Prediction Markets: The Arbitrage & Value Betting Playbook

What Are Prediction Markets?

A prediction market is an exchange where traders buy and sell binary contracts tied to real-world outcomes. Each contract resolves to either $1.00 (or 100 cents) if the event happens, or $0.00 if it doesn't. Because payouts are binary, the price you pay is a direct implied probability. A contract trading at 63 cents means the market thinks there's a 63% chance the event happens. That's it. No odds conversion, no overround, no fiddling with American vs decimal formats.

Two venues dominate the space in 2026. Kalshi is a CFTC-regulated US exchange priced in cents (0 to 100), with a clean REST and WebSocket API, RSA-authenticated automated trading, and fees that scale by volume tier (typically 3% to 7% on winnings). Polymarket runs a Central Limit Order Book (CLOB) on Polygon L2 with prices in dollars (0.00 to 1.00), settles trades in USDC, and charges no exchange fees (traders pay Polygon gas instead). Polymarket is larger by volume and covers thousands of markets; Kalshi is the friendlier venue for US-based traders.

The space used to be bigger. PredictIt operated under a CFTC no-action letter for years and was a favorite of academics, but it wound down its US political markets in 2023 with residual markets still settling through 2024. Manifold Markets runs on play money and is useful as a forecasting sentiment signal, not an execution venue. For the mechanics of how the cross-venue arbitrage math works and why prediction markets beat sportsbooks at scale, our prediction market arbitrage basics post covers the fundamentals with worked examples.

What makes prediction markets structurally different from everything else in the retail trader stack is the resolution mechanism. A Kalshi contract on "Will the Fed raise rates by 50bps in March?" resolves based on a predefined data source at a predefined time. There's no bookmaker judgment, no grading ambiguity, and no post-event dispute process (usually). Contracts settle atomically at expiry, funds move within hours, and the entire lifecycle from listing to settlement is governed by contract terms you can read before you ever place a trade. That predictability is part of why institutional money has been flowing into the space since Kalshi's S&P 500 binary contracts launched.

Why Prediction Markets Beat Traditional Sportsbooks

Every serious sportsbook bettor runs into the same problem: the books limit you. Win consistently on a DraftKings or FanDuel account and within a few weeks the max stake drops, promotions vanish, and eventually the account closes. The industry term is "gubbing," and it's the single biggest reason most retail edge strategies fail to scale. Prediction markets don't do this. There's no house to gub you, because the other side of your trade is just another trader on the exchange.

Attribute	Traditional Sportsbooks	Prediction Markets
Account limits	Limited or closed after wins	Unlimited, no gubbing
Liquidity source	Bookmaker sets lines	Orderbook, peer-to-peer
Margin / vig	4% to 10% (soft) to 2% to 3% (sharp)	0% (Kalshi fees separate)
Short outcomes (bet "No")	Usually requires exchange	Native (buy No contract)
Market coverage	Sports only	Politics, macro, weather, sports
Order types	Usually market only	GTC, GTD, FOK, FAK
Settlement	Post-event, manual payout	Atomic at resolution

The practical impact of these differences is huge. On a prediction market, you can buy "No" on a team winning just by buying the opposite contract, which means you can short any outcome without a lay-betting exchange. Order types like FOK (fill-or-kill) and FAK (fill-and-kill) let you specify execution conditions that matter for a bot, whereas most sportsbooks only accept a blind market order. And the orderbook is visible: you can see depth at each price level before you size an order, which completely changes how you think about liquidity risk.

The other hidden benefit is market coverage. Sportsbooks only book sports, and within sports they only book the lines that generate the most hold. Prediction markets cover political elections, Fed rate decisions, CPI prints, weather events, tech milestones, and cultural events alongside full sports coverage. This widens the opportunity set for value betting and arbitrage well beyond what a pure sportsbook strategy can ever reach.

One more thing worth flagging: fees on prediction markets are transparent and flat, not embedded in the price. Kalshi publishes its fee tiers openly (the top retail tier caps at 7% on winnings, with institutional tiers lower). Polymarket posts zero exchange fees but you pay Polygon gas, which typically runs fractions of a cent per trade. On a sportsbook, the "fee" is the overround baked into the line, and you have no way to know what your true hold is without devigging every price. The shift from hidden vig to explicit fees is a huge deal for any trader running systematic math on expected value.

