Jiajing Sun

Jiajing Sun

PhD, CFA charterholder, FIMA. Associate Professor and the Deputy Director of the Department of Statistics and Data Science, School of Economics and Management, University of the Chinese Academy of Sciences.

Can Generative AI Preserve Tail-Risk Networks? Evidence from Cryptocurrency Markets

Published:

Status: Working paper. Revision requested, Management & Marketing.

Authors:

Jiajing Sun, Ruting Wang, and Junda Wu. Corresponding author: Ruting Wang.

Summary and Contribution:

This paper asks whether generated financial scenarios preserve the tail-risk relationships used in network-based risk monitoring and portfolio decisions. It distinguishes three layers of evaluation: conventional distributional and risk diagnostics; preservation of a conditional lower-tail network, assessed against sampling uncertainty in the empirical network; and realised outcomes of an allocation rule based on network rankings. The networks describe conditional lower-tail associations, not causal transmission.

Data and Methods:

The main application uses hourly data for ten cryptocurrencies over 2020–2025. A stationary bootstrap, a GARCH-t copula model, and a denoising diffusion probabilistic model (DDPM) are compared in a rolling-window design. The analysis estimates directed tail-risk networks using penalised quantile regression and evaluates network-based allocation rules. Kraken prices, a second asset panel, and controlled simulations provide additional evidence.

Main Findings:

Conventional diagnostics, network preservation, and allocation outcomes give distinct assessments of the generators. GARCH-t can pass an overall distributional comparison while failing to preserve the network; DDPM can reproduce the network in individual panels despite failing the joint conventional-diagnostic criterion. Network differences change weights and volatility exposure, but after ex ante risk matching, the empirical expected-shortfall differences are not statistically distinguishable from zero. Controlled simulations show that ranking errors can be costly when tail hubs differ chiefly in asymmetric downside dependence.

Keywords:

Generative AI; synthetic financial data; tail-risk networks; quantile LASSO; cryptocurrency; expected shortfall.

Recommended citation: Sun, J., Wang, R., & Wu, J. (2026). Can Generative AI Preserve Tail-Risk Networks? Evidence from Cryptocurrency Markets. Working paper.