Paper: arXiv 2602.07048

Authors: Sumin Kim, Minjae Kim, Jihoon Kwon, Yoon Kim, Nicole Kagan, Joo Won Lee, Oscar Levy, Alejandro Lopez-Lira, Yongjae Lee, Chanyeol Choi

Abstract

Prediction markets provide a unique setting where event-level time series are directly tied to natural-language descriptions, yet discovering robust lead-lag relationships remains challenging due to spurious statistical correlations. We propose a hybrid two-stage causal screener to address this challenge: (i) a statistical stage that uses Granger causality to identify candidate leader-follower pairs from market-implied probability time series, and (ii) an LLM-based semantic stage that re-ranks these candidates by assessing whether the proposed direction admits a plausible economic transmission mechanism based on event descriptions. Because causal ground truth is unobserved, we evaluate the ranked pairs using a fixed, signal-triggered trading protocol that maps relationship quality into realized profit and loss (PnL). On Kalshi Economics markets, our hybrid approach consistently outperforms the statistical baseline. Across rolling evaluations, the win rate increases from 51.4% to 54.5%. Crucially, the average magnitude of losing trades decreases substantially from 649 USD to 347 USD. This reduction is driven by the LLM’s ability to filter out statistically fragile links that are prone to large losses, rather than relying on rare gains. These improvements remain stable across different trading configurations, indicating that the gains are not driven by specific parameter choices. Overall, the results suggest that LLMs function as semantic risk managers on top of statistical discovery, prioritizing lead-lag relationships that generalize under changing market conditions.

Complexity vs Empirical Score

  • Math Complexity: 3.5/10
  • Empirical Rigor: 6.0/10 (capped from 7.5: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper uses standard statistical time-series methods (Granger causality, VAR, log-odds transformation) combined with LLM semantic filtering to discover and trade lead-lag relationships in prediction markets. It features strong empirical validation using rolling evaluations and practical PnL/risk metrics, demonstrating a practical implementation of LLMs for quantitative risk management.

Research Flowchart

  flowchart TD
    A[Research Goal: Filter Spurious Lead-Lag<br>Relationships in Prediction Markets] --> B[Data Input: Kalshi Economics<br>Probability Time Series & Event Descriptions]
    
    B --> C[Stage 1: Statistical Screening<br>Granger Causality Identifies Candidate Pairs]
    
    C --> D[Stage 2: Semantic Filtering<br>LLM Re-ranks Pairs Based on Economic Logic]
    
    D --> E[Evaluation: Signal-Triggered<br>Fixed Trading Protocol]
    
    E --> F[Key Outcomes:<br>- Win Rate: 51.4% to 54.5%<br>- Avg Loss: $649 to $347<br>- LLM Functions as Semantic Risk Manager]

Cited References (23)

  1. Lead–lag detection and network clustering for multivariate time series with an application to the US equity market (2022) — link
  2. Prediction market accuracy in the long run (2008)
  3. Uncertainty assessment and false discovery rate control in high-dimensional Granger causal inference (2017)
  4. An econometric analysis of nonsynchronous trading (1990)
  5. One security, many markets: Determining the contributions to price discovery (1995)
  6. Prediction Markets (2004)
  7. Investigating causal relations by econometric models and cross-spectral methods (1969)
  8. Granger-causality in the presence of structural breaks (2008)
and 15 more
  1. Vector autoregressive-based Granger causality tests in the presence of instabilities (2019) — link
  2. Understanding spurious regressions in econometrics (1986)
  3. Data-snooping, technical trading rule performance, and the bootstrap (1999)
  4. Exact recovery of Granger causality graphs with unconditional pairwise tests (2023)
  5. Comparing predictive accuracy (2002)
  6. Forecasting in the presence of instabilities: How we know whether models predict well and how to improve them (2021)
  7. Distribution of the Estimators for Autoregressive Time Series with a Unit Root (1979) — link
  8. Testing for unit roots in autoregressive-moving average models of unknown order (1984)
  9. Time series analysis (2020)
  10. Large language models are zero-shot time series forecasters (2023)
  11. Time-LLM: Time Series Forecasting by Reprogramming Large Language Models (2024) — link
  12. Judging llm-as-a-judge with mt-bench and chatbot arena (2023)
  13. A Critical Review of Causal Reasoning Benchmarks for Large Language Models (2024) — link
  14. Causalbench: A comprehensive benchmark for evaluating causal reasoning capabilities of large language models (2024)
  15. Compare Models (2026)

0 of 23 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.