IOSG: Making Probability an Asset, Forecasting Market Intelligence Agent

Original Article Title: "IOSG Weekly Brief | Making Probability an Asset: Prediction Market Agent Outlook #315"

Original Article Author: Jacob Zhao, IOSG Ventures

In our past Crypto AI series reports, we have consistently emphasized the viewpoint: the most practical application value in the current crypto space is mainly focused on stablecoin payments and DeFi, with the Agent being the key interface of AI industry-facing users. Therefore, in the trend of Crypto and AI integration, the two most valuable paths are: in the short term, AgentFi based on existing mature DeFi protocols (basic strategies such as lending, liquidity mining, as well as advanced strategies such as Swap, Pendle PT, funding rate arbitrage), and in the medium to long term, Agent Payment centered around stablecoin settlement, relying on protocols such as ACP/AP2/x402/ERC-8004.

Prediction Markets have become an industry trend that cannot be ignored in 2025, with the annual total trading volume soaring from approximately $9 billion in 2024 to over $40 billion in 2025, achieving over 400% year-on-year growth. This significant growth is driven by multiple factors: macro political events bringing uncertain demand, maturation of infrastructure and trading patterns, and breakthroughs in the regulatory environment (Kalshi's legal victory and Polymarket's return to the U.S.). Prediction Market Agents are showing an early form in early 2026 and are expected to become a new product form in the field of agents in the coming year.

Prediction Markets: From Betting Tool to "Global Truth Layer"

A Prediction Market is a financial mechanism centered around trading on the outcome of future events, where the contract price essentially reflects the market's collective judgment of the event's probability. Its effectiveness stems from the combination of collective wisdom and economic incentives: in an environment of anonymous, real-money betting, dispersed information is rapidly integrated into a price signal weighted by fund intent, thereby significantly reducing noise and false judgments.

▲ Prediction Market Nominal Trading Volume Trend Chart Data Source: Dune Analytics (Query ID: 5753743)

By the end of 2025, the forecast market is predicted to have formed a duopoly dominated by Polymarket and Kalshi. According to Forbes, the total trading volume in 2025 is estimated to reach around $44 billion, with Polymarket contributing approximately $21.5 billion and Kalshi around $17.1 billion. Weekly data from February 2026 shows Kalshi's trading volume ($25.9B) surpassing Polymarket's ($18.3B), approaching a 50% market share. Kalshi has achieved rapid expansion due to a previous legal victory in an election contract case, a compliance-first-mover advantage in the U.S. sports prediction market, and relatively clear regulatory expectations. Currently, the development paths of the two have clearly diverged:

· Polymarket adopts a hybrid CLOB architecture with "off-chain matching, on-chain settlement" and a decentralized settlement mechanism, establishing a global, non-custodial, high-liquidity market. After returning to compliance in the U.S., it has formed an "onshore + offshore" dual-track operating structure;

· Kalshi integrates into the traditional financial system, connects via API to mainstream retail brokers, attracts Wall Street market makers to deeply participate in macro and data-driven contract trading, and its product is subject to traditional regulatory processes, with relatively lagging long-tail demand and unexpected events.

In addition to Polymarket and Kalshi, other competitive participants in the forecast market field are mainly developing along two paths:

· One is the Compliance Distribution Path, embedding event contracts into the existing accounts and clearing systems of brokers or large platforms. Relying on channel coverage, compliance qualifications, and institutional trust advantages (such as Interactive Brokers × ForecastEx's ForecastTrader, FanDuel × CME Group's FanDuel Predicts), compliance and resource advantages are significant, but products and user scale are still in early stages.

· The other is the Crypto-native On-chain Path, represented by Opinion.trade, Limitless, and Myriad, achieving rapid scaling through token mining, short-term contracts, and media distribution, emphasizing performance and capital efficiency. However, their long-term sustainability and risk resilience are yet to be validated.

The traditional financial compliance gateway and the performance advantages of crypto-native solutions together form a diverse competitive landscape in the forecast market ecosystem.

