A GAN Data Augmentation approach for trading applications in European Carbon Emission Allowances

In the last years, energy markets have shown a great volatility with high prices' variations. Most of the machine learning algorithms implemented to forecast the market evolution and to optimize the trading strategies are data-driven and data-intensive; thus, it is important to have a sufficient number of training samples to avoid the overfitting issue. In this context, Generative Adversarial Networks (GANs) have become very popular in the recent years, especially in computer vision, where they are used to manipulate and generate video and images. This methodology has been applied also for time series analysis, especially in the financial context for generating synthetic data. In this work, the application of GANs for generating intraday open-high-low-close prices and related traded volumes for European Carbon Emissions Allowances futures. In particular, a new encoding and decoding process has been introduced to allow the generation of longer time series with significantly higher quality, making the generated data useful for practical applications.