I Put $1,000 Into Prediction Markets for 30 Days: Polymarket vs Kalshi Fees, Resolution Fights, and What I Actually Earned

On September 1, 2026, I moved $1,000 into prediction markets: $500 to Polymarket international (USDC I already held on Polygon) and $500 to Kalshi via ACH. On October 1 I withdrew $939 — down $61, or 6.1%. Few trades have taught me more per dollar.

This is the diary: exact fee formulas at nine price points, the 15-minute Bitcoin markets that ate my first week, a $300M resolution fight I watched live, a failed arbitrage, and a maker book that paid rebates while losing money on adverse fills. Every number carries a source and date; none of it is advice — you'll see why I refuse to call this yield.

The setup: $1,000 across two rails that look nothing alike

Kalshi is a CFTC-regulated designated contract market (since November 2020); customer dollars sit at JPMorgan and BNY Mellon, and FDIC covers those banks, not Kalshi. Polymarket international runs on Polygon PoS; the April 28, 2026 V2 cutover switched collateral from USDC.e to pUSD, so my USDC needed one swap. Gas cost cents — covered in our gas fees vs yield guide — and my keys never left my wallet, per the wallet security checklist.

Day one's lesson came before any trade: debit deposits on Kalshi cost 2%, so I waited two days for free ACH and saved $10. pUSD is young collateral, so I sized it like the stablecoin depeg risk guide suggests.

For the full breakdown, see our Polymarket vs Kalshi comparison, the Polymarket and Kalshi reviews, and the best prediction markets page.

How big is this thing, really? Polymarket has cleared $33.5B lifetime volume across 3.07M traders (Polydata, Oct 1–6, 2026), but the post-World Cup hangover shows: September volume ran $734.8M single-sided versus June's $5.31B peak, and monthly actives fell from 750k+ to roughly 283,300 (Token Terminal, via Decrypt, Oct 6, 2026). I cite single-sided volume where I can, because nominal or "matched" figures double-count each trade. Kalshi out-traded Polymarket for the first time in April 2026 ($14.81B vs $9.01B, Galaxy/CoinShares); by September the US split ran Kalshi 80%, Polymarket US 12%, Polymarket international 6%.

The fee table that decides every trade

Kalshi charges takers 0.07 × contracts × π(1−π), rounded up to the cent; Polymarket international charges shares × feeRate × p(1−p) with 2026 coefficients of 0.07 crypto, 0.05 sports, 0.04 politics/finance/tech, and 0 geopolitics. Polymarket US charges a flat 10-basis-point taker fee, $0.001 minimum, from July 1, 2026. Since p(1−p) peaks at 50¢, coin-flip markets are the most expensive place to express a view — the curve bends exactly like the utilization kink in lending rate models.

Per 100 contracts at 50¢: $1.75 on Kalshi or Polymarket crypto, $1.25 sports, $1.00 politics, $0.05 Polymarket US — identical curves, free geopolitics, a 3.5%-of-notional toll per side. The takeaway: before you bet a coin flip, remember you are paying the maximum fee to do it.

Thirty days on the books: fees, a $300M resolution fight, and the bot race

Week one went to 15-minute Bitcoin and Solana price markets. A Stanford/SMU paper (June 30, 2026, arXiv 2606.31675) documents spot order-flow spikes in the final 10 seconds of 5-minute BTC contracts, largely absent in 15-minute markets. I picked the 15s; it didn't save me — spread, the 0.07 crypto taker fee, one-tick-late reversals — and the week bled out in small increments. Five- and 15-minute crypto markets settle on Chainlink, not UMA: no resolution drama, just friction — the legal cousin of oracle manipulation attacks and MEV sandwich attacks.

Week two I slowed down: the 2026 midterms boards, a Fed-cut ladder, and the CLARITY Act passage market that sat around 34% on July 29. Best trade of the month: Senate GOP at 62¢ after a filing-cycle shift, sold into 71¢ for +$18.50; the Fed ladder lost $9.75 on a hot payrolls print. NBA Finals and Nvidia earnings markets I left alone — a violent September weekend reinforced the crash weekend playbook rule: don't chase moves you can't explain.

