Makroekonomik veriler ile NPL tahmini: Bir Türkiye uygulaması
NPL FORECASTING WITH MACROECONOMIC DATA: THE TURKEY APPLICATION
- Tez No: 955596
- Danışmanlar: DR. ÖĞR. ÜYESİ MEHMET ALİ ERGÜN
- Tez Türü: Yüksek Lisans
- Konular: Endüstri ve Endüstri Mühendisliği, Industrial and Industrial Engineering
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2025
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Lisansüstü Eğitim Enstitüsü
- Ana Bilim Dalı: Endüstri Mühendisliği Ana Bilim Dalı
- Bilim Dalı: Endüstri Mühendisliği Bilim Dalı
- Sayfa Sayısı: 82
Özet
Bu araştırma, takipteki krediler (NPL) oranının makroekonomik göstergeler yardımıyla tahmin edilmesini amaçlayan kapsamlı bir çalışmadır. Finansal istikrarın korunması ve kredi risklerinin etkin yönetimi açısından kritik öneme sahip olan bu konu, 2015-2023 yılları arasındaki dokuz yıllık dönemin aylık verileri kullanılarak analiz edilmiştir. Çalışmada, toplam 61 farklı makroekonomik değişkenin NPL oranları üzerindeki etkisi sistematik olarak incelenmiştir. Bu değişkenler arasında enflasyon göstergeleri (TÜFE, ÜFE), faiz oranları (politika faizi, gösterge faizler), ekonomik büyüme göstergeleri (GSYİH), işsizlik oranları, döviz kurları, konut ve otomotiv sektörü verileri, borç stoklarına ilişkin veriler ve çeşitli güven endeksleri yer almaktadır. Araştırmanın özgün yanlarından biri, makroekonomik şokların NPL oranları üzerindeki etkilerinin hangi zaman diliminde en belirgin şekilde ortaya çıktığını tespit etmeye yönelik sistematik yaklaşımdır. Bu amaçla, 3 aylık periyotlarla 24 aya kadar uzanan farklı gecikme sürekleri test edilmiştir. Her bir gecikme periyodu için ayrı modeller oluşturularak performans karşılaştırmaları yapılmıştır. Analiz sonuçları, makroekonomik faktörlerin NPL oranları üzerindeki etkisinin 9 aylık gecikme ile en güçlü seviyeye ulaştığını ortaya koymuştur. Bu gecikme süresi ile oluşturulan model, eğitim veri seti için %91, test veri seti için %86 oranında açıklayıcılık gücü sergilemiştir. Bu bulgular, ekonomik şokların bankacılık sektörü üzerindeki yansımalarının belirli bir zaman dilimi gerektirdiğini ve bu sürecin yaklaşık dokuz ay sürdüğünü göstermektedir. Çalışmanın regresyon varsayımları testleri (normallik, homojenlik, otokorelasyon kontrolü) sonucunda, oluşturulan modelin istatistiksel açıdan geçerli ve güvenilir olduğu doğrulanmıştır. Bu durum, modelin pratik uygulamalarda kullanılabilirliğini desteklemektedir. Araştırma sonuçları, bankacılık sektörü için çeşitli stratejik öneriler sunmaktadır. Erken uyarı sistemlerinin geliştirilmesinde dokuz aylık öngörü süresinin dikkate alınması, dinamik karşılık politikalarının bu zaman dilimi çerçevesinde şekillendirilmesi ve kredi portföy optimizasyonunda makroekonomik risk faktörlerinin sistematik olarak değerlendirilmesi temel öneriler arasında yer almaktadır. Düzenleyici otoriteler açısından ise makro ihtiyati politikaların zamanlaması ve stres testlerinin tasarımında bu gecikme etkisinin göz önünde bulundurulması önem taşımaktadır. Bu çalışma, Türkiye bankacılık sektörü özelinde NPL tahmininde makroekonomik faktörlerin rolünü sistematik olarak analiz eden araştırmalardan biri olma özelliği taşımakta ve gelecekteki akademik çalışmalar ile sektörel uygulamalar için sağlam bir temel oluşturmaktadır.
