Prediction markets allow users to trade on outcomes of real-world events, but are prone to fragmentation with overlapping questions, implicit equivalences, and hidden contradictions across markets. We present an agentic AI (AAI) pipeline that autonomously recovers cross-market structure from contract text before prices enter the analysis. The workflow first clusters markets into coherent topical groups using natural-language understanding over contract text and metadata, and then identifies contracts within each cluster, but from different event markets, that exhibit strong dependence or leader--follower relationships. We evaluate this system, along with a natural language inference (NLI) benchmark, on a large prediction market dataset from early 2026. Using resolved outcomes to evaluate identified relations, we find that AAI-identified relations are 62.8% consistent with exchange-recorded settlements, whereas the NLI benchmark only achieves 40.6% accuracy. Within clusters, the AAI output is sparse and also remarkably compatible as a signed graph with a frustration rate of 0.324%. As an application, we show how discovered relations inform semantics-based trading strategies on prediction markets. One such strategy yields 14.12% net ROI after fees in a two-month period in 2026. Overall, we demonstrate the potential for agentic AI as a structural discovery layer for prediction markets.
Figures & tables
AType
Field
Description
Market
event_title
Title of the event associated with the market
market_info
Market question and resolution criteria provided as input
MarketRelation
mkt_info_leader
Exact input text for the antecedent market
mkt_info_follower
Exact input text for the consequent market
direction
Boolean relation sign: true for the same outcome and false for opposite outcomes
strength
Relationship strength: high, medium, or low
Table 1. Agentics input and output ATypes and their target fields used in the workflow.
Relation type
Leader proposition
Follower proposition
Agent rationale
Logical, +
Trump is impeached and convicted by the Senate before Jan. 20, 2029
Trump leaves office before Jan. 20, 2029
Conviction removes the President, so leader-Yes implies follower-Yes.
Economic, +
Fed target upper bound exceeds 3.5% after the March meeting
Three-month Treasury par yield exceeds 3.5% at quarter end
The short yield is strongly influenced by the policy-rate target.
Exclusion, −
Crockett wins the Democratic primary by at least 9%
The general election is Talarico versus Paxton
Crockett’s nomination rules out a matchup requiring Talarico as nominee.
Table 2. Abbreviated high-strength Agentic relations from the March output. Questions are shortened only for presentation; endpoint resolution uses the full exact text.
Metric
NLI
AAI
Difference/test
Resolution (all)
40.6%
62.8%
+22.2 [14.0,36.9] pp
Shared resolved
25.7%
57.0%
block p=.0018
Positive only
60.6%
66.7%
+6.2 [ − 10.6,17.3] pp
Negative only
40.5%
39.0%
− 1.5 [ − 14.3,16.8] pp
Frustration L/∣E∣
5.88–16.16%
0.324%
NLI bounded; AAI exact
Table 3. Latest-snapshot pooled outcome tests and February–March graph diagnostics. Intervals are cluster-bootstrap 95% CIs for AAI minus NLI.
Parameter
Description
Value
gˉh
Minimum edge for high-strength relations
0.75
gˉm
Minimum edge for medium-strength relations
0.60
Bh
High-strength base notional
$800
Bm
Medium-strength base notional
$50
pˉ
Minimum eligible contract price
$0.03
C
Maximum exposure per cluster
$2,500
Table 4. Parameters of the semantic trading strategy.
Metric
Value
Metric
Value
Net PnL
$1,412.21
ROI
14.12%
Sharpe
4.46
Max drawdown
− 33.56%
Trades
46
Win rate
60.87%
Invested
$3,121.89
Fees
$190.44
Avg. PnL/trade
$30.70
Avg. hold
12.5 days
Table 5. Results of an investable performance estimate. Sharpe uses daily realized PnL divided by initial capital.