Scenario Based Anomaly Detection in Financial Institutions: A Study on the Turkish Factoring Sector

The increase in the number and speed of online and mobile transactions in the financial sector generates various risks and monitoring costs. Some of these risks include fraud risk, credit risk, database errors, operational problems and churn risk. In this study, scenario-based anomaly detection analysis for factoring transactions is used to identify these risks at an early stage without establishing a supervised statistical model. In anomaly detection, observations at the check, customer or customer representative level whose characteristics deviate from the main cluster are defined as outliers. The characteristics in scenarios are selected based on the experience of factoring experts. The deviations of the characteristics from the main cluster are calculated by the Mahalanobis, Minimum Covariance, and Orthogonolized Gnanadesikan-Kettenring distances. The data used in this study are comprised of check-level factoring transactions of a factoring company between 2018-2020 and the check and risk reports issued by the Credit Registration Bureau and are detected as outliers by using 7 different risk scenarios. The study also shows that the outlier detection threshold can be optimized within the framework by considering the model errors and the monitoring budget of the financial institution. The developed model can detect risk carrying anomalies in almost every financial transaction in the banking, factoring, leasing, and insurance sectors and can be also employed by the financial regulatory and supervisory institutions.