Prediction markets as bets on events: what they can and cannot do

In depth12 min read · 16 sources

Prediction markets are exchanges for bets on events: a contract pays $1 if, say, a candidate wins the election, and nothing otherwise. Its price can be read as the probability the market assigns to the event. This works well surprisingly often, but not always: fees, interest, thin markets, large bets and the preference for long shots distort prices. Here you will learn how these markets work, what research says about their accuracy, how they are regulated and where their limits lie. No advertising for platforms and no investment advice.

How a prediction market works

The basic building block is the event contract with two outcomes: Yes pays $1 if the event occurs, No pays $1 if it does not. A pair of Yes and No therefore always pays exactly $1. That is why their prices add up to roughly $1. If Yes costs $0.62, the market is saying, roughly: 62% probability. Economically, this is a digital option on an event rather than on a price. Besides such all-or-nothing contracts, there are contracts that pay out a share, such as a party's share of the vote. Depending on the contract, according to Wolfers and Zitzewitz (2004), you can read off probabilities, means or medians of the market's expectation.

Trading takes place in two ways. In an order book, buy and sell orders meet as on a stock exchange; if there are few participants, the bid-ask spread is wide. An automated market maker, by contrast, always quotes a price. A well-known one is the logarithmic market scoring rule (LMSR), which goes back to the market scoring rules of Hanson (2003): a cost function determines what each additional unit costs, the price rises with the quantity bought, and the operator can lose at most an amount known in advance.

Types of markets

• Elections and politics: the Iowa Electronic Markets of the University of Iowa trade contracts on US elections; researchers have evaluated them for the presidential elections since 1988 (Berg et al. 2008). As early as between 1868 and 1940, there were large, often well-organised markets in the US for bets on presidential elections (Rhode and Strumpf 2004).
• Economic data and central banks: contracts on interest rate decisions, inflation or labour market figures. Similar expectations are embedded in interest rate futures, options and the yield curve, with far more money behind them.
• Sport: economically close to sports betting and, in the US, the biggest point of contention between platforms and states (see below).
• Crypto platforms and regulated exchanges: some platforms run on blockchains and digital tokens outside traditional exchange supervision, while others are registered in the US as exchanges with the futures market regulator, the CFTC. In 2022, the CFTC imposed a $1.4 million penalty on the operator of the crypto platform Polymarket for offering event-based binary options without registration, and required it to wind down non-compliant markets (CFTC 2022).

Regulation: the US, Germany and the EU

US. The CFTC treats event contracts as swaps that may only be offered on registered trading venues (CFTC 2022). How far its jurisdiction extends and what the states may prohibit is currently being decided by the courts:

• Election contracts: in September 2023, the CFTC prohibited the exchange Kalshi from offering contracts on which party will control Congress. It deemed them unlawful, a form of gaming and contrary to the public interest. The federal district court in Washington, D.C. overturned this on 6 September 2024: the contracts concerned elections, which are neither unlawful nor gaming. The appeals court denied a stay on 2 October 2024 (Davis Wright Tremaine 2024); on 5 May 2025, the CFTC moved to withdraw its appeal (Finance Magnates 2025).
• Sports contracts: several states regard sports contracts as sports betting under their gambling laws. On 4 April 2026, the federal Court of Appeals for the Third Circuit ruled 2 to 1 in favour of Kalshi against New Jersey: sports contracts on an exchange registered with the CFTC are swaps, and federal law pre-empts the state's gambling law (Skadden 2026). On 28 August 2026, the Ninth Circuit ruled against Kalshi in the Nevada case, and in September 2026 the Sixth Circuit ruled in favour of Ohio and Tennessee: Kalshi had not shown that the contracts were swaps, and even if they were, federal law would not pre-empt the gambling laws. On 2 September 2026, New Jersey asked the Supreme Court to review the case; proceedings in the Fourth Circuit were still pending at the end of September (The Block 2026).
• As of October 2026: the courts are split, and the CFTC claims exclusive jurisdiction. The outcome is open.

Germany and the EU. To our knowledge, there is no specific regulation for prediction markets here. Roughly speaking: bets fall under gambling law, in Germany under the federal states' Interstate Treaty on Gambling (Glücksspielstaatsvertrag); derivatives such as options fall under financial market law and BaFin supervision. Whether an event contract is a bet or a financial instrument depends on how it is structured and is a legal question to be decided case by case. Our assessment: before any trade, check whether the provider is licensed in Germany. Without a licence, client money and legal protection are uncertain. This is not legal advice.

What research shows: strengths

Accuracy. According to Wolfers and Zitzewitz (2004), the forecasts of such markets are usually fairly accurate and beat most moderately sophisticated benchmark forecasts. Berg et al. (2008) compared the Iowa Electronic Markets in the US presidential elections from 1988 to 2004 with 964 polls:

For forecasts more than 100 days before the election, the market was significantly better in every election (Berg et al. 2008). Even before the era of scientific polling, the betting markets of 1868 to 1940 predicted elections remarkably well and were fairly efficient, despite participants' limited information and active attempts to manipulate the odds (Rhode and Strumpf 2004).

Aggregating information. The idea behind it: many participants have dispersed information, and anyone who puts money on the line has an incentive to trade on their true assessment. That is the efficient market hypothesis in miniature. In Science in 2008, 22 researchers led by Kenneth Arrow called for prediction markets to be freed from unnecessary government restrictions as a research tool (Arrow et al. 2008).

