l*********s
发帖数: 5409 | 1 Say a bunch of loans with varying features (borrower,servicer, location,
credit score etc.), with time passing borrowers will lose money on a portion
due to forclosure.
High-risk loans often have property insurance to mitigate the loss situation
. The insurers pay the less amount of actual loss or a predefined sum at
orgination.
However, due to recent economic downturn, many insurers are increasing the
rate of policy cancellation. So the insurance coverage is random as well.
What is the appropri |
l***a
发帖数: 12410 | 2 doesn't a linear regression work?
portion
situation
lot
【在 l*********s 的大作中提到】
: Say a bunch of loans with varying features (borrower,servicer, location,
: credit score etc.), with time passing borrowers will lose money on a portion
: due to forclosure.
: High-risk loans often have property insurance to mitigate the loss situation
: . The insurers pay the less amount of actual loss or a predefined sum at
: orgination.
: However, due to recent economic downturn, many insurers are increasing the
: rate of policy cancellation. So the insurance coverage is random as well.
: What is the appropri
|
l*********s
发帖数: 5409 | 3 the event of compensation by insurer is random ; the amount is also
dependent on the actual loss. I don't think a plain linear regression is
appropriate.
【在 l***a 的大作中提到】
: doesn't a linear regression work?
:
: portion
: situation
: lot
|
b******1
发帖数: 367 | 4 the question is too big. there are simple ways as well as complicated ways
which depend on your data, available tools and human resource.
【在 l*********s 的大作中提到】
: Say a bunch of loans with varying features (borrower,servicer, location,
: credit score etc.), with time passing borrowers will lose money on a portion
: due to forclosure.
: High-risk loans often have property insurance to mitigate the loss situation
: . The insurers pay the less amount of actual loss or a predefined sum at
: orgination.
: However, due to recent economic downturn, many insurers are increasing the
: rate of policy cancellation. So the insurance coverage is random as well.
: What is the appropri
|
z**k
发帖数: 378 | 5 你做MBS么？
【在 l*********s 的大作中提到】
: the event of compensation by insurer is random ; the amount is also
: dependent on the actual loss. I don't think a plain linear regression is
: appropriate.
|
l*********s
发帖数: 5409 | 6 No,just an intern.
To further clarify, the question is that the data of actual insurance
payment is unknown, so what is a sensible way to estimate the underlying
true influence of various factors on mortgage loss with linear regression?
I think it certainly is not as difficult as generalized/nonlinear regression
, no?
【在 z**k 的大作中提到】
: 你做MBS么？
|
c******n
发帖数: 4965 | 7 linear regression IS used for random outcomes ....
use a 1/0 variable to mark compensation , doesn't this work?
【在 l*********s 的大作中提到】
: the event of compensation by insurer is random ; the amount is also
: dependent on the actual loss. I don't think a plain linear regression is
: appropriate.
|
l*********s
发帖数: 5409 | 8 Thanks for reminder, maybe I can circumvent the issue by discounting loss at
pool level.
【在 c******n 的大作中提到】
: linear regression IS used for random outcomes ....
: use a 1/0 variable to mark compensation , doesn't this work?
|
|
b*******r
发帖数: 152 | 9 so the question is to model the mortgage loss or the loss claim? |
l*********s
发帖数: 5409 | 10 underlying loss, which is experienced loss + insurance coverage; however,the
actual coverage/claim is unkown.
Is there a SAS procedure to do regression with left censoring scheme and variable critical values (maximum payout defined in policies)?
【在 b*******r 的大作中提到】
: so the question is to model the mortgage loss or the loss claim?
|
