Paper: arXiv 2311.18717

Authors: Brett Hemenway Falk, Gerry Tsoukalas, Niuniu Zhang

Abstract

Existing studies on crypto wash trading often use indirect statistical methods or leaked private data, both with inherent limitations. This paper leverages public on-chain NFT data for a more direct and granular estimation. Analyzing three major exchanges, we find that ~38% (30-40%) of trades and ~60% (25-95%) of traded value likely involve manipulation, with significant variation across exchanges. This direct evidence enables a critical reassessment of existing indirect methods, identifying roundedness-based regressions à la Cong et al. (2023) as most promising, though still error-prone in the NFT setting. To address this, we develop an AI-based estimator that integrates these regressions in a machine learning framework, significantly reducing both exchange- and trade-level estimation errors in NFT markets (and beyond).

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced statistical methods and machine learning (boosting, deep learning) with hyperparameter optimization, indicating moderate-to-high math complexity. It is highly empirical, using massive public on-chain NFT data from three exchanges, implementing direct filters, and reporting specific performance metrics (AUC scores, error rates), making it data- and implementation-heavy.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Can AI directly detect NFT wash trading using<br>public on-chain data?"] --> B["Methodology<br>Empirical analysis of NFT trades across<br>3 major exchanges (NFT marketplace data)"]
  
  B --> C["Computational Process<br>1. Baseline: Statistical models (e.g., roundedness regressions)<br>2. AI Estimator: Machine learning framework integrating<br>regressions to correct exchange & trade-level errors"]
  
  C --> D["Key Findings/Outcomes<br>• Direct Evidence: ~38% of trades / ~60% of value manipulated<br>• Exchange variation: significant differences detected<br>• ML estimator significantly reduces estimation errors vs.<br>traditional statistical methods<br>• Public data enables granular, repeatable analysis"]
  
  style A fill:#e1f5e1
  style D fill:#fff2cc

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes an AI‑based estimator that combines roundedness regressions with machine‑learning to detect NFT wash trading and reduce estimation errors.

Source sentence
we develop an AI-based estimator that integrates these regressions in a machine learning framework, significantly reducing both exchange- and trade-level estimation errors in NFT markets (and beyond).

Assumes 5

  • Filters 1‑3 are extremely conservative and unlikely to misclassify organic behavior as wash trading. model
    Source sentence
    We believe our Filters 1-3 are extremely conservative, and unlikely to classify organic behavior as wash trading.
  • Only trades paid with ETH (or wETH) are included to ensure price consistency. data
    Source sentence
    the 0.39% of transactions not conducted in ETH-type currencies (e.g., ETH, wETH) to ensure consistency in our analysis.
  • Roundness of trade price (last non‑zero digit >1% of price) indicates low wash‑trading activity and serves as a benchmark. model
    Source sentence
    Cong et al. (2023) categorize a trade price as “round” if its last nonzero digit exceeds 1% of the total price.
  • Upstream Ether transfers are limited to external EOAs to conservatively capture potential collusion. data
    Source sentence
    we limited the transfers to be of “external” type, i.e., Ether transfers from Externally Owned Accounts (EOAs).
Show 1 more
  • The threshold τ can be optimized to minimize aggregate prediction error. model
    Source sentence
    We refer to this approach as “τ-Regression” where the threshold τ is the outcome of an optimization.

Data

  • LooksRare NFT sales · Alchemy getNFTSales API crypto-onchain public
    11:04, December 29th, 2021 to 07:56, Apr · LooksRare marketplace transactions
    Source sentence
    For LooksRare, our data covers all transactions from 11:04, December 29th, 2021 to 07:56, April 13th, 2023.
  • Blur NFT sales · Alchemy getNFTSales API crypto-onchain public
    04:49, October 19th, 2022 to 17:14, Marc · Blur marketplace transactions
    Source sentence
    For Blur, our data covers all transactions from 04:49, October 19th, 2022 to 17:14, March 7th, 2024.
  • OpenSea NFT sales · Alchemy getNFTSales API crypto-onchain public
    19:57, February 18th, 2022 to 13:46, Mar · OpenSea marketplace transactions
    Source sentence
    For OpenSea, our data covers all transactions from 19:57, February 18th, 2022 to 13:46, March 20th, 2024.

