Bayes in 30 Seconds
Bayesian updating says your new probability estimate should combine your prior belief with the evidence provided by new information. If you think an event has a 50% chance of happening and you receive information that would be 3 times more likely if the event were going to happen than if it were not, your updated estimate should be roughly 75%. The math is straightforward but the intuition is what matters: the more surprising the evidence, the more it should shift your estimate.
The formula itself is P(A|B) = P(B|A) × P(A) / P(B), where A is your hypothesis and B is the new evidence. But forget the notation for a moment. The core insight is that evidence matters more when it would be unlikely under alternative scenarios. If you're trying to figure out whether a candidate will win an election, a poll showing them ahead by 2 points in a typically close state tells you less than a poll showing them ahead by 10 points in a state they usually lose.
How Prediction Markets Approximate Bayesian Updating
When a news article is published that is relevant to a prediction market contract, traders who read the article update their probability estimates and trade accordingly. If the article provides strong evidence in one direction, prices move significantly. If it provides weak or ambiguous evidence, prices barely move. The market's price change in response to new information approximates Bayesian updating across the aggregate of all participants.
This is why prediction markets are such efficient information processors. Each participant is performing their own (possibly informal) Bayesian update, and the market price reflects the weighted average of all these updates, where the weighting is by capital committed.
Consider what happened when FTX collapsed in November 2022. Prediction markets on crypto regulation immediately shifted from around 35% chance of major new regulations within 12 months to over 70% within hours. Traders were essentially performing a collective Bayesian update: given that a major exchange had collapsed due to alleged fraud, how much more likely was aggressive regulatory action compared to the baseline scenario?
The speed of this adjustment reveals something important about market-based updating versus individual updating. While any single trader might anchor too heavily on their prior beliefs or overreact to dramatic news, the market as a whole tends to find a middle ground that approximates optimal Bayesian updating. Traders who consistently update poorly lose money and reduce their influence on prices.
When Markets Get Bayesian Updating Wrong
Markets are not perfect Bayesian updaters, though. They exhibit predictable patterns of over and under-reaction that mirror individual cognitive biases. Research by behavioral economists has identified several systematic errors that show up in prediction market pricing.
Anchoring appears when markets adjust too little from previous price levels, even when evidence is strong. You can spot this in prediction markets that move slowly toward obvious conclusions. During the 2020 election, some markets took days to fully incorporate clear polling trends, suggesting traders were anchored to previous price levels rather than updating appropriately to new information.
Overreaction happens when markets swing too far in response to vivid but ultimately limited evidence. A single dramatic poll or news story can cause price movements that subsequent information reveals were excessive. The key is distinguishing between evidence that looks important and evidence that actually is important from a probabilistic standpoint.
Base rate neglect shows up when markets fail to properly weight how common or rare certain types of events actually are. A prediction market on whether a specific startup will reach unicorn status might overweight positive news about the company's latest product launch while underweighting the base rate that fewer than 1% of startups ever reach billion-dollar valuations.
Reading Market Reactions as Information
The market's reaction to information is itself informative. If a seemingly important news event causes no price movement, the market is telling you the information was already priced in. If a seemingly minor event causes a large price movement, something about it was more significant than you realized, and investigating why the market moved is more valuable than assuming the market overreacted.
This principle becomes especially useful when using tools like Blockcircle's whale tracking system. When large traders move into or out of positions without obvious public catalysts, they may be responding to information that hasn't been widely processed yet. The absence of obvious news makes their trading behavior more informative, not less.
Sometimes the most valuable information comes from what doesn't happen. When prediction markets fail to move in response to news that seems significant, that non-reaction tells you something about how sophisticated participants actually view that information. During earnings seasons, for instance, prediction markets on company-specific outcomes often remain stable despite headlines that sound dramatic, suggesting the actual content was already anticipated.
Signal vs. Noise in Information Processing
Proper Bayesian updating requires distinguishing between signal and noise in new information. Signal is information that actually changes the probability of the outcome you're predicting. Noise is information that feels relevant but doesn't actually shift the underlying probabilities.
A concrete example: if you're trading on whether a particular bill will pass Congress, a news story about a senator giving a speech supporting the bill might feel significant. But if that senator was already known to support the bill, the speech is mostly noise. However, if the speech reveals new details about potential amendments or coalition-building, those details might be genuine signal.
The momentum tracking tools on Blockcircle can help identify when price movements reflect genuine new information versus temporary noise. Sustained directional movement with increasing volume typically indicates signal. Sharp moves that quickly reverse often indicate noise or overreaction to information that was less significant than it initially appeared.
Common Errors in Updating
Several systematic errors in probability updating are well-documented. Anchoring: adjusting too little from the prior estimate, even when the evidence is strong. Overreaction: adjusting too much for weak evidence, especially when it is vivid or emotionally salient. Base rate neglect: ignoring how common or rare the event type is when evaluating specific evidence. And confirmation bias: updating more for evidence that confirms your existing view than for evidence that contradicts it.
Being aware of these errors does not eliminate them, but it helps. When you receive new information about a prediction market contract, explicitly asking "how much should this actually shift my estimate?" and comparing that to how much you want to shift it reveals the direction and magnitude of your bias.
Confirmation bias deserves special attention because it's so pervasive in prediction market trading. Traders naturally pay more attention to information that confirms their existing positions and discount information that contradicts them. This creates opportunities for traders who can process contradictory information more objectively.
One practical approach is to actively seek out the strongest arguments against your current position before making any trade adjustments. If you're long on a particular outcome, specifically look for the most credible evidence pointing toward alternative outcomes. This helps counteract the natural tendency to cherry-pick supporting information.
The Practical Application
For prediction market trading, Bayesian updating provides a framework for deciding when to adjust positions. When genuinely surprising information arrives (information you would not have expected regardless of which outcome occurs), your position should change. When expected information arrives (information that was already priced in), your position should not change, even if the news is dramatic-sounding.
The prediction market mispricing engine can help identify when current prices appear to deviate from what Bayesian updating would suggest given available information. Markets that seem to be under-reacting to cumulative evidence or over-reacting to recent dramatic events often present trading opportunities.
Consider developing a systematic approach to information processing. Before trading on any new information, ask yourself: What was my prior probability? What is the likelihood of observing this information under different scenarios? How much should this shift my estimate? Only after working through this framework should you consider position adjustments.
The key insight is that not all information is created equal. A single data point that contradicts a strong trend deserves less weight than multiple independent sources pointing in the same direction. Bayesian updating provides the mathematical framework for weighting information appropriately, but developing good intuition for information quality takes practice.
Explore these tools on Blockcircle: Prediction Markets Mispricing Engine