How to Detect Insider Trading in Prediction Markets: A Guide to Analyzing Polymarket Data

Overview

Prediction markets like Polymarket allow users to bet on the outcome of events—from election results to military actions. While these platforms are often touted for their ability to aggregate information, they are not immune to manipulation. A recent analysis by the Anti-Corruption Data Collective (ACDC), a nonprofit research group, uncovered a stark anomaly: long-shot bets (those of $2,500 or more at odds of 35% or lower) on military and defense-related markets win an average of 52% of the time. In contrast, the average win rate across all politics markets is 25%, and across all markets on the platform it is just 14%.

This discrepancy strongly suggests insider trading—individuals with non-public information are betting on events they know have a higher probability than the market price reflects. This guide will walk you through how to replicate ACDC’s methodology, interpret the results, and understand why this matters for regulatory oversight.

Prerequisites

Before diving into the analysis, you’ll need the following:

• Basic understanding of prediction markets – Familiarity with concepts like odds, market prices, and liquidity.
• Access to Polymarket data – You can use the Polymarket API or third-party tools like Dune Analytics to query historical trade data.
• Data analysis skills – Proficiency in Python (pandas, numpy) or SQL for filtering and aggregating betting records.
• Knowledge of statistical concepts – Understand win rates, confidence intervals, and simple hypothesis testing.

Step-by-Step Instructions

1. Define Long-Shot Bets and Thresholds

ACDC defined a “long-shot bet” as any single wager of $2,500 or more placed at odds of 35% or less (i.e., the market assigned a probability no higher than 35%). This thresholds isolates bets that are both large in size and low in implied probability—precisely the kind that would yield outsized returns if based on inside information.

Code Example (Python):

import pandas as pd

# Assume df has columns: 'amount', 'odds', 'win_flag'
long_shots = df[(df['amount'] >= 2500) & (df['odds'] <= 0.35)]
print(f"Filtered {len(long_shots)} long-shot bets")

2. Compare Win Rates Across Categories

Next, you compute the win rate (percentage of bets that correctly predicted the outcome) for three categories:

• Military/Defense markets – e.g., “Will the US strike in Yemen by June?”
• All politics markets – including elections, policy decisions, etc.
• All markets on Polymarket – this serves as the baseline.

Code Example:

categories = {
    'military': military_df,
    'politics': politics_df,
    'all': all_df
}

for cat, df in categories.items():
    long_shots = df[(df['amount'] >= 2500) & (df['odds'] <= 0.35)]
    win_rate = long_shots['win_flag'].mean()
    print(f"{cat}: {win_rate:.2%}")

You should observe the pattern: military long-shots win ~52%, politics ~25%, all ~14%.

3. Identify Anomalous Patterns

The 52% win rate is not just high—it is more than 3.7 times the platform’s average. To confirm the anomaly is statistically significant, you can perform a chi-squared test or compare confidence intervals.

Using scipy:

from scipy.stats import chi2_contingency

# Contingency table for military vs. all markets
observed = pd.DataFrame([
    [military_wins, military_losses],
    [all_wins, all_losses]
])
chi2, p, dof, expected = chi2_contingency(observed)
print(f"p-value: {p:.6f}")  # Should be < 0.05

4. Analyze Specific Markets

Drill down into individual military/defense markets. Look for markets where the number of long-shot bets and win rate are highest. Pay attention to timing: if large bets are placed just before a government announcement, that’s a red flag.

Visualization tip: Create a bar chart showing win rates per market category. Use colors to highlight military markets.

5. Interpret Results

The findings indicate that a small number of individuals are consistently profiting from insider knowledge about military actions—information that is otherwise classified or non-public. This undermines the fairness of prediction markets and raises serious ethical and legal questions. Regulators like the CFTC may need to step in, as they have with sports betting insider trading cases.

Common Mistakes

• Ignoring sample size – A few lucky bets can skew win rates. ACDC’s analysis covered thousands of trades; your own should too.
• Confusing correlation with causation – High win rates could be due to better analysis, not necessarily insider info. Check if the bets are placed right after breaking news (e.g., a classified leak).
• Not adjusting for multiple comparisons – If you test many categories, you might find false positives. Use Bonferroni correction or similar.
• Overlooking market manipulation – Wash trading or spoofing can artificially move odds. Filter out suspicious accounts (e.g., new accounts with only large bets).

Summary

By following the steps above, you can independently verify the findings of the Anti-Corruption Data Collective: long-shot bets on Polymarket’s military and defense markets win at an extraordinary rate (52%) compared to the platform average (14%). This pattern is a clear indicator of insider trading. The implications are severe—if left unchecked, insider betting can distort political and military decisions, eroding trust in both prediction markets and democratic processes.

Last updated: March 2025