AI-powered prop firms tighten Polymarket and Kalshi spreads

Proprietary trading firms, quantitative shops and automated AI agents have increased live capital on prediction markets Kalshi and Polymarket, tightening bid-ask spreads and reducing the straightforward arbitrage opportunities once available to casual traders. Activity has concentrated around high-profile macro events such as the Federal Reserve’s July 28–29 rate decision.

Traders are using bonds, currencies, crypto and contracts that settle on central bank announcements to stake probabilities before and after the Fed statement. Economists polled ahead of the meeting expected the Fed to hold rates at 3.50%–3.75%, and market prices put that outcome in the high 80s percent range, leaving a narrow band of uncertainty where price moves still matter.

Market makers, funded-trading firms and automated agents now serve as counterparties that retail traders previously supplied. Infrastructure and brokerage firms have built access layers that connect institutions to these venues. Several investment firms and trading platforms have advertised specialist roles focused on prediction markets. Kalshi reported annualized volume of $178 billion after six months of growth, said institutional activity rose roughly 800% and the platform completed its first customized block trade. Combined monthly volume across Kalshi and Polymarket reached $13.7 billion in June and topped $11 billion in July.

Industry participants and researchers point to event contracts as a cleaner test of forecasting skill because each binary contract resolves to a defined outcome. Louis Régis, founder of on-chain prop firm Propr and a former quantitative trader, argued that the contracts make trader selection more rigorous and described himself as ‘confident about the direction, not the magnitude.’ Benchmarks show that detecting a persistent two-percentage-point forecasting edge requires roughly 350 resolved binary predictions and proving a one-point edge needs about four times more data.

A controlled experiment that let six advanced models trade autonomously on Kalshi and Polymarket for two months found the models lost capital on Kalshi, with declines between 16% and 30.8%, and posted a smaller average loss of 1.1% on Polymarket. A separate working paper on converting forecasts into profit noted that forecast accuracy only turns into expected profit with a proper betting strategy and enough liquidity to execute trades.

Propr uses a staged evaluation to convert signals into funded accounts. Traders and AI agents can qualify for accounts up to $100,000 and scale to $300,000 across multiple accounts after testing. The firm copies a small share of signals onto live venues as funded trades while simulating the rest internally; payouts are settled on-chain in USDC.

Corporate treasuries are also testing event contracts to hedge exposures such as tariffs and regulatory risk. Those hedges require counterparties willing to quote both sides of a market continuously and at size, and they depend on firms that can compare related contracts across venues and correct prices quickly when lines go stale.

Research and industry reports outline possible outcomes: more professional liquidity could tighten spreads and deepen order books, or faster firms could capture mispricings and leave casual traders at a disadvantage. A working paper published in January 2026 found that Polymarket often led Kalshi in price discovery when liquidity and activity were higher. Upcoming macro releases-the Bureau of Economic Analysis advance GDP estimate on July 30 and the July employment report on Aug. 7-will create additional opportunities for traders and models to test execution and forecasting under live conditions.