A Data Science Approach to Predict the Impact of Collateralization on Systemic Risk

Abstract

In this paper, we simulate and analyze the impact of financial regulations concerning the collateralization of derivative trades on systemic risk - a topic that has been vigorously discussed since the financial crisis in 2007/08. Experts often disagree on the efficacy of these regulations. Compounding this problem banks regard their trade data required for a full analysis as proprietary. We adapt a simulation technology combining advances in graph theory to randomly generate entire financial systems sampled from realistic distributions with a novel open source risk engine to compute risks in financial systems under different regulations. This allows us to consistently evaluate, predict and optimize the impact of financial regulations on all levels - from a single trade to systemic risk - before it is implemented. The resulting data set is accessible to contemporary data science techniques like data mining, anomaly detection and visualization. We find that collateralization reduces the costs of resolving a financial system in crisis, yet it does not change the distribution of those costs and can have adverse effects on individual participants in extreme situations.

Keywords: big data, graph theoretic models, data science, machine learning, Python, C , random graph generation, stochastic Linear Gauss-Markov model, Monte Carlo simulation, financial risk analytics, systemic risk, collateralizations, variation margin, initial margin, open source risk engine, financial regu

JEL Classification: G01, G18, G28, G32, G38, C12, C15