What the research says about losing money in prediction markets and online betting
Most of what circulates about prediction markets is anecdote: a screenshot of a win, a thread about a whale. Below are the numbers that come from measurement, each with the study it came from and the date we last read it.
Most participants lose
| Finding | Number | Source |
|---|---|---|
| Share of Polymarket (international) wallets with settled positions that lost money | about 84% | On-chain wallet analyses of the public Polymarket book, 2025 to 2026; see the bot-market research in our outline for the specific dashboards. The two U.S. venues publish no account-level data. |
| Our own momentum rule on Coinbase, 30-day replay | 24% winners, −1.86% a trade | We bought the pump candle 1,172 times |
| Our own “final seconds” rule, live test | 0 fills possible in 368 windows | We found free money in our own data |
Why most lose is arithmetic, not character: a two-sided market returns the stake minus fees on average, and the fees on Kalshi and Polymarket US run 3.5% to 4.8% of a 50¢ trade for an ordinary account (Kalshi, Polymarket US). Whoever quotes both sides collects; whoever clicks pays.
What legal online betting did to household finances
The largest study to date followed the credit records of 4.38 million U.S. adults from March 2016 to June 2023 across states that did and did not legalize sports betting. It measured whole-state populations, not just bettors, so the effects below are diluted by everyone who never placed a bet.
| In states that legalized online sports betting | Change within about four years |
|---|---|
| Likelihood of a personal bankruptcy filing | +25% to +30% |
| Debt in collections | +8% |
| Credit card delinquencies | +8% |
| Auto loan delinquencies | +8% |
| Credit scores | about −2% |
| States with in-person betting only | effects very small or not statistically significant |
Source: Brett Hollenbeck (UCLA Anderson), Poet Larsen and Davide Proserpio (USC), “The Financial Consequences of Legalized Sports Gambling,” SSRN working paper 4903302, 2024, using the University of California Credit Panel. The authors suggest the strongest effects fall on people from their late twenties to mid-thirties. The online channel, not betting itself, carried the damage, which is the channel every prediction market uses.
Prediction markets are not sports betting in law, and no study of this size exists for them yet; they are, however, online, always open, and settle in minutes, which are the properties this study found did the harm.
How many people this is
The National Council on Problem Gambling estimates that in any given year about 2.5 million U.S. adults meet the criteria for a severe gambling problem and a further 5 to 8 million have mild or moderate problems. Most never contact anyone about it. If the app has become hard to put down, you are not an outlier; you are in a group the size of a large city, and the line is free: 1-800-GAMBLER, call or text, 24 hours, or help in your state.
What we will add as it is measured
- Account-level outcomes on the U.S. venues, if either venue ever publishes them, or as soon as our own Reality Check extension can show you yours.
- The fee drag per venue per month from our recorder, as a running figure.
- Any peer-reviewed study of prediction-market participants’ finances; none exists at this writing.
Sources read 8 November 2026: UCLA Anderson Review summary of Hollenbeck, Larsen and Proserpio (2024); Front Office Sports, 29 July 2024; National Council on Problem Gambling. Nothing here is medical, legal or financial advice.