Prediction Markets for Public Policy: Should Governments Experiment?
You sit in a policy room on a hard Monday. A storm is coming. The team must choose: act fast and risk error, or wait and risk harm. A hand goes up. “What if we ask a market?” The room is quiet. Not a stock market. A small, focused market that prices a public outcome. A price that moves when new facts land. A signal you can read by noon.
This piece asks a plain question: should governments try prediction markets? Not for sport. Not for hype. For better signals, sooner. Below is a clear guide: what they are, where they fit, when they fail, and how to run a safe pilot. The tone is blunt. The steps are concrete. No magic, no wishful thinking.
What Are Prediction Markets (Really)?
A prediction market lets people buy and sell small claims on a yes/no outcome or a number. The price shows the crowd’s view of odds, in real time. If the outcome happens, the claim pays out. If not, it does not. People have skin in the game, so they tend to share and seek better info.
How is this not a poll? A poll asks what you think. A market asks what you will stake. How is this not an expert panel? A panel talks. A market forces a price that reflects all views at once. It does not need a long meeting to update. It can move in minutes when a new report drops.
Do they work? There is good evidence on prediction market accuracy in practice across many domains. They tend to beat or match polls and simple models when there is fresh, diverse info and clear rules.
Limits do exist. Some people may try to push prices to shape a story. Thin markets can be noisy. Public policy has legal and ethics rules that markets must follow. Data leaks and conflicts can harm trust. A pilot must plan for these from day one.
A Short, Messy History: Wins, Walkbacks, and Near-Misses
One early case was the Iowa Electronic Markets. It ran small, research-grade markets on elections. It showed that even with low stakes, prices can track reality well. It also showed the need for clear caps and a research frame.
In the early 2000s, a defense unit backed a market for policy signals. Critics framed it as “betting on bad events.” The project died fast. The lesson: optics matter as much as stats. A good idea can fail if the story is wrong.
Later, the intel research arm ran the IARPA’s Aggregative Contingent Estimation program. It did not run a cash market but tested ways to combine human forecasts. Results were strong, and led to better methods to blend signals from many people.
In New Zealand, a university-linked market faced new rules and then closed. See the regulator notice on New Zealand’s iPredict closure. In the U.S., some event markets have also faced stops. The pattern is clear: the law is in flux, and risk control is key.
Are They Any Good? What the Evidence Says
Across cases, markets are best when the question is clear, the event is near-term, and traders have varied info. They can be very fast to update, and they can beat simple polls. They are not magic, but they are often a strong signal to add to your mix.
On human forecasting, we have data from long runs. See Good Judgment research on forecasting accuracy. With training and scoring, groups can reach high, steady skill. That means you can blend a market with a trained crowd and gain both speed and depth.
Design and liquidity matter. A dead market is a bad signal. Clear rules, small but real stakes, and market makers help. See how one platform tracks this in Metaculus accuracy and scoring. You want incentives that reward true, not loud.
Where do markets do poorly? When a policy body controls the outcome or the data, prices can be biased. When the world is calm and slow, a simple model or baseline may be enough. When stakes are huge or sensitive, the optics risk may outweigh any gain.
The Government’s Real Question: Where Do Markets Fit in the Policy Toolkit?
Governments already use many tools. There are public polls, expert panels, superforecaster pools, impact studies, and foresight work. See broad OECD guidance on regulatory policy evaluation for a map of good practice. Each tool has a place. The choice is about task, time, and risk.
Use a market when you need a live signal on a near-term, external event. Use a forecast pool when you need reasoned, source-linked views and clear track records. Use an experiment when you must prove cause and effect. For impact study basics, see the World Bank’s Impact Evaluation in Practice.
Where Prediction Markets Fit Among Policy Tools
| Prediction Markets | Near-term, well-defined events with diverse info | Hours to days | High when liquid [1] | Low–Medium | Medium–High | Medium–High (jurisdiction-specific) | Thin liquidity, manipulation, bad framing | — |
| Polling | Public mood and stated intent | Days to weeks | Medium; often lagging | Medium | Low–Medium | Low | Wording effects, nonresponse bias | When you need representative attitudes |
| Expert Panel | Deep domain synthesis | Weeks | Medium; variable | Medium | Low–Medium | Low | Groupthink, anchoring | When tacit knowledge is key |
| Superforecasting Tournament | Trackable questions with rationales | Days to weeks | High with training [2] | Medium | Low–Medium | Low–Medium | Overfitting to scoring rules | When you need auditable reasoning |
| RCT / Impact Evaluation | Proving cause and effect | Months to years | High (internal validity) | High | Low–Medium | Medium | Scope creep, attrition | When decisions hinge on causal proof [3] |
| Strategic Foresight | Long-term scenarios and options | Weeks to months | Not an accuracy tool; frames choices | Medium | Low | Low | Vague outputs if not scoped | When exploring futures, not odds |
[1] See NBER survey on market accuracy. [2] See Good Judgment’s research pages. [3] See the World Bank guide on impact evaluation.
Blueprint for a Small, Safe Pilot
Start narrow. Pick one policy-relevant, near-term, external event. Example: “Will at least 85% of service tickets be resolved within 48 hours in Q4?” That ties to a real KPI and a date. Avoid events your team can directly control without clear rules.
Keep stakes small and rules sharp. Cap positions. Use identity checks. Publish how you will handle new data, late events, or changes. Use a market maker or liquidity subsidy so the price can move. Log all rules up front.
Set guardrails on who can trade. Exclude staff with direct control of the outcome. Ask legal and comms to review the plan. For methods and horizon scanning, government teams can borrow ideas from the UK Government’s Futures Toolkit for policymakers.
Define success before launch. For example:
- Lead time gained vs. your current signal (days).
