Thesis Defence – Elif Feyzioğlu

Thesis Defence – Elif Feyzioğlu

Asst. Prof. Dr. Emrah Ahi– Advisor

Date: 12.05.2026

Time: 16:00

Location: Özyeğin University Altunizade Campus - Classroom ALT 102

“Forecasting Corporate Credit Spread Changes with Time-Series Models and Machine Learning: Evidence from Turkish USD Bonds”

Asst. Prof. Dr. Emrah Ahi, Özyeğin University

Asst. Prof. Dr. Levent Güntay, Özyeğin University

Asst. Prof. Dr. Rıza Ergün Aysal, Bilgi University

Abstract:

This thesis introduces prediction analysis of credit-spread changes by implementing time series model and machine learning method: random forest. The study focuses on eighty-three US dollar based corporate bonds issued by twelve companies listed on Borsa Istanbul. The influenced factors of credit spread change are determined firm financial indicators, global and local market indicators and bond-market series. The weekly data set is first implemented by performing Autoregressive Model with exogenous regressors (ARX), the Autoregressive moving-average model with exogenous regressors (ARMAX) and Vector Autoregression (VAR) models’ predictions at time horizon t to assess the explanatory power of independent variables. Furthermore, the analysis is expanded by using expanding window forecast of time series models. It is considered both time horizon t and t+4. To see whether nonlinear relations provide better prediction performance, Random Forest model is performed. Model performances are assessed based on benchmark relative measures like Theil’s U2, out-of-sample R-squared and Diebold-Mariano statistics while comparing random walk and historical mean benchmark. Model performances’ error metrics which are MSE, RMSE, MAE, MAPE, sMAPE and MASE are considered due to benchmark models. Finally, we evaluate relative importance of independent variables by categorizing them as credit spread, firm financials, global and local markets. The forecast models rerun by dropping these categorized variables to see their effectiveness on forecast models. It is considered based on RMSE and out-of-sample R-squared due to full variable model results.

Keywords: Prediction, Forecasting time series model, Random Forest, ARX, ARMAX, VAR, Time Series Models, Relative Importance

Bio:       

Elif Feyzioğlu, having graduated from the SUNY Economics Department at Istanbul Technical University in 2010, currently works at the Fibabank as an Financial Quality and Governance Manager.