// QUANT STRATEGIES · PILLAR GUIDE

Quantitative trading on prediction markets: a field guide

2026-07-16 · Mithril

Prediction markets in 2026 look a lot like a young asset class: two liquid venues (Kalshi, CFTC-regulated; Polymarket, crypto-settled), thousands of binary contracts on elections, macro prints, weather, sports, and pop culture, and order books thin enough that a careful individual quant can still find edges an index fund never could. This is the field guide: how the market is structured, where edge actually comes from, what data exists, and why execution is a first-class part of the strategy rather than an afterthought.

Market structure in one section

Both venues run a central limit order book on binary contracts that pay $1 (Kalshi) or 1 USDC (Polymarket) if an event resolves YES, zero otherwise. Price is implied probability: a 37¢ Kalshi contract or a 0.37 Polymarket token is the book's consensus that the event has a 37% chance.

Three structural facts drive everything else:

1. Bounded payoffs. Every position has a known worst case. There is no gap risk beyond your premium — risk management is about probability, not tail exposure.
2. Hard expiry with a binary jump. Prices don't drift to fair value; they snap to 0 or 1 at resolution. Being early and being wrong look identical until the event.
3. Thin books. Outside of headline markets, top-of-book depth is often a few hundred to a few thousand dollars. Your fill price, not your model, frequently decides whether a trade was good. See slippage in thin prediction markets.

Where edge comes from

1. Cross-venue relative value

The same real-world event trades on both venues, usually at slightly different prices. Sometimes the gap exceeds fees and depth constraints and becomes a true arbitrage; more often it's a relative-value signal — one venue leads, the other follows. The classic mechanics, including the fee math and the resolution-criteria traps that eat naive arbs, are covered in cross-venue arbitrage between Kalshi and Polymarket. The prerequisite infrastructure is unglamorous: you need to know that Kalshi ticker X and Polymarket condition Y are the same event, which is a mapping problem — see unified market IDs.

2. Event modeling

The purest form of quant edge here: build a better probability estimate than the market. Macro prints (CPI, payrolls, Fed decisions) reward people who can turn nowcasting models into calibrated probabilities. Weather markets reward ensemble-forecast literacy. Sports markets reward the same models the betting industry has refined for decades. The market's advantage is aggregation; your advantage is specialization. A model that is well calibrated in one narrow domain, run systematically, beats a generalist's intuition everywhere.

The discipline that matters most is calibration measurement: track your predicted probabilities against outcomes (Brier score or log loss) before you size up. A model that "feels right" but is 5 points miscalibrated loses money against a 3-point spread.

3. Liquidity provision

Thin books mean wide spreads, and wide spreads pay whoever is willing to quote. On Kalshi the economics tilt further in the maker's favor because resting orders generally pay no trading fee on most markets while takers pay the formula fee. Quoting both sides of a binary, managing inventory, and stepping away around news is a real business — the mechanics, the adverse selection problem near resolution, and the spread-capture math are in market making on prediction markets 101.

4. Flow and microstructure

Retail-heavy order flow has patterns: longshot bias (tail prices trade rich), momentum chasing after news, and stale quotes on the venue that isn't currently in the headlines. These are smaller, faster edges that mostly accrue to whoever has the better data pipeline and lower latency — worth knowing about, harder to build a first strategy on.

Data: what exists and what doesn't

Live data is good. Historical data is the weak point of the ecosystem.

• Kalshi serves markets, order books, and trades over REST and WebSocket at api.elections.kalshi.com/trade-api/v2, including candlestick history for markets. Full historical order-book depth is not something you can simply download — if your strategy depends on book state, start recording now.
• Polymarket splits metadata and prices (Gamma, gamma-api.polymarket.com, no auth) from the order book and trading (clob.polymarket.com). Because settlement is on-chain, trade history is reconstructible from Polygon, and several third-party datasets index it — but book snapshots are again something you mostly have to capture yourself.

Backtesting on prediction markets

This deserves its own warning label. Backtests here fail in ways equity backtests don't:

• Fill assumptions dominate. In a book with $400 at the touch, "I would have bought 5,000 contracts at the mid" is fiction. Backtest against recorded depth, or haircut aggressively.
• Resolution criteria are part of the data. Two venues' versions of "the same" market can resolve differently. A backtest that treats them as identical will book profits that never existed.
• Survivorship in market selection. It's easy to backtest only the markets you remember being liquid. Pull the full market list per period, not the ones that made news.
• Fees and gas. Kalshi's taker fee peaks exactly where the interesting trading happens (near 50¢). Model it per fill, not as a flat haircut — the formula is in Kalshi fees explained.

A practical starting stack: record top-N book levels on both venues for the markets you care about, store trades, and replay against a simple queue-aware fill simulator. It's a week of work that upgrades every strategy you test afterwards.

Prediction markets vs sportsbooks vs options

Quants arrive here from two neighboring worlds, and both analogies are half-right.

vs sportsbooks: A sportsbook sets a price and takes the other side; you are always trading against the house's margin (the vig), you cannot be the maker, and winners get limited. Prediction markets are exchanges: you trade against other participants, you can quote instead of take, and edge is rewarded with fills rather than bans. The flip side is that liquidity is yours to find — nobody is obligated to offer you a price in size.

vs options: A binary contract is economically a cash-or-nothing digital option, so options intuition transfers: price ≈ probability ≈ the N(d2)-ish quantity, and sensitivity to new information peaks near 50¢ and near expiry. What doesn't transfer: there's no underlying to delta-hedge against, no volatility surface to trade, and no continuous payoff — you are trading the event itself. Risk management therefore leans on sizing and correlation control rather than Greeks, which is exactly why Kelly sizing on binary contracts is the sizing framework of choice here.

Execution is alpha

On instruments this thin, execution quality isn't operational hygiene — it's a return stream. The same signal, executed naively vs carefully, can differ by more than the signal's entire edge:

• Venue choice is a pricing decision. The displayed-cheaper venue is often the net-more-expensive one after Kalshi's taker fee or Polymarket's spread and gas. Fee-aware comparison on every order is the whole idea behind smart order routing.
• Working beats crossing. In a thin book, posting inside the spread and waiting captures the spread instead of paying it — and on Kalshi it generally avoids the taker fee entirely.
• Measure it. If you can't attribute PnL between signal and slippage, you can't improve either. That's the case for execution receipts on every fill.

This is also where tooling earns its keep. Mithril exists precisely for this layer: one API over both venues that routes each order for the best net price after fees, works orders in thin books, enforces hard risk limits server-side, and hands back an execution receipt per fill — your number, not a marketing claim. The API docs cover it if you get to that stage.

The spokes

This guide is the hub of our quant series. The deep dives:

• Cross-venue arbitrage between Kalshi and Polymarket
• Market making on prediction markets 101
• Position sizing on binary contracts: Kelly and its discounts
• Execution quality on prediction markets
• Slippage in thin prediction markets

Start with a narrow domain you know, measure your calibration, size with a Kelly discount, and treat execution as part of the strategy from day one.

Market structure and fees change. Details above reflect public documentation as of July 2026 — always confirm against the venues' own docs before trading real money.