Generation of Synthetic Financial Time Series with GANs - Casper Hogenboom

During his master thesis research, Casper has been working on financial time-series generation with use of Generative Adversarial Networks (GANs). The unparalleled success of GANs in generating realistic synthetic images has initiated an entirely new field of research. However, due to the inherent difficulty of evaluating synthetic generated data instances, this progress has been largely limited to applications like images and video, where visual inspection can serve as a guide. In this research, we will extend state-of-the-art concepts within GAN training to the (financial) time series domain. The challenging task is to evaluate the synthetically generated time-series, which need to have statistical properties as close as possible to the historic data. This talk will give an insight into difficulties encountered during this research, and discuss the proposed solutions.