h*******t
发帖数: 214 | 1 能否具体谈谈是做什么的，然后什么skillset比较重要。
正在做market risk, 考虑是否要转行去做counterparty credit risk. |
A***o
发帖数: 351 | 2 You may read Steven Zhu's paper about CCR. it covers the essence of CCR
modeling. |
d*****r
发帖数: 2583 | 3 no difference, why bother?
【在 h*******t 的大作中提到】
: 能否具体谈谈是做什么的，然后什么skillset比较重要。
: 正在做market risk, 考虑是否要转行去做counterparty credit risk.
|
L*******t
发帖数: 2385 | 4 计算CVA需要做expecation和time integral
这个对trade很多的银行来说岂不是一个nightmare。。。
【在 d*****r 的大作中提到】
: no difference, why bother?
|
s******e
发帖数: 1751 | 5 why?
【在 L*******t 的大作中提到】
: 计算CVA需要做expecation和time integral
: 这个对trade很多的银行来说岂不是一个nightmare。。。
|
L*******t
发帖数: 2385 | 6 计算太复杂啦
【在 s******e 的大作中提到】
: why?
|
t********t
发帖数: 1264 | 7 不能hedge这种计算有任何意义么？除了满足监管
【在 L*******t 的大作中提到】
: 计算CVA需要做expecation和time integral
: 这个对trade很多的银行来说岂不是一个nightmare。。。
|
L*******t
发帖数: 2385 | 8 有个model，有个好方法能计算price和Greeks。
这个应该就是hedge的第一步吧。
一旦市场放开了，就可以实际运作啦。
而且。。。满足监管需要也是很重要的吧。。
【在 t********t 的大作中提到】
: 不能hedge这种计算有任何意义么？除了满足监管
|
s******e
发帖数: 1751 | 9 谁说的?没听说过。
【在 L*******t 的大作中提到】
: 计算太复杂啦
|
L*******t
发帖数: 2385 | 10 没人说，这是看steve zhu的文章有感而发。
【在 s******e 的大作中提到】
: 谁说的?没听说过。
|
|
|
s*******h
发帖数: 1361 | 11 Steve Zhu's paper is actually the easy part. The complexity can explode once
you try to implement it in real world. You are about to solve one of the
hardest yet mandatory cross-asset-class problems that each bank faces.
Smaller firms may not be able to set up the problem even if they want to.
- How can you generate scenarios across different asset classes and
instruments in each netting set and across all netting sets? Are simple
difussion processes enough to model individual assets, and are simple
copulas enough to capture the essense of tail risks? Will it be different
for the gold price and dollar spot in a rising rate or falling rate
environment?
- How can you reduce the dimemsion of the whole universe while maintaining
the ability to accurately price each derivative for each node on each
simulation path. Industrial standard may be a few factors for each asset
class but even a simple correlation structure can be challenging to estimate
as the number of factors can still outnumber assets in each netting sets,
and idios can be significant for some netting sets.
- How can you calculate all greeks fairly accurately at a reasonable speed (
not even dreaming about real time). Methods such as automatic
differentiation require a large-scale implementation to all pricing models
in the firm, and singlarities can still break the system if not properly
handled (e.g. digital options).
- Even with the most aggressive modