[R-SIG-Finance] transition matrices and "robustness"

[Sean O'Riordain]

Thanks for that Debashis,

Well the transition matrices (TM)  themselves are interesting - e.g.
we might have a 6% chance of a NORM going OD, and a 86% chance that an
OD pays up and goes NORM again.  So I was looking at these TMs
breaking out different industry sectors or loan purposes etc... as
they give us a better and more objective handle on the relative
problem size in different groups of loans.

Then there is the possibility further down the road of modeling the
entire book in the near term - i.e. over the next three months.  Or
even looking at how the book might be cut into similar clusters... but
I've no clue how I might do this - I can understand the concept of
linear cluster "distance" but not when the "distance"  is a transition
matrix! (computational intensity aside!)

Many thanks again,
Sean



On Wed, Nov 5, 2008 at 7:55 PM, Debashis Dutta <dutt.debashis at gmail.com> wrote:
> Dear Sean,
>
> CreditMetrics  package in R is a good resource. However, it would convenient
> for me to comment further if you please be a little specific regarding  your
> end purpose.
>
> Kind Regards,
> Debashis
>
>
> On 05/11/2008, Sean O'Riordain <seanpor at acm.org> wrote:
>>
>> Good evening,
>>
>> I'm quite new to the risk modeling arena and recently I starting to
>> look at some loans and their state transitions. For simplicity I will
>> describe a cut-down version - NORM=normal, OD means between 1 and 89
>> days after missing a payment and NPL means that there is a payment
>> which is more than 89 days overdue.  My base data is the daily state
>> transition list and the end of month state of the loan book.  I
>> calculate the 3x3 probability transition matrix using a generator
>> matrix method - loosely based on [1].
>>
>> For transition  / generator matrices, has anybody any suggestions for
>> how I might look at the robustness / sensitivity of the calculations?
>> My own thought was to calculate the matrix say N times leaving out a
>> sample of 1/N of the input loans, and then look at the summary() type
>> stats for each point in the matrix - thoughts? or is this just daft
>> talk?  In a single variable I can plot the output density and look at
>> the summary() data - but I have no clue really how to do this for a
>> transition matrix.
>>
>> Then there is the question of time varying transition matrices and how
>> to understand them and even <gasp> estimate where the matrix is headed
>> over the next three months?
>>
>> Has anybody any suggestions as to where I might look for other ideas
>> on where to go with this or has anybody done this type of loan book
>> credit risk modeling?
>>
>> Many thanks in advance!
>>
>> Best regards,
>> Sean O'Riordain
>> Dublin, Ireland.
>> seanpor at acm.org
>>
>>
>> [1] Lando, D., and T. Skodeberg (2002). "Analyzing ratings transitions
>> and rating drift with continuous observations." Journal of Banking and
>> Finance 26: 423-444.
>>
>> _______________________________________________
>> R-SIG-Finance at stat.math.ethz.ch mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
>> -- Subscriber-posting only.
>> -- If you want to post, subscribe first.
>
>