IOSG; Turning Probability into Assets: Predictive Market Agent Outlook

Author: Jacob Zhao @IOSG

In our previous Crypto AI research reports, we have consistently emphasized that the most practically valuable applications in the current crypto space are primarily concentrated in stablecoin payments and DeFi, with agents serving as the key user interface for the AI ​​industry. 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 and liquidity mining, as well as advanced strategies such as Swap, Pendle PT, and funding rate arbitrage); and in the medium to long term, Agent Payment centered around stablecoin settlement and relying on protocols such as ACP/AP2/x402/ERC-8004.

Prediction markets have become an undeniable new industry trend in 2025, with total annual trading volume surging from approximately $9 billion in 2024 to over $40 billion in 2025, representing a year-on-year growth of over 400%. This significant growth was driven by multiple factors: demand arising from uncertainty caused by macro-political events, the maturation of infrastructure and trading models, and a breakthrough in the regulatory environment (Kalshi's victory and Polymarket's return to the US). Prediction market agents are expected to take early shape in early 2026 and are poised to become a new product form in the agent field within the coming year.

1. Prediction Markets: From Betting Tools to the “Global Layer of Truth”

Prediction markets are financial mechanisms that facilitate trading around the outcomes of future events. Contract prices essentially reflect the market's collective judgment on the probability of these events occurring. Their effectiveness stems from the combination of collective wisdom and economic incentives: in an environment of anonymity and real-money betting, dispersed information is rapidly integrated into price signals weighted by willingness to pay, thereby significantly reducing noise and false judgments.

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

By the end of 2025, the prediction market had largely formed a duopoly dominated by Polymarket and Kalshi. According to Forbes, total trading volume in 2025 reached approximately $44 billion, with Polymarket contributing about $21.5 billion and Kalshi about $17.1 billion. Weekly data from February 2026 showed that Kalshi's trading volume ($25.9 billion) had surpassed Polymarket's ($18.3 billion), approaching 50% market share. Kalshi achieved rapid expansion thanks to its previous legal victory in the election contract case, its first-mover advantage in compliance in the US sports prediction market, and relatively clear regulatory expectations. Currently, the development paths of the two have clearly diverged:

• Polymarket adopts a hybrid CLOB architecture of "off-chain matching and on-chain settlement" and a decentralized settlement mechanism to build a global, non-custodial, highly liquid market. After returning to the United States in compliance with regulations, it has formed a dual-track operation structure of "onshore + offshore".

• Kalshi integrates into the traditional financial system by connecting to mainstream retail brokerages via APIs, attracting Wall Street market makers to participate deeply in macro and data-driven contract trading. However, its products are subject to traditional regulatory processes, resulting in a relative lag in long-tail demand and response to unforeseen events.

Aside from Polymarket and Kalshi, other competitive players in the prediction market have primarily developed along two paths:

• One approach is the compliant distribution path, which embeds event contracts into the existing accounts and clearing systems of securities firms or large platforms. It leverages advantages such as channel coverage, compliance qualifications, and institutional trust to establish advantages (e.g., ForecastTrader from Interactive Brokers × ForecastEx, and FanDuel Predicts from FanDuel × CME Group). While these approaches offer significant compliance and resource advantages, the product and user scale are still in their early stages.

• The second is the Crypto native on-chain path, represented by Opinion.trade, Limitless, and Myriad. It achieves rapid scaling through points mining, short-cycle contracts, and media distribution, emphasizing performance and capital efficiency. However, its long-term sustainability and risk control robustness still need to be verified.

The traditional financial compliance entry point and the native performance advantages of cryptography together constitute the diverse competitive landscape of the prediction market ecosystem.

Prediction markets superficially resemble gambling, essentially being zero-sum games. However, the core difference lies in whether they possess positive externalities: aggregating dispersed information through real-money transactions, publicly pricing real-world events, and forming a valuable signal layer. The trend is shifting from game theory to a "global truth layer"—with the integration of institutions like CME and Bloomberg, event probabilities have become decision metadata that can be directly accessed by financial and corporate systems, providing more timely and quantifiable market-based truths.

From a global regulatory perspective, the compliance paths for prediction markets are highly divergent. The United States is the only major economy that has explicitly included prediction markets in its financial derivatives regulatory framework. Markets in Europe, the United Kingdom, Australia, and Singapore generally regard them as gambling and tend to tighten regulations, while China and India have completely banned them. The future global expansion of prediction markets will still depend on the regulatory frameworks of each country.

