💰 quantitative finance

Detecting unusual trading patterns on cryptocurrency exchanges by means of complexity measures

This study proposes a diagnostic framework using complexity and statistical-structure measures on high-frequency trade data to detect artificial transaction patterns, revealing a significant anomaly on the Bitget exchange for BTC and ETH in mid-2025 characterized by a surge in low-volume trades that suggests potential market manipulation.

Original authors: Jakub Zwydak, Marcin Wątorek, Jarosław Kwapień, Stanisław DroĊdĊ

Published 2026-07-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jakub Zwydak, Marcin Wątorek, Jarosław Kwapień, Stanisław DroĊdĊ

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the global financial system as a giant, chaotic dance floor where millions of people are trading digital tokens. In the world of cryptocurrency, this dance never stops; it happens 24/7 across different clubs (exchanges) around the world. But here's the tricky part: sometimes, the music seems to speed up not because more people are dancing, but because a few dancers are spinning in circles just to make the room look crowded. This is called "wash trading," and it's a way to fake excitement, tricking everyone into thinking a party is bigger and more popular than it really is.

To catch these fakers, scientists use a special kind of detective work called "complexity analysis." Think of it like listening to the rhythm of the music. In a real, healthy market, the ups and downs of prices, the number of trades, and the volume of money moving follow a complex, messy, but predictable pattern—like a jazz band improvising together. If someone is faking the activity, the rhythm might suddenly become too perfect, too random, or just "off" in a way that doesn't match the natural chaos of a real crowd. This paper uses these rhythm-checking tools to see if any crypto exchanges are secretly spinning in circles.

The researchers in this study decided to put four major crypto exchanges—Binance, Bitget, Kraken, and KuCoin—under the microscope. They looked at three specific cryptocurrencies: Bitcoin (BTC), Ethereum (ETH), and XRP. They didn't just watch the prices; they counted every single trade, measured the volume of coins moving, and tracked how often trades happened, minute by minute, from April 1 to June 30, 2025. Their goal wasn't to catch a specific person or account, but to see if the statistical heartbeat of the market sounded normal or if it had developed a weird, artificial arrhythmia.

What they found was a massive red flag at one specific club: Bitget. While the other exchanges (Binance, Kraken, and KuCoin) kept a steady, natural rhythm, Bitget suddenly started acting strange around mid-May 2025. Specifically, for Bitcoin and Ethereum, the number of trades skyrocketed. It looked like a frenzy of activity. But here's the catch: the amount of money actually changing hands didn't go up, and the prices didn't get more volatile. It was as if the DJ suddenly started playing 1,000 tiny, tiny beats per minute, but the dancers weren't actually moving any further.

The scientists used a toolkit of "complexity measures" to prove this wasn't just a lucky coincidence. They checked the "autocorrelation," which is like seeing if the next beat in the song is related to the previous one. On Bitget, the connection between trades broke down; the trades became so frequent and random that they lost their natural structure. They also looked at "multifractality," a fancy way of describing how the market's complexity changes over different time scales. On Bitget, this complex, multi-layered structure collapsed into something simple and flat, almost like a machine generating random noise instead of a human crowd reacting to news.

Furthermore, they measured the "entropy," or the unpredictability of the patterns. On Bitget, the transaction patterns became strangely irregular and less repeatable, suggesting that the trades weren't following the usual logic of buyers and sellers. When they compared the number of trades to the actual volume of coins, the link snapped. On normal exchanges, more trades usually mean more volume. On Bitget, after mid-May, they could have thousands of trades with almost no volume, like a factory churning out empty boxes.

The study also ruled out a few things. They confirmed that this weirdness wasn't happening to XRP on Bitget, which suggests the problem wasn't a glitch in the entire exchange's software. They also saw that the prices on Bitget were still moving in sync with the other exchanges, meaning the "fake" trades weren't actually moving the market price; they were just filling up the transaction log.

So, what does this all mean? The paper suggests that after May 21, 2025, Bitget's Bitcoin and Ethereum markets were flooded with a "noise-like" component. It's highly likely that these were artificial trades—perhaps generated by bots or automated scripts designed to make the exchange look busier than it was. The authors are careful to say they can't prove it was "wash trading" in the legal sense because they don't have access to the private accounts behind the trades. However, the statistical evidence is overwhelming: the market's natural, complex heartbeat was replaced by a mechanical, repetitive thud.

In the end, this research shows that you can't just look at the price to know if a crypto market is healthy. You have to listen to the rhythm of the trades. If the number of transactions goes up but the volume and price chaos stay the same, you might just be watching a digital party where the dancers are spinning in place to make the room look full. The authors conclude that these complexity-based tools are powerful new ways to spot these hidden anomalies, helping to keep the crypto dance floor honest.

