This study proposes a novel model that integrates the generative adversarial network (GAN) with the value at risk (VaR) measuring method. The objective is to investigate the efficacy of the VaR method in addressing the issue of excessive financialization in enterprises. Firstly, the related concepts and calculation principles of VaR in the financial field are explained, and the autoregressive conditional heteroscedastic (ARCH) family-based VaR calculation method and the basic structure of GAN under the deep learning are introduced. Then, the GAN algorithm is employed to optimize and train the initialization network, transformation network, and structure network of the GAN algorithm. Finally, the optimized GAN is applied to the VaR measurement of 300 stocks in Shanghai and Shenzhen stock market. GAN demonstrates the ability to handle unbalanced data samples, sample minority data, and fit the overall distribution of minority samples. GAN introduces a groundbreaking method for data processing, and its integration with manual efforts yields significant improvements in practical applications. Moreover, GAN demonstrates a positive impact on data set training, offering reliable potential for advancement and serving as a valuable point of reference. In conclusion, the combination of GAN under deep learning with VaR showcases a dependable practicability in assessing the risks associated with excessive financialization.