Who wins and who loses in prediction markets? Evidence from Polymarket

Profit concentration & inequality

• Commenters highlight that top-1%-capture-~75%-of-profits matches broader power-law patterns (OnlyFans, economy, wealth models).
• Yard-sale / Boltzmann-style models are cited as analogies: repeated random exchanges tend to extreme concentration.
• Debate over whether we should deliberately “fight” such power laws via policy; some label that as akin to communism, others argue for progressive/“sigmoid” taxation.
• Several argue real-world payoffs reward capital and decision-making more than “hard work,” reinforcing inequality.

Sources of edge and trading behavior

• The paper’s result that winning traders mostly provide liquidity via favorable limit orders matches many readers’ prior: markets transfer wealth from impatient, less-informed takers to patient makers.
• Some suspect many top accounts are essentially arbitrage or liquidity bots rather than “forecasters from first principles.”
• There’s discussion of cross-venue arbitrage: some see it as a real skill and primary profit source; others say basic API connectivity is easy and real edge is still informational.
• Authors note they have not yet studied cross-venue matching or capital locked in positions; incorporating capital reuse would likely make liquidity providers look even better.

Insider trading and information

• Several think insiders must exist, especially for events under direct human control (press conference length, taped shows, etc.).
• Authors state insiders likely trade but don’t account for a large share of total profits, and are hard to identify because opportunities are one-off and accounts can be rotated.

Market structure, resolution, and comparisons

• Prediction markets are emphasized as zero-sum (before fees), unlike the stock market which many see as having real growth, dividends, and capital formation; others counter that even equity markets are ultimately redistributive within a fixed money supply.
• Polymarket is described as user-vs-user with the platform taking fees, unlike sportsbooks where the house bears risk and bans sharp winners.
• Sports markets appear especially profitable for sophisticated traders, possibly because many users bet with identity/loyalty, pushing prices away from “reality.”
• Some long-horizon markets lack explicit time discounting, which may systematically hurt traders willing to overpay for far-dated outcomes.

Resolution, gray areas, and governance

• Resolution rules can be ambiguous; examples are given where mispronunciation or fine-print criteria led to controversial rulings.
• UMA-based “independent” resolution is viewed by some as largely cosmetic, with accusations that platforms still effectively choose outcomes.
• Questions arise about famous disputed markets and whether corruption in oracles is an overblown concern versus a real structural risk.

Baselines, skill vs luck, and persistence

• Readers ask what profit distribution would look like if everyone bet randomly; authors say many shapes are possible depending on assumptions but their simulations confirm that high concentration is unsurprising even without extreme skill.
• Monthly performance shows weak persistence; some interpret this as sample selection rather than robust trader skill, echoing classic “coin-flipping contest” analogies.

User losses, warnings, and regulation

• Many stress that most participants lose money, comparing prediction markets to lotteries, CFDs, Vegas, and sportsbooks’ “vig.”
• Suggestions include mandatory risk disclosures similar to EU CFD warnings or cigarette labels (“you will lose money on this app”).
• Some worry about broader societal damage if political or economic insiders can profitably manipulate or hedge via these platforms; calls for tighter regulation or bans appear alongside more neutral/curious takes.

Meta: AI-generated comments and community norms

• A substantial subthread debates whether some comments are LLM-generated “slop,” how to detect them (stylistic tells), and whether banning AI-written posts is good policy.
• Some argue AI comments should be quietly downvoted; others defend explicit calling-out and strict enforcement, citing time-wasting and low-quality content concerns.