The 3 Edge Sources in Prediction Markets

If you reduce prediction market trading to its actual edge sources, there are three. They're not mutually exclusive. The best traders run all three simultaneously and let their bot decide which one is hottest at any given moment.

Cross-venue arbitrage

The same event, priced on both Kalshi and Polymarket, doesn't always agree. When the combined price of buying Yes on one venue and No on the other comes in under 100 cents ($1.00), you've got a guaranteed-profit arbitrage. Typical edges run 2% to 5% on sports and 3% to 8% on politics, and they close within 15 to 60 seconds. Watch these arbs appear and expire in real time on our free live scanner. The playbook is in our Kalshi vs Polymarket arbitrage guide.

Value betting

Single-side +EV trading against a sharp reference line. You devig Pinnacle or PS3838 to get a no-vig probability, compare it to the Kalshi or Polymarket price, and bet one side when the gap crosses your edge threshold. Value betting on prediction markets is higher-variance than arbitrage but pays bigger edges (5% to 15%) and works on markets that don't have an arb counterpart. Our prediction market value betting playbook covers the full math.

Sharp-vs-soft arbitrage

Pinnacle and PS3838 reprice within milliseconds of any news. Kalshi and Polymarket lag. During that 30 to 90 second window, the prediction market shows a stale price while the sharp book already reflects the new information. A cross-book bot buys the correct side on the prediction market before it catches up, then hedges at the new sharp price. This is the highest-edge strategy in the stack when it hits. Full architecture breakdown in our prediction market arbitrage primer.

Arbitrage on Prediction Markets

Arbitrage is the bread-and-butter strategy for most prediction market bots. Risk-free profit, no forecasting skill required, and the math is simpler than traditional sports arbing because every contract normalizes to $1.00. Here's the reading path.

Value Betting on Prediction Markets

Value betting is the newest and most under-covered area of prediction market strategy. Unlike arbitrage, it requires a real theory of fair probability, and that theory comes from sharp sportsbook lines. Combine a devigged Pinnacle probability with a Kalshi or Polymarket price and you get a rigorous +EV signal. These four guides walk the full stack.

Trading Bots & Automation

Manual edge chasing doesn't work on prediction markets. Arbitrage gaps close within 15 to 30 seconds, and value bets against sharp lines live for 30 to 90 seconds before the market catches up. A human opening two browser tabs takes longer than the entire life of the edge. Automation is the whole game. Here are the product landing pages plus the reviews of the bot category.

Free Prediction Market Tools

Before you trade real money, you need to be sure the math works. These four calculators run entirely in your browser (no signup, no data collection) and cover the core scenarios: fee modeling, devigging, +EV evaluation, and cross-venue arb sizing.

Why Claw Arbs

Most "arbitrage tools" in this category do one job: detection. They scan venues, find edges, and show you a dashboard. You're expected to open two browser tabs, eyeball the numbers, and place both legs manually before the edge closes. That approach stopped working on prediction markets years ago, because edges on Kalshi vs Polymarket close in 15 to 30 seconds. If you're not executing in milliseconds, you're not capturing anything.

Claw Arbs is the only tool that detects AND executes across prediction markets AND sportsbooks in the same stack. Twelve venues come pre-installed: Kalshi (US-regulated, WebSocket feed), Polymarket (WebSocket CLOB feed on Polygon L2), Polymarket US (CFTC-regulated US exchange via API), Predict.fun (prediction market via API), SX Bet (P2P prediction market via API), Sportmarket (sharp broker), Stake.com, BC.Game, and Bet105 (crypto and reduced-juice sportsbooks), Cloudbet, Pinnacle (PS3838), and BetInAsia (sharp aggregator). The WebSocket paths on Kalshi and Polymarket put price-to-edge-check latency at roughly 25 milliseconds, and ~20ms DOM observation means any sportsbook with a web interface can be added without waiting for an API.

Under the hood, three arb engines run in parallel. A cross-venue engine handles Kalshi vs Polymarket. A cross-book engine handles sharp-vs-soft edges against Pinnacle and BIA with a 14-step safety pipeline (kill switch, scanning pause, view-only, halt-if-naked, sharp freshness, min edge, max edge, min depth, token check, match rate limit, cooldown, execution mode, execute). An intra-market engine handles Polymarket bracket arbs. All three share a single naked leg tracker, so if any engine has a leg 1 fill without a hedge, every engine halts new trades on that match until the exposure resolves. This is the architecture that stops amateur bots from stacking multiple unhedged positions during volatile games and blowing up.