2. Architecture design of predictive market intelligent agents

Currently, prediction market agents are entering their early practical phase. Their value lies not in "more accurate AI predictions," but in amplifying the efficiency of information processing and execution in prediction markets. Prediction markets are essentially information aggregation mechanisms, where prices reflect collective judgments about the probability of events. Real-world market inefficiencies stem from information asymmetry, liquidity constraints, and attentional limitations. The appropriate positioning for prediction market agents is executable probabilistic portfolio management: transforming news, rule texts, and on-chain data into verifiable pricing biases, executing strategies faster, more disciplined, and at lower cost, and capturing structural opportunities through cross-platform arbitrage and portfolio risk control.

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

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

• The analysis layer uses LLM and ML to identify mispriced items and calculate the edge;

• The strategy layer converts Edge into positions through the Kelly Criterion, phased position building, and risk control;

• The execution layer completes multi-market order placement, slippage and gas optimization, and arbitrage execution, forming a highly efficient and automated closed loop.

3. Strategy Framework for Predictive Market Agents

Unlike traditional trading environments, prediction markets differ significantly 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 clearly defined rules, coded functionality, and that align with their structural advantages. The following analysis will focus on three aspects: target selection, position management, and strategy structure.

Predicting market target selection

Not all prediction markets possess tradable value. Their participation value depends on: settlement clarity (whether the rules are clear and the data source is unique), liquidity quality (market depth, spreads, and trading volume), insider risk (degree of information asymmetry), time structure (expiration dates and event timing), and the trader's own informational advantage and professional background. Only when most of these dimensions meet the basic requirements can a prediction market be considered a viable entry point. Participants should match their own strengths with the market characteristics.

• Human core strengths lie in markets that rely on expertise, judgment, and the integration of ambiguous information, and have relatively flexible time windows (measured in days or weeks). Typical examples include political elections, macroeconomic trends, and corporate milestones.

• The core advantages of AI agents lie in their reliance on data processing, pattern recognition, and rapid execution in markets with extremely short decision windows (measured in seconds or minutes). Typical examples include high-frequency encrypted pricing, cross-market arbitrage, and automated market making.

• Unsuitable markets: Markets dominated by insider information or purely random/highly manipulated, which do not offer an advantage to any participant.

Position management in prediction markets

The Kelly Criterion is the most representative money management theory in repeated game scenarios. Its goal is not to maximize the return of a single game, but to maximize the long-term compound growth rate of capital. Based on the estimation of win rate and odds, this method calculates the theoretically optimal position ratio, improving capital growth efficiency under the premise of positive expected value. It is widely used in quantitative investment, professional betting, 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

• Market prediction 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 validity of the Kelly Criterion relies heavily on accurate estimates of true probabilities and odds. In reality, traders struggle to consistently and accurately grasp true probabilities. Therefore, professional gamblers and market participants tend to employ more actionable, rule-based strategies that rely less on probability estimation.

• Unit System (Unit Betting): This method involves dividing funds into fixed units (such as 1%) and betting different numbers of units based on confidence level. The unit limit automatically constrains the risk of each bet and is the most common practice.

• Flat Betting: A fixed percentage of funds is used for each bet, emphasizing discipline and stability. It is suitable for risk-averse or low-confidence environments.

• Confidence Tiers: Preset discrete position tiers and set absolute upper limits to reduce decision-making complexity and avoid the pseudo-precision problem of the Kelly model.

• The Inverted Risk Approach: Starting with the maximum acceptable loss, the position size is determined by working backward, forming a stable risk boundary based on risk constraints rather than expected returns.

For market prediction agents, strategy design should prioritize feasibility and stability rather than pursuing theoretical optimality. The key lies in clear rules, concise parameters, and tolerance for judgment errors. Under these constraints, the tiered confidence method combined with a fixed position cap is the most suitable general position management solution for PM agents. This method does not rely on precise probability estimation but instead divides opportunities into finite tiers based on signal strength and assigns corresponding fixed positions; even in high-confidence scenarios, a clear upper limit is set to control risk.

Market prediction strategy selection

From a strategic structure perspective, market prediction can be mainly divided into two categories: deterministic arbitrage strategies characterized by clear and coded rules, and speculative directional strategies that rely on information interpretation and directional judgment. In addition, there are market-making and hedging strategies that are mainly operated by professional institutions and require high levels of capital and infrastructure.

Deterministic arbitrage strategy

• Resolution arbitrage: Resolution arbitrage occurs when the outcome of an event is largely determined, but the market has not yet fully priced it in. Profits primarily come from information synchronization and execution speed. This strategy has clear rules, low risk, and is fully coded, making it the core strategy most suitable for agents to execute in prediction markets.

