Market Making on Prediction Markets: The Power User Guide

# Market Making on Prediction Markets: The Power User Guide **Market making on prediction markets means continuously posting both buy (YES) and sell (NO) limit orders to profit from the bid-ask spread, while managing the risk that your probability estimates are wrong.** Done well, it generates consistent passive returns and earns platform rebates. Done poorly, it bleeds capital to better-informed traders in seconds. This guide is written for traders who already understand how prediction markets work and want to extract serious edge from the market-making role. We'll cover spread math, inventory management, adverse selection defense, automation, and the metrics that separate profitable market makers from expensive liquidity donors. --- ## Why Market Making on Prediction Markets Is Different Traditional market making on stocks or crypto involves assets with continuous price histories, derivatives for hedging, and ultra-tight spreads measured in fractions of a percent. Prediction markets are different in almost every important way. - **Binary payoffs**: Contracts resolve to $1 (YES wins) or $0 (NO wins). There's no "sort of" right. - **Event risk is lumpy**: A single news item can instantly shift the fair value by 30+ cents. - **Thin order books**: Many markets have fewer than 50 active participants, meaning your quotes actually move the market. - **Time decay asymmetry**: As resolution approaches, liquidity typically dries up and spreads widen — the opposite of options theta behavior. These features create both the opportunity and the danger. A 3-cent spread on a contract trading near 50¢ sounds small. But if you're posting $500 on each side and an informed trader hits you after a news event you missed, that one fill can cost more than a week of spread income. --- ## Understanding the Core Math: Spreads, Edges, and Expected Value Before you automate anything or post a single quote, you need to internalize the math. ### The Basic Spread Equation If you post a bid at **P_b** and an ask at **P_a**, your gross edge per round-trip is: ``` Edge = P_a - P_b ``` But your *net* edge after adverse selection is: ``` Net Edge = (P_a - P_b) - (2 × α × |P_true - P_mid|) ``` Where **α** is the fraction of your fills that come from informed traders and **P_true** is the true fair value at fill time. In practice, on liquid markets like Polymarket's major political contracts, informed-trader ratios can run **30–50%** of order flow. That's why a 4-cent spread that looks profitable in a spreadsheet often breaks even or worse in live trading. ### Sizing Your Spread to the Market A useful rule of thumb used by experienced market makers: | Market Condition | Suggested Minimum Spread | Rationale | |---|---|---| | High-volume political event, 50¢ fair value | 4–6 cents | High adverse selection risk | | Low-volume niche market, 50¢ fair value | 8–14 cents | Thin book, slow fills | | Near resolution (< 48 hrs), high certainty | 2–3 cents | Minimal event risk remains | | Near resolution (< 48 hrs), uncertain | Avoid or 15+ cents | Binary jump risk extreme | | Breaking news period | Pause or 20+ cents | Information asymmetry maximal | This table is a starting point, not a rulebook. Your actual spread should be calibrated to your real fill data over at least 200 trades. --- ## Inventory Management: The Silent Killer of Market Makers Most traders who blow up as market makers don't lose on individual trades — they lose because they accumulate a **skewed inventory** without realizing it until the position is too large to unwind at a reasonable price. ### Why Inventory Accumulates Every time you're filled, you move from market maker to position holder. If you sell YES at 52¢ and nobody buys it back from you, you're now long YES. If that happens 20 times in one direction, you're carrying $1,000+ of directional exposure you didn't intend to have. This happens fastest in **mean-reverting markets** where the price bounces back after you've been hit — a pattern worth studying in the context of [scaling up mean reversion strategies with a $10K portfolio](/blog/scale-up-mean-reversion-strategies-with-a-10k-portfolio). ### Inventory Control Techniques 1. **Set hard inventory limits per market**: Never hold more than X% of your total market-making capital as directional exposure in a single contract. Most professionals use 10–20%. 2. **Skew your quotes dynamically**: If you're long YES, move your ask down (to sell faster) and your bid down (to slow further YES accumulation). The formula: shift quotes by *k × net_inventory* where k is your aggressiveness parameter. 3. **Use time-based inventory resets**: If you're still carrying a skewed position after N hours, widen spreads dramatically until the position reverts. 