Make Probability an Asset: A Forward Look at Predictive Market Agents

Author: Jacob Zhao, IOSG

In our previous Crypto AI series reports, we have consistently emphasized the viewpoint that the most practically valuable scenarios in the current crypto space are primarily concentrated in stablecoin payments and DeFi, while Agents are the key interface for AI industries facing users.

Therefore, in the trend of the integration of Crypto and AI, the two most valuable paths are: AgentFi, based on existing mature DeFi protocols (basic strategies such as lending, liquidity mining, and advanced strategies like Swap, Pendle PT, and funding rate arbitrage) in the short term, and Agent Payment, focusing on stablecoin settlement and relying on protocols such as ACP/AP2/x402/ERC-8004 in the medium to long term.

Prediction markets have become an industry trend that cannot be ignored by 2025, with annual total trading volume surging from approximately $9 billion in 2024 to over $40 billion in 2025, achieving more than 400% year-on-year growth.

This significant growth is driven by multiple factors: the demand for uncertainty brought by macro-political events, the maturity of infrastructure and trading models, and the regulatory environment beginning to thaw (Kalshi's legal victory and Polymarket's return to the U.S.). Prediction Market Agents are expected to show early forms in early 2026 and may become an emerging product form in the agent field in the coming year.

I. Prediction Markets: From Betting Tools to "Global Truth Layer"

Prediction markets are financial mechanisms that trade on the outcomes of future events, where contract prices essentially reflect the market's collective judgment on the probability of events occurring. Their effectiveness stems from the combination of collective intelligence and economic incentives: in an environment where anonymous, real-money betting occurs, dispersed information is rapidly integrated into price signals weighted by capital willingness, significantly reducing noise and false judgments.

By the end of 2025, prediction markets have basically formed a duopoly dominated by Polymarket and Kalshi. According to Forbes, the total trading volume in 2025 is expected to reach approximately $44 billion, with Polymarket contributing about $21.5 billion and Kalshi about $17.1 billion. Data from February 2026 shows that Kalshi's trading volume ($25.9B) has surpassed Polymarket's ($18.3B), nearing a 50% market share. Kalshi has achieved rapid expansion due to its prior legal victory in election contracts, its compliance-first advantage in the U.S. sports prediction market, and relatively clear regulatory expectations. Currently, the development paths of the two have shown clear differentiation:

• Polymarket adopts a hybrid CLOB architecture with "off-chain matching and on-chain settlement" and a decentralized settlement mechanism, creating a global, non-custodial high-liquidity market, forming an "onshore + offshore" dual-track operational structure after compliance returns to the U.S.;

• Kalshi integrates into the traditional financial system by connecting with mainstream retail brokers through API, attracting Wall Street market makers to participate deeply in macro and data-driven contract trading. Its products are subject to traditional regulatory processes, with long-tail demand and sudden events relatively lagging.

In addition to Polymarket and Kalshi, other competitive participants in the prediction market space are developing along two main paths:

• One is the compliance distribution path, embedding event contracts into existing accounts and clearing systems of brokers or large platforms, relying on channel coverage, compliance qualifications, and institutional trust to establish advantages (e.g., Interactive Brokers × ForecastEx's ForecastTrader, FanDuel × CME Group's FanDuel Predicts). Compliance and resource advantages are significant, but the product and user scale are still in the early stages.

• The other is the crypto-native on-chain path, represented by Opinion.trade, Limitless, and Myriad, which leverage point mining, short-cycle contracts, and media distribution to achieve rapid scaling, emphasizing performance and capital efficiency, but their long-term sustainability and risk control robustness remain to be validated.

The two paths of traditional financial compliance entry and crypto-native performance advantages together form a diverse competitive landscape for the prediction market ecosystem.

On the surface, prediction markets resemble gambling, but the essential difference lies in whether they have positive externalities: by aggregating dispersed information through real-money transactions, they publicly price real events, forming a valuable signal layer. The trend is shifting from gaming to a "global truth layer"—as institutions like CME and Bloomberg get involved, event probabilities have become decision metadata that can be directly invoked by financial and corporate systems, providing more timely and quantifiable market truths.

