ECB on Approach to Discover Plausibility Checks for Reporting Data

The approach serves business experts in two ways: one of the ways is by inspiring business experts to formulate new plausibility checks, which can be implemented in reporting systems and thus have permanent, positive effects on data quality; the second way is by making it possible to conduct ad hoc investigations into specific observations of anomalous or suspicious values. Overall, the approach has shown its benefits for the use cases considered and will be used for investigative data quality management in the future. The evaluation also provided some insights into potential extended applications. ECB has already started to investigate other sets of templates for identifying new plausibility checks. Preliminary results showed that in the context of the COREP framework, too, it was able to identify latent relationships that may serve as basis for new rules. An interesting next step will be to extend the investigations to relationships that span data from both the FINREP and COREP templates. At present, such relationships are not subject to any checks, so they represent a promising line of investigation. Another extended application regarding data is to cover more reporting periods and to look for seasonal patterns. 

Finally, there are some technical improvements that might be worth investigating. One main area for improving the performance of the predictive model is to make a better distinction in the data between missing values and semantic zero values. For instance, the analysis considered differentiating between models for predicting the presence of a data point and those for predicting its actual values. This would make it easier to identify implausible values that actually represent missing or non-reported data. Another area of work on future extensions is a broader evaluation of other models for the regression analysis, also taking into consideration ensemble methods. This might lead to further improvements in prediction quality and needs to be accompanied by explainable artificial intelligence methods.

Related Link: Paper (PDF)

Keywords: Europe, EU, Banking, Plausibility Checks, Reporting, FINREP, COREP, SUBA, Data Quality Checks, Machine Learning, Artificial Intelligence, Regtech, ECB