• Dutch Book Arbitrage: Dutch Book arbitrage leverages the structural imbalance created by the deviation of the sum of prices of mutually exclusive and complete sets of events from the probability conservation constraint (∑P≠1) to lock in directionless risk and return through portfolio construction. This strategy relies solely on the relationship between rules and prices, has low risk, and is highly rule-based, making it a typical deterministic arbitrage form suitable for automated agent execution.

• Cross-platform arbitrage: Cross-platform arbitrage profits by capturing pricing discrepancies of the same event across different markets. It carries lower risk but requires high levels of latency and parallel monitoring. This strategy is suitable for agents with infrastructure advantages, but increasing competition leads to a continuous decline in marginal returns.

• Bundle arbitrage: Bundle arbitrage exploits pricing discrepancies between related contracts. The logic is clear, but opportunities are limited. This strategy can be executed by an agent, but it requires some engineering expertise in rule parsing and combination constraints, and agent adaptability is moderate.

Speculative directional strategies

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

• Signal Following: This strategy generates profits by following the historically high-performing accounts or funds. The rules are relatively simple and can be automated. Its core risks lie in signal degradation and misuse, thus requiring filtering mechanisms and strict position management. It is suitable as a supplementary strategy to an agent.

• Unstructured/Noise-driven strategies: These strategies heavily rely on emotions, randomness, or participation behavior, lack a stable and reproducible edge, and have unstable long-term expected values. Due to their difficulty in modeling and extremely high risk, they are unsuitable for systematic agent execution and are not recommended as long-term strategies.

High-frequency price and liquidity strategies (Market Microstructure): These strategies rely on extremely short decision windows, continuous quotes, or high-frequency trading, placing extremely high demands on latency, models, and capital. While theoretically suitable for agents, they are often limited by liquidity and competitive intensity in prediction markets, making them suitable only for a few participants with significant infrastructure advantages.

Risk Control & Hedging: These strategies do not directly pursue returns but rather aim to reduce overall risk exposure. They have clearly defined rules and objectives, and operate long-term as a foundational risk control module.

Overall, suitable strategies for agents in prediction markets are concentrated in scenarios with clear rules, coded patterns, and low subjective judgment. Certainty arbitrage should be the core source of profit, supplemented by structured information and signal-following strategies. High-noise and emotion-driven trading should be systematically excluded. The long-term advantage of agents lies in their highly disciplined, high-speed execution and risk control capabilities.

4. Predicting the business models and product forms of intelligent agents in the market.

The design of ideal business models for predictive market agents offers opportunities for exploration in different directions at different levels:

• The infrastructure layer provides multi-source real-time data aggregation, a Smart Money address database, a unified prediction market execution engine and backtesting tools, and charges B2B companies to obtain stable revenue unrelated to prediction accuracy.

• The Strategy layer introduces community and third-party strategies to build a reusable and evaluable strategy ecosystem. It also captures value through invocation, weighting, or execution sharing, thereby reducing dependence on a single Alpha.

• In the Agent/Vault layer, smart agents directly participate in live trading through entrusted management, relying on transparent on-chain records and a strict risk control system to collect management fees and performance fees.

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

• Entertainment/Gamification Model: By lowering the barrier to entry through intuitive interaction similar to Tinder, it possesses the strongest user growth and market education capabilities, making it an ideal entry point for breaking into new markets. However, it needs to be integrated into subscription or action-oriented product monetization.

• Strategy subscription/signal model: This model does not involve fund custody, is regulatory-friendly, has clear responsibilities, and offers a relatively stable SaaS revenue structure, making it the most feasible commercialization path at the current stage. Its limitations lie in the ease with which strategies can be copied, the inefficiencies in execution, and the limited long-term revenue ceiling. However, a semi-automated approach using "signals + one-click execution" can significantly improve user experience and retention.

• The Vault custody model offers advantages in scale and execution efficiency, resembling asset management products in form. However, it faces multiple structural constraints, including asset management licenses, trust barriers, and centralized technology risks. Its business model is highly dependent on market conditions and sustained profitability. Unless it possesses long-term performance and institutional backing, it is not suitable as a primary approach.

Overall, the diversified revenue structure of "infrastructure monetization + strategic ecosystem expansion + performance participation" helps reduce reliance on the single assumption that "AI will continue to outperform the market." Even if Alpha converges as the market matures, underlying capabilities such as execution, risk control, and settlement will still have long-term value, thereby building a more sustainable business loop.