Technical Summary: Detecting Unusual Trading Patterns on Cryptocurrency Exchanges by Means of Complexity Measures

Problem Statement
Cryptocurrency markets are particularly vulnerable to market manipulation, specifically wash trading, where artificial transactions inflate reported volume and liquidity without genuine economic risk transfer. On centralised exchanges (CEXs), detecting such activity is challenging because public trade-level data lacks trader identities or account ownership information. Consequently, individual transactions cannot be directly classified as artificial. Instead, detection must rely on identifying statistical deviations from the complex, non-linear regularities typically observed in genuine market microstructure. Standard price-based indicators often fail to reveal these anomalies, as price dynamics may remain consistent across exchanges due to arbitrage, even when trading activity is artificially manipulated.

Methodology
The study proposes a diagnostic framework utilizing complexity and statistical-structure measures derived from high-frequency, tick-by-tick transaction data. The analysis focuses on three primary variables aggregated into 1-minute time series: log-returns ( $R$ ), trading volume ( $V$ ), and the number of transactions ( $N$ ).

The methodology employs a suite of econophysics tools to characterise the scaling organisation and cross-dependencies of these variables:

Multifractal Detrended Fluctuation Analysis (MFDFA) and Multifractal Detrended Cross-Correlation Analysis (MFCCA): These methods are used to characterise the scaling properties of individual time series and the scale-dependent cross-correlations between pairs of variables ( $|R|$ vs. $V$ , $|R|$ vs. $N$ , and $V$ vs. $N$ ). This approach separates genuine temporal structure from local non-stationarities and trends.
Approximate Entropy (ApEn) and Sample Entropy (SampEn): These metrics quantify the regularity and unpredictability of the time series. Higher entropy values indicate less repeatable, more irregular dynamics, while lower values suggest more predictable, repetitive patterns.
Rolling Window Analysis: To identify time-localised regime changes, the aforementioned measures (specifically detrended cross-correlation coefficients and entropy) are calculated within rolling windows (7 days) across the sample period.
Change-Point Detection: A formal statistical procedure is applied to the transaction-count series to identify structural breaks, dividing the data into pre- and post-anomaly subsamples for comparative analysis.

The study analyses data for Bitcoin (BTC), Ethereum (ETH), and XRP traded on Binance, Bitget, Kraken, and KuCoin from April 1 to June 30, 2025.

Key Results
The analysis reveals a pronounced, exchange-specific anomaly on Bitget concerning BTC and ETH, emerging after mid-May 2025 (specifically May 21, 2025). While price dynamics (log-returns) remained consistent across all exchanges, the statistical organisation of trading activity on Bitget for these two assets deviated significantly from the norms observed on Binance, Kraken, and KuCoin.

Key findings regarding the Bitget anomaly include:

Decoupling of Volume and Transaction Count: There was a sharp increase in the number of transactions ( $N$ ) without a proportional increase in traded volume ( $V$ ) or return fluctuations ( $|R|$ ).
Distributional Shift: The distribution of transaction counts shifted from a heavy-tailed, power-law-like behaviour (typical of genuine markets) to a near-Gaussian distribution.
Weakened Correlations: Autocorrelations of the transaction-count series weakened significantly, and detrended cross-correlations between $N$ and both $V$ and $|R|$ dropped to near-zero levels, indicating a loss of synchronisation between transaction intensity and market impact.
Loss of Multifractality: The multifractal organisation of the transaction-count series collapsed, exhibiting monofractal scaling characteristics, suggesting a homogenisation of the process across scales.
Increased Irregularity: Approximate and Sample Entropy values for the transaction-count series increased sharply, indicating that short-term patterns became less repeatable and more noise-like.
Low-Volume Dominance: Decomposition of the data revealed that the post-anomaly period was dominated by numerous low-volume trades. When the smallest transactions were filtered out, periods of no trading re-emerged, confirming that the apparent continuity of activity was driven by micro-transactions.
Asset Specificity: This regime change was specific to BTC and ETH on Bitget; XRP on the same exchange and all assets on other exchanges did not exhibit these characteristics.

Significance and Claims
The paper claims that complexity-based indicators are effective diagnostic tools for detecting exchange-specific trading anomalies that remain hidden when relying solely on price-based measures. The study demonstrates that the statistical relationships between transaction counts, traded volume, and return fluctuations serve as a robust benchmark for "usual" market behaviour.

The authors interpret the observed features on Bitget—specifically the noise-like component in trading activity, the decoupling of transaction counts from volume, and the structural breakdown of cross-correlations—as consistent with artificially increased transaction counts, potentially indicative of wash trading. However, the paper explicitly states that these findings do not provide direct proof of wash trading, as the analysis is based on aggregated public data without access to account-level or order-book information. Instead, the study concludes that the transaction-count process for BTC and ETH on Bitget after May 21, 2025, is statistically distinct from its prior behaviour and from the behaviour observed on other major exchanges, suggesting a fundamental alteration in the internal organisation of market activity.

More like this