ICT in Business Method for automated reconciliation of risk rating data – With a design and implementation at ING

Acknowledgements First of all, I would like to thank my first supervisor, Dr. Emiel Caron, for his valuable guidance, support and recommendations during my master thesis and my second supervisor, Dr. Hans Le Fever, for assisting me in the initial phase of the master thesis and for his recommendations. In addition, I would like to thank my manager at ING, Anabel Almagro, for giving me the opportunity to work on this project. Special thanks should be given to two colleagues of mine at ING, Rene Vermaat and Gokulram Krishnamurthy, for their continuous assistance and guidance throughout this project. Last but not least, I would like to thank my parents, Panagiotis and Ekaterini, and my sister, Maria, for their endless support and encouragement throughout my study. Business Summary In this study, a method is designed that automates the integration and reconciliation of risk rating data obtained from credit-rating agencies. Data reconciliation is considered to be part of data integration and, in the business intelligence context, deals with the decision whether data descriptions from different sources point to the same real world entity. Due to syntactic, structural and semantic variation, much of the reconciliation process is done manually, which increases the cost of maintaining data quality. ING, as many other organizations in the financial sector, is concerned with this problem as it uses risk ratings for risk management. After conducting literature review on data reconciliation and risk ratings, we used ING's resources and risk rating data from risk rating agencies, such as Moody's, Fitch and Standard and Poor's, to identify the main challenges of data integration and afterwards design an automated method for reconciliation. The main focus of this method was the reduction of the manual effort and the securement of the quality of the data. The conceptual design was implemented using ING's infrastructure and evaluated using golden data sets. At the last stage, data quality requirements for onboarding a vendor were identified in order to facilitate the reconciliation process in the future. Our method has very high precision, recall and F1 score values and it matches on average 93.79% issuers.