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

Abstract: Artificial transaction generation remains an important source of potential market manipulation on cryptocurrency exchanges, as it may distort reported liquidity and reduce market transparency. This study proposes a diagnostic framework for detecting unusual trading patterns based on complexity and statistical-structure measures derived from high-frequency trade-level data. The analysis considers log-returns, trading volume, and transaction counts, using tail distributions, autocorrelation functions, multifractal characteristics, approximate entropy, and detrended cross-correlations. The methodology is applied to BTC, ETH, and XRP traded on Binance, Bitget, KuCoin, and Kraken over the period from April 1 to June 30, 2025. The results reveal a pronounced anomaly on Bitget for BTC and ETH after mid-May 2025. The number of transactions increases sharply, but there is no proportional increase in traded volume or return fluctuations. This regime is characterised by numerous low-volume trades, weaker autocorrelations, reduced multifractal organisation, higher short-pattern irregularity, and weaker cross-correlations involving the transaction-count series. These features are consistent with a noise-like component in trading activity and may indicate artificially increased transaction counts, although they do not provide direct proof of wash trading. The findings show that complexity-based indicators can be useful for detecting exchange-specific trading anomalies that remain hidden in price-based measures.

• The paper identifies a dramatic anomaly in Bitget’s BTC and ETH transaction counts, indicating likely wash trading activity.
• It employs advanced techniques, including MFDFA, MFCCA, and entropy measures, to analyze 1-min aggregated trading data effectively.
• Results reveal a decoupling between trading volume and transaction counts, challenging traditional price-based market surveillance.

Detecting Anomalous Trading Activity on Cryptocurrency Exchanges via Multiscale Complexity Analysis

Context and Motivation

Manipulation of reported trading activity—most notably via wash trading—remains a critical concern in the cryptocurrency market, particularly among centralized exchanges with opaque internal matching engines. This study systematically investigates how complexity-based measures applied to high-frequency trading data can diagnose deviations from typical market microstructure, thus identifying the signature of potential artificial activity beyond standard price-based diagnostics (2607.13916). The analysis focuses on Bitcoin (BTC), Ethereum (ETH), and XRP across Binance, Bitget, Kraken, and KuCoin for Q2 2025, leveraging advanced techniques from the econophysics toolkit, including heavy-tail analysis, multifractal detrended fluctuation analysis (MFDFA), multifractal detrended cross-correlation analysis (MFCCA), and both approximate (ApEn) and sample entropy.

Methodological Framework

A central novelty of this work is the implementation of detrending-based multifractal and cross-correlation techniques to parse 1-minute aggregated time series—log-returns, trading volume, and transaction counts—in the presence of strong non-stationarity and inhomogeneity. The methodology includes:

• Distributional Analysis: Characterizing deviations from heavy-tailed norms via CCDFs for returns, volume, and count distributions.
• MFDFA/MFCCA: Estimation of the multiscaling exponents and corresponding singularity spectra for all variables, probing the presence and breakdown of multifractality and scale-dependent correlations.
• Time-localized Entropy Metrics: Rolling-window ApEn/SampEn to characterize short-term predictability and phase transitions in the dynamics.
• Change-Point Detection: Rigorous statistical testing to locate regime shifts in the time series.
• Cross-sectional and cross-exchange comparative diagnostics to anchor findings within empirical stylized facts.

Empirical Results

Exchange-specific Activity Regularities and Deviations

For BTC, ETH, and XRP, price discovery across the four investigated venues exhibited high inter-exchange synchrony at the 1-min scale—return distributions consistently manifested heavy tails very close to the inverse cubic law (tail exponent $\gamma\approx3$), and cross-exchange return correlations remained robust. However, pronounced divergence arises in trading volume and transaction-count series, with Bitget showing the most anomalous behaviour.

Differences across exchanges are evident in the average transaction size and frequency (see Table 1 of the manuscript), but are most stark in the temporal activity patterns.

Figure 1: Evolution of the cumulative log-returns, trading volume, and the number of transactions for BTC across the four exchanges, highlighting the abrupt post-May regime on Bitget.

Complexity Diagnostics for Regime Detection

The most significant finding is the detection of a pronounced anomaly for Bitget (BTC and ETH) after May 21, 2025. This regime is characterized by:

• Transaction counts per minute sharply increase, but with no commensurate rise in traded volume or volatility.
• The statistical organization of the transaction count process transitions from heavy-tailed and highly autocorrelated to almost Gaussian and nearly memoryless.
• Multifractal structure—evident across all other series—is destroyed: the fluctuation functions for transaction counts on Bitget become almost monofractal post-transition.
• There is a phase transition in entropy: ApEn and SampEn of the transaction counts jump to high values, denoting a marked increase in local randomness and a breakdown of repeatable microstructure patterns.
• Detrended cross-correlation coefficients between transaction counts and both volume and volatility (|returns|) drop dramatically, breaking the canonical microstructural link between activity and price formation.

