04 / MARKET INTEGRITY

How much prediction-market volume may be artificial? ~25%

A Columbia study’s estimated Polymarket wash-trading patterns, category differences and why detection models are not proof of every individual trade.

The short reading

A research model flagged a material share of historic Polymarket activity as resembling wash trading. The estimates concern detected patterns, not adjudicated misconduct by every account or every market.

This is research and context, not a market recommendation or a statement of UK product availability.
At a glance~25%of historic Polymarket volume estimated as likely wash trading in research coverageView original source
Measure / 0145%sports category estimateShare of historic sports-market volume flagged by the study’s method.View original source
Measure / 0217%election category estimateNot a statement about all election markets worldwide.View original source
Measure / 033%crypto category estimateReported share for this category in the analysed data.View original source
01 / Analysis

What wash trading means here

Wash trading describes transactions intended to create the appearance of genuine market activity without comparable independent economic risk. The Columbia researchers used network patterns to identify suspected activity in on-chain trading data. A detection model is an estimate; it cannot by itself determine intent for every wallet.

02 / Analysis

What the Columbia findings say

Columbia Business School describes a study of coordinated trades that may inflate Polymarket activity. Coverage of the study reports that about a quarter of volume over the analysed three-year period matched the method’s wash-trading signals. This is a platform-and-period-specific estimate, not a blanket claim that a quarter of all prediction-market activity is fake.

03 / Analysis

Why categories differ

Reported category estimates range from around 45% of historic sports-market volume to 17% for elections, 12% for politics and 3% for crypto. The composition of traders, incentives and data availability may differ by topic. The categories should be read as outputs of the same study’s detection approach, not independent verified counts of wrongdoing.

04 / Analysis

Why apparent activity matters

Volume can create an impression of confidence and liquidity that may not match the executable order book. A market price should be checked against quoted depth, spread, settlement rules and independent evidence. Even a correctly measured volume figure is not itself a forecast-quality score.

Figure / wash-categories

Estimated likely wash trading by category

Estimated likely wash trading by categoryHorizontal bar chart. % of category volume; exact figures appear in the data table below. MEASURE % OF CATEGORY VOLUME Sports 45%Elections 17%Politics 12%Crypto 3%
Chart data: Estimated likely wash trading by category
Measure% of category volume
Sports45%
Elections17%
Politics12%
Crypto3%
Share of historic Polymarket category volume flagged in coverage of the Columbia study.

Reading note. Algorithmic estimates, not judicial findings. Category definitions and observation periods are those used by the study.

Method / How to read this

Methodology.

The research institution describes the detection method; category values and overall estimate are attributed to contemporary reporting on that study. These are estimates based on an algorithm and historical Polymarket data. The outlet and research team should be consulted for later revisions.

Reference / Original record

Sources.

  1. 01
    Polymarket volume inflated by artificial activity, study finds (opens in a new tab)
    Research papers

    Funder: Columbia Business School research coverage; consult the linked study for disclosures.

    Describes an algorithmic detection study; a flagged pattern is not a court finding against any particular trader.

  2. 02
    Polymarket volume inflated by wash trading, Columbia study finds (opens in a new tab)
    Reporting

    Funder: Crypto.news coverage of Columbia researchers’ findings.

    Reports the approximate overall and category percentages attributed to the study.

  3. 03
    Columbia study estimates Polymarket wash trading (opens in a new tab)
    Reporting

    Funder: Decrypt coverage of the Columbia study.

    Reports variation over time in algorithmically identified activity; estimates depend on the detection method.

Follow the original documents for definitions, scope and subsequent revisions. Browse the full source library ↗

Questions / In context

Questions worth asking

01Does this prove every flagged trade was fraudulent?

No. Pattern detection cannot determine intent or adjudicate individual trades; the values are study estimates.

02Does wash trading change the probability of an event?

It may distort visible activity and possibly prices, but a flagged volume share alone cannot tell you the correct probability for a particular market.

03Is the 25% estimate for every prediction platform?

No. It concerns historic Polymarket activity examined using one research method.