Costs

Not applicable The paper reports no trading or execution result; the cost the extractor found is a model parameter.

The paper-text extraction found a cost parameter inside a model (Filters consider trading fees to the marketplace and Ethereum gas costs, but costs are not incorporated into headline error metrics.); the classifier found no reported trading result, so costs are not applicable.

Finds 2

  • Approximately 38% of NFT trades and 60% of traded value across three major exchanges likely involve wash‑trading. statistical
    Source sentence
    we find that ∼38% (30–40%) of trades and ∼60% (25–95%) of traded value likely involve manipulation
  • Error rates drop dramatically from 1‑Regression (45.6% on Blur) to τ‑Regression (31.0%) to ML models (0.4%). performance
    Source sentence
    Table 6: Absolute Estimation Errors Across Platforms and Methods

Checked by others

Contradicted by

  • NFT Wash Trading Detection (2023)
    On the question: What proportion of NFT trades involve wash trading?
    Compare the findings

    This paper: Approximately 38% of NFT trades and 60% of traded value across three major exchanges likely involve wash‑trading.

    That paper: On average 0.136% of transactions, 0.157% of tokens, and 0.109% of wallets are involved in wash trading, generating total price manipulation, sales, and repurchase profit of around $900K, $1.1M, and -$1.6M respectively.

    A finds 0.136% of transactions, B finds ~38% of trades; the estimates differ by orders of magnitude, reflecting different samples/metrics.

  • Abnormal Trading Detection in the NFT Market (2023)
    On the question: What proportion of NFT trades involve wash‑trading?
    Compare the findings

    This paper: Approximately 38% of NFT trades and 60% of traded value across three major exchanges likely involve wash‑trading.

    That paper: The wash trader percentage in the NFT market is estimated at 5.38% based on the clustering results.

    A reports 5.38% wash traders; B reports ~38% of trades (and 60% of value) as wash‑trading, showing a large discrepancy.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (22)

  1. Tabnet: Attentive interpretable tabular learning (2021)
  2. Using machine learning Meta-Classifiers to detect financial frauds (2022) — link
  3. Blur continues momentum with incentive program (2023) — link
  4. Detecting wash trade in the financial market (2014) — link
  5. Most Popular NFT Marketplaces by Market Share (2023)
  6. Crypto Wash Trading (2023) — link
  7. How Much Wash Trading Is Really Happening on Blur? (2023) — link
  8. Pricing Dynamics and Herding Behavior of NFTs (2023) — link
and 14 more
  1. NFT Wash Trading on Ethereum ( 2022)
  2. Selection-Neglect in the NFT Bubble (2023) — link
  3. Fraud Transaction Recognition: A Money Flow Network Approach (2015) — link
  4. Ethereum ETL ( 2023 )
  5. A Game of NFTs : Characterizing NFT Wash Trading in the Ethereum Blockchain (2023) — link
  6. What’s Not There: Odd Lots and Market Data (2014)
  7. DEXs Gain Market Share As Faith In Centralized Crypto Players Erodes (2022) — link
  8. NFT Wash Trading: Quantifying suspicious behaviour in NFT markets (2022) — link
  9. NFT Wash Trading (2023)
  10. Direct Evidence of Bitcoin Wash Trading (2024) — link
  11. Trading volume manipulation and competition among centralized crypto exchanges (2020)
  12. The Dark Side of NFTs: A Large-Scale Empirical Study of Wash Trading (2024) — link
  13. Market Manipulation in Non-Fungible Token Markets (2024)
  14. Tax-Loss Harvesting with Cryptocurrencies (2021)

0 of 22 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.