Weaknesses and criticism

• Reading polls correctly: Erikson and Wlezien (2008) counter that market prices predict the result on election day, whereas a poll measures sentiment on the day of the survey. The two cannot simply be compared. If poll leads are properly discounted, forecasts from polls beat the vote share markets of the Iowa Electronic Markets from 1988 to 2004, and winning probabilities derived from polls beat the all-or-nothing markets. Traders saw more uncertainty in the campaign than the polls justified.
• Favourite-longshot bias: in horse racing, long shots are regularly overvalued and favourites undervalued (favourite-longshot bias). Snowberg and Wolfers (2010) attribute this to misperceived probabilities, as described by the prospect theory of Kahneman and Tversky (1979), not to a love of risk. A contract at $0.03 may then stand for a probability below 3%.
• Long horizons: according to Page and Clemen (2013), prediction markets are fairly well calibrated when little time remains, but clearly biased for events far in the future: likely events are too cheap, unlikely ones too expensive. Taking the time value of money into account, this can only be exploited with a low discount rate.
• Thin markets and large bets: where little money is at stake, a single person moves the price, as the LMSR example shows. In October 2024, a French trader bet so much on Donald Trump via four accounts that potential payouts of around $46 million accumulated; the platform gave Trump 62% at the time, while polls showed a neck-and-neck race. After its own investigation, the platform found no evidence of manipulation and said it had been a bet based on personal conviction (Reuters 2024). Whether a price reflects information or financial muscle is hard to tell from the outside.
• Insider trading on events: anyone who knows or influences the outcome (a team member, an official, a decision-maker in a company) can bet on it. For the market, this cuts both ways: the price becomes more accurate, but the other participants lose to the informed traders, as in the model of Kyle (1985). More on the ban in securities markets under insider trading.
• Wash trading and volume: in wash trading, the same party trades with itself. This inflates turnover without any opinions actually meeting. Our assessment: especially on lightly supervised platforms, volume figures should be read with caution; high turnover is no proof of accuracy.
• Gambling addiction and consumer protection: sports contracts resemble sports betting. In the Sixth Circuit, the court confronted Kalshi with its own earlier concession that its sports contracts had no inherent economic significance (The Block 2026). Anyone trading there is, as a rule, not hedging a risk but betting, with all the dangers that gambling brings. Before fees, the whole thing is a zero-sum game; after fees, participants lose on average.

Worked example: price, fee and interest

A price is only a probability if buying costs nothing and money tied up earns no interest. Neither is true.

For advanced readers: when is a price a probability?

A risk-neutral trader with no costs and no interest buys as long as their probability q is above the price P. With a fee f and interest rate r over the term T, buying only pays off from:

For buying No at the price 1 − P, the mirror image applies: q < 1 − (1 − P + f) × (1 + r)^T. In between lies a band in which nobody trades. With P = 0.62, f = 0.01, r = 5% and T = 1, it ranges from 59.05% to 66.15%. The price then only reveals that the marginal traders lie roughly within this band, not the exact probability.

Three further effects come on top. Risk aversion: anyone betting a lot demands a premium; as with options, prices are then closer to a risk-neutral valuation than to the true probability. Limited capital: traders with a strong opinion can only bet as much as they have. Perception: small probabilities are overestimated (Snowberg and Wolfers 2010). As an assessment, this means: if you use prices as probabilities, stick to short horizons and liquid markets, and read extremes below 10% or above 90% with caution.

For investors, what markets expect about economic data and interest rates is often more important than election bets. Interest rate futures and options also contain such expectations, likewise as risk-neutral prices rather than pure probabilities (option pricing).

Limitations and common mistakes

• Price equals truth: a price of 62% does not mean that the event will occur, but that it should fail to occur in about 38 out of 100 such cases. Calibration can only be checked across many events.
• One outcome as proof: if the market was right once and the polls were wrong, that proves little. Only many forecasts show which is better calibrated.
• High turnover as a seal of quality: volume says nothing about accuracy and can be inflated by wash trading.
• Prediction markets as an investment: contracts do not earn a risk premium like equities. Anyone who takes part is betting. Attention-driven trading and herd behaviour, as in speculative bubbles, are possible here too.
• Overlooking the legal situation: what is permitted in the US may be illegal gambling in Germany, and vice versa.
• Forgetting other sources: polls, models and alternative data often carry the same information. A prediction market is one source among several.

How Portfolio Lab calculates it

To be honest: it does not. Portfolio Lab has no data from prediction markets and no tool for calculating with or trading event contracts. The closest thing is the Option pricer: it also values digital options. A cash-or-nothing call pays a fixed amount if the price at expiry is above the strike, and costs amount × e^(−rT) × N(d₂) (option pricer methodology). N(d₂) is the risk-neutral probability that the option ends in the money, calculated with the Black-Scholes model and the volatility entered. It is not a probability derived from bets. As with prediction markets: a price is discounted and contains risk premia; as a probability, it is only an approximation. Probabilities priced into the yield curve and options are shown on the Market expectations tab of Macro (see market expectations).

This article explains a method; it is not investment or tax advice. Worked examples are simplified and rounded. All terms in the glossary.