Tim VintisKlassifikationsverfahren zur Risikobewertung von Jahresabschlüssen

About this book deutschenglish

For a large number of external users, information from the annual financial statements serves as an essential source of information for decision-making. Despite numerous control bodies, accounting scandals occur regularly, which put the current control mechanisms to the test. At the same time, auditors are confronted with more complex corporate systems and data as a result of digitalization, making more efficient audits imperative.

The question of whether or not financial statements have been manipulated can be understood statistically as a classification problem. Classification methods of machine learning are able to make adequate risk assessments, on the basis of which a possible case selection of financial statements to be audited can be made and which allows the scope of the audit to be determined in consideration of the risk.

In this study, various machine learning methods as well as the combination of these methods by ensemble methods are presented and their quality is empirically tested on the basis of the area under the curve (AUC) of receiver operating characteristic curves (ROC curves). The data basis are financial statements of German companies for which a non-qualified audit opinion has been issued or an enforcement finding has been announced. The group of non-defective financial statements was identified using a k-nearest-neighbor based matching procedure.

The results show that a high classification quality can be achieved especially by using ensemble methods. The practical application of the methods is illustrated using the case of Wirecard AG. Taking into account further criteria such as simplicity, understandability and implementability, the possible applications in the current control system of annual financial statements are discussed.

Ihr Werk im Verlag Dr. Kovač

Möchten Sie Ihre wissenschaftliche Arbeit publizieren? Erfahren Sie mehr über unsere günstigen Konditionen und unseren Service für Autorinnen und Autoren.

Dissertation veröffentlichen