A statistical analysis of the Goldman Sachs closing price (EN)

You can find the unpublished paper here.

Abstract

In this report we try to conduct a classical statistical analysis of fi- nancial time-series to 1) study the Goldman Sachs closing price using ARIMA/GARCH processes, and 2) leverage these methods to better the prediction quality of several machine and deep learning models, mainly recurrent neural networks (RNN). We also discuss the possibility of using ARIMA/GARCH and Fourier analysis as features to generate realistic financial time-series through a Generative Adversarial Network [4] [6] ar- chitecture inspired by [3]. The code and the data are available in [1].

References

[1] https://github.com/abdollahrida/deeplearning/tree/master/rgan-pytorch/map565.

[2] https://github.com/ckmarkoh/gan-tensorflow.

[3] Crist ́obal Esteban, Stephanie L. Hyland, and Gunnar R ̈atsch. Real-valued (medical) time series generation with recurrent conditional gans, 2017.

[4] Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Gen- erative adversarial networks, 2014.

[5] Sepp Hochreiter and Ju ̈rgen Schmidhuber. Long short-term memory. Neural Comput., 9(8):1735–1780, November 1997.

[6] Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets, 2014.