From the perspective of global regulatory status, the compliance paths for prediction markets are highly differentiated. The U.S. is the only major economy that explicitly includes prediction markets in its financial derivatives regulatory framework, while markets in Europe, the UK, Australia, and Singapore generally view them as gambling and tend to tighten regulations. Countries like China and India completely prohibit them, meaning the future global expansion of prediction markets still relies on the regulatory frameworks of various countries.

II. Architecture Design of Prediction Market Agents

Currently, Prediction Market Agents are entering an early practical stage, where their value lies not in "AI predictions being more accurate," but in amplifying information processing and execution efficiency within prediction markets. Prediction markets are essentially information aggregation mechanisms, where prices reflect collective judgments on event probabilities; inefficiencies in real markets stem from information asymmetry, liquidity, and attention constraints.

The reasonable positioning of prediction market agents is executable probabilistic portfolio management: transforming news, rule texts, and on-chain data into verifiable pricing deviations, executing strategies in a faster, more disciplined, and cost-effective manner, and capturing structural opportunities through cross-platform arbitrage and portfolio risk control.

An ideal prediction market agent can be abstracted into a four-layer architecture:

• The information layer aggregates news, social, on-chain, and official data;

• The analysis layer uses LLM and ML to identify mispricing and calculate Edge;

• The strategy layer converts Edge into positions through Kelly's formula, batch building, and risk control;

• The execution layer completes multi-market ordering, slippage and Gas optimization, and arbitrage execution, forming an efficient automated closed loop.

III. Strategy Framework of Prediction Market Agents

Unlike traditional trading environments, prediction markets have significant differences in settlement mechanisms, liquidity, and information distribution, and not all markets and strategies are suitable for automated execution. The core of prediction market agents lies in whether they are deployed in scenarios with clear rules, codable structures, and structural advantages. The following analysis will unfold from three aspects: target selection, position management, and strategy structure.

Target Selection in Prediction Markets Not all prediction markets have tradable value; their participation value depends on: clarity of settlement (whether rules are clear, whether data sources are unique), quality of liquidity (market depth, spreads, and trading volume), insider risk (degree of information asymmetry), time structure (expiration time and event rhythm), and the trader's own information advantages and professional background. Only when most dimensions meet basic requirements does a prediction market have the foundation for participation, and participants should match their advantages with market characteristics:

• Human core advantages: relying on expertise, judgment, and integration of ambiguous information, and markets with relatively wide time windows (measured in days/weeks). Typical examples include political elections, macro trends, and corporate milestones.

• AI Agent core advantages: relying on data processing, pattern recognition, and rapid execution, and markets with extremely short decision windows (measured in seconds/minutes). Typical examples include high-frequency crypto prices, cross-market arbitrage, and automated market making.

• Non-adaptive areas: markets dominated by insider information or purely random/highly manipulated markets do not constitute advantages for any participants.

Position Management in Prediction Markets The Kelly Criterion is the most representative capital management theory in repeated gaming scenarios, aiming not to maximize single-instance returns but to maximize the long-term compound growth rate of capital. This method estimates the optimal position ratio based on win rates and odds, enhancing capital growth efficiency under the premise of positive expectations, and is widely used in quantitative investing, professional gambling, poker, and asset management.

• The classic form is: f\^* = (bp - q) / b

• Where f∗ is the optimal betting ratio, b is the net odds, p is the win rate, and q=1−p.

• Prediction markets can be simplified to: f\^* = (p - market\price) / (1 - market\price)

• Where p is the subjective true probability, and market_price is the market implied probability.

The theoretical effectiveness of the Kelly Criterion highly depends on the accurate estimation of true probabilities and odds. In reality, traders find it difficult to consistently grasp true probabilities accurately. In practice, professional gamblers and prediction market participants tend to adopt more executable, less probability-dependent rule-based strategies:

• Unit System: dividing capital into fixed units (e.g., 1%), investing different unit numbers based on confidence levels, and automatically constraining single-bet risk through unit limits, is the most common practical method.

