A0917
Title: Portfolio selection based on anomaly detection using GANs
Authors: Yongjae Lee - Ulsan National Institute of Science and Technology (Korea, South) [presenting]
Abstract: The application of Generative Adversarial Networks (GANs) for anomaly detection in stock time series data is explored, and subsequently, employ this approach for portfolio selection across industry sectors. Recently, generative models have garnered substantial interest in the realm of artificial intelligence, with GANs standing out due to their distinctive dual architecture, comprising a generator that creates synthetic data and a discriminator that differentiates between genuine and fake data. Capitalizing on this feature, various GAN-based anomaly detection models have emerged. Our research introduces a novel method for devising an industry sector portfolio grounded on the anomaly detection signals derived from GANs applied to stock time series data. The GANs are specifically trained for each industry sector, generating anomaly signals that are consolidated to assemble the final portfolio. This approach enables the intuitive identification of sectors facing challenges and facilitates the adjustment of the portfolio in response, ultimately offering enhanced explainability to investors.
