I've made a mistake in the implementation of Pykthin Multi-factor adjustment which I'm fairly certain comes from me not understanding the model completely. The model was developed to drastically reduce computational time by using an analytic approach. When I try to implement it I found that it is much slower than the Monte Carlo approach that it is to replace....

The portfolio I'm using consists of 10 000 exposures. The way i understand Pykhtin's framework this means i must look at the all the asset correlations between these as follows:

$$\rho_{ij}^Y=\frac{r_i r_j \sum_{k=1}^N\alpha_{ik} \alpha_{jk}-a_i a_j}{\sqrt{(1-a_i^2)(1-a_j^2)}}$$

I guess that there is some symmetry that could make it a bit easier (i'm thinking n over k) but that is still a correlation matrix with 10 000 x 10 000 elements. As it stands now i cannot calculate this in matrix form due to lack of memory and using a loop takes forever...

I'm hoping someone is familiar with the model and knows where I'm wrong.

Here is a link to his paper that I'm using: "Multi-factor adjustment", by Michael Pykhtin, Risk, March 2004 http://www.ressources-actuarielles.net/EXT/ISFA/1226.nsf/0/e9c944cf9ab30ac9c12577b4001e0342/$FILE/Pykhtin-Multi-fractor%20adjustment.pdf

Edit: I'm now thinking that since I have a small number of variations of credit scores (8) I can calculate all the combinations of factors (sectors) and credit scores 8x13 for me and most calculations could be classed as one of these and easily summed up.

Edit 2: I think this might have worked. Essentially i made my portfolio of 10 000 exposures to a portfolio of 8x13 (credit scores x sectors) and summed up the appropriate variables so that this made sense. Then I just preceded with the calculations.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.