• Flat Betting: using a fixed capital ratio for each bet, emphasizing discipline and stability, suitable for risk-averse or low-confidence environments.

• Confidence Tiers: pre-setting discrete position tiers and setting absolute limits to reduce decision complexity and avoid the pseudo-precision issues of the Kelly model.

• Inverted Risk Approach: starting from the maximum tolerable loss to backtrack position size, forming stable risk boundaries based on risk constraints rather than return expectations.

For prediction market agents, strategy design should prioritize executability and stability rather than pursuing theoretical optimality. The key lies in clear rules, simple parameters, and fault tolerance for judgment errors. Under this constraint, the confidence tier method combined with fixed position limits is the most suitable general position management scheme for PM Agents. This method does not rely on precise probability estimates but divides opportunities into limited tiers based on signal strength and corresponds to fixed positions; even in high-confidence scenarios, it sets clear limits to control risk.

Strategy Selection in Prediction Markets From a structural perspective, prediction market strategies can be broadly divided into two categories: deterministic arbitrage strategies characterized by clear rules and codability, and speculative directional strategies relying on information interpretation and directional judgment; additionally, there are market-making and hedging strategies primarily led by professional institutions, which require high capital and infrastructure.

Deterministic Arbitrage Strategies

• Resolution Arbitrage: Resolution arbitrage occurs when the event outcome is basically determined, but the market has not fully priced it yet. The profit mainly comes from information synchronization and execution speed. This strategy has clear rules, lower risk, and can be fully coded, making it the most suitable core strategy for agents in prediction markets.

• Dutch Book Arbitrage: Dutch Book arbitrage exploits structural imbalances formed when the sum of prices of mutually exclusive and complete event sets deviates from the probability conservation constraint (∑P≠1), locking in directionless risk returns through combination positions. This strategy relies solely on rules and price relationships, has lower risk, and can be highly regulated, making it a typical deterministic arbitrage form suitable for automated execution by agents.

• Cross-Platform Arbitrage: Cross-platform arbitrage profits by capturing pricing deviations of the same event across different markets, with lower risk but high requirements for latency and parallel monitoring. This strategy is suitable for agents with infrastructure advantages, but increased competition has led to declining marginal returns.

• Bundle Arbitrage: Bundle arbitrage trades on pricing inconsistencies between related contracts, with clear logic but limited opportunities. This strategy can be executed by agents but has certain engineering requirements for rule parsing and combination constraints, making agent adaptability moderate.

Speculative Directional Strategies

• Information Trading: This type of strategy revolves around clear events or structured information, such as official data releases, announcements, or adjudication windows. As long as the information source is clear and the triggering conditions are definable, agents can leverage speed and discipline in monitoring and execution; however, when information turns into semantic judgment or contextual interpretation, human intervention is still required.

• Signal Following: This strategy gains profits by following accounts or capital behaviors with historically superior performance, with relatively simple rules and automated execution. Its core risk lies in signal degradation and being reverse-utilized, thus requiring filtering mechanisms and strict position management. It is suitable as an auxiliary strategy for agents.

• Unstructured / Noise-driven: This type of strategy heavily relies on emotions, randomness, or participant behavior, lacking a stable and replicable edge, with long-term expected values being unstable. Due to difficulties in modeling and extremely high risks, it is not suitable for systematic execution by agents and is not recommended as a long-term strategy.

Market Microstructure Strategies: These strategies rely on extremely short decision windows, continuous quoting, or high-frequency trading, requiring high demands for latency, models, and capital. Although theoretically suitable for agents, they are often limited by liquidity and competition intensity in prediction markets, making them suitable only for a few participants with significant infrastructure advantages.

Risk Control & Hedging Strategies: These strategies do not directly pursue profits but are used to reduce overall risk exposure. They have clear rules and objectives and operate as long-term underlying risk control modules.

Overall, the strategies suitable for agent execution in prediction markets are concentrated in scenarios with clear rules, codability, and weak subjective judgment, where deterministic arbitrage should serve as the core source of returns, while structured information and signal-following strategies serve as supplements, and high-noise and emotion-driven trading should be systematically excluded. The long-term advantage of agents lies in their high discipline, speed of execution, and risk control capabilities.

