Data Science for Financial Services

Financial Analytics and Research
An human-like, artificially intelligent robot holding a tablet.

Regulatory Impact Analysis

The financial industry has been affected by a large number of new regulations and regulatory changes within the recent past. Governments and federations make use of regulatory impact analysis in order to assess whether new rules will bring the desired effects. Yet, current tools for regulatory impact analysis depend on quantitative data although many policy decisions are not measurable by quantitative data but result in a lot of unstructured data. For this reason, the goal of this research project is to provide an exemplary approach for analyzing the regulatory impact of policy decisions which are not measurable by quantitative data. This exemplary approach for analyzing regulatory impact based on unstructured data are the recently enforced best execution requirements under the new European regulation for financial markets called Markets in Financial Instruments Directive II (MiFID II).

In today’s fragmented securities markets, the selection of the appropriate trading venue for an order is one of the key tasks of investment firms to achieve the best possible result for their clients (“best execution”). To protect investors from the potential downside of market fragmentation, the European regulator defined principles for investment firms concerning best execution in MiFID I and strengthened as well as extended those standards in MiFID II. Aspects of best execution cover – among others – price, costs, speed, as well as likelihood of execution and settlement. Based on this regulation, all investment firms that execute orders on behalf of clients have to provide execution policies, which build the basis of this research project. Specifically, the project aims to extract textual features from the best execution policies using natural language processing in order to determine whether and how the policies changed with the introduction of MiFID II and to relate these features to differences between firms’ best execution policies and their actual executions.

Principal Investigator: Prof. Dr. Peter Gomber

Project Members: Dr. Benjamin Clapham, Jens Lausen

Decision Support for Aggregating Analyst Estimations

Analyst reports of sell-side firms intend to provide buy-side firms with relevant information, e.g., buy/sell recommendations, of analyzed companies to make informed investment decisions. Regarding the actual investment decisions, however, portfolio managers are often confronted with large amounts of data, i.e., analyst estimates, containing information of heterogeneous quality that have to be processed by the portfolio manager herself. Therefore, a statistical learning system would help to make improved investment decisions by predicting the accuracy of each single analyst estimate that is too costly or presumably not feasible by human decision making.

In this project, we apply a machine learning framework that approximates the information quality of analyst estimates. We create a decision support system (DSS) that is able to quantify information quality of analyst reports, reduces the information load that stems from the large amount of analysts’ forecasts, and finally delivers investment recommendations. Then, we evaluate the machine learning approaches by implementing different trading strategies, which take advantage of the DSS recommendation.

Principal Investigator: Prof. Dr. Peter Gomber

Research Assistants: Sven Panz, Timo Schäfer

People working together and celebrating achievement


The following sponsors support efl - the Data Science Institute Frankfurt