Risk Management 4.0 – Risk Analysis of Investment Assets

The core idea of the collaborative research project “Risk Management 4.0” is the exploratory analysis regarding the possibility of machine learning for predictive causal analysis of risk factors across investment portfolios. Such automated predictive risk analysis is also in demand by the German Federal Financial Supervisory Authority (BaFin) in its consultation paper of July 16, 2018. Regulatory policies at the EU and FRG level on the application of machine learning by financial institutions aim to make these applications transparent and comprehensible. Consequently, the scientific findings of the explainability of the causality of machine learning with production data of a financial institution are of overriding interest for the financial system. The research collaboration includes several sub-projects in addition to the exploratory testing phase for descriptive analysis of risk data and potential analysis based on it. In particular, the following items under the theme “Risk Identification, Warning and Data Optimization”, derived among others from future regulatory needs, will be investigated within the SDI-C project: the ex-ante analysis of causal relationships and prediction of risk drivers of funds and securities based on them, as well as the ex-post analysis of risk drivers that led to risk limit breaches.

The goal is to develop an automated early warning system to identify key risk drivers and causally analyze their impact across multiple securities and investment assets. This early warning system is intended to assist risk managers in the day-to-day detailed analysis and communication of risk drivers with portfolio managers, clients, and the regulator.

The scientific contribution is the explainability of causality between risk factors driving risks in investment assets. Given future regulatory requirements, these findings are of paramount importance to the financial system. In addition, structured financial data differs significantly from industry production data, so explainability of causality is the key challenge to creating regulatory required transparency. Upstream is research into automated optimization of big data in the financial industry


For the first phase of the exploratory data analysis for the feasibility study, which precedes the actual research project, the data includes 167 daily risk ratios of 50 investment assets in the period from January 2017 to September 2018. This data will be gradually expanded with additional investment assets over a longer period of time. In addition, the risk ratios will be supplemented with market data from Thomson Reuters for the securities included in the investment assets. The gradual addition of data serves to expand the cluster analyses and train the machine learning. Also, in practice, other data sources are being connected and linked because this is mandatory for financial services providers due to regulatory requirements, such as BCBS 239.

The exploitation plan foresees that KIT and DWS Investment GmbH share the research results for initial applications and further develop both jointly, as well as individually internally, for applications based on them that condition the explainability of causality between risk drivers in financial data. Joint scientific publications are planned.

Project period

01.11.2019 – 30.04.2020


Ployplearn Ravivanpong, Karlsruher Institut für Technologie (ployplearn.ravivanpong@kit.edu), KIT
Dr. Till Riedel, Karlsruhe Institut für Technologie (till.riedel@kit.edu), KIT
Dr. Pascal Stock (pascal.stock@dws.com), DWS Investment GmbH