5. Case studies of predictive market agents

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

We divide our current ecosystem into three layers: Infrastructure, Autonomous Agents, and Prediction Market Tools.

Infrastructure layer

#

Polymarket Agents Framework

Polymarket Agents, an official developer framework from Polymarket, aims to address the standardization of engineering aspects related to "connectivity and interaction." This framework encapsulates market data acquisition, order creation, and basic LLM (Local Management Model) API calls. It solves the problem of "how to place orders with code," but leaves core trading capabilities—such as strategy generation, probability calibration, dynamic position management, and backtesting systems—basically undeveloped. It's more like an officially recognized "access specification" than a finished product with alpha returns. Commercial-grade agents still need to build their own complete investment research and risk control kernels on top of this framework.

#

Gnosis Predictive Market Tools

Gnosis Prediction Market Agent Tooling (PMAT) provides full read/write support for Omen/AIOmen and Manifold, but only grants read-only access to Polymarket, creating a clear ecosystem barrier. It is suitable as a cornerstone for agent development within the Gnosis ecosystem, but its practicality is limited for developers whose primary focus is Polymarket.

Polymarket and Gnosis are currently the only prediction market ecosystems that have explicitly productized "Agent development" into an official framework. Other prediction markets, such as Kalshi, still mainly rely on APIs and Python SDKs, requiring developers to complete key system capabilities such as strategies, risk control, operation, and monitoring themselves.

Autonomous Agent

Most of the current "prediction market AI agents" are still in the early stages. Although they are called "Agents", their actual capabilities are still far from being able to automate closed-loop trading. They generally lack an independent and systematic risk control layer and have not incorporated position management, stop loss, hedging and expected value constraints into the decision-making process. The overall productization level is low and a mature system that can be operated in the long term has not yet been formed.

#

Olas Predict

Olas Predict is currently the most productized prediction market AI agent ecosystem. Its core product, Omenstrat, is built on Omen within the Gnosis ecosystem, employing an FPMM and decentralized arbitration mechanism at its core. It supports small-amount, high-frequency interactions, but is limited by insufficient liquidity in the Omen single market. Its "AI prediction" primarily relies on general LLM, lacking real-time data and systematic risk control, resulting in significant historical win rates across different product categories. In February 2026, Olas launched Polystrat, extending Agent capabilities to Polymarket—users can set strategies using natural language, and the Agent automatically identifies probability deviations in the settlement market within four days and executes trades. The system operates locally on Pearl, uses a self-hosted Safe account, and controls risk through hard-coded restrictions, making it the first consumer-grade autonomous trading agent for Polymarket.

#

UnifAI Network Polymarket Strategy

Polymarket offers an automated trading agent with a core tail-risk-taking strategy: scanning for and buying near-settlement contracts with an implied probability >95%, aiming to capture a 3-5% price spread. On-chain data shows a win rate close to 95%, but returns vary significantly across different asset classes, indicating that the strategy is highly dependent on execution frequency and asset class selection.

#

NOYA.ai

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

Prediction Market Tools

Current market prediction analysis tools are insufficient to constitute a complete "market prediction agent." Their value is mainly concentrated in the information and analysis layers of the agent's architecture, while trade execution, position management, and risk control still need to be borne by the trader. From a product perspective, they are more in line with the positioning of "strategy subscription/signal assistance/research enhancement" and can be regarded as an early prototype of a market prediction agent.

Through a systematic review and empirical screening of projects included in Awesome-Prediction-Market-Tools, this paper selects representative projects that have already achieved preliminary product form and use cases as case studies for the research report. These mainly focus on four areas: analysis and signal layer, alert and whale tracking systems, arbitrage discovery tools, and trading terminals and aggregated execution.

#

Market analysis tools

• Polyseer is a research-based market prediction tool that employs a multi-agent architecture (Planner/Researcher/Critic/Analyst/Reporter) for bilateral evidence gathering and Bayesian probabilistic aggregation, outputting structured research reports. Its advantages lie in its transparent methodology, engineered processes, and fully open-source and auditable nature.

• Oddpool: Positioned as a "Bloomberg terminal for prediction markets," it provides cross-platform aggregation, arbitrage scanning, and real-time data dashboards for Polymarket, Kalshi, CME, and other platforms.

• Polymarket Analytics: A global data analytics platform from Polymarket, it systematically displays trader, market, position, and transaction data. With clear positioning and intuitive data, it is suitable for basic data queries and research references.

• Hashdive: A data tool for traders that uses Smart Score and multi-dimensional Screener to quantitatively screen traders and the market, making it practical for "smart money identification" and copy trading decisions.

