There's a sentence that changed how I think about markets: markets don't reward correct predictions — they reward correct updates.
Read that again. The distinction is everything.
A correct prediction means you were right about what happened next. A correct update means you incorporated new information faster and more accurately than the consensus. These are not the same thing. You can have a correct update and still be wrong about the outcome. You can have a correct prediction and still lose money if the market had already priced it in.
The Bayesian framework is built around updates, not predictions. And that framing turns out to be far more useful.
The Formula Most Investors Ignore
In plain language: your updated belief in a hypothesis, given new evidence, equals how likely that evidence would be if your hypothesis were true — multiplied by your prior conviction — divided by the overall probability of seeing that evidence at all.
What this means practically: new information should move your beliefs in proportion to how diagnostic it is. Evidence that is equally likely regardless of whether your thesis is true should not move your beliefs much. Evidence that would only appear if your thesis were true should move them significantly.
Most investors do neither of these things correctly. They either overweight new information (noise trading) or underweight it (anchoring). The Bayesian framework gives you a principled approach to calibration.
The Frequentist Investor
The dominant mode of investment thinking is what I'd call frequentist by temperament: form a view, collect supporting evidence, defend the view, update only under extreme duress.
This approach fails in a specific, predictable way. The investor builds a thesis. The thesis gets published, discussed, debated. It becomes part of their identity. At that point, new contrary evidence stops being incorporated cleanly — it gets filtered through the lens of defending the prior thesis. The update function breaks.
You can watch this happen in real time during earnings calls. A company misses estimates. The bull analyst explains why the miss doesn't matter. A company beats estimates. The bear analyst explains why the beat is unsustainable. Both responses are the brain's way of preventing a clean update. The analysis follows the prior conclusion rather than informing it.
"The update function breaks the moment a thesis becomes part of your identity. The analysis starts following the conclusion rather than informing it."
What a Bayesian Investor Actually Does
A genuine Bayesian approach to investing requires three things most investors resist.
First: explicit priors. Before any position, state your thesis with a probability estimate. Not "I think this will go up." Something like: "I estimate 65% probability that this company will achieve $100M ARR within 18 months, based on current growth rate and pipeline visibility." The number forces precision. Precision forces honesty about what you actually believe.
Second: diagnostic evidence lists. Before entering a position, identify what evidence would significantly update your thesis in either direction. What would make you more confident? What would make you less? This list becomes your update protocol. When new information arrives, you run it against the list — not against your current P&L.
Third: regular posterior updates. Set a cadence — quarterly for long-term positions, more frequently for short-term — to explicitly recalculate your conviction level given what you've learned. Not "does the thesis still feel right?" That's emotional. "What new evidence have I received, how diagnostic is each piece, and what is my updated probability estimate?" That's Bayesian.
Obsidian Quant as a Bayesian System
The architecture of Obsidian Quant is, at its core, a Bayesian belief-updating machine. Every price signal, every volume pattern, every volatility regime shift is treated as evidence that updates the probability distribution of where we are in the market cycle.
No position is entered based on conviction alone. Every entry is based on an estimated edge that is recalculated on each new evidence batch. When the evidence shifts — when the market starts behaving differently than the model expects — the position sizing updates accordingly. The model doesn't argue with the market. It updates.
This is what algorithmic trading gets right that discretionary trading often gets wrong: the update function is clean. It isn't contaminated by ego, identity, or the sunk cost of a previously expressed view. The prior is replaced by the posterior on every iteration.
The Calibration Problem
The hardest part of Bayesian investing isn't the framework. It's calibration.
A well-calibrated investor is right about 70% of things they're 70% confident about, right about 90% of things they're 90% confident about, and so on. Most investors are overconfident: they express 80% confidence in things they're right about 55% of the time. The overconfidence means they update too little when they're wrong — because they can't believe they were that wrong.
Calibration is built through deliberate tracking. Keep a prediction journal. Every time you make a significant investment decision, record your thesis and your explicit probability estimate. Review it quarterly. Where are you overconfident? Where are you underconfident? The answers are usually surprising and consistently useful.
The Bayesian investor isn't trying to be right. They're trying to be right faster than the consensus — which means being wrong quickly, updating cleanly, and compounding on the edge that comes from better information processing rather than better initial prediction. That edge doesn't erode. It compounds.