University of Twente Student Theses


Estimation of Asset Correlation for Residential Mortgages

Klück, A (2020) Estimation of Asset Correlation for Residential Mortgages.

[img] PDF
Abstract:Asset correlations are key parameters for banks to determine their economic capital (EC). Their importance arises from industry experience that the economy at large affects changes in the creditworthiness of borrowers. This thesis focuses on methods to estimate these parameters. Next, we apply these methods to Dutch residential mortgage default rates. Furthermore, mortgage characteristics and ratings divide the portfolio into groups to evaluate differences among groups of borrowers in terms of their asset correlations. The other part of the thesis explores the relationship between macroeconomic variables and default rates, by developing regression models. In addition, macroeconomic scenarios provide prospective default rates and correlation parameters derived from these rates. An auxiliary Vasicek model assesses the estimators' performance and consistency on generated default rates. From the latter it can be concluded that all estimators underestimate the asset correlation parameter by 20 to 50% depending on the time series length. The estimator (MM5) that accounts for autocorrelation and finite time series, shows the lowest underestimation. The asset correlation estimates on portfolio level range from A to B% indicating relatively low sensitivity of defaults to macroeconomic conditions. Inclusion of stress scenarios raises the estimates to C to D%. All in all, the results show that banks should be careful adopting asset correlation estimates and conservatism is needed to mitigate the hazard of using underestimated parameters as this yields disproportionate EC.
Item Type:Essay (Master)
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:31 mathematics, 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