The Prediction Market, superficially similar to gambling, is essentially a zero-sum game. However, the core difference lies in whether it has positive externality: through real-money transactions, it aggregates and disseminates information, publicly prices real-world events, and forms a valuable signaling layer. The trend is shifting from a game to a "global truth layer" — with the participation of institutions such as CME and Bloomberg, event probabilities have become decision-making metadata that can be directly accessed by financial and corporate systems, providing a more timely, quantifiable market truth.

From the perspective of global regulation, the compliance path of Prediction Markets is highly differentiated. The United States is the only major economy that clearly includes Prediction Markets in the financial derivative regulatory framework. In markets such as Europe, the UK, Australia, and Singapore, Prediction Markets are generally considered gambling and regulatory scrutiny is increasing. On the other hand, countries like China and India completely prohibit Prediction Markets. The future global expansion of Prediction Markets still relies on the regulatory frameworks of each country.

Architecture Design of Prediction Market Intelligent Agents

Currently, Prediction Market Intelligent Agents are in the early stages of practical application. Their value lies not in "more accurate AI predictions" but in amplifying information processing and execution efficiency within Prediction Markets. Prediction Markets are essentially information aggregation mechanisms, where prices reflect the collective judgment of event probabilities. Market inefficiencies in reality stem from information asymmetry, liquidity, and attention constraints. The proper positioning of Prediction Market Intelligent Agents is Executable Probabilistic Portfolio Management: transforming news, rule text, and on-chain data into verifiable pricing deviations, executing strategies faster, more systematically, and at lower costs, capturing structural opportunities through cross-platform arbitrage and portfolio risk management.

An ideal Prediction Market Intelligent Agent can be abstracted into a four-layer architecture:

· Information Layer aggregates news, social, on-chain, and official data;

· Analysis Layer identifies mispricing and calculates Edge using LLM and ML;

· Strategy Layer transforms Edge into positions through the Kelly criterion, staged position building, and risk management;

· Execution Layer completes multi-market orders, slippage and Gas optimization, arbitrage execution, forming an efficient automated closed loop.

Prediction Market Agent Strategy Framework

Unlike traditional trading environments, prediction markets exhibit significant differences in settlement mechanisms, liquidity, and information distribution, and not all markets and strategies are suitable for automated execution. The core of a prediction market agent lies in whether it is deployed in a scenario that is clear in rules, codable, and aligns with its structural advantages. The following sections will analyze the target selection, position management, and strategy structure.

Prediction Market Target Selection

Not all prediction markets have tradable value, and their participation value depends on: clarity of settlement (clarity of rules, uniqueness of data source), quality of liquidity (market depth, spread, and volume), insider risk (degree of information asymmetry), time structure (expiration time and event rhythm), and the trader's own information advantage and professional background. Only when most dimensions meet the basic requirements does a prediction market have the foundation for participation, and participants should align with their own strengths and market characteristics:

· Human Core Advantage: Relies on expertise, judgment, and integration of ambiguous information, and operates in a relatively long time window (measured in days/weeks) market. Typical examples include political elections, macro trends, and corporate milestones.

· AI Agent Core Advantage: Relies on data processing, pattern recognition, and rapid execution, and operates in a market with an extremely short decision window (measured in seconds/minutes). Typical examples include high-frequency crypto pricing, cross-market arbitrage, and automated market making.

· Non-Adaptive Areas: Markets dominated by insider information or purely random/highly manipulative markets that do not provide an advantage to any participant.

Prediction Market Position Management

The Kelly Criterion is the most representative money management theory in repeated game scenarios. Its goal is not to maximize single-event returns but to maximize the long-term compounded growth rate of funds. This method, based on estimating win rate and odds, calculates the theoretically optimal position size to enhance capital growth efficiency under the premise of a positive expectation and is widely used in quantitative investing, professional gambling, poker, and asset management fields.

