Credit ratings are expensive to obtain and there is a risk of bias in the credit rating process. It is therefore interesting to analyze the credit rating process to see if it can be automated. In this thesis it has been investigated how the credit ratings of firms in the manufacturing industry as well as in the retail industry can be modelled using neural networks and logistic regression. Data sets of US credit ratings from Standard and Poor’s for the period 1 January 2013 - 31 December 2017 were used together with publicly available accounting data. For both industries the neural network outperformed the logistic regression, in both terms of accuracy and Cohen’s kappa. In most previous research using neural networks to predict credit ratings, focus have solely been on the predictive performance. In this thesis the neural networks was further analyzed using the connection weight approach. The analysis showed that total assets is a especially important explanatory variable for Standard & Poors’s credit ratings on manufacturing firms. For the retail firms the cash flow from operations to current liabilities was found to be important. Furthermore, earnings per share was found to be among the most important variables for both industries.
|Educations||MSc in Applied Economics and Finance, (Graduate Programme) Final Thesis|
|Number of pages||51|