carries his name. The formal version involves conditional probabilities and looks intimidating on a whiteboard. The practical version is simpler than that:
Start with a belief. Update it when you get evidence. Don’t overreact, don’t underreact: adjust in proportion to how strong the evidence actually is.
That’s the whole idea. Three components:
Prior: what you believed before the new information arrived
Evidence: the new data point, signal, or event
Posterior: your updated belief, which becomes the new prior for the next round
The posterior is never final. It’s just the prior for whatever comes next. This is the part people miss: Bayesian thinking isn’t a one-time calculation, it’s a loop that never closes.
Why This Is Harder Than It Sounds
If updating beliefs in proportion to evidence were natural, everyone would already do it. They don’t, for a few predictable reasons.
Anchoring. The first number or opinion you hear does disproportionate work on everything that follows. A prior gets set too early and too hard, and subsequent evidence gets bent to fit it rather than the other way around.
Confirmation bias. People notice evidence that supports the prior and wave off evidence that contradicts it. In Bayesian terms, this means treating disconfirming evidence as weaker than it is and confirming evidence as stronger than it is — which corrupts the update before it even happens.
Recency bias. The opposite failure: throwing out a well-supported prior because of one loud, recent data point. A single surprising game, poll, or headline gets treated as if it invalidates months of prior evidence.
Certainty bias. Most people are uncomfortable saying “I think this is 65% likely.” They want to round to 100% or 0%, because probability feels like hedging and hedging feels like weakness. It isn’t. Precision about uncertainty is a strength, not an evasion.
A working Bayesian doesn’t try to eliminate these biases through willpower. They build a process that structurally resists them.
What This Looks Like in Practice
Forget the formula. Here’s the operating discipline:
Write the prior down before you look at new information. If you don’t commit to a number first, you can’t tell whether new information actually moved you or whether you’re just rationalizing where you already wanted to land.
Ask what the evidence would look like if you were wrong. Not just “does this confirm what I think”: actively look for the version of events that would prove the opposite. If you can’t articulate what disconfirming evidence would even look like, you’re not forecasting, you’re narrating.
Size the update to the strength of the evidence, not the volume of it. A single high-quality data point should move you more than ten low-quality ones. Loud is not the same as informative.
Separate the outcome from the process. A good process can produce a bad outcome, and a bad process can produce a good outcome, especially in small samples. Judging a forecast purely by whether it was “right” teaches you nothing about whether your reasoning was sound. Track the calibration, not just the box score.
Never let the posterior calcify into a new anchor. The update you make today is provisional. Tomorrow’s evidence gets weighed against it the same way today’s evidence was weighed against yesterday’s belief. There’s no finish line where you’re “done” updating.
Where This Actually Matters
This isn’t an academic exercise. It’s the operating system behind every domain where the reward goes to whoever handles uncertainty better than the field:
Markets don’t reward being right. They reward being less wrong, faster than everyone else pricing the same event.
Medicine runs on this exact structure: pretest probability, new evidence from a scan or a lab result, updated diagnosis. Doctors who anchor too hard on an initial impression miss things.
Competitive strategy: in business or in sports depends on updating your read of an opponent as new information comes in without abandoning a sound plan over one bad possession or one bad quarter.
Any prediction market is, structurally, a mechanism for aggregating thousands of individual Bayesian updates into a single number. The price is the crowd’s posterior. It moves because someone, somewhere, got new evidence and updated.
The Uncomfortable Part
Bayesian thinking doesn’t feel good, and that’s worth saying plainly. It requires holding a belief you’re not fully certain of, assigning it a number that admits the uncertainty, and then being willing to change that number when the evidence says so. Sometimes reversing a position you argued for confidently a week earlier.
Most people would rather be confidently wrong than accurately uncertain. Confidence feels like competence. It isn’t. Calibration is competence. The forecaster who says “I think this is 70% likely, here’s what would change my mind” is doing harder, more honest work than the one who says “this is definitely happening”: even when the second person turns out to be right, because they were right for the wrong reason, and reasons are what generalize to the next forecast.
The point of Bayesian thinking isn’t to be right more often on any single call. It’s to build a process where being wrong teaches you something, being right doesn’t make you overconfident, and your beliefs track reality a little more closely with every piece of evidence you’re willing to actually look at.
That’s the whole game. Everything else is commentary.
Axiom Forecasting Group applies this discipline to probabilistic markets: treating every price as a question, not an answer.