Publication: İstatistiksel makine öğrenme yöntemleriyle bankaların derecelendirilmesi : Türkiye bankacılık sistemi uygulaması
Abstract
Derecelendirme kuruluşları tarafından yapılan finansal derecelendirmeler, özel ya da kamu şirketlerine veya devlete verilen kredilerin geri ödenmemesi riskini değerlendirmek için kullanılan bir araçtır. İyi bir derecelendirmeye sahip olmak, söz konusu şirketin finansal sağlığını ve güvenilirliğini ortaya koymak açısından önemlidir. Bu tezde, Türkiye Cumhuriyeti'ndeki bankaların finansal oranları kullanılarak derecelendirilmesi amaçlanmaktadır. Bu amaçla istatistiksel öğrenme yöntemlerinden yararlandık. Türkiye bankacılık sisteminde 2001 yılında varlık ve bilanço denetleme sisteminde yapılan değişikliklere kadar pek çok finansal kriz yaşanmıştır. Bu krizler neticesinde banka iflasları gözlemlenmiştir. Banka iflasları genel olarak ekonomik çöküntüye yol açabileceğinden bankaların düzenli olarak izlenmesi çok önemlidir. Genelde bu izleme bankaların yıllık bilançolarının takibi üzerinden yapılmaktadır. Yıllık bilançosuna bakarak, bir bankanın iflas edip etmediğini açıkça söyleyebiliriz ancak bankalar arasında hangi bankanın diğerinden daha fazla çökmeye yakın olduğunu söylemek zor olabilir. Bu çalışmanın amacı bankaların başarı düzeyleri açısından sıralanmalarını sağlayan bir derecelendirme sistemi önermektir. Bu çalışmada kullanılan veriler Türkiye Bankalar Birliği'nin (TBB) web sitesinden elde edilmiştir. TBB'den gelen veriler iki türe ayrıldığı için (iki bin bir yılından önceki bilançolardan ve iki bin bir yılından sonra bilançolardan oluşan veriler) her şeyden önce bu verilerin uyumunu kontrol edip gereken düzeltmeleri yaptık. Sonra uygun veri düzenleme ve temizleme yöntemleri uygulayarak temel istatistiksel hesaplama işlemleri yaptık. Ardından, literatürde daha önce yapılmış bilimsel çalışmaları inceleyip alternatif bir derecelendirme modeli oluşturmak için istatistiksel öğrenme yöntemlerinden yararlandık.
Financial ratings made by rating agencies are a tool used to assess the risk of non-repayment of loans given to private or public companies or to governments. Having a good rating is important to demonstrate the financial health and reliability of the company in question. In this thesis, it is aimed to rate banks of the Republic of Turkey using financial ratios. For this purpose, statistical learning methods were used. Up to the balance sheet assets and audit changes made to the Turkey Banking System in 2001, there has been too much financial crisis. As a result of these crises, bankruptcies were observed. Regular monitoring of banks is very important, as bankruptcy can lead to economic downturn in general. In general, this monitoring is carried out by following the annual balance sheets of the banks. We can clearly say whether one bank went bankrupt or not by looking at their annual balance sheet, but it can be difficult to tell which bank is close to collapse among several banks. The aim of this study is to propose a rating system that allows to rank banks in terms of their success levels. The data used in this study were obtained from the Banks Association of Turkey’s (TBB) web site. Since the data from the TBB is divided into two types (balance sheets before the year two thousand one and balance sheets after the year two thousand one), first of all, the compliance of these data was checked and necessary corrections were made. Then, basic statistical calculations were performed by applying classical data editing and cleaning methods. Afterwards, statistical learning methods was used in order to form an alternative rating model by examining the previous scientific studies in the documentatons.
Financial ratings made by rating agencies are a tool used to assess the risk of non-repayment of loans given to private or public companies or to governments. Having a good rating is important to demonstrate the financial health and reliability of the company in question. In this thesis, it is aimed to rate banks of the Republic of Turkey using financial ratios. For this purpose, statistical learning methods were used. Up to the balance sheet assets and audit changes made to the Turkey Banking System in 2001, there has been too much financial crisis. As a result of these crises, bankruptcies were observed. Regular monitoring of banks is very important, as bankruptcy can lead to economic downturn in general. In general, this monitoring is carried out by following the annual balance sheets of the banks. We can clearly say whether one bank went bankrupt or not by looking at their annual balance sheet, but it can be difficult to tell which bank is close to collapse among several banks. The aim of this study is to propose a rating system that allows to rank banks in terms of their success levels. The data used in this study were obtained from the Banks Association of Turkey’s (TBB) web site. Since the data from the TBB is divided into two types (balance sheets before the year two thousand one and balance sheets after the year two thousand one), first of all, the compliance of these data was checked and necessary corrections were made. Then, basic statistical calculations were performed by applying classical data editing and cleaning methods. Afterwards, statistical learning methods was used in order to form an alternative rating model by examining the previous scientific studies in the documentatons.