The week that changed my risk model wasn't about my trades. The Strategy (formerly MicroStrategy) Bitcoin-sale market asked whether the company would sell BTC between May 26 and 31, 2026. It sold 32 BTC in that window (roughly $2.5M at about $77,135 average), but the disclosing 8-K went public June 1 — after the deadline. YES spiked from around 10% toward 80%; a trader, willo2, bought roughly 700,000 YES at about 76¢ (~$500,000). Polymarket posted "additional context" requiring confirmation within the window; after two disputes it resolved NO a third time. Volume topped $300M (Galaxy).

The event happened inside the window, yet the market resolved NO — defensible under the rules, brutal against the news. UMA's optimistic oracle: a proposer posts a roughly $750-bonded answer, challengers have about two hours, disputes go to a token-holder vote — roughly 24–48 hours of debate plus about 48 hours of voting, so 4–6 days total. UMA reports 99.8% of requests resolve undisputed. Prior fights: the Zelenskyy suit market (about $237M, 2025) and the April 2025 Ukraine minerals market resolved YES with no signed deal, aided by whale concentration — roughly 25% of votes from three accounts, two wallets holding over 50% of UMA voting power. Polymarket refused refunds. My fix: I cut every news-based position to a size I could lose entirely.

Then I chased the "hidden yield" as a maker. Polymarket pays maker rebates — 20% of taker fees on crypto, 15% sports, 25% politics — plus a Taker Rebate Program (live May 28, 2026) returning 3–50% of taker fees by 30-day volume tier. LP rewards cover only about 5,700 of roughly 50,000 live markets; cumulative incentives run near $128M — 54.3% of all trading fees (Odaily, Sep 30, 2026). A CFTC advisory on August 12, 2026 warned rewards drive wash trading; Kalshi terminated its volume-incentive program in October 2026 after ETH perps wash trades. It's the shape Hyperliquid points farmers and airdrop farming readers know.

So I ran a small maker book: weekly BTC-above-X, about $250 at risk. I earned $9.40 in rebates and got filled adversely by more than that whenever BTC moved — net −$11.60. Same lesson as real yield vs emissions: incentives aren't income. Idle cash earns 0% on both platforms.

Last test: cross-venue arbitrage. An IMDEA study of 86 million trades found about $40M per year extracted from Polymarket, almost all by bots, windows shrinking from roughly 5 minutes in 2024 to about 30 seconds in 2026. Retail edges of 0.8–6% exist before the killer risk: one leg fills, the other doesn't. Mid-month, a Fed-cut contract quoted $0.62 on Polymarket versus $0.56 on Kalshi — the June 2026 shape. My Kalshi leg cleared after the gap had converged; thirty seconds isn't enough for a human with two browser tabs. It's delta-neutral yield farming as a race against colocation — perpetual funding rate math covers the same game.

Fees bleed silently — every 15-minute taker fill paid the 0.07 curve — and one stale quote erased three careful weeks: late in week four I sold a weekly BTC contract against a quote that moved first (−$28.90, $4.85 of it fees). Net: −$61.00 on $1,000, or −6.1%, with $29.30 of taker fees — nearly half the loss.

Where the "yield" actually goes

Galaxy's data: 69.2% of retail accounts lose, and in a $13.76B-volume study, 3% of accounts captured roughly 27% of profits. The advertised "hidden yield" is real and almost entirely captured by bots; humans get the entertainment, the resolution drama, and the tax confusion — the trap pattern from our yield traps red flags checklist.

My verdict: prediction markets are trading and entertainment, not yield. There is no APY (see APY vs APR), idle cash pays 0%, and the only durable edge requires becoming the bot. I keep small balances for event hedging — a Fed ladder is a useful macro expression. For actual yield, price it with our yield calculator, screen venues with the yield risk grader, read the DeFi yield reality check, and track rates in the stablecoin APY tracker. If you want stable, audited income instead of event bets, compare Ethena vs Sky, read the Ethena and Sky reviews, and follow the base yield guide.

Taxes, legality, and whether this belongs in your portfolio

Kalshi issues 1099s; Polymarket international issues nothing — but no form doesn't mean no tax. The IRS has zero event-contract guidance, leaving three treatments: capital gains (most common), Section 1256 60/40 (uncertain), and gambling income (worst — the OBBBA caps gambling loss deductions at 90% of winnings in 2026). I logged every fill with timestamp and venue — the discipline our DeFi taxes guide covers; reconstructing a Polygon history in April is not a plan.