Paper trading mode runs the full pipeline against live data without risking capital, so you can validate your fee configuration, your event matching, and your execution timing before going live. The same codebase handles paper and live with a single settings toggle. Once you're happy with paper, flip the switch and the same engine starts placing real orders with identical logic.

Claw Arbs ships as a single async Python process that runs on your own machine or VPS. Your API keys stay with you. Your trades fire through your own accounts. Nothing routes through a third party. The community also shares configs for additional sportsbooks beyond the five pre-installed venues, so if someone's built a scraper for your favorite book, you can drop it in. Download it at clawarbs.com and start capturing edges instead of just watching them.

The practical difference in daily workflow is stark. A detection-only tool like OddsJam gives you a feed of edges and a stopwatch. You're racing against other humans and other bots to place two orders before the gap closes. Your hit rate on flagged opportunities is probably 10% to 20% after you account for missed fills, naked legs, and fee surprises. With an execution-aware stack, the bot takes the trade the instant edge passes the 14-step pipeline. Your hit rate on flagged opportunities climbs to 70% to 85%, your naked leg rate drops to the low single digits, and your actual PnL starts matching the theoretical edge your detector shows. That gap between "detected edge" and "captured edge" is the entire point of building an execution-aware system instead of a dashboard.

Frequently Asked Questions

What's the difference between prediction markets and sportsbooks?

Sportsbooks take the other side of your bet as the house, bake a 4% to 10% margin into the odds, and limit or close accounts once you win consistently. Prediction markets like Kalshi and Polymarket run peer-to-peer orderbooks where traders set the prices, so the margin is effectively 0% (Kalshi charges a separate fee on winnings, Polymarket charges gas). They don't gub you, they let you short outcomes by buying No, and they cover politics, macro, and weather markets sportsbooks won't touch.

How does value betting work on Kalshi vs a traditional sportsbook?

On a sportsbook, you hunt for soft lines mispriced against a sharp reference (usually Pinnacle) and bet them before they correct. On Kalshi you do the same thing, but the math is cleaner: Kalshi prices are already implied probabilities (cents out of 100), so you just compare them to devigged Pinnacle probability and bet when Kalshi is offering at least 2% to 3% of edge. Kalshi won't close your account for winning, which means you can actually scale the strategy. See our Kalshi value betting breakdown for the full playbook.

Can I arbitrage Polymarket against a sportsbook?

Yes, and it's one of the highest-edge setups in the stack. Polymarket trades the same NBA, NFL, and soccer events as Pinnacle and PS3838 but often lags sharp price moves by 30 to 90 seconds. When Pinnacle moves first, you can buy the correct side on Polymarket before it catches up. You need a bot that devigs the sharp line, checks orderbook depth on Polymarket, and fires both legs concurrently. Manual execution is too slow. Our Polymarket arbitrage bot architecture breaks down the exact pipeline.

Why do sharp bookmaker lines (Pinnacle, PS3838) matter for prediction market value betting?

Pinnacle operates at roughly 2% margin versus 4% to 10% on soft books. After you strip the vig, Pinnacle's implied probability is the closest public proxy to the true probability of an event. Because Kalshi and Polymarket don't always price in real time against sharp information, the gap between the devigged Pinnacle number and the Kalshi or Polymarket price is your value bet edge. Without a sharp reference, you're just guessing at fair value. Full method in our devigging prediction markets guide.

Is automated prediction market trading legal?

On Kalshi, yes. Kalshi is CFTC-regulated, its terms of service explicitly permit API trading, and many institutional accounts run bots against its order book. On Polymarket, the protocol restricts US IPs, so traders outside the US run bots freely while US-based traders typically stick to Kalshi. PS3838 and BetInAsia also allow automated trading within their terms. See our arbitrage legality post for a jurisdiction-by-jurisdiction breakdown.

How does Claw Arbs compare to OddsJam or BetBurger for prediction markets?

OddsJam and BetBurger show you a dashboard of detected edges and expect you to place the trades yourself. That doesn't work on prediction markets because edges close in 15 to 30 seconds. Claw Arbs connects to Kalshi, Polymarket, Pinnacle (PS3838), BetInAsia, and Cloudbet via WebSocket and headless browser, detects edges in roughly 25 milliseconds, and executes both legs automatically. It's the only tool in the category that actually places the orders, not just the only one that finds them. Compare architectures in our best Kalshi trading bot guide.