Evaluating model performance in SME scoring: A comparison of logistic regression and survival analysis

This doctoral thesis aimed to develop an estimation method for the probability of default (PD) in small and medium-sized enterprises (SMEs) by integrating a comprehensive set of variables (firm-specific, credit bureau, financial, macroeconomic, demographic, and other variables) into survival and logistic models. Using advanced techniques such as bucketing and data transformations, the study led to the development of regression models by identifying influential variables. This would allow credit risk to be assessed in a more dynamic, time-sensitive manner. Traditional credit scoring models always rely on financial ratios and historical performance; they hardly put enough emphasis on wide economic and environmental factors affecting the survival of SMEs. The literature suggests that survival analysis needs to be considered in credit scoring models to obtain a more accurate assessment of SMEs in terms of the risk of defaults. In the thesis, the model generated by survival analysis did not outperform existing models but has some advantages over them. Since the model estimates the probability of survival across the entire dataset, not just a specific time period as in logistic regression, this model can be used with significantly less data cleaning. Survival analysis has advantages over Logistic regression by predicting when the loan is default. In this thesis, while the observation period is between 2010-2020, the data is tested in 12-month periods which is generally accepted in scoring models. In further research, this period can be longer durations (24, 36, 48 months) in order to capture the real predictive power of logistic and survival models in different time intervals. Additionally, by continuously updating with real-time data and feedback mechanisms, institutions can improve their ability to more accurately predict default probabilities over time. This adaptability is vital given the rapidly changing economic landscape where macroeconomic shocks can significantly impact SMEs.