- Mean Brier score vs. benchmark forecasts.
- Number of real decisions changed by the market signal.
- Stakeholder trust score in the post-pilot survey.
Plan a clean stop date. Publish a plain-language after-action note.
Ethics, Optics, and Public Trust
Words shape how people see your pilot. Avoid “betting.” Use “forecast” and “information market.” Be clear that the aim is better public outcomes, not profit. State that any real stakes are small and set for signal quality, not gain.
Be open by default. Share the question text, the rules, and the audit trail. Publish the price history and the final score. Invite outside review. Small acts of candor beat glossy slides.
Protect against manipulation. Ban conflicts. Monitor for wash trades. Freeze or void a market if rules break. Keep a short and public incident response plan.
What Would Change My Mind
I would shift my view if, in careful tests, markets trailed trained forecasters and simple models by a wide margin across tasks. I would also shift if the cost in legal work, optics, and time outweighed any gain in signal speed and quality. If that happens, park the tool and use other methods.
Sidebar: How This Differs from Gambling (and Why It Matters)
A prediction market is an information tool. It is built to get a better signal, not to entertain. Design choices force clear rules, low limits, and a public goal. Gambling sites aim to entertain and often set house edges. That is a different aim and risk profile. This line must be clear in your comms.
If you or your readers ever touch real-money platforms of any kind, first check licenses, limits, and consumer care. Independent reviews can help you scan these basics fast. One neutral, practical starting point is this short guide to blackjack online casinos. Read the fine print. Know your local laws. This is not financial advice.
For a deeper theory angle, read Robin Hanson’s foundational proposal on futarchy—a bold idea to “vote on values, bet on beliefs.” You do not need to endorse it to learn from its logic on incentives and information.
Regulatory Landscape: A Moving Target
In the U.S., the key body is the CFTC. It explains how it views event contracts here: CFTC on understanding event contracts. The note shows that law treats some events like commodities under certain terms. Terms and scope matter a lot.
There have been actions against unregistered, event-based venues. Read one case in this CFTC enforcement action on unregistered event-based markets. For a public pilot, you want clear falls within research, sandbox, or other safe paths. Get legal advice before you start.
Outside the U.S., rules differ. Some countries allow small, research-grade markets. Others do not. In all cases, you need counsel and a written compliance plan. Err on the side of narrow scope, open data, and low stakes.
Mini-Case Design Sketch: Two Pilots Governments Could Run This Year
Pilot 1: Digital service uptime. Question: “Will the citizen portal sustain 99.5% uptime next month?” Traders: vetted IT staff from other agencies, vendors without control of this portal, and external SRE experts. Use a liquidity subsidy for the first week. Link the price to a risk dashboard. If the price falls below 70%, trigger a review and a pre-planned fix window.
Pilot 2: Supply risk. Question: “Will hospital X see a stock-out of drug Y in Q3?” Traders: buyers from nearby systems, logistics analysts, and vetted suppliers with no direct control. Combine the market with a forecast pool that posts short rationales. If the price crosses 60% odds, place a safety order. For wider horizon work, browse the European Commission’s strategic foresight resources to frame backup plans.
For both pilots, write a one-page “if-then” plan: if the price crosses a level, then leaders do X. This turns a signal into action. It also lets you judge the value of lead time in plain terms.
FAQ: Practicalities People Actually Ask
- Is it legal? It depends on where you are and how you design it. Some paths exist for research-grade pilots. Talk to counsel early. See the CFTC links above for the U.S.
- Can insiders game it? If someone can change the outcome, set strict rules or exclude them. Log all trades. Set low caps. Pause the market if rules break.
- What if the market shows an “awkward” price? Publish and explain. Say what will change the price. Use it as a prompt to seek new info, not as a verdict.
- What does it cost? Small pilots can be cheap in cash, but not in care. The main costs are design, legal review, and good comms.
- Should civil servants trade? Often no, if they can shape the outcome. If they cannot, set a code of conduct and disclosure rules.
- How do we judge success? Compare forecast scores to your benchmark, count decisions changed, and measure lead time gained. Run a short post-mortem and publish it.
- How do we close a market? State the end rule before launch. On close, post the source for the outcome, the final price, and the score.
Bottom Line and Next Steps
Should governments try prediction markets? Yes—carefully. Use them for tight, near-term, policy-relevant questions. Keep stakes small, rules clear, and audits open. Blend them with forecast pools and your current risk tools. Do one or two clean pilots, learn, and then decide how to scale or stop.
Next steps:
- Pick one narrow, external event tied to a real decision.
- Draft rules, caps, and a data/audit plan.
- Pre-brief legal and comms. Plan your public story.
- Set success metrics. Define a stop rule. Publish the results.
If you do this, you will know more, sooner—at a fair cost and with less drama.
How We Sourced This
This article draws on public research and regulator notes checked in May 2026. Key sources include NBER reviews of market accuracy, Good Judgment studies, platform accuracy pages, and policy guides from OECD, the World Bank, and UK government. See links in the text. Laws and rules change; confirm details with your legal team.
Sources and Further Reading
- Wolfers & Zitzewitz on prediction market accuracy (NBER)
- Iowa Electronic Markets (University of Iowa)
- IARPA ACE program
- FMA NZ on iPredict closure
- Good Judgment research
- Metaculus accuracy and scoring
- OECD regulatory policy resources
- World Bank: Impact Evaluation in Practice
- UK Government Futures Toolkit
- Hanson: Futarchy
- CFTC on event contracts
- CFTC enforcement press release
- European Commission foresight resources
Disclaimers: For information only. Not legal, financial, or investment advice. Laws differ by place and change over time. Always consult qualified counsel before any pilot. If you follow links to third-party sites, review their terms and privacy policies.
Last updated: May 2026