Reflections on Text Mining Approaches in Corporate Failure Prediction based on German Financial Statements

by Tobias Nießner
Cumulative thesis
Date of Examination:2023-04-25
Date of issue:2023-05-05
Advisor:Prof. Dr. Matthias Schumann
Referee:Prof. Dr. Jan Muntermann
Referee:Prof. Dr. Stefan Dierkes
Persistent Address: http://dx.doi.org/10.53846/goediss-9874

 

 

Files in this item

Name:Dissertation_PRINT.pdf
Size:3.11Mb
Format:PDF

The following license files are associated with this item:

Abstract

English

The field of corporate risk assessment is in a state of flux due to society's increased expectations of statistical methods, which are classified as part of the field of machine learning. While in earlier times multivariate methods were used to calculate bankruptcy probabilities, nowadays a new picture is emerging. The analysis of structured data alone is no longer sufficient to reflect diverse developments on the market and in companies and to find an explanation for them accordingly. The analysis of unstructured data is increasingly coming into focus, as increased computing capacities now also make it possible to evaluate and transform it into relevant key figures. This dissertation starts exactly at this point and shows to what extent the classical quantitative oriented balance sheet analysis can be automatically supplemented by the qualitative analysis of unstructured data from annual financial statements. Using the example of a data mining project, not only text mining approaches known from the literature are classified, but also interdisciplinary approaches are examined in order to reflect causal explanations for the financial situation of a company in addition to correlations relevant for machine learning. In this context, it should be emphasized that so far hardly any specified text mining approaches exist that explicitly deal with the focused use case, and the broad literature has so far mainly addressed the adaptation of well-known methods. The paper shows its added value in reflecting on the design of text mining approaches in forecasting the financial performance of companies, pointing out major weaknesses in the research field and especially in the development of AI-based solutions. In summary, the inclined reader is presented with considerations that critically, but forward-looking, suggest ways for a structured merging of exploratory approaches in terms of refocusing the research field. Practitioners are given direct hints how text mining methods can be implemented as well as used to make unstructured data usable for their statistical models and to transform them in the sense of transparency. It is also critically questioned to what extent time analyses and accordingly company profiles can represent a new status quo and to what extent the analysis of unstructured data can be adapted in this respect.
Keywords: Text Mining, Financial Statement Analysis, Corporate Failure Prediction
 
Statistik