A country's sovereign risk rating is an indicator that expresses the risk that foreign investors are subjected to when purchasing bonds of the country. The rating is issued by rating agencies, independent government companies or private companies. However, rating agencies have been widely criticized for their lack of transparency in rating processes.
A credit rating is an assessment of each sovereign government's creditworthiness and focuses on political and economic risks and can be both quantitative and qualitative. It takes into account the borrower's intrinsic financial strength, including traditional credit factors such as management quality, market position and diversity, financial flexibility, transparency, regulatory environment, and the issuer's ability to meet its financial obligations across business cycles. including sovereign risks such as vulnerability to political developments, monetary and fiscal policies and, in rare cases, risk of convertibility and transfer of foreign currency. The greater the risk that investors take in acquiring a sovereign government bond, the lesser the government's ability to make this acquisition attractive and, therefore, attract foreign capital. As a result, the higher the premium paid to investors to compensate them for taking on this risk.
Ratings are important not only because some of the largest debt issuers are sovereign governments, but also because according to their attributions, fundraising by states, cities or private companies located in these countries is affected. The large amount of research aimed at predicting sovereign ratings, the relevance of the results obtained in recent years, as well as the need for independent validation due to the lack of transparency are motivations for the work developed under the Bachelor's thesis of Computer Engineering student Diego Ramon, which was an extension of projects developed by Bruno Frascarolli, Professor of Economics (UFPB), in which two studies were carried out using machine learning methods developed under our supervision and that of Professor Bruno.
The first study aimed to predict the sovereign rating from World Bank development indicators using the random forest method (RF). The second verified the hypothesis that the 2008 crisis resulted in a structural rupture in sovereign risk assessments and the degree of development of countries influenced different assessments by agencies.
The dataset with information about country development that was used in the experiment initially had instances corresponding to 137 different countries, with historical data comprising 38 years, from 1958 to 2017, with 22 sovereign classifications.
After applying pre-processing techniques to prepare the data for ML methods and applying PCA to reduce the number of attributes, the original dataset with 3489 instances and 644 attributes was reduced to 3489 instances, with each instance containing 294 attributes.
In addition to the dimensionality reduction provided by the use of principal component analysis, we also applied a clustering technique to the problem's output classes. Thus, there was a reduction in the number of classes through their grouping, seeking an eventual increase in accuracy.
Results show accuracy values of up to 98.28% in the prediction problem. Additionally, the p-value of the hypothesis test showed that there was a structural rupture in the evaluations after the 2008 financial crisis, and that agencies started to assess countries with developed economies in different ways. These results indicate that the agencies' inability to predict the crisis has led to a change in the assessment methodology.
The data suggest that rating agencies assess countries with transition economies and developed economies in different ways. However it is important to point out some details about these results.
In the case of transitioning economies, most of these countries belong to the organization of the Commonwealth of Independent States, a group that includes 11 countries that belonged to the former Soviet Union. Most of these countries have their own contexts that can invalidate these results. For example, Ukraine went through an onset of civil war, which led to the separation of Crimea, that is geopolitical factors may be the cause of different treatment by risk agencies.
As for developed economies, a possible explanation for a different treatment by risk agencies is also the 2008 financial crisis, which forced risk agencies to change their methodology, more precisely, to increase the weights of quantitative inputs in the models. These adjustments led to a decrease in the grades of developed economies. This can also be verified through the large difference between the mean and standard deviation indicating volatility for developed economies.
https://repositorio.ufpb.br/jspui/bitstream/123456789/16065/1/DRBS17072018.pdf