• Polyfactual focuses on AI-powered market intelligence and sentiment/risk analysis, embedding analysis results into the trading interface via a Chrome extension, and is geared towards B2B and institutional user scenarios.

• Predly: An AI-powered mispricing detection platform that identifies pricing discrepancies between Polymarket and Kalshi by comparing market prices with AI-calculated probabilities. The platform claims an 89% accuracy rate in alerts and focuses on signal discovery and opportunity screening.

• Polysights: Covers 30+ market and on-chain metrics, and uses Insider Finder to track unusual behavior such as new wallets and large single bets, making it suitable for daily monitoring and signal discovery.

• PolyRadar: A multi-model parallel analysis platform that provides real-time interpretation, timeline evolution, confidence scores, and source transparency for single events, emphasizing multi-AI cross-validation and positioning analysis tools.

• Alphascope: An AI-driven predictive market intelligence engine that provides real-time signals, research summaries, and probability change monitoring. It is still in its early stages and focuses on research and signal support.

#

Alert/Whale Tracking

• Stand: Clearly define whale copy trading and provide high-confidence action alerts.

• Whale Tracker Livid: Productizing Whale Position Changes

#

Arbitrage discovery tools

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

• PolyScalping: A real-time arbitrage and scalping analysis platform for Polymarket, supporting full market scanning every 60 seconds, ROI calculation and Telegram push notifications. It can also filter opportunities by liquidity, spreads and trading volume, and is geared towards active traders.

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

• Prediction Hunt: A cross-exchange prediction market aggregation and comparison tool that provides real-time price comparison and arbitrage identification for Polymarket, Kalshi, and PredictIt (refreshes approximately every 5 minutes), focusing on information symmetry and market inefficiency detection.

#

Transaction Terminal/Aggregated Execution

• Verso: An institutional-grade prediction market trading terminal supported by YC Fall 2024, offering a Bloomberg-style interface, real-time tracking of 15,000+ contracts from Polymarket and Kalshi, in-depth data analysis, and AI-powered news intelligence, targeting professional and institutional traders.

• Matchr: A cross-platform prediction market aggregation and execution tool covering 1,500+ markets. It achieves optimal price matching through intelligent routing and plans automated profit strategies based on high-probability events, cross-market arbitrage, and event-driven strategies, focusing on execution and capital efficiency.

• TradeFox: A professional prediction market aggregator and Prime Brokerage platform powered by Alliance DAO and CMT Digital, offering advanced order execution (limit orders, stop-loss and take-profit orders, TWAP), self-managed trading, and multi-platform smart routing. Targeting institutional traders, it plans to expand to platforms such as Kalshi, Limitless, and SxBet.

6. Summary and Outlook

Currently, prediction market agents are in the early exploratory stage of development.

1. Market Foundation and Essential Evolution: Polymarket and Kalshi have formed a duopoly structure, providing ample liquidity and a solid foundation for building intelligent agents around them. The core difference between prediction markets and gambling lies in positive externalities. By aggregating dispersed information through real transactions, public pricing of real-world events is achieved, gradually evolving into a "global truth layer."

2. Core Positioning: Predictive market agents should be positioned as executable probabilistic asset management tools. Their core task is to transform news, rule texts, and on-chain data into verifiable pricing biases, and to execute strategies with greater discipline, lower costs, and cross-market capabilities. The ideal architecture can be abstracted into four layers: information, analysis, strategy, and execution. However, its 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 strategy perspective, deterministic arbitrage (including settlement arbitrage, probability conservation arbitrage, and cross-platform spread trading) is best suited for automated execution by intelligent agents, while directional speculation can only serve as a supplement. In position management, feasibility and fault tolerance should be prioritized; a tiered approach combined with a fixed position cap is most suitable.

4. Business Model and Prospects: Commercialization is mainly divided into three layers: the infrastructure layer generates stable B2B revenue through data-driven infrastructure; the strategy layer monetizes through third-party strategy calls or revenue sharing; and the Agent/Vault layer participates in live trading under transparent on-chain risk control constraints and collects management and performance fees. Corresponding forms include entertainment-oriented entry points, strategy subscriptions/signals (currently the most feasible), and high-barrier Vault hosting. The "infrastructure + strategy ecosystem + performance participation" model represents a more sustainable path.

Despite the emergence of diverse attempts in the prediction market agent ecosystem, ranging from underlying frameworks to upper-level tools, there are currently no mature, replicable, standardized products in key dimensions such as strategy generation, execution efficiency, risk control, and business closed loop. We look forward to the iteration and evolution of prediction market agents in the future.