Özet (Çeviri)
The banking sector serves as the cornerstone of financial systems, playing a critical role in overall economic health. Financial stability of these institutions affects not only their individual performance but also holds significant importance for macroeconomic stability. One of the critical risks faced by banks is credit risk, measured through non-performing loans (NPL) ratios. The NPL ratio represents one of the fundamental indicators used in evaluating the quality of a bank's credit portfolio and assessing the bank's financial health. High NPL ratios can lead to increased credit losses for banks, reduced profitability, and ultimately decreased capital adequacy ratios. The prolonged continuation of this situation can result in systemic risks within the banking sector and contribute to economic crises. Changes in macroeconomic conditions directly impact NPL ratios, as fluctuations in macroeconomic indicators such as interest rates, employment indicators, economic growth, inflation, and exchange rates affect borrowers' repayment capacity, leading to increases or decreases in credit risk depending on the importance and magnitude of change in the relevant parameter. This study aims to forecast NPL ratios using macroeconomic variables for a bank operating in Turkey. The research examines the effects of 61 different macroeconomic variables on NPL ratios using monthly data from 2015-2023. The study employs Partial Least Squares (PLS) regression methodology, creating models with different time lags to determine the optimal model. The importance of this study lies in using macroeconomic variables to forecast NPL ratios, which are critical indicators for the banking sector, and testing the predictability of these forecasts for a specific time period. The research addresses several key questions: What is the relationship between macroeconomic variables and NPL ratios? Which macroeconomic variables are more effective in predicting NPL ratios? What is the optimal time lag for NPL prediction? How effective is the PLS regression model in predicting NPL ratios? How do the predictive power and accuracy of the constructed models vary? The study utilizes monthly data from January 2015 to June 2023 for a bank operating in Turkey, combined with macroeconomic data. The dataset consists of one dependent variable (NPL ratio) and 61 independent variables (macroeconomic indicators). Some independent variables were used directly as obtained, while others were derived from raw data through calculations such as addition, subtraction, or converting quarterly data to monthly using linear interpolation methods. The dependent variable is the bank's monthly NPL ratio, calculated as: NPL Ratio = (Non-Performing Loans / Total Loans) × 100. Non-performing loans include credits where principal or interest payments have been delayed for more than 90 days. A crucial aspect in this calculation was the write-off adjustment process, which is significant for ensuring the target variable remains unmanipulated. The 61 independent variables consist of macroeconomic indicators including inflation indicators (CPI, PPI), interest rates (policy rate, benchmark rates), economic growth indicators (GDP), unemployment rates, exchange rates, housing and automotive sector data, debt stock data, and various confidence indices. These variables were obtained from sources such as the Central Bank of the Republic of Turkey (CBRT), Turkish Statistical Institute (TURKSTAT), and official association websites. The original aspect of the research involves a systematic approach to determine the time period in which macroeconomic shocks' effects on NPL ratios emerge most prominently. For this purpose, different lag periods extending up to 24 months in 3-month intervals were tested. Separate models were created for each lag period, and performance comparisons were conducted. The PLS regression method was chosen due to its advantages in handling multicollinearity issues common among macroeconomic variables and its effectiveness when the number of independent variables exceeds the sample size. PLS regression transforms independent variables into components that best explain the dependent variable and models the relationship between these components. Additional methodologies were employed for comparison, including Extreme Gradient Boosting (XGBoost), Random Forest algorithms, and moving average time series forecasting. Model performance was evaluated using R² (coefficient of determination), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE) metrics. The analysis revealed that macroeconomic factors' impact on NPL ratios reaches its strongest level with a 9-month lag. The model created with this lag period demonstrated an explanatory power of 91% for the training dataset and 86% for the test dataset. These findings indicate that economic shocks' reflections on the banking sector require a specific time period, and this process lasts approximately nine months. The variable importance analysis for the 9-month lag PLS model revealed that unemployment rate, indicative-policy gap, oil prices, and indicative-policy area were the most effective variables in NPL prediction. The unemployment rate emerged as the most important variable with a VIP score of 1.78, followed by the indicative-policy gap (1.73) and oil prices (1.64). These results support the hypothesis that NPL ratios tend to increase during economic contraction periods. Cross-validation tests confirmed the robustness of the model results. When the dataset was randomly divided into four folds, with each fold serving as a test set while others formed