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

Battle Count: 56

Highly relevant for quantitative traders operating in cryptocurrency markets. The paper provides a framework for assessing exchange reliability and detecting venues where reported liquidity may be artificially inflated. This is critical for execution quality, slippage estimation, and venue selection. The finding that price-based indicators alone cannot detect such anomalies (returns remain consistent across exchanges) is particularly important for algo traders who rely on price signals. The complexity-based approach could be integrated into trading infrastructure for real-time exchange monitoring. However, the paper is primarily diagnostic/descriptive rather than providing actionable trading signals.

The methodology involves multiple sophisticated statistical techniques (MFDFA, MFCCA, ApEn, SampEn, change-point detection) that require careful implementation. The rolling-window analysis with multiple parameter choices (embedding dimensions, tolerance radii, scale ranges) adds complexity. Processing tick-by-tick data and aggregating to 1-min intervals requires efficient data handling. The framework combines 6+ complementary analyses that must be coordinated. However, many of these methods have established implementations in Python (e.g., multifractal libraries) and MATLAB. The main challenge is parameter selection and interpretation rather than pure computational complexity.

The paper provides detailed methodological descriptions including all equations for MFDFA, MFCCA, ApEn, and SampEn. Data are publicly available through exchange APIs (Binance, Bitget, KuCoin, Kraken). Parameter choices are specified (m=2, τ=1, r=0.2σX, polynomial order l=2, rolling window 10080 min with 1440 min step). MATLAB findchangepts is referenced for change-point detection. However, no code repository is provided, and the analysis involves many parameters that could affect results.