Deep Reinforcement Learning Algorithms for Risk Management

Deep Reinforcement Learning Algorithms for Risk Management - DREAM

The DREAM project is funded by the Croatian Science Foundation, for a period of 4 years.

The goal of this project is to develop a novel class of risk-sensitive reinforcement learning algorithms in dynamic environments with applications in financial risk management. The objectives also include the implementation of state space representation models that extract information from time series data by exploiting latent factors and design of portfolio optimization algorithms based on proposed methods.


News

Deep Neural Networks for Behavioral Credit Rating

New paper under the title “Deep Neural Networks for Behavioral Credit Rating" has recently been published in the journal Entropy, as a part of a special issue: Human-Centric AI: The Symbiosis of Human and Artificial Intelligence.

Within this paper, a deep neural network model for behavioral credit rating was proposed. The proposed deep neural network was trained on two different datasets: the first one contains information on loans between 2009 and 2013 (during the financial crisis) and the second one from 2014 to 2018 (after the financial crisis); combined, they include more than 1.5 million examples. The proposed network outperformed multiple benchmarks and was evenly matched with the XGBoost model. Long-term credit rating performance is also presented, as well as a detailed analysis of the reprogrammed facilities’ impact on model performance.

To see more, please visit the publisher website. The paper is open access and is available for download.


Cluster-Specific Latent Factor Estimation in High-Dimensional Financial Time Series

New paper under the title “Cluster-Specific Latent Factor Estimation in High-Dimensional Financial Time Series" has recently been published in the journal IEEE Access.

In this paper high-dimensional financial time series with pervasive and cluster-specific latent factors are considered. A clustering and latent factor estimation method is proposed, together with a model selection algorithm, based on the spectral properties of asset correlation matrices and asset graphs. A simulation study with known data generating processes demonstrates that the proposed method outperforms other clustering methods and provides estimates with a high degree of accuracy. Moreover, the proposed methods were applied to datasets of asset returns from global financial markets using a backtesting approach. The results demonstrate that the clustering approach and estimated latent factors yield relevant information, improve risk modelling and reduce volatility in optimal minimum variance portfolios.

To see more, please visit the publisher website. The paper is open access and is available for download.


Project team

  • Assoc. prof. Zvonko Kostanjčar, PhD – project leader
  • Prof. Branko Jeren, PhD – senior researcher
  • Asst. prof. Davor Zoričić, PhD – senior researcher
  • Asst. prof. Petra Posedel Šimović, PhD – senior researcher
  • Stjepan Begušić, PhD – researcher
  • Fredi Šarić – PhD student
  • Tessa Bauman – PhD student

Publications