1、Hypothetical Market Data Scenario Generation UsingGenerative AISyed Hashim SHAH,Joel VIKLUND,Minoru AOKI,Japan Securities Clearing CorporationAurora Solutions K.K.Nomura Research Institute,Ltd.August 4,2025AbstractGenerative AI and AI have revolutionized a wide range of industries in recent years.Ho

2、wevermajor use cases have been limited to Large Language Models(LLMs)such as chatbots,codeassistants and text analysis.In this white paper we explore the use of generative AI for riskmanagement in the context of central counterparties(CCPs).Since the world of finance is ruledby exact numbers,we want

3、 to look at generative models which can handle and generate numer-ical data.In the paper we will first introduce common AI and machine learning concepts andthen explore how to use generative AI models to create synthetic yet realistic market data witha focus on 3-Month TONA Futures contracts.We comp

4、are variational autoencoder(VAE)andprincipal component analysis(PCA)to generate synthetic data and analyze the generated sce-narios.Subsequently,we use the generated synthetic market data to estimate the risk profile ofvarious portfolios by calculating Expected Shortfall Value-at-Risk(ES-VaR).The re

5、sults showedthat the VAE generated scenarios were more diverse and affected the ES-VaR more than the PCAgenerated scenarios.1.JSCC does not intend to use generative AI to create scenarios for initial margin calculation.Thisdocument is only for research purposes.2.The English version is the authorita

6、tive version.DisclaimerTable of Contents1Introduction11.1Role of CCP and Risk Management.11.2CCPs Absorb Risk.11.3Value at Risk.21.4Generative AI for Market Scenarios.32Life,Machine Learning,AI and Generative AI42.1Definitions.42.2AI/Machine Learning is Curve Fitting.52.3Supervised,Unsupervised,and