Welcome to d-fine

Due to growing concerns about the stability of the financial markets, the Central Bank of Ireland decided in 2010 to establish a more risk-based supervisory process that focuses on systemically important institutions. Rolling out this new approach required the implementation of a new software solution capable of supporting more than 400 supervisors in the daily monitoring of more than 10,000 financial institutions. The aim was to ensure the highest standards of stability, expandability and information security.

d-fine played a key role in the initial requirements analysis and system implementation. Key to the success of the project was the combination of in-depth knowledge of supervisory processes and their underlying risk methodology and excellent IT architecture design. The outcome is a scalable and highly efficient solution that is in use by various other internationally active supervisory authorities and is continuously being enhanced.

[ DOWNLOAD CENTRAL BANK OF IRELAND RESOLUTION ]

Energy turnaround means that we are relying more and more on electricity from renewable sources - and have to cope with the naturally fluctuating feed-in quantities, which in many cases, cannot be forecast with certainty even in the short term. The rapid price fluctuations also pose a challenge for direct marketers, who want to sell electricity from solar and wind farms profitably on the stock exchange around the clock, separately for every quarter of an hour. A complex task for human electricity traders.

The solution is provided by an algorithm that determines the marketing strategy with the optimal risk-return ratio for every quarter of an hour from the current forecast and its uncertainty and translates it into trading orders on the so-called intraday market. Our method is based on the idea that fluctuating feed-in quantities shift the power supply curve against the demand curve and thus lead to price movements. Supply and demand curves, in turn, can be observed in the so-called day-ahead market of the exchange on the previous day. Finally, the link with a model of forecast uncertainty for different weather regimes results in a stochastic distribution of trading success, which we optimise by means of the choice of marketing strategy depending on the trader's risk appetite. The algorithm is part of our customized IT solutions package for direct marketers, which map and automate the entire energy trading process in a highly efficient manner – from the feed-in forecast to optimization and bidding on EPEX Spot.

Owing to numerous cases of market manipulation, banks must systematically monitor the communication of traders. Our client, a major international bank, has commissioned d-fine to develop an innovative model to prevent, or subsequently identify cases of fraud and initiate legal measures. The solution to this easily formulated problem is proving to be extremely complex. The volume of information to be evaluated exceeds the personnel capacities of the compliance department at hundreds of globally active merchants and suggests a largely automated implementation.

The development of the methodology is based on a combination of expert feedback and mathematical-statistical models with the aim of identifying suspicious events in the noise of all conversations as optimally as possible. This requires two essential building blocks: Natural Language Processing (NLP) and Machine Learning. After pre-processing, NLP techniques (e. g. lexicons, maximum entropy part-of-speech tagging and word embeddings) transform text content into a multidimensional structure that allows quantitative analysis. Machine learning techniques (e. g. Boosted Decision Trees and Deep Neural Networks) train the model to recognize and quantify deviations from "typical" chat behavior. In order to support the process flow, which has to be accompanied by manual control of the conversations that the model classifies as critical, an intuitive visualization concept was developed. It displays model predictions in an interactive, multi-user capable web interface and allows the user to intervene in model decisions (supervised learning). The newly developed method is unique in the banking sector and achieves significantly better results than the commercial software previously used. It significantly reduces the bank's efforts to reliably uncover market manipulation and insider trading and fulfils all requirements required by the supervisory authority.

The steadily growing importance of centrally processed financial transactions to reduce counterparty default risks since the financial crisis, poses a challenge for clearing houses to measure and transparently present the risk potential of ever larger and more complex portfolios. Against this background, our client – one of the world's leading clearing providers – commissioned the design, development and implementation of a "state-of-the-art" risk model that determines risk coverage requirements (margins) for clearing participants (including banks, insurance companies and hedge funds). The newly developed in-house solution provides real-time risk measurement with exact valuation of all financial instruments. The IT architecture of the model consists of a high-performance GPU-based computing core and system components for generating, evaluating and analysing risk scenarios. The seamless and stable integration of the methodology into the client's complex process and IT landscape as well as the coordination of the internationally operating project teams played a key role in the success of the project. The solution, which has been further enhanced together with the client under increasingly demanding regulatory conditions, can be described as a reference for modern risk measurement and management.