Explainable Artificial Intelligence Methods in FinTech Applications

The increasing amount of available data and access to high-performance computing allows companies to use complex Machine Learning (ML) models for their decision-making process, so-called ”black-box” models. These ”black-box” models typically show higher predictive accuracy than linear models on complex data sets. However, this improved predictive accuracy can only be achieved by deteriorating the explanatory power. ”Open the black box” and make the model predictions explainable is summarised under the research area of Explainable Artificial Intelligence (XAI). Using black-box models also raises practical and ethical issues, especially in critical industries such as finance. For this reason, the explainability of models is increasingly becoming a focus for regulators. Applying XAI methods to ML models makes their predictions explainable and hence, enables the application of ML models in the financial industries. The application of ML models increases predictive accuracy and supports the different stakeholders in the financial industries in their decision-making processes. This thesis consists of five chapters: a general introduction, a chapter on conclusions and future research, and three separate chapters covering the underlying papers. Chapter 1 proposes an XAI method that can be used in credit risk management, in particular, in measuring the risks associated with borrowing through peer-to-peer lending platforms. The model applies correlation networks to Shapley values and thus the model predictions are grouped according to the similarity of the underlying explanations. Chapter 2 develops an alternative XAI method based on the Lorenz Zonoid approach. The new method is statistically normalised and can therefore be used as a standard for the application of Artificial Intelligence (AI) in credit risk management. The novel ”Shapley-Lorenz”-approach can facilitate the validation of model results and supports the decision whether a model is sufficiently explained. In Chapter 3, an XAI method is applied to assess the impact of financial and non-financial factors on a firm’s ex-ante cost of capital, a measure that reflects investors’ perceptions of a firm’s risk appetite. A combination of two explanatory tools: the Shapley values and the Lorenz model selection approach, enabled the identification of the most important features and the reduction of the independent features. This allowed a substantial simplification of the model without a statistically significant decrease in predictive accuracy.