The discipline of predicting company defaults is of great economic significance both due to the consequences for affected businesses and individuals, and due to the implications for investment and lending activity. This thesis investigates the discipline through an academic, practical and methodological lens, with the aim of contributing to the improvement of the predictive techniques available. The thesis evaluates three hypotheses deducted from the pertinent theory. In the first hypothesis, it examines to what extent the classic academic models of default probability have discriminative power. In the second and third hypothesis, it considers through what measures the practical model of credit risk can be improved. In the analysis of to what extent the classic academic models of default probability have discriminative power, the thesis applies the original frameworks of Altman, Ohlson, Merton and their respective reestimations on a modern portfolio. The portfolio distinguishes itself from the related literature with a scope simultaneously covering a wide range of OECD countries and multiple industries. This analysis concludes that the classic academic approaches to credit risk do have significant discriminative power in a contemporary setting. The subsequent analysis of through what measures the practical model of credit risk can be improved is split into two parts. The first is academically driven and expand upon the practical model by combining the accountingbased and market-based paradigms of the field. This framework is labelled the “Default Model of Synthesis”, and the thesis finds that it is superior to the model relied upon by practitioners. The second is methodologically driven and transcends a machine learning algorithm into the sphere of default probability. The research concludes that the practical model can be improved significantly by applying a random forest methodology. This finding also serves as a platform for discussing the implications of machine learning in the practical discipline of default probability. The discussion points toward interpretability and “institutional stickiness” as non-exclusive explanations for the neglection of machine learning amongst practitioners. Ultimately, the research contributes to the field of default probability both practically and academically. First, it conceptualizes a framework that relies on market and accounting theory in synthesis. Second, it extends the practical model to default probability through the application of a random forest algorithm.
|Educations||MSc in Finance and Accounting, (Graduate Programme) Final ThesisMSc in Applied Economics and Finance, (Graduate Programme) Final Thesis|
|Number of pages||120|