IV. Business Models and Product Forms of Prediction Market Agents

The ideal business model design for prediction market agents has different exploratory directions at different levels:

• Infrastructure Layer: Providing multi-source real-time data aggregation, Smart Money address databases, a unified prediction market execution engine, and backtesting tools, charging B2B fees to obtain stable income unrelated to prediction accuracy;

• Strategy Layer: Introducing community and third-party strategies to build a reusable, assessable strategy ecosystem, capturing value through invocation, weighting, or execution sharing, thereby reducing dependence on a single Alpha.

• Agent / Vault Layer: Agents participate in real trading in a trustee management manner, relying on on-chain transparent records and strict risk control systems, charging management fees and performance fees for realization capabilities.

The product forms corresponding to different business models can also be divided into:

• Entertainment / Gamification Model: Lowering participation barriers through intuitive interactions similar to Tinder, possessing the strongest user growth and market education capabilities, serving as an ideal entry point for breaking into new circles, but needing to transition to subscription or execution-based product monetization.

• Strategy Subscription / Signal Model: Not involving capital custody, regulatory-friendly, with clear responsibilities, and a relatively stable SaaS income structure, this is the most feasible commercialization path at the current stage. Its limitation lies in the ease of strategy replication and execution losses, with a limited long-term income ceiling, which can be significantly improved through a semi-automated form of "signal + one-click execution" to enhance experience and retention.

• Vault Custody Model: Possessing scale effects and execution efficiency advantages, this form is close to asset management products but faces multiple structural constraints such as asset management licenses, trust barriers, and centralized technical risks, with the business model highly dependent on market conditions and sustained profitability. Unless there are long-term performance and institutional endorsements, it is not advisable as the main path.

Overall, a diversified income structure of "infrastructure monetization + strategy ecosystem expansion + performance participation" helps reduce reliance on the single assumption of "AI continuously outperforming the market." Even if Alpha converges as the market matures, underlying capabilities such as execution, risk control, and settlement still hold long-term value, thus constructing a more sustainable business closed loop.

V. Project Cases of Prediction Market Agents

Currently, Prediction Market Agents are still in the early exploration stage. Although the market has seen diverse attempts from underlying frameworks to upper-layer tools, a set of standardized products that are mature in strategy generation, execution efficiency, risk control systems, and business closed loops has yet to form.

We categorize the current ecological landscape into three levels: Infrastructure Layer, Autonomous Trading Agents, and Prediction Market Tools.

Infrastructure Layer # Polymarket Agents Framework Polymarket Agents is a developer framework officially launched by Polymarket, aimed at solving the engineering standardization issues of "connection and interaction." This framework encapsulates market data acquisition, order construction, and basic LLM calling interfaces. It addresses the question of "how to place orders with code," but leaves core trading capabilities—such as strategy generation, probability calibration, dynamic position management, and backtesting systems—largely blank. It is more like an officially recognized "access specification" rather than a finished product with Alpha returns. Commercial-grade agents still need to build a complete investment research and risk control core based on this.

# Gnosis Prediction Market Tools Gnosis Prediction Market Agent Tooling (PMAT) provides complete read and write support for Omen/AIOmen and Manifold, but only offers read-only access to Polymarket, showing a clear ecological barrier. It is suitable as a foundational development base for agents within the Gnosis system, but has limited practicality for developers primarily focused on Polymarket.

Polymarket and Gnosis are currently the only prediction market ecosystems that have clearly productized "Agent development" into official frameworks. Other prediction markets like Kalshi still mainly remain at the API and Python SDK level, requiring developers to fill in key system capabilities such as strategy, risk control, operation, and monitoring on their own.

Autonomous Trading Agents Currently, the "prediction market AI Agents" in the market are mostly still in the early stages. Although they are labeled as "Agents," their actual capabilities are significantly distant from automated closed-loop trading that can be delegated, generally lacking independent, systematic risk control layers, and have not incorporated position management, stop-loss, hedging, and expected value constraints into decision-making processes, resulting in a low overall degree of productization and not forming a mature system that can operate long-term.

