Why One Prediction Market Is Never Enough

Every prediction market has an opinion. But any single market is just one data point; it reflects the particular mix of traders, liquidity, and incentives on that one platform at that one moment. Polymarket skews toward crypto-native traders. Kalshi has a different regulatory structure and a different user base. Metaculus draws from forecasters who think explicitly about calibration. Each one has a different blind spot.

The research on forecasting is fairly unambiguous on this point; aggregated predictions from multiple sources consistently outperform any single source. It is the same reason that the average of many polls beats any individual poll, and why ensemble models in machine learning tend to beat individual models. Diversity of inputs reduces noise.

The FatNarwhal Consensus Aggregator takes probability estimates from multiple markets or sources and blends them into a single consensus probability. It is not just a simple average; it lets you weight each source by how much you trust it, which is the right way to aggregate if some sources have better track records than others.

How to Use It

Head to fatnarwhal.com/consensus and start adding sources for a question you care about.

Say you want a consensus probability on whether the Fed will raise rates at the next meeting. You find four sources; Polymarket at 22%, Kalshi at 19%, a superforecaster estimate at 25%, and your own research at 18%. Enter each one with a weight you assign based on your trust in that source.

If you have no strong prior on which to trust more, equal weights are fine and the tool defaults to that. If you think one source is substantially more reliable, bump its weight up and the consensus will shift accordingly. The tool shows you the weighted average, the range across your sources, and the variance, which is a useful signal on its own; high variance across sources means the sources genuinely disagree, which is worth knowing before you trade.

Try it with sources that span a wide range and watch the consensus stabilize in the middle. Then try removing the outlier and see how much it moves. If one source has an outsized effect, that is a sign it is either very far from the others or you gave it a very high weight.

The Math Behind It

The weighted consensus is straightforward:

$$\text{consensus} = \frac{\sum_i w_i\, p_i}{\sum_i w_i}$$

Where each probability_i is the estimate from source i and weight_i is how much you trust it relative to the others.

If all weights are equal it reduces to a simple arithmetic mean. If weights differ, the result is a weighted average that pulls more toward the more-trusted sources.

More sophisticated aggregation methods, like the geometric mean of odds or the logarithmic opinion pool, handle the combination differently and can perform better in specific settings. The geometric mean of odds tends to be less influenced by extreme estimates and is worth understanding as an alternative. The tool shows the simple weighted mean as the headline but surfaces the geometric mean as a comparison so you can see when they diverge meaningfully.

When to Use It and When Not To

Use it any time you have multiple probability estimates for the same event and want a principled single number to work from. It is particularly useful before sizing a prediction market position; rather than anchoring on whichever market you happened to check first, ground your estimate in the full available information.

The main limitation is obvious; the tool can only be as good as the inputs. If all your sources are watching the same news and have the same information, aggregating them gives you false confidence in a consensus that is actually just one view repeated multiple times. True aggregation benefit requires genuine diversity of information and methodology across sources.

Also, be thoughtful about weighting. Defaulting to equal weights is not always right; if one source is a superforecaster team with a proven track record and another is a new market with thin liquidity, they should not carry the same weight.

Try It

Find an event with live prices on at least two prediction platforms and go to fatnarwhal.com/consensus. Add your sources, weight them, and build your consensus. Compare the result against the individual sources; you will almost certainly have more conviction in the number you get out than in any single input.

The wisdom of crowds works. So does the wisdom of markets. Combining both works better.