Reconstruction Rating Model of Sovereign Debt by Logical Analysis of Data

Here this thesis follows two distinct objectives. First, it reconstructs the rating system of the Fitch credit agency. Second, it analyses the role played by the COVID-19 pandemic on the Fitch agency`s sovereign debt rating. Dataset and main features of both objects were collected by the World Bank (WB) and the International Monetary Fund (IMF). Out of 250 countries, after feature engineering, 67 countries have been trained to figure out the hidden patterns of different ratings and to verify the result, 39 countries were studied as the test set for the thesis. Reconstructed patterns of each rating have been demonstrated as decision trees. The Logical Analysis of Data as a supervised learning technique classifies the countries into 16 labels of the Fitch rating agency. The reconstructed rating method`s consequences were compared to the published countries` ratings by Fitch and approximately more than 85% of the matched ratings over 4 years confirmed the high accuracy of our method. The second object of the thesis analyzed the impact of COVID-19 for the initial months of the pandemic which has been observed to mostly downgraded and some upgraded countries` ratings by Fitch. The high accuracy of the Logical Analysis of data was approved for the second part of the study also.