Simulating Correlated Financial Market Data

Transfer project

The project’s aim, with the cooperation of Uniper and ML2R, is to develop a method for the simulation of correlated financial market data, which is to be applied to the energy and commodity markets. For more project details please take a look at the project summary below.

 

Project Partners

Uniper SE

Uniper is a public-listed international energy company with over 12,000 employees and operations in over 40 countries, including the core European markets. The transfer project will be carried out in cooperation with the quantitative methods department of Uniper Global Commodities. Contact: PD Dr. Pascal Heider

 

TU Dortmund, Faculty of Informatics and ML2R

The Competence Centre of Machine Learning Rhein-Ruhr is one of four BMBF-funded, state-wide hubs that aim to bring the development of artificial intelligence and machine learning in Germany to a global scale. Partners of ML2R are the Technical University of Dortmund, Fraunhofer Institute for Intelligent Analysis- and Information systems IAIS in Saint Augustin, University of Bonn, as well as Fraunhofer Institute for Material Flow and Logistics IML in Dortmund. Contact: Dr. Nico Piatkowski 


Project Summary

The project’s aim, with the cooperation of Uniper and ML2R, is to develop a method for the simulation of correlated financial market data, which is to be applied to the energy and commodity markets. This is where application-related specifications will play a significant role. There will be a comprehensive review of the literature on the applicability of classical and modern methods. Of particular interest will be the implementation and evaluation of several Machine Learning methods (e.g. Generative Adversarial Networks GANs). Accompanying this task is the question of measuring the simulation’s goodness of fit, which will be discussed both in theory and in practical application.

Starting Point of the Transfer Project

Several studies have contributed to making this collaborative project possible. Posch/Ullmann/Wied (2019) focus on structural changes in high dimensional time series, while Bücher/Posch/Schmidtke (2018) work on the simulation of key indicators in financial risk management. The combination of both approaches is what will help to tackle the problem of simulating time series.

On the other hand, Piatkowski (2019) uses hyperparameters-free methods and can thus contribute to explaining the results of deep neural networks. This maximum likelihood-based approach allows for the use of classical likelihood measures to determine the goodness of fit. Piatkowski/Lee/Morik (2013) apply spatiotemporal probabilistic models to model distributed sensor data. Again, combining both approaches forms the theoretical basis for the synthesis of correlated financial market data by means of generative models.


Print Friendly, PDF & Email