The European Central Bank (ECB) published a paper that presents a data-driven approach, based on machine learning, for discovering new plausibility checks on reporting data. The paper aims to provide a further tool to support the development of new checks that extend the validation rules to automatically detect potential quality issues requiring further investigation. The presented approach uses large amounts of historical data to identify patterns of interest in past observations. These patterns are suitable for discovering data anomalies and can serve as a basis for defining new checks. The paper illustrates how such patterns are used by business experts to refine their data quality framework. The paper also provides suggestions for further work that could be done to improve technical performance and prediction quality.
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
The European Banking Authority (EBA) published four draft principles to support supervisory efforts in assessing the representativeness of COVID-19-impacted data for banks using the internal ratings based (IRB) credit risk models.
The Bank for International Settlements (BIS) Innovation Hub updated its work program, announcing a set of projects across various centers.
Certain members of the U.S. Senate Committee on Banking, Housing, and Urban Affairs issued a letter to the Securities and Exchange Commission (SEC)
The European Insurance and Occupational Pensions Authority (EIOPA) published a consultation paper on the advice on the review of the securitization prudential framework in Solvency II.
The Prudential Regulation Authority (PRA) issued a statement on PRA buffer adjustment while the Bank of England (BoE) published a notice on the statistical reporting requirements for banks.
The Federal Financial Supervisory Authority of Germany (BaFin) proposed to amend the “Capital Investment Conduct And Organization Ordinance” and issued a draft circular on the minimum resolvability requirements for resolution planning.
The European Banking Authority (EBA) proposed guidelines, for the resolution authorities, on the publication of the write-down and conversion and bail-in exchange mechanic, with the comment period ending on September 07, 2022.
The Financial Services Authority of Indonesia (OJK) is strengthening cooperation with the Australian Prudential Regulation Authority (APRA) and the Japanese Financial Services Agency (JFSA)
The European Parliament and the Council published Regulation 2022/868 on European data governance (Data Governance Act).
The European Banking Authority (EBA) published phase 2 of its reporting framework 3.2. The technical package supports the implementation of the updated reporting framework by providing standard specifications