Ülke Kredi Notlarını Etkileyen Faktörlerin Çeşitli Sınıflandırma Analizleri ile İncelenmesi
Ayşe Mine Örender, Selay Giray YakutKredi derecelendirmeleri, Standard and Poor’s Corporation, Moody’s Yatırımcı Servisi ve Fitch Ratings gibi uluslararası derecelendirme kuruluşları tarafından sağlanan kredi riskinin alfabetik göstergeleridir. Kredi notları hükümetlerin kamu borcunu zamanında geri ödeme kabiliyetinin ve istekliliğinin bir değerlendirmesi olduğundan, yatırımcılar, borç veren kuruluşlar ve ilgili piyasa katılımcıları, yayınlanan raporlar doğrultusunda yatırım kararları alabilmektedir. Bu nedenle verilen notlar oldukça önemlidir. Bu çalışmada, 85 ülkenin 2017 yılına ait verisi için lojistik regresyon analizi ve yapay sinir ağları tekniklerinden yararlanılarak Moody’s kredi derecelendirme kuruluşunun ülke kredi notlarını verirken baskın olarak hangi faktörleri ele aldığı belirlenmiş ve verilen kredi notlarına göre ülkeler yatırım yapılabilirlik durumuna göre sınıflara ayrılmıştır. Analizsonucunda, kişi başına düşen gayrisafi yurtiçi hasıla (GSYİH), enflasyon, genel hükümet faiz dışı dengesi / GSYİH, devlet borcu, dış ödemeler ve resmi Forex rezervleri değişkenleri istatistiksel olarak anlamlı bulunmuş, lojistik regresyon modelinin doğru sınıflandırma oranının %90,6 ve yapay sinir ağları modelinin doğru sınıflandırma oranının %88 olduğu sonucuna varılmıştır. Türkiye zaman zaman yatırım “yapılabilir ülkeler” kategorisinde yer alsa da, kredi derecelendirme kuruluşu Moody’s, 2018 Ağustos ayında Türkiye’nin kredi notunu Ba2’den Ba3’e, 2019 Haziran ayında ise B1’e düşürerek not görünümünü durağandan negatife düşürmüştür. Analiz sonucunda da buna paralel olarak kredi notları açısından Türkiye’nin “yatırım yapılamaz” sınıfına dahil edildiği belirlenmiştir.
Investigation of Factors Affecting Sovereign Ratings by Various Classification Analyses
Ayşe Mine Örender, Selay Giray YakutCredit ratings are alphabetical indicators of credit risk provided by international rating agencies such as Standard and Poor’s, Moody’s, and Fitch. Since credit ratings are an assessment of a government’s ability to repay the public debt on time, investors, lenders, and market participants can make investment decisionsin line with published reports. Therefore, the scores given are very important. In this study, the factors handled by Moody’s as sovereign ratings were determined using Logistic Regression Analysis and Artificial Neural Networks for the data of 85 countries for 2017 and the countries were divided into classes according to investment grade. As a result of the analysis, per capita GDP, inflation, general government primary balance / GDP, government debt, external payments, official forex reserves were found to be statistically significant. The correct classification rate of the logistic regression model was found to be 90%, whereas the correct classification rate of the artificial neural network model was found to be 88%. In August 2018, Moody’s downgraded Turkey’s credit rating from ‘’Ba2’’ to ‘’Ba3’’ and in June 2019 to B1 so the rating outlook dropped from stable to negative. Similarly, the analysis concluded that Turkey had been placed onto the list of non-investable countries.
In this study, using logistic regression analysis and artificial neural network techniques on the data of 85 countries for 2017, the factors that were taken into account by Moody’s credit rating agency while determining a country’s credit ratings were determined, and the countries were divided into classes according to investmentability based on the granted credit scores . Within the scope of the study, the data set for year 2017 was taken from Moody’s official website. Moody’s had organized the data under four main headings, which included economic structure and performance, government financing, external payments and debt, monetary-external vulnerability and liquidity indicators. The data set, which included a total of 138 world countries, had been reorganized under the headings of economic structure and performance, state financing, external payments and debt to form the largest data set with common variables. Then, lost observations and extreme values were excluded from the analysis and a final data set consisting of 85 different countries was created . Here, it is also considered that the sample size should be at least five times the number of variables observed. According to studies in the literature 13 independent variables (Per capita GDP, Real GDP, Inflation, Gross Domestic Savings / GDP, Economic Gap, Government Revenue / GDP to Government Expenditures / GDP, General Government Primary Balance / GDP, General Government Debt, External Payments and Debt, Real Effective Exchange Rate, Current Account Balance / GDP, Net Foreign Direct Investment / GDP) are determined as Official Forex Reserves. For the dependent variable, a two category dependent variable was determined as Y = 1, meaning investment can be made, and Y = 2, meaning investment cannot be made , by making use of the investability level determined by Moody’s for long-term credit rating. Missing observations were not included in the analysis and contradictory observations, namely Luxembourg, Ecuador and South Africa, were excluded from the data set. After this, logistic regression analysis and artificial neural network statistical analysis methods were applied to the data set. Logistic regression is one of the alternative techniques that can be used in cases where the dependent variable is categorical as in socio-economic issues. According to the results of the logistic regression application, while the total correct classification rate is 90.6%, according to the predicted results, the correct classification rate of noninvestable countries is 87.9% and the correct classification rate of investable countries is 92.3%. Artificial neural networks are computer-based systems that are developed with the inspiration of the neurons of the human brain and do not require traditional skills. In other words, artificial neural networks are computer networks that try to simulate the neural cell (neurons) networks of the biological central nervous system (Yegnanarayana, B. 2009), (Öztemel E., 2003). In the application of artificial neural networks, the data set containing a total of 85 observations was randomly divided into 2 sets as training data containing 69 observations and test data containing 16 observations. As a result of the application, the training data accuracy rate was found to be 0.88 and the test data accuracy rate was found to be 0.56. In addition, the area under the ROC curve for the training data was 0.90 and the area under the ROC curve for test data was 0.61. A value close to 1.00 indicates a good classification, while a value of 1.00 indicates an excellent classification. In this study, two different methods were used with the most up-to-date data and a sample representing all world countries. Based on the literature review, it was observed that inflation, financial openness, per capita income, technological development, public income, public expenditure, borrowing, current account balance, democracy, adoption of laws and order and less corruption were most effective in granting country credit ratings. Parallel to previous studies and in addition to the application of logistic regression analysis, the most important factors determining whether countries can be invested in or not according to the 2017 country credit rating are GDP, Inflation, General Government Primary Balance / GDP, General Government Debt, Foreign Payments-Debt and Official Forex Reserves. According to the results of the analysis, it was observed that the correct classification rate of the logistic regression model is 90.6% and that of the artificial neural network model is 88%. Also in terms of credit scores it was concluded that Turkey was included in the list of non-investable countries.
