Rank Is All You Need: Robust Estimation of Complex Causal Networks

Abstract

Financial networks can be constructed using statistical dependencies found within price series of speculative assets. Inference generally involves multivariate predictive modelling to reveal causal and correlational structures within the time series data, but difficulties frequently arise due to the highly unstable nature of these markets. The complex interplay of social and economic factors results in erratic behaviour, producing data that rarely adheres to theoretical assumptions. It remains unclear if these violations impact the constructed networks, and if so, whether robust alternatives produce more informative results. This study introduces the Rank-Vector-Autoregression model, demonstrating its capacity to produce robust cryptocurrency networks aligned with economic rationale. Our rank method achieves superior classification compared to the standard approach for various types of simulated data, particularly when including adversarial abnormalities. When applied to a dataset of 261 cryptocurrency return series, our method produces a network containing fewer, but more strongly market-correlated links, and increased connectivity within the mean-reversion subset. Applying our method to the squared deviations produces a comparatively dense volatility network, suggesting that significant price coupling occurs in higher order moments. Our results demonstrate the use of a robust and scalable technique for obtaining accurate causality networks in finance.