4. **Define a "stop making" threshold**: If your net directional position exceeds 2× your normal max, stop quoting entirely and focus on unwinding. --- ## Adverse Selection Defense: Don't Feed the Informed Traders Adverse selection — getting picked off by traders who know more than you — is the defining challenge of market making. Here's how to fight back. ### Information Latency Monitoring Build or subscribe to an alert system that tracks news sources relevant to your active markets. If you're making markets on "Will the Fed cut rates in September?", you need to know about FOMC releases *before* you get filled on stale quotes. [PredictEngine](/) includes real-time signal feeds that can be integrated into your quoting logic to pause or widen spreads when high-impact events are detected. ### Quote Freshness Rules Never let a quote sit in the book for more than a defined window without refreshing it. For political markets, many market makers refresh every **30–60 seconds** during high-activity periods and every **5–10 minutes** during quiet periods. ### Order Flow Toxicity Scoring Track your **fill rate directional bias**. If 70%+ of your YES fills happen before price moves up, and 70%+ of your NO fills happen before price moves down, your order flow is toxic — informed traders are finding you. Responses include: - Widening spreads by 2–3 cents - Reducing quote size by 50% - Temporarily withdrawing from that market entirely This kind of [algorithmic order book analysis for prediction markets](/blog/algorithmic-order-book-analysis-for-prediction-markets) is what separates hobbyist market makers from systematic operators. --- ## Building a Market Making System: Step-by-Step Here's how to build a basic but functional market making setup from scratch: 1. **Define your universe**: Start with 3–5 markets maximum. Prioritize markets with >$50K total volume and clear, verifiable resolution criteria. 2. **Establish fair value models**: For each market, build or adopt a probability model. This can be as simple as averaging Polymarket, Metaculus, and Manifold prices, or as complex as an LLM-based news signal model (see our [beginner's guide to LLM-powered trade signals](/blog/beginners-guide-to-llm-powered-trade-signals-for-q2-2026)). 3. **Set initial spread parameters**: Use the table above as a starting point. Log every parameter change and its effect on PnL. 4. **Build quote generation logic**: Your system should output bid/ask prices every refresh cycle based on fair value, inventory position, and current spread parameters. 5. **Integrate order management**: Connect to the exchange API to cancel stale orders, post new ones, and track fills in real time. 6. **Implement inventory tracking**: Every fill updates your inventory. Your quoting logic should read current inventory before each quote refresh. 7. **Set kill switches**: Define conditions under which the system stops quoting entirely — e.g., unusual fill rate, connectivity issues, or scheduled high-impact events. 8. **Paper trade for 2 weeks**: Run your system in simulation before going live. Log everything: fills, spreads, inventory levels, PnL by market. 9. **Go live with 20% of intended capital**: Let the real system run for 4 weeks before scaling. Real order flow behaves differently than simulations. 10. **Review and iterate weekly**: Measure your **spread capture rate**, **adverse selection ratio**, and **inventory turnover**. Adjust parameters based on data, not intuition. For a parallel perspective on systematic trade execution, the [trader playbook on prediction market economics and limit orders](/blog/trader-playbook-economics-prediction-markets-limit-orders) is required reading before you go live. --- ## Platform-Specific Considerations ### Polymarket Polymarket uses a **CLOB (Central Limit Order Book)** model powered by a matching engine on Polygon. Key facts for market makers: - Minimum order size: $1 - No native maker rebates (as of mid-2025), but passive fills face no taker fee - Gas costs on Polygon are negligible but still real at high frequency - Books are public and scrapable via API — essential for competitive quoting ### Other Platforms Platforms like Manifold, Metaculus, and Kalshi each have different mechanisms. Kalshi operates under **CFTC regulation**, which affects position limits and counterparty structure. Some platforms use **AMM (Automated Market Maker)** models rather than order books, which changes the math entirely — you provide liquidity to a pool rather than posting discrete quotes. If you're also active in sports prediction markets, the dynamics differ substantially from political or financial contracts. Our [deep dive into NFL season predictions](/blog/deep-dive-into-nfl-season-predictions-a-step-by-step-guide) covers how sports-specific information asymmetry affects your edge. --- ## Measuring Success: Key Metrics for Market Makers | Metric | Definition | Healthy Range | |---|---|---| | Spread Capture Rate | % of posted spread actually captured | 