Analyzing Cointegration in Cryptocurrency Markets: A Network-Based Approach

This article examines the cointegration relationships among 472 cryptocurrency tokens listed on the Binance Exchange, utilizing historical price data from January 2021 to January 2025. The study employs a network-based methodology to identify interconnected tokens, revealing significant relationships that may not be apparent from raw time-series data.

Nature

The research begins with a cointegration analysis of each token pair, employing the ordinary least squares (OLS) method to estimate cointegration coefficients. The stationarity of residuals is assessed using the Augmented Dickey-Fuller (ADF) test, with 16% of all possible links showing stationarity at a 5% significance level. This analysis serves as a preliminary screening for potential trading pairs, although the authors caution that no statistical test can guarantee out-of-sample cointegration.

Nature

To enhance the clarity of the identified relationships, the study applies two advanced filtering methods: the Planar Maximally Filtered Graph (PMFG) and the Triangular Maximally Filtered Graph (TMFG). The PMFG retains significant edges while ensuring a planar structure, preserving both three-node and four-node cliques. The TMFG builds a fully triangulated graph, focusing on three-node cliques. These methods are preferred over traditional Minimum Spanning Tree (MST) techniques as they maintain a richer internal structure, capturing meaningful market relationships.

Nature

The study employs three community detection algorithms—Louvain, Infomap, and Stochastic Block Models (SBM)—to identify clusters of highly interconnected tokens. These algorithms help reveal groups of assets influenced by similar market drivers. Additionally, the research calculates centrality measures using the X and Y indexes, which classify nodes based on their connectivity patterns. Central nodes are identified as those with lower X + Y values, while peripheral nodes exhibit higher values.

Nature

The research introduces a pairs trading strategy based on the assumption that historically correlated assets will revert to their mean relationship over time. The spread between two tokens is computed, and trading signals are generated based on the Hurst Exponent (H), which indicates the likelihood of mean-reversion. The strategy specifies conditions for buying or selling pairs, incorporating a rolling window of 28 days for calculations.

Nature

This study provides a structured view of the interconnectedness of cryptocurrency tokens, offering insights into potential trading strategies based on cointegration relationships. The findings underscore the importance of advanced network analysis techniques in understanding market dynamics and developing quantitative trading strategies.

Nature

• “This network-based approach provides a structured view of interconnected tokens, revealing relationships that may not be immediately evident from the raw time-series data.” — Research Team

Nature

• “The PMFG retains the most significant edges while ensuring a planar structure, preserving both three-node (triangle) and four-node (quadrangle) cliques, making it more informative than simpler filtering methods like the Minimum Spanning Tree (MST).” — Research Team

Nature

• “Full size image As stated before, pairs trading is a quantitative trading strategy based on the assumption that two assets that historically move in sync will revert to their historical relationship over time.” — Research Team

Nature

No conflicting reports were identified in the sources. However, the study acknowledges that while statistical tests can indicate potential cointegration, they cannot provide definitive proof, highlighting a gap in the assurance of out-of-sample cointegration.

Nature