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

· Where f∗ is the optimal bet ratio, b is the net odds, p is the win rate, and q=1−p

· The prediction market can be simplified as: 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 probability and odds. In reality, traders find it challenging to consistently grasp the true probability. In practical operations, professional gamblers and prediction market participants tend to adopt rule-based strategies with higher executability and lower reliance on probability estimation:

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

· Flat Betting: Using a fixed fund proportion for each bet, emphasizing discipline and stability, suitable for risk-averse or low-confidence environments.

· Confidence Tiers: Setting predefined discrete position tiers and absolute limits to reduce decision complexity, avoiding the pseudoprecision issue of the Kelly model.

· Inverted Risk Approach: Starting from the maximum tolerable loss to deduce position size, originating from risk constraints rather than return expectations, forming a stable risk boundary.

For prediction market agents, strategy design should prioritize executability and stability rather than pursuing theoretical optimality. The key is to have clear rules, concise parameters, and tolerance for judgment errors. Under this constraint, combining Confidence Tiers with a fixed position limit is the most suitable general position management scheme for the PM Agent. This method does not rely on precise probability estimation but categorizes opportunities into limited tiers based on signal strength and corresponds to fixed positions; even in high-confidence scenarios, a clear limit is set to control risk.

Strategy Selection for Prediction Markets

From a structural perspective, prediction markets can mainly be divided into two major categories: deterministic arbitrage strategies characterized by clear rules and codability as features, and speculative directional strategies relying on information interpretation and directional judgment. Additionally, there are market making and hedging strategies dominated by professional institutions with higher capital and infrastructure requirements.

Deterministic Arbitrage Strategy

· Resolution Arbitrage: Resolution Arbitrage occurs when the outcome of an event is mostly determined, but the market has not fully priced it in yet. The main source of profit comes from information synchronization and execution speed. This strategy has clear rules, low risk, and can be entirely coded. It is considered the core strategy most suitable for Agent execution in a predictive market.

· Dutch Book Arbitrage: Dutch Book Arbitrage exploits the structural imbalance formed by the sum of prices of mutually exclusive and exhaustive event sets deviating from the probability conservation constraint (∑P≠1). By combining positions to lock in risk-free profits, this strategy solely relies on rules and price relationships. It has low risk, high level of rule-based structure, and is a typical form of deterministic arbitrage suitable for automated Agent execution.

· Cross-Platform Arbitrage: Cross-Platform Arbitrage profits from price deviations of the same event across different markets. It has low risk but requires high demands on latency and parallel monitoring. This strategy is suitable for Agent execution with infrastructure advantages, but intensified competition leads to continuously decreasing marginal returns.

· Bundle Arbitrage: Bundle Arbitrage involves trading based on pricing inconsistencies among related contracts. While the logic is clear, opportunities are limited. This strategy can be executed by Agents, but it requires certain engineering demands for rule parsing and portfolio constraints, with a moderate level of Agent adaptability.

Speculative Directional Strategy

· Information Trading Strategy: This type of strategy revolves around specific events or structured information, such as official data releases, announcements, or arbitration windows. As long as the information source is clear and trigger conditions are definable, Agents can leverage speed and discipline in monitoring and execution. However, when information shifts to semantic judgment or scenario interpretation, human intervention is still necessary.

· Signal Following Strategy: This strategy seeks returns by following accounts or fund behaviors with historically superior performance. The rules are relatively simple and can be automated. The main risk lies in signal degradation and being exploited in the opposite direction, thus requiring filtering mechanisms and strict position management. It is suitable as an ancillary strategy for Agents.

· Unstructured / Noise-driven Strategy: This type of strategy heavily relies on emotion, randomness, or crowd behavior, lacking a stable and replicable edge with an unstable long-term expected value. Due to its difficulty to model, high risk, it is not suitable for systematic Agent execution and is not recommended as a long-term strategy.

· Market Microstructure Strategy: This type of strategy depends on a very short decision window, continuous quoting, or high-frequency trading, requiring high performance in latency, modeling, and capital. Although theoretically suitable for Agents, in practice, it is often constrained by liquidity and competitive intensity in predicting markets, only suitable for a few participants with significant infrastructure advantages.

