University of Twente Student Theses


Non-linearity issues in probability of default modelling

Klinkers, L.R. (2017) Non-linearity issues in probability of default modelling.

[img] PDF
Abstract:The purpose of this paper is to investigate the accuracy of predicting the probability of default with logistic regression and whether the linearity assumption is violated when multiple risk drivers are included in the model. Violation of the linearity assumption will cause a deviation between predicted PDs and observed PDs. Correcting for this deviation will increase the prediction accuracy of the PD model and therefore the regulatory capital calculation of the Rabobank will more accurately reflect the risks. We suggest making an adjustment to the transformation of client score to PD. This adjustment allows us to identify whether the linearity assumption is violated and estimates the size of the correction that is needed. The great benefit is that the ranking performance based on the creditworthiness of the clients remains the same. The correction is applied before the transformation to probability, so only the absolute value of the PD is affected to improve the prediction accuracy. The average PD prediction error improved from 16% to 4% by correcting the log-odds. The calculated PDs for each clients will therefore represent the corresponding risks, which is essential for efficient capital allocation and RAROC measures.
Item Type:Essay (Master)
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:83 economics
Programme:Industrial Engineering and Management MSc (60029)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page