They priced rate decisions days early—but political deadlines stayed uncertain until the final hours.
Prediction markets are often described as real-time forecasting machines. But “real time” does not mean the same thing for every type of event.
In a new analysis of aggregated historical prediction-market data, we compared market expectations at three fixed points: seven days, 24 hours and one hour before an event closed. The matched headline sample covered 38 events between March 2024 and May 2026: nine interest-rate decisions, 25 inflation releases and four non-overlapping government-funding deadlines.
Scheduled economic events were largely priced early. Political deadlines still carried substantial uncertainty in the final day.
The central finding was not simply that probabilities became more accurate as an event approached. It was that different kinds of uncertainty appeared to resolve on very different schedules:
Throughout this article, “confidence” means the market probability assigned to the outcome that later occurred. It is a retrospective measure for comparing when information arrived—not a guarantee that a displayed probability was correct or tradable at that price.
- Interest-rate markets were already highly confident a week before the decision.
- Inflation forecasts showed no measurable improvement during the final hour.
- Government-funding deadlines remained much less certain until the final day.
Average probability assigned to the outcome that eventually occurred, measured at three fixed horizons.
rate decisions led by 32.2 percentage points: rates 91.0%, funding 58.8%.
| Horizon | Rate decisions | Government-funding deadlines |
|---|---|---|
| 7 days | 91% | 58.8% |
| 24 hours | 94.5% | 83.2% |
| 1 hour | 94.3% | 99% |
Rate expectations were largely settled a week ahead
Across the nine interest-rate decisions with complete observations at all three horizons, the outcome that eventually occurred was also the top-priced outcome in every case seven days before the decision.
The average probability assigned to the eventual outcome was:
- 91.0% seven days before the decision
- 94.5% 24 hours before the decision
- 94.3% one hour before the decision
Most of the available information therefore appeared to be incorporated before the final day. Average confidence increased by 3.5 percentage points between seven days and 24 hours, then was effectively unchanged during the final 23 hours.
This does not establish that rate markets are always correct. Nine meetings remain a limited sample, and the period contained several decisions for which one outcome was already strongly favored. But within this matched cohort, the final hour added very little information.
Inflation forecasts did not become sharper in the final hour
For inflation, we reconstructed a market-implied median forecast from contracts tied to a range of possible release values. The analysis included 25 releases with prices available at all three horizons: 12 monthly inflation readings and 13 year-over-year readings.
Across all 25 releases, the implied median landed within 0.1 percentage point of the realized interval in:
- 84.0% of cases seven days before release
- 92.0% of cases 24 hours before release
- 92.0% of cases one hour before release
The median absolute error was 0.061 percentage point seven days ahead, 0.065 point 24 hours ahead and 0.067 point one hour ahead. In other words, the final hour did not make the median forecast more precise in this sample.
Monthly inflation was easier for the market than the year-over-year figure. All 12 monthly forecasts landed within 0.1 point of the realized interval at each horizon. For year-over-year inflation, the corresponding rate improved from 69.2% seven days ahead to 84.6% one day ahead, then remained unchanged during the final hour.
Share of market-implied median forecasts landing within 0.1 percentage point of the realized interval.
84% landed within 0.1 percentage point; median absolute error was 0.061 pp.
| Horizon | Within 0.1 percentage point | Median absolute error |
|---|---|---|
| 7 days | 84% | 0.061 pp |
| 24 hours | 92% | 0.065 pp |
| 1 hour | 92% | 0.067 pp |
The largest miss followed a disruption to the normal data calendar
The largest seven-day inflation miss in the sample occurred around the delayed November 2025 year-over-year CPI release.
The market-implied median remained close to 3.0% from seven days before the release through the final hour. The official figure was 2.7%, producing a miss of roughly 0.3 percentage point against the midpoint of the traded outcome interval.
This was not a routine release. The U.S. Bureau of Labor Statistics postponed the November report following a lapse in appropriations, and its technical notes said that October survey data could not be collected retroactively. The episode suggests that market forecasts may be most vulnerable when the underlying information-production process itself has been disrupted. See the BLS release and revised release schedule.
That interpretation is descriptive, not causal. A larger sample of disrupted releases would be required to determine whether unusual data-collection conditions systematically reduce forecast accuracy.
Political deadlines followed a much later information cycle
Government-funding markets presented a contrasting pattern. After removing overlapping formulations of the same deadline, four events had complete, comparable observations at all three horizons.
The average probability assigned to the eventual outcome rose from:
- 58.8% seven days before the deadline
- 83.2% 24 hours before the deadline
- 99.0% one hour before the deadline
The top-priced outcome was correct in three of the four cases seven days ahead and in all four cases at 24 hours and one hour.
The political sample is small, so it should be treated as an exploratory comparison rather than a population-wide estimate. Even so, its timing was notably different from the rate-decision sample. Scheduled macro events were largely priced early; political bargaining produced much larger revisions close to the deadline.
One plausible explanation is that macro expectations are built from continuously arriving public indicators, official communications and analyst models. Political deadlines often depend on negotiations whose decisive information arrives through late statements, procedural votes or last-minute agreements. The data are consistent with that explanation, but do not prove it.
What the pattern means for using prediction-market probabilities
The results suggest that the information value of a market probability depends on both the event and the observation horizon.
For scheduled macroeconomic events, the week-ahead price may already contain most of the available signal. Watching the final hour can create the impression of precision without materially improving the forecast.
For political deadlines, the opposite may be true. A week-ahead probability can still represent genuine uncertainty rather than a settled consensus, and the final 24 hours may contain the most important repricing.
Prediction-market prices should therefore be read as time-sensitive summaries of available information—not as static forecasts and not as guarantees. The useful question is not only “What probability does the market show?” but also “What type of event is this, and when does its uncertainty usually resolve?”
Methodology in brief
From raw contracts to comparable events
Settlement clarity, event volume, hourly price coverage and consistent timing determined which records entered the headline analysis.
- 1,457contracts screened
- 107events passed initial checks
- 38events in matched headline sample
Pariflow Research screened 1,457 historical prediction-market contracts from publicly accessible market-data interfaces. The study did not use individual trader, wallet or account data.
Contracts were filtered for settlement clarity, minimum reported event volume, hourly price availability and consistent event timing. The final headline analysis used only events with valid observations at all three horizons. Overlapping formulations of the same government-funding deadline were deduplicated.
For mutually exclusive rate outcomes, contract probabilities were normalized at the event level. For inflation releases, threshold probabilities were fitted to a monotonic curve and used to estimate the market-implied median. Results were compared with the final resolved outcome or outcome interval.
Main limitations
- The nine rate decisions and four funding deadlines are small samples.
- The estimates are descriptive; the study does not report confidence intervals or hypothesis tests.
- Requiring complete observations at all three horizons can favor better-covered, more liquid contracts and introduce selection bias.
- Contract availability differs by event type and time period.
- Prices can reflect transaction costs, spread and liquidity constraints as well as beliefs.
- Inflation outcomes are reconstructed as intervals; midpoint error is an approximation rather than the exact official print.
- This is a descriptive observational study. It does not establish why a probability changed or prove that one event family is inherently more forecastable.
The underlying records were aggregated before reporting. No venue-level ranking was performed, and Pariflow does not claim ownership of the underlying market data. The published chart aggregates are available as a machine-readable CSV. A detailed source inventory and extraction log can be made available to editors for verification.
This research is informational and does not constitute financial, trading or investment advice.
Artem Goryushin
Fintech expert, business analyst
Artem is a fintech expert and business analyst with experience in prediction markets and financial analytics.