# Olas Predict Olas Predict is the most productized prediction market agent ecosystem currently. Its core product Omenstrat is built on Gnosis's Omen, using FPMM and decentralized arbitration mechanisms at the underlying level, supporting small-scale high-frequency interactions, but limited by insufficient liquidity in the Omen single market. Its "AI predictions" mainly rely on general LLMs, lacking real-time data and systematic risk control, with historical win rates showing significant differentiation across categories. In February 2026, Olas launched Polystrat, extending agent capabilities to Polymarket—users can set strategies in natural language, and agents automatically identify probability deviations in markets settling within four days and execute trades. The system controls risk through local Pearl operation, self-hosted Safe accounts, and hard-coded limits, making it the first consumer-grade autonomous trading agent aimed at Polymarket.

# UnifAI Network Polymarket Strategy Provides an automated trading agent for Polymarket, focusing on tail risk-bearing strategies: scanning near-settlement contracts with implied probabilities >95% and buying them, aiming for a 3-5% price difference. On-chain data shows a win rate close to 95%, but returns vary significantly across categories, with the strategy highly dependent on execution frequency and category selection.

# NOYA.ai NOYA.ai attempts to integrate "research---judgment---execution---monitoring" into an agent closed loop, with architecture covering intelligence, abstraction, and execution layers. It has currently delivered Omnichain Vaults; the Prediction Market Agent is still in development and has not formed a complete mainnet closed loop, remaining in the vision validation phase.

Prediction Market Tools Currently, prediction market analysis tools are not sufficient to constitute a complete "prediction market agent," with their value primarily concentrated in the information and analysis layers of the agent architecture, while trading execution, position management, and risk control still need to be borne by traders themselves. From a product form perspective, they align more with the positioning of "strategy subscription / signal assistance / research enhancement," and can be seen as early forms of prediction market agents.

Through systematic sorting and empirical screening of projects included in Awesome-Prediction-Market-Tools, this report selects representative projects that already possess preliminary product forms and use cases as case studies. They mainly focus on four directions: analysis and signal layers, alert and whale tracking systems, arbitrage discovery tools, and trading terminals and aggregated execution.

# Market Analysis Tools

• Polyseer: A research-oriented prediction market tool that employs a multi-agent division of labor architecture (Planner / Researcher / Critic / Analyst / Reporter) for bilateral evidence collection and Bayesian probability aggregation, outputting structured research reports. Its advantages lie in transparent methodology, process engineering, and complete open-source auditability.

• Oddpool: Positioned as the "Bloomberg terminal of prediction markets," providing cross-platform aggregation, arbitrage scanning, and real-time data dashboard terminals for Polymarket, Kalshi, CME, etc.

• Polymarket Analytics: A global data analysis platform for Polymarket, systematically displaying trader, market, position, and transaction data, with clear positioning and intuitive data, suitable as a basic data query and research reference.

• Hashdive: A data tool for traders, quantifying the selection of traders and markets through Smart Score and multi-dimensional Screeners, practical for "smart money identification" and follow-on decision-making.

• Polyfactual: Focused on AI market intelligence and sentiment/risk analysis, embedding analytical results into trading interfaces via a Chrome extension, leaning towards B2B and institutional user scenarios.

• Predly: An AI mispricing detection platform that identifies pricing deviations in Polymarket and Kalshi by comparing market prices with AI-calculated probabilities, with the official claim of an alert accuracy rate of 89%, positioned for signal discovery and opportunity screening.

• Polysights: Covers 30+ markets and on-chain indicators, tracking anomalies such as new wallets and large one-way bets with Insider Finder, suitable for daily monitoring and signal discovery.

• PolyRadar: A multi-model parallel analysis platform providing real-time interpretations, timeline evolution, confidence scoring, and source transparency for single events, emphasizing multi-AI cross-validation, positioned as an analysis tool.

• Alphascope: An AI-driven prediction market intelligence engine providing real-time signals, research summaries, and probability change monitoring, still in the early stages, leaning towards research and signal support.