45–65% | | Adverse Selection Ratio | % of fills followed by adverse move | < 35% | | Inventory Turnover | How often position reverts to flat | Every 4–12 hrs | | Fill Rate | % of posted quotes that get filled | 5–20% per session | | Sharpe Ratio (annualized) | Risk-adjusted return | > 1.5 | | Max Drawdown | Worst peak-to-trough PnL | < 15% of capital | Track these weekly. If your adverse selection ratio creeps above 40%, something has changed in your market — either your model is stale, informed traders have found your quoting pattern, or an event has shifted the true probability and you haven't updated. --- ## Advanced Techniques for Serious Market Makers ### Cross-Market Hedging If you're making markets on "Will Candidate A win State X?" and "Will Candidate A win the election overall?", these markets are correlated. You can use fills in one to hedge inventory in the other — effectively running a **delta-neutral portfolio** across related prediction markets. ### Latency Arbitrage Defense Some traders specifically target slow market makers. If you notice you're being filled milliseconds before price moves, you're being latency-arbed. Solutions: shorten your quote refresh interval, reduce quote size, or use randomized refresh timing to avoid pattern exploitation. ### NLP-Driven Quote Adjustment More sophisticated setups use real-time NLP on news feeds to dynamically adjust fair value estimates. When a significant article is detected, the system instantly widens spreads or pauses quoting. This is covered in depth in our [advanced NLP strategy compilation via API](/blog/advanced-nlp-strategy-compilation-via-api-a-deep-dive). --- ## Frequently Asked Questions ## What capital do I need to start market making on prediction markets? You can technically start with as little as $500, but **$5,000–$10,000** is a more realistic minimum to spread across 3–5 markets meaningfully. Below that threshold, transaction costs and minimum order sizes make it hard to diversify your inventory risk properly. ## How much can a market maker realistically earn on prediction markets? Experienced market makers report **annualized returns of 20–60%** on deployed capital, though this varies enormously by market conditions, competition, and strategy sophistication. These figures assume active management and a well-tuned adverse selection defense — passive or poorly monitored systems typically underperform significantly. ## Is market making on prediction markets legal? In most jurisdictions, yes — prediction market trading including market making is legal for retail participants on platforms like Polymarket. **Kalshi operates under CFTC regulation** in the US, which imposes specific compliance requirements. Always verify the regulatory status of any platform you trade on and consult a financial or legal professional if in doubt. ## What's the difference between market making and arbitrage on prediction markets? **Market making** involves posting two-sided quotes and profiting from the bid-ask spread with minimal directional exposure. **Arbitrage** involves exploiting price discrepancies between platforms for a near-risk-free profit. Both are systematic strategies, but they have very different risk profiles — market making has inventory risk while arbitrage has execution and timing risk. You can explore the latter at [/polymarket-arbitrage](/polymarket-arbitrage). ## How do I handle market making around scheduled news events? The standard practice is to **withdraw your quotes 15–30 minutes before** a scheduled high-impact event (Fed announcements, election results, earnings releases) and resume quoting 5–10 minutes after the market has repriced to the new information. Running live quotes through binary events is one of the fastest ways to take large, unintended losses. ## Can I automate market making on prediction markets? Yes, and at scale you almost need to. Manual market making is viable for 1–2 markets but becomes error-prone beyond that. Most serious market makers use bots connected to exchange APIs. [PredictEngine](/) provides API access and tooling specifically designed to support automated quoting strategies for prediction markets. --- ## Get Serious About Market Making with PredictEngine Market making on prediction markets rewards precision, discipline, and good tooling. If you've made it through this guide, you have the conceptual foundation. What you need next is a platform that gives you real-time data, API access, and the analytics to track your edge over time. [PredictEngine](/) is built for exactly this kind of power-user trading. From live order book feeds to signal integration and portfolio-level PnL tracking, it's the infrastructure layer that turns good market making strategy into consistent, measurable results. [Explore PredictEngine's features and pricing](/pricing) and see how serious traders are building systematic edges on prediction markets today.