A comparison of two coupling schemes for modeling dependent credit rating transitions

Speaker(s): 
Kaniovskyi Yuriy
Date: 
Monday, November 5, 2012 - 2:00pm
Location: 
Spandauer Strasse 1, Room 23

Two coupling methods for Markov chains are considered: KP, the model suggested by Kaniovski and Pflug (2007), and WH, its modification by Wozabal and Hochreiter (2012). It is shown that they generate statistically distinguishable distributions of dependent credit rating transitions. n particular, the variance of the distribution of defaults for the KP variant is typically higher than for the WH modification. Using a Standard and Poor's data set covering 30 OECD countries from 1991 through 2006, the parameters of both the KP and WH model are estimated. The debtors are classified into two non-default credit classes. Two partitions of the pool of debtors are considered: with 6 and with 12 industry sectors. The maximum likelihood estimates obtain by the MATLAB optimization software: the Interior Point Algorithm (IP) and the Least Squares Algorithm (LS).