Generating_Financial_Time_Series_with_GAN

KO | EN

gitlite — search

#javascript #typescript #go #android #linux #docker #macos #windows #cli #awesome-list #pytorch #security #mcp #open-source

Generating_Financial_Time_Series_with_GAN

(★ 18)

No description available.

Generating_Financial_Time_Series_with_GAN Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Generating Financial Time Series with Generative Adversarial Networks

This project is part of the "Machine Learning for Finance" course conducted by Romuald Elie at ENSAE Paris. In this project we explored 
different Generative Adversarial Networks architectures in order to generate financial Time Series. For a complete description of the methodology please refer 
to our pdf report.

The lack of available data makes the training of Machine Learning algorithms sometimes difficult in Finance. Hence, there is a great interest 
in the development of Time Series generators as a data augmentation method. However, Financial Time Series exhibits different properties that are often difficult to reproduce with simple models. A good generator needs to be able both to get the properties of the general distribution of the data (Moments for instance),
but also to get its temporal properties (Autocorrelation structures).

Generative Adversarial Networks have been introduced by Goodfellow et al. in 2014. His novative approach was to propose a game theory framework, by training simultaneously two neural networks in an adversarial setting. The GAN architecture has provided impressive results in Image related fiels (Face generation for instance), but to our knowledge it has not been used successfully in finance yet. Our aim was to exhibit some properties of this framework for this kind of problems.

Along with the Neural Networks we implement several methods for the training process and the output evaluation.
They can be divided into the following categories:
* Generation of Training Samples (utils file)
* Implementation of Generator and Discriminator Neural Networks, following a given architecture (Lin_GAN, DCGAN, RGAN, CvGan,...)
* Implementation of a general training framework for all architectures (GAN file)
* Statistical tests to verify the adequacy of the properties of the generated samples with the ones of samples from the true distribution

We also provide some pre-trained models that can be loaded using functions from the utils module. An experimental report is also provided in a Notebook format. Our experiments where conducted in order to sample from the distribution of the CBOE VIX Index in a time period from 2014 to 2019, when stationarity holds. The interest of this distribution is to combine a Gaussian process with recurring random spikes, hence it is very easy to evaluate the generated samples visually.

Generating_Financial_Time_Series_with_GAN

Generating_Financial_Time_Series_with_GAN Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Generating_Financial_Time_Series_with_GAN

Generating_Financial_Time_Series_with_GAN Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README