By: Igor (@justigor)

Prediction markets are usually framed as places to take a view. You buy one side, you wait, and eventually reality settles the score. That framing works, but it leaves a lot of value on the table.

Markets can also be tools for thinking.

The most consistently accurate forecasters don’t win by making bold calls and defending them. They win by breaking questions apart, anchoring on base rates, and updating continuously as new information arrives. Markets don’t automatically enforce that discipline, but they can reward it if you approach them the right way.

This is a practical guide to applying that style of forecasting to modern prediction markets, using a concrete example: US Non-Farm Payrolls. Not to prove a point, but to show how markets can support better reasoning when used deliberately.

A more comprehensive theoretical guide may follow. This is the applied version.

1. Start by making the question explicit

Consider the question:

“What will US Non-Farm Payrolls print next month?”

This should immediately raise a red flag. This is not a yes/no question. It’s a distributional one.

Payroll prints vary month to month. They have seasonality, revisions, fat tails during regime changes, and asymmetric surprises. Before touching a market, the first step is to describe the shape of the answer, not just a number.

What would count as a weak print? What would be extreme upside? Where do most outcomes usually cluster?

If you can’t answer those, you’re not forecasting yet. You’re guessing.

2. Take the outside view first

The outside view asks: what normally happens?

For Non-Farm Payrolls, this means looking at historical distributions over relevant periods. Most months fall within a relatively stable band. Big surprises happen, but they’re rare and often tied to structural shifts.

This step matters because markets are very good at pulling people away from base rates. Narratives, headlines, and social amplification feel more informative than history, even when they aren’t.

Anchoring on the outside view gives you a baseline that is hard to hand-wave away later.

3. Decompose the drivers

Only after anchoring on history does it make sense to look inside the current situation.

For payrolls, that might include:

• Recent jobless claims

• Wage growth and hours worked

• Sector-specific hiring signals

• Business surveys

• Seasonality and recent revisions

You don’t need a precise model. The point is structure. Each component nudges the distribution slightly. Some widen it. Some shift the centre. Some mostly affect the tails.

If your reasoning collapses into “this feels strong” or “this feels weak,” you’ve lost the thread.

4. Express belief as a distribution, not a call

Most prediction markets compress all of this reasoning into a single directional trade. That can work, but it discards information.

A better mental model is: where would I place probability mass if I had to draw it?

Early on, that distribution should be wide. As evidence arrives, it should narrow. The mean might move, but so should your confidence.

Markets that support buying and selling throughout the lifecycle make this natural. You’re rewarded not just for being right at the end, but for tightening the distribution before others do.

“Early” here doesn’t mean first in time. It means early relative to the market’s certainty.

5. Update incrementally, not emotionally

Good forecasters update often and in small steps.

A single data point rarely justifies a wholesale belief reversal. More often, it moves the distribution a few percent or changes the weight of the tails.

A useful habit is to write down why you updated:

• Was it a leading indicator?

• A correlated signal?

• Something genuinely new?

Markets tempt people to overreact because price moves feel like information. Sometimes they are. Often they’re just feedback loops.

If you can’t articulate why something should move the distribution, it probably shouldn’t.

6. Track confidence, not just outcomes

One of the most unintuitive lessons from forecasting research is that being “right” is not the same as being well-calibrated.

Two people can both predict 200k jobs. One does so with high confidence. The other with humility. Over time, the second forecaster is often more useful, even if their point estimates match.

Markets tend to hide this distinction. P&L collapses confidence and correctness into a single number. That’s fine for trading, but it’s a poor teacher.

If you want to improve, track how confident you were when you made the call, not just whether it paid off.

7. Remember that forecasting is a team sport

Small, diverse groups consistently outperform individuals in forecasting tasks. Not because they agree, but because they surface blind spots and force updates.

Markets are meant to be that aggregation layer. When designed well, they let independent beliefs collide, update, and converge.

But this only works if changing your mind is cheap. If revising a view is punished or awkward, people cling to identity instead of information.

Markets that allow continuous participation and symmetric entry and exit align better with how collective forecasting actually improves.

8. Where this leaves us

None of this requires exotic tools. The habits alone outperform most casual market participation.

But market design can make these habits easier or harder to apply. Some markets optimise for simplicity and fast resolution. Others optimise for expressivity and belief updating. Neither is universally better.

Non-Farm Payrolls is a good example of where distributions matter, confidence evolves, and updating is the real skill. Not every question is like that.

Prediction markets don’t just reveal what people think. At their best, they teach people how to think under uncertainty.

Igor leads research at @functionspaceHQ an open-source project exploring market-led resolution and novel economic instruments for prediction markets