· Risk Control & Hedging Strategy: This type of strategy does not directly pursue returns but is used to reduce overall risk exposure. With clear rules and objectives, it operates as a fundamental risk control module in the long run.

Overall, the strategies suitable for Agent execution in predictive markets focus on scenarios that are rule-based, codable, and involve weak subjective judgment, where deterministic arbitrage should serve as the core source of profit, structured information, and signal-following strategies as supplements, and high-noise and emotion-driven trading activities should be systematically excluded. The long-term advantage of Agents lies in high-discipline, high-speed execution, and risk control capabilities.

Predictive Market Agent Business Model and Product Form

The ideal business model design of a predictive market agent explores different directions at different levels:

· Infrastructure Layer, providing multi-source real-time data aggregation, Smart Money address database, a unified predictive market execution engine, and backtesting tools, charging B2B fees to generate stable income independent of prediction accuracy;

· Strategy Layer, introducing community and third-party strategies to build a reusable, evaluable strategy ecosystem, and capturing value through invocation, weighting, or execution revenue sharing to reduce reliance on a single Alpha.

· Agent / Vault Layer, where agents participate in live trading directly in a trust-managed manner, relying on on-chain transparent records and a strict risk management system, charging management and performance fees for monetization.

And the product form corresponding to different business models can also be divided into:

· Gamification / Gaming Mode: By reducing the participation threshold through intuitive interaction like Tinder, it has the strongest user growth and market education capabilities, making it an ideal entry point for breaking the circle. However, it needs to transition to a subscription or execution-based product monetization.

· Strategic Subscription / Signal Mode: It does not involve fund custody, is regulation-friendly, and has clear rights and responsibilities. The SaaS revenue structure is relatively stable, making it the most feasible commercialization path at the current stage. Its limitations include easy replication of strategies, execution loss, limited long-term revenue ceiling, which can be significantly improved through a "signal + one-click execution" semi-automated form to enhance the experience and retention.

· Vault Custody Mode: It has the advantages of economies of scale and execution efficiency, with a form similar to asset management products. However, it faces multiple structural constraints such as asset management licenses, trust thresholds, and centralized technological risks. The business model highly relies on market conditions and continuous profitability. Unless it has long-term performance and institutional endorsement, it is not suitable as the main path.

Overall, a multi-dimensional revenue structure of "Infrastructure Monetization + Strategic Ecosystem Expansion + Performance Participation" helps reduce reliance on a single assumption of "AI continuously outperforming the market." Even if Alpha converges with market maturity, fundamental capabilities such as execution, risk control, and settlement still hold long-term value, thereby building a more sustainable business loop.

Case Study of Prediction Market Agents

Currently, Prediction Market Agents are still in the early exploration stage. Although the market has seen a variety of attempts from underlying frameworks to upper-level tools, a standardized product that is mature in strategy generation, execution efficiency, risk control system, and business loop has not yet been formed.

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

Infrastructure Layer

· Polymarket Agents Framework

Polymarket Agents

An official framework released by Polymarket with the aim of addressing the engineering standardization issue of "connection and interaction." This framework encapsulates market data retrieval, order construction, and a basic LLM call interface. It solves the problem of "how to place orders with code," but leaves significant gaps in core trading capabilities such as strategy generation, probability calibration, dynamic position management, and backtesting systems. It is more like an officially recognized "access specification" rather than a finished product with Alpha returns. Business-level Agents still need to build a complete research and risk control core on top of this foundation.

· Gnosis Prediction Market Tool

The Gnosis Prediction Market Agent Tooling (PMAT) provides full read/write support for Omen/AIOmen and Manifold but only offers read-only access for Polymarket, showing a clear ecosystem barrier. It is suitable as a foundational tool for Gnosis ecosystem Agents, but its utility is limited for developers focusing on Polymarket as the main battlefield.

