The Application of Generative Adversarial Networks to Simulate Market Data for Predictive Analytics
, Cohen & Steers
I'll demonstrate the application of a generative adversarial network (GAN) in developing synthetic financial datasets that can be used as training data across many classes of machine learning models. GANs can complement and augment the value of Monte Carlo simulations and replicate regime-specific conditions to better prepare models for more robust predictive analytics. Using a generator and discriminator, the model will transition the simulated data so that it converges to an empirical distribution in a Nash or Quasi-Nash process, preserving more of the real-world temporal characteristics consistent with the targeted market regime.