# Alerts/Whale Tracking

• Stand: Clearly positioned for whale following and high-confidence action alerts.

• Whale Tracker Livid: Productizing whale position changes.

# Arbitrage Discovery Tools

• ArbBets: An AI-driven arbitrage discovery tool focusing on Polymarket, Kalshi, and sports betting markets, identifying cross-platform arbitrage and positive expected value (+EV) trading opportunities, positioned for high-frequency opportunity scanning.

• PolyScalping: A real-time arbitrage and scalp analysis platform for Polymarket, supporting full market scans every 60 seconds, ROI calculations, and Telegram push notifications, with opportunities filtered by liquidity, spreads, and trading volume, leaning towards proactive traders.

• Eventarb: A lightweight cross-platform arbitrage calculation and alert tool covering Polymarket, Kalshi, and Robinhood, with focused functionality and free usage, suitable as a basic arbitrage assistant.

• Prediction Hunt: A cross-exchange prediction market aggregation and comparison tool, providing real-time price comparisons and arbitrage identification for Polymarket, Kalshi, and PredictIt (approximately 5-minute refresh), positioned for information symmetry and market inefficiency discovery.

# Trading Terminals/Aggregated Execution

• Verso: An institutional-level prediction market trading terminal supported by YC Fall 2024, providing a Bloomberg-style interface, covering real-time tracking of 15,000+ contracts from Polymarket and Kalshi, deep data analysis, and AI news intelligence, positioned for professional and institutional traders.

• Matchr: A cross-platform prediction market aggregation and execution tool covering 1,500+ markets, achieving optimal price matching through smart routing, and planning automated yield strategies based on high-probability events, cross-market arbitrage, and event-driven approaches, positioned for execution and capital efficiency.

• TradeFox: A professional prediction market aggregation and prime brokerage platform supported by Alliance DAO and CMT Digital, offering advanced order execution (limit orders, take profit/stop loss, TWAP), self-custody trading, and multi-platform smart routing, positioned for institutional-level traders, with plans to expand to Kalshi, Limitless, SxBet, and other platforms.

VI. Conclusion and Outlook

Currently, Prediction Market Agents are in the early exploration stage of development.

1. Market Foundation and Essence Evolution: Polymarket and Kalshi have formed a duopoly structure, providing sufficient liquidity and scenario foundations for building agents around them. The core difference between prediction markets and gambling lies in positive externalities, aggregating dispersed information through real transactions to publicly price real events, gradually evolving into a "global truth layer."

2. Core Positioning: Prediction Market Agents should be positioned as executable probabilistic asset management tools, with the core task of transforming news, rule texts, and on-chain data into verifiable pricing deviations, executing strategies with higher discipline, lower costs, and cross-market capabilities. The ideal architecture can be abstracted into four layers: information, analysis, strategy, and execution, but its actual tradability highly depends on the clarity of settlement, quality of liquidity, and degree of information structuring.

3. Strategy Selection and Risk Control Logic: From a strategy perspective, deterministic arbitrage (including resolution arbitrage, probability conservation arbitrage, and cross-platform price difference trading) is most suitable for automated execution by agents, while directional speculation can only serve as a supplement. In position management, executability and fault tolerance should be prioritized, with the tiered method combined with fixed position limits being the most suitable.

4. Business Models and Prospects: Commercialization mainly divides into three layers: the infrastructure layer obtaining stable B2B income through data execution infrastructure, the strategy layer monetizing through third-party strategy invocation or sharing, and the Agent/Vault layer participating in real trading under transparent on-chain risk control constraints, charging management and performance fees. Corresponding forms include entertainment entry, strategy subscription/signals (currently the most feasible), and high-threshold Vault custody, with "infrastructure + strategy ecosystem + performance participation" being a more sustainable path.

Although diverse attempts have emerged in the prediction market agent ecosystem, from underlying frameworks to upper-layer tools, a mature, replicable standardized product has yet to appear in key dimensions such as strategy generation, execution efficiency, risk control, and business closed loops. We look forward to the iteration and evolution of prediction market agents in the future.