Polymarket international paid a $1.4M CFTC settlement in 2022 and geo-blocks US IPs; VPN use violates its terms. Kalshi covers every state except Michigan and Nevada (Sep 2026) — but lost a Utah ruling August 5, 2026, the New York AG is suing, and fights continue in NV, MA, NJ, MD, CT, TN, and AZ; Minnesota made unlicensed event betting a felony from August 1, 2026. Check your state the day you deposit.

Does this belong in your portfolio? Only as a tiny, deliberate slice — sized with our portfolio allocator and reviewed in a monthly portfolio review, never as core yield. Event contracts are not lending, not staking, and not the stablecoin yield rotation under the Fed you can actually plan around. The same risk discipline applies across new DeFi chains for yield and the DeFi yield portfolio allocation guide: if it can't survive a dispute, a stale quote, or a bot, it is entertainment, not allocation. For insurance-style protection on the rest of your book, the year with Nexus Mutual shows what real coverage looks like; for boring core income, the best stablecoin yield guide and yield discovery tool are where I park the rest.

Sources and further reading

• Polymarket Docs — platform rules and market structure.
• Polymarket API docs — fees FAQ — the p(1−p) fee formula and category coefficients.
• UMA — how the optimistic oracle works — the 2-hour challenge window and 4–6 day dispute path.
• Galaxy Research — the Strategy Bitcoin-sale resolution dispute (May–June 2026) — the flagship $300M+ case.
• DeFiLlama — Polymarket — volume history and TVL.
• Stanford/SMU paper (arXiv 2606.31675) — order-flow evidence in short-duration BTC markets.

Frequently asked questions

What fees does Polymarket charge in 2026?

Polymarket international charges makers nothing and takers a formula fee: shares × feeRate × p(1−p), where p is the share price. The 2026 coefficients are 0.07 crypto, 0.05 sports, 0.04 politics and finance, and 0 geopolitics, so a 100-share taker fill at 50¢ costs $1.75 in crypto, $1.25 in sports, and $1.00 in politics. Polymarket US (QCX) instead charges a flat 10-basis-point taker fee with a $0.001 minimum, effective July 1, 2026.

Is Polymarket legal for US users?

Polymarket international is not open to US users: after a $1.4M CFTC settlement in 2022 the site geo-blocks US IPs, and using a VPN to trade violates its terms. US residents have two compliant routes: Polymarket US, launched through the $112M QCEX acquisition and live since December 2, 2025, and Kalshi, a CFTC-regulated exchange available in every state except Michigan and Nevada as of September 2026 — though Utah ruled against Kalshi on August 5, 2026 and state-level fights continue.

How does a Polymarket resolution dispute work?

Markets resolve through UMA's optimistic oracle. A proposer posts an answer with a bond of roughly $750, and challengers have about two hours to dispute. A dispute escalates to UMA's token-holder vote — roughly 24–48 hours of debate plus about 48 hours of voting, so 4–6 days total. UMA reports 99.8% of requests resolve undisputed, but contested ones can resolve against the obvious news: the June 2026 Strategy Bitcoin-sale market resolved NO a third time after two disputes. Fast 5- and 15-minute crypto markets settle on Chainlink instead.

Can you arbitrage the same market between Polymarket and Kalshi?

Sometimes, but rarely profitably by hand. An IMDEA study of 86 million trades found about $40M per year extracted from Polymarket, almost entirely by bots, and arbitrage windows shrank from roughly five minutes in 2024 to about 30 seconds by 2026. Realistic gross edges run 0.8–6% before the killer risk: one leg fills and the other doesn't. My Fed-cut attempt — YES at $0.62 on Polymarket versus $0.56 on Kalshi, the same shape as a June 2026 example — closed the gap before my second leg filled.

Do you pay taxes on prediction market winnings?

In the US, yes — but the rules are genuinely unsettled. Kalshi issues 1099 forms; Polymarket international issues nothing, yet the income is still reportable. The IRS has published no event-contract guidance, leaving three possible treatments: capital gains (the most common approach), Section 1256 60/40 (uncertain), and gambling income (the worst — 2026's OBBBA caps gambling loss deductions at 90% of winnings). Track every fill yourself with timestamps and prices; reconstructing a Polygon trade history in April is misery.

Related: the venue deep dives — Polymarket, Kalshi, and our Polymarket vs Kalshi comparison; the tax side in DeFi taxes 2026; the boring-dollar alternative in stablecoin yield rotation; and risk discipline in DeFi yield portfolio allocation.