Application of Generative Adversarial Networks for Financial Data
CUI Yihao,LIU Sen,YE Guangnan
DOI: https://doi.org/10.11959/j.issn.2096-109x.2024047
2024-01-01
Abstract:Data,recognized as a fundamental strategic resource and key production factor for a nation,has served as the foundational resource and innovation engine for economic and social development.The financial industry,characterized by its data-intensive and technology-driven nature,necessitates the optimal allocation of data assets to facilitate industrial upgrading.However,financial data commonly exhibits issues such as uneven distribution,in-formation asymmetry,and data silos,which have prevented data from fully realizing its value.To address these challenges,various generative models have been actively adopted by financial institutions to synthesize highly real-istic data,thereby breaking down data barriers and monopolies,and shaping the future trend of the financial indus-try.Among these models,Generative Adversarial Networks(GANs)have emerged as particularly popular,demon-strating impressive performance across various fields and showing great potential in generating financial tabular data,financial time series,and detecting financial fraud.The advantages of the GAN model compared with other generative models in the financial field were analyzed.The GAN models that have been applied to the financial field since Generative Adversarial Networks were proposed in 2014 were presented,and the principles of each model were introduced.The application practice of the GAN model in generating financial tabular data,generating financial time series,and financial fraud detection,as well as other financial data fields,was explored.Finally,the challenges and development direction of GANs for the future were discussed,taking into account the actual situa-tion in China.