the training set, the model maintained consistent performance across all folds. However, when the division was made chronologically (periodic test set separation), the model performance significantly deteriorated, indicating that the model requires observations from each period in the training set for successful results. Comparison with alternative methodologies showed that while Random Forest and XGBoost algorithms achieved high training performance, they suffered from overfitting problems. The Random Forest model achieved training R² values close to 1.0 but showed significantly lower test performance, with the best test R² being 0.90 for the 21-month lag. Similarly, XGBoost models exhibited even more severe overfitting, with training R² values approaching 1.0 but inconsistent test performance. The three-month moving average method showed limited success, with only 3-month and 6-month lag models achieving positive R² values, though these were still considered unsuccessful. The 9-month lag moving average model yielded an R² of -1.23, indicating poor predictive capability. Regression assumption tests confirmed the statistical validity and reliability of the PLS model. The Shapiro-Wilk test for normality (p-value = 0.508 > 0.05) indicated that residuals follow a normal distribution. The Breusch-Pagan test for homogeneity (p-value = 0.99 > 0.05) showed that residual variance is homogeneous. The Durbin-Watson test (DW = 2.17) indicated no significant autocorrelation among residuals. The research results offer various strategic recommendations for the banking sector. In developing early warning systems, consideration of the nine-month forecast period is crucial. Dynamic provisioning policies should be shaped within this timeframe, and macroeconomic risk factors should be systematically evaluated in credit portfolio optimization. For regulatory authorities, it is important to consider this lag effect in the timing of macroprudential policies and stress test design. The 9-month lag finding is consistent with international literature. Similar studies by Louzis et al. (2012) in Greece and Bofondi and Ropele (2011) in Italy found comparable lag effects of 6-12 months and 3-9 months, respectively. This consistency suggests that the transmission mechanism of macroeconomic shocks to banking sector credit quality follows similar patterns across different economies. The study's forecasts indicate that NPL ratios are expected to stabilize in the upcoming 9-month period, with projected values ranging between 3.6% and 3.9%. This stabilization suggests that recent macroeconomic improvements may begin to positively impact credit quality with the identified lag. The superior performance of PLS regression over machine learning algorithms in this context highlights the importance of choosing appropriate methodologies for specific data characteristics. While machine learning algorithms often show promise in financial forecasting, the presence of multicollinearity and limited sample size in this study favored the dimension reduction approach of PLS regression. This study contributes to the literature by being among the first to systematically analyze the role of macroeconomic factors in NPL prediction specifically for the Turkish banking sector. The comprehensive approach using 61 macroeconomic variables and the systematic examination of lag effects provide valuable insights for both academic research and practical applications in risk management. Future research directions include sectoral disaggregation of NPL ratios, incorporation of bank-specific characteristics, and exploration of alternative modeling techniques. The study's limitations include the focus on a single bank and the limited time series length, which could be addressed in future research with broader institutional coverage and longer observation periods. The findings establish a solid foundation for developing more sophisticated early warning systems and risk management frameworks in the Turkish banking sector, contributing to enhanced financial stability and more effective regulatory oversight.
Benzer Tezler
- Ticari bankalarda likiditenin kârlılık üzerine etkisi: Türk bankacılık sektörü üzerine bir uygulama
The effect of liquidity on profitability in commercial banks: Evidance from Turkish banking sector
AHMET KARAKAŞ
- Sorunlu kredilere etki eden faktörler: Farklı gelişmişlik düzeyindeki ülkeler üzerine bir araştırma
Factors affecting non-performing loans: A study on countries with different levels of development
NUR ESRA BEKERECİ
Doktora
Türkçe
2023
İşletmeKahramanmaraş Sütçü İmam Üniversitesiİşletme Ana Bilim Dalı
PROF. DR. MUHAMMET MUSTAFA KISAKÜREK
- Makroekonomik faktörlerin Türk bankacılık sektörünün takipteki kredileri üzerine etkisi
The impact of macroeconomic factors on the non-performing loan of the Turkish banking sector
HALİLCAN KARAKAYA
Yüksek Lisans
Türkçe
2025
EkonomiDokuz Eylül Üniversitesiİktisat Ana Bilim Dalı
PROF. DR. SELİM ŞANLISOY
- Takipteki kredilerin ve makroekonomik göstergelerin doğrusal olmayan ekonometrik yöntemler ile analizi
Analysis of following loans and macroeconomic indicators with nonlinear econometric methods
MERVE BİLGİN ÖNCEL
Yüksek Lisans
Türkçe
2023
BankacılıkMarmara ÜniversitesiEkonometri Ana Bilim Dalı
DOÇ. DR. SELİN DEVRİM ÖZDEMİR YAZGAN
- Makroekonomik ve bankaya özgü faktörlerin ticari bankalarda kredi riski üzerindeki etkisi - Türkiye örneği
Macroeconomic and bank specific factors' impact on credit risk in commercial banks – The case of Turkey
NİGAR HACIYEVA
Yüksek Lisans
Türkçe
2019
BankacılıkHacettepe Üniversitesiİşletme Ana Bilim Dalı
PROF. DR. SEMRA KARACAER