Rolling-window analysis pinpoints the timing and persistence of this anomaly—it is exchange and asset specific (affecting BTC and ETH, but not XRP), and cross-exchange correlations in $N$ collapse only for Bitget.

Microstructural and Cross-Asset Perspectives

Scatter plots and subsequent cross-correlation analysis reveal that post-May, Bitget records a distinct decoupling: the number of transactions increases via extremely low-volume trades with negligible impact on observed volatility or aggregate trading volume. After filtering out micro-lot trades, intermittent trading and inactivity return, confirming that the surge in activity is a consequence of artificially induced micro-trades.

This effect is not a concomitant change in reporting, as it is absent in XRP and in equivalent measures for other exchanges. Cross-asset correlations in $N$ for BTC and ETH on Bitget remain near zero post-regime change—there is no evidence of coordinated information flow.

Figure 2: Relationship between trading volume and transaction counts in 1-min intervals, with Bitget post-May 21 showing regime separation and heavy fragmentation for BTC and ETH.

Figure 3: Relationship between log-returns and transaction counts, indicating the dissociation of price volatility from activity spikes on Bitget in the post-transition period.

Change-Point Analysis

Change-point algorithms precisely localize the transition to May 21, 2025; decompositions show that this shift embeds both a dramatic increase in "active seconds" (minutes containing at least one trade per second) and a moderate rise in trade-per-second, yet the volume per trade collapses.

Rolling and subsample analyses of distributional shape, autocorrelation structure, entropy metrics, and multifractal spectrum all reinforce the interpretation that the elevated transaction counts in Bitget2 are structurally different—dominated by micro-sized, statistically independent trades.

Theoretical and Practical Implications

• Implication for Wash Trading Detection: The observed breakdown of complexity measures—multifractality, autocorrelation, cross-correlation, and entropy regularity—is highly consistent with a noise-generating mechanism not reflected in market-driven dynamics. While public transaction feeds do not provide direct trader or order-linkage, the identified pattern is congruent with platform-driven artificial trade generation (e.g., wash trading or activity padding), as modeled in recent literature [Cong et al., Management Science 2023; Pennec et al., FRL 2021].
• Limitations: The analytic protocol does not constitute direct proof of manipulation, as order-level and account-level data are absent; however, it robustly constrains the space of possible microstructural explanations, ruling out typical liquidity-driven or market-wide shocks.
• Market Surveillance: Complexity-based monitoring augments standard price- or volume-based surveillance by quantitatively exposing decoupling between activity statistics and genuine economic trading, thereby serving as an early-warning or auditing tool.

Implications for Future AI and Market Surveillance Methodology

• Complexity measures and information-theory-based diagnostics provide a quantifiable, unsupervised method for regime or anomaly detection, generalizable to other opaque or fragmented trading venues.
• AI/ML systems can be trained on an expanded set of multiscale statistical and multifractal signatures (not just simple technical indicators or returns-based features) for more robust detection of anomalous/unnatural activity.
• The clear demonstration that price-based statistics are insensitive to exchange-specific manipulation underscores the need for joint analysis of volume, count, and higher-order complexity features in regulatory and forensic algorithms.

Figure 4: Relationship between log-returns and trading volume in 1-min intervals; regular structure is preserved on all exchanges, but only volume-activity links break down on Bitget post-anomaly.

Conclusion

By deploying an array of advanced complexity measures, this study provides strong statistical evidence that abnormal transaction count behaviour on Bitget (BTC and ETH after May 21, 2025) originates from a decoupling between market activity and price dynamics, concurrent with a destruction of underlying multifractal and memory structure. This indicates the likely presence of artificially generated activity—potentially wash trading—that escapes conventional price-based monitoring. These findings advocate for the integration of multiscale complexity diagnostics in ongoing market surveillance and highlight the limitations of relying on price or volume statistics alone for anomaly detection. The framework is broadly applicable to real-time anomaly detection and can be algorithmically updated as more granular (order-level, account-level) data becomes available.

---

Reference:

"Detecting unusual trading patterns on cryptocurrency exchanges by means of complexity measures" (2607.13916)