| Abstract [eng] |
Credit rating agencies are responsible for issuing credit ratings for countries and companies. This rating asseses the debtor`s ability to repay the debt on time and probability that the obligations will not be met. Approximately 95% of the credit rating industry is managed by the 3 largest credit rating agencies (often referred to as the Big Three): Standard & Poor's, Moody's and Fitch. As a result, competition in the credit rating industry is very limited. The "Big Three" agencies are constantly criticized for their decisions. For instance, before the financial crisis of 2007 - 2008, agencies failed to properly assess the risks of financial instruments. This lead to a significant impact on the global economic downturn. These days, agencies are also criticized for disproportionately managing their credit ratings, making borrowing more expensive and encouraging people to "tighten their belts." The Master's Final degree project examines the relationship between changes in ratings issued by credit agencies and the real financial situation of banks. The aim of the thesis is to study the changes in ratings made by the Big Three credit agencies using historical corporate financial data and machine learning algorithms. Thesis also analyzes whether the ratings made by the credit agencies are based on real financial indicators or whether they are made through media channels and political decisions. The work used real data obtained from the Bloomberg software system. Machine training models were constructed by applying 89 financial variables. Detection, classification, and survival analysis tasks were performed to make informed predictions. The detection models predict whether the credit rating of the investigated bank will decrease or will not due to a change in certain financial indicators over a period of 1 year. Logistic regression, random forest and support vector models were used to perform the detection task. The changes in Moody's rating were most accurately predicted. A random forest model with a ROC AUC of 0.92 was used for this prediction. The classification task analyzes whether, after a change in financial ratios, the credit rating of the analyzed bank will decrease, remain unchanged or increase over a period of 1 year. The classification problem is solved by applying classifiers of random forests and support vectors. All 3 classes were most reliably separated by the random forest model when analyzing Fitch's results. The overall accuracy of the model was 65.336%. Using the survival analysis method, the average daily probability that the credit rating would decline was calculated. We analyze the period up to 1 year after the changes in the related financial indicators. Survival forest and Cox models were used to solve the problem. The survival analysis models evaluated the changes in the ratings of all 3 credit agencies very similarly. |