Polymarket and Gnosis are currently the prediction market ecosystems that have explicitly productized "Agent development" into an official framework. Other prediction markets like Kalshi mainly remain at the API and Python SDK level, requiring developers to independently supplement key system capabilities such as strategies, risk control, operation, and monitoring.

Autonomous Agents

Current "Prediction Market AI Agents" in the market are still mostly in the early stages. Although named "Agents," their actual capabilities are significantly far from an empowered automated closed-loop trading system, generally lacking an independent, systematic risk control layer. They do not incorporate position management, stop-loss, hedging, and expected value constraints into the decision-making process, and their overall level of productization is relatively low, with no mature long-running system yet formed.

· Olas Predict

Olas Predict is the most highly productized prediction market intelligent agent ecosystem currently. Its core product, Omenstrat, is based on Gnosis's internal Omen, built on FPMM and decentralized arbitration mechanism, supporting small-scale high-frequency interactions but limited by Omen's insufficient single-market liquidity. Its "AI prediction" mainly relies on a general LLM, lacks real-time data and systematic risk control, and shows significant differentiation in historical win rates across categories. In February 2026, Olas launched Polystrat, expanding Agent capabilities to Polymarket—users can set strategies using natural language, and the Agent automatically identifies probability deviations in markets settling within 4 days and executes trades. The system runs locally through Pearl, uses self-hosted Safe accounts and hard-coded limit controls to manage risk, making it the first consumer-grade autonomous trading Agent tailored to Polymarket.

· UnifAI Network Polymarket Strategy

Provides a Polymarket automated trading Agent with a core tail risk acceptance strategy: scans for nearby settlement contracts with implied probabilities >95% and buys in to target a 3–5% spread. On-chain data shows a win rate close to 95%, but returns are significantly diversified 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 loop, with an architecture covering an intelligence layer, an abstraction layer, and an execution layer. Omnichain Vaults have already been delivered; the Prediction Market Agent is still in the development stage, not yet forming a complete mainnet loop, and the overall project is in a vision validation period.

Prediction Market Tools

Current prediction market analysis tools are not sufficient to constitute a complete "prediction market agent." Their value mainly lies in the information and analysis layers of the agent architecture, while trade execution, position management, and risk control still need to be borne by the trader. In terms of product form, it is more in line with the positioning of "strategy subscription / signal assistance / research enhancement" and can be seen as an early prototype of a prediction market agent.

Through a systematic review and empirical selection of projects included in Awesome Prediction Market Tools, this article selects representative projects that already have a preliminary product form and usage scenarios as research report cases. It mainly focuses on four directions: analysis and signal layers, alert and whale tracking systems, arbitrage discovery tools, and trading terminals and aggregate execution.

· Market Analysis Tools

Polyseer: A research-oriented prediction market tool that adopts a multi-agent division of labor architecture (Planner / Researcher / Critic / Analyst / Reporter) for bilateral evidence collection and Bayesian probability aggregation, producing structured research reports. Its strengths lie in methodological transparency, process engineering, and being fully open-source and auditable.

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

Polymarket Analytics: Globalized Polymarket data analytics platform, systematically displaying trader, market, position, and transaction data, with a clear focus and intuitive data, suitable as a basic data query and research reference.

Hashdive: Trader-focused data tool that quantitatively filters traders and markets through Smart Score and multi-dimensional Screener, practical for "smart money identification" and copy-trading decisions.

Polyfactual: Focuses on AI market intelligence and sentiment/risk analysis, embedding analysis results into the trading interface via a Chrome extension, leaning towards B2B and institutional user scenarios.

Predly: AI mispricing detection platform, identifying mispricing between Polymarket and Kalshi by comparing market prices with AI-calculated probabilities, with an alleged alert accuracy rate of 89%, positioned for signal discovery and opportunity screening.

Polysights: Covers 30+ markets and on-chain indicators, with an Insider Finder to track new wallets, large one-way bets, and other abnormal behaviors, suitable for daily monitoring and signal discovery.

PolyRadar: Multi-model parallel analysis platform providing real-time interpretation, timeline evolution, confidence scoring, and source transparency for a single event, emphasizing multi-AI cross-validation and positioned as an analysis tool.

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

· Alerts/Whale Tracking

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

Whale Tracker Livid: Productizing whale position changes.

· Arbitrage Discovery Tools

ArbBets: 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 at a high-frequency opportunity scanning layer.

PolyScalping: Real-time arbitrage and scalping analysis platform for Polymarket, supporting full-market scans every 60 seconds, ROI calculation, and Telegram push notifications, and allowing opportunity filtering based on dimensions such as liquidity, spread, and volume, catering to active traders.

Eventarb: Lightweight cross-platform arbitrage calculation and alert tool, covering Polymarket, Kalshi, and Robinhood, feature-focused, free to use, suitable as a basic arbitrage aid.

Prediction Hunt: Cross-exchange prediction market aggregation and comparison tool, providing real-time price comparison and arbitrage identification for Polymarket, Kalshi, and PredictIt (refreshed approximately every 5 minutes), focused on information symmetry and market inefficiency discovery.

· Trading Terminal/Aggregated Execution

Verso: Institution-grade prediction market trading terminal supported by YC Fall 2024, offering a Bloomberg-style interface, covering real-time tracking of 15,000+ contracts on Polymarket and Kalshi, deep data analysis, and AI news intelligence, targeted at professional and institutional traders.

Matchr: Cross-platform prediction market aggregation and execution tool, covering 1,500+ markets, achieving optimal price matching through intelligent routing, and designing automated revenue strategies based on high-probability events, cross-market arbitrage, and event-driven opportunities, positioned at the execution and capital efficiency layer.

TradeFox: 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 orders, TWAP), self-custody trading, and multi-platform smart routing, targeting institutional traders, with plans to expand to platforms such as Kalshi, Limitless, and SxBet.

Summary and Outlook

Currently, the Prediction Market Agent is in the early stages of development exploration.

1. Market Foundation and Essence Evolution: Polymarket and Kalshi have formed a duopoly, building intelligent agents around them with sufficient liquidity and scene foundation. The core difference between prediction markets and gambling lies in positive externality. Through real trades aggregating dispersed information, they publicly price real-world events, gradually evolving into a "global truth layer."

2. Core Positioning: Prediction Market Agents should be positioned as executable probability asset management tools. Their core task is to transform news, rule text, and on-chain data into verifiable pricing deviations, and execute strategies with higher discipline, lower cost, and cross-market capability. The ideal architecture can be abstracted into four layers: information, analysis, strategy, and execution, but their actual tradability highly depends on the clarity of settlement, the quality of liquidity, and the degree of information structuring.

3. Strategy Selection and Risk Control Logic: From a strategic perspective, deterministic arbitrage (including settlement arbitrage, probability conservation arbitrage, and cross-platform price difference trading) is best suited for automated execution by agents, while directional speculation can only serve as a supplement. In position management, executable and fault-tolerant considerations should take precedence, with a step-ladder method combined with a fixed position limit being most suitable.

4. Business Model and Prospects: Commercialization is mainly divided into three layers: the infrastructure layer obtains stable B2B income through data execution infrastructure, the strategy layer monetizes through third-party strategy calls or profit sharing, and the Agent/Vault layer participates in spot trading under on-chain transparent risk constraints and charges management fees and performance fees. Corresponding forms include entertainment entry points, strategy subscription/signals (currently the most feasible), and high-threshold Vault custody. "Infrastructure + Strategy Ecosystem + Performance Participation" is a more sustainable path.

Although various attempts have emerged in the Prediction Market Agent ecosystem from underlying frameworks to upper-layer tools, in key dimensions such as strategy generation, execution efficiency, risk control, and commercial closed-loop, there is currently no mature, replicable standardized product. We look forward to the iteration and evolution of Prediction Market Agents in the future.

Original Article Link