Codes for my thesis project: replicating and modifying Quant GANs.
Python files for neural network creation/training, preprocessing and metrics can be found in the backend package. Models ought to be trained in Colab or using a PC with a CUDA capable GPU.
Replication/testing:
- S&P 500 Quant GAN training
- Stylized facts of S&P 500 generated paths
- Mode collapse of Quant GANs on S&P 500 generated paths
- Training and testing for mode collapse of Quant GANS on the stocks compromising the Dow Jones index
- Generating multiple returns (MSFT/AAPL) with plausible dependencies
∆CoVaR related:
- Training and testing conditionally on JP Morgan*
- Training and testing conditionally on us publicly listed bank stocks
Experiments found in the appendix:
* Please note that this notebook makes use of an older CGAN implementation. I didn't have the time to use the new CGAN from the backend package here. In order to keep congruent with my thesis I will keep it like this.