What Is Forecasting? Estimate the Outcome Before You Consider the Price

Forecasting is estimating what will happen in the future using the information available now. That information can include historical data, current evidence and knowledge of upcoming events. A forecast describes what you expect—not what you want to happen or what you plan to do. Forecasting: Principles and Practice makes this distinction between forecasts, goals and plans.

Forecasts can estimate a quantity, such as next month’s sales, or the probability of an event. In a Yes/No prediction market, the useful question is usually: “What is the probability that this contract will resolve Yes?”

That is different from simply picking a winner—and different again from deciding whether to buy.

How forecasting works

Forecasting methods range from statistical models that use historical patterns to structured human judgment. The appropriate method depends on the data available and how predictable the subject is; a complicated model is not automatically necessary. Forecasting: Principles and Practice explains these choices and limitations.

For a beginner assessing an event market, use this sequence:

1. Define the exact outcome and deadline. What must happen for Yes to win? Which source will establish it?
2. Find a relevant starting point. How often have genuinely comparable events happened? This historical frequency is a base rate, not a final answer.
3. Assess current evidence. Explain why this case should differ from the starting point. Consider evidence against your preferred outcome, too.
4. Record and revise the estimate. Write down your probability, assumptions and date. Update when relevant information changes, then compare the forecast with the eventual result.

Clear definitions, documented assumptions and evaluation are central to structured judgmental forecasting.

One example: from forecast to market decision

Suppose a hypothetical market asks:

Will Company A make Product X commercially available by December 31, 2026?

Assume its full rules require availability on the company’s official website by 23:59 UTC, and explicitly exclude announcements and preorders.

You need to forecast commercial availability under those rules—not whether the company will hold a launch event. This distinction matters: Polymarket’s resolution documentation specifies that markets have rules covering the resolution source, end date and edge cases.

For this invented example, suppose 12 of 20 comparable product launches met their deadlines:

Starting frequency = 12 ÷ 20 = 60%.

You then review company-specific evidence and judge that it modestly improves the outlook, producing a 65% forecast. That five-percentage-point adjustment is an assumption based on judgment, not something the historical count proves. A small or poorly matched comparison group would also weaken the starting estimate.

Now suppose the Yes order book shows:

Item	Hypothetical value
Best bid: what a buyer offers	$0.61
Best ask: what a seller requests	$0.65
Midpoint	$0.63
Bid–ask spread	$0.04
Assumed purchase fee per contract	$0.01

A displayed midpoint is not necessarily your purchase price. Polymarket’s order-book documentation explains that an immediate purchase starts at the best available ask, while an immediate sale starts at the best available bid. Larger orders can exhaust those quotes and fill at worse prices.

Assume one contract is available at that ask, pays $1 for Yes and $0 for No, and you hold it through settlement with no further costs. The fee here is invented for teaching, not a platform fee quote.

• Total purchase cost: $0.65 + $0.01 = $0.66.
• Expected payout using your forecast: 65% × $1 + 35% × $0 = $0.65.
• Expected net result: $0.65 − $0.66 = −$0.01.

That expected result is a probability-weighted average, not the actual settlement result: you would gain $0.34 if Yes wins or lose $0.66 if No wins.

Your forecast favors Yes, but the purchase has a negative expected result under your own assumptions. The break-even probability is 66%, not the displayed 63%. Do not add the four-cent spread again: you already used the ask as your entry price.

How to tell whether your forecasts are useful

A 65% forecast still allows a 35% chance of No. One loss does not, by itself, prove the forecast was unreasonable.

Instead, keep a record across many questions. Among outcomes you assign approximately 65%, roughly 65% should occur if those forecasts are well calibrated, allowing for sampling variation. Calibration means the match between predicted probabilities and outcome frequencies, as described in this CIRANO research summary.

You can practise forecasting without risking money. Before turning an estimate into a position, use the pre-trade reading checklist to check the rules, executable price, liquidity and costs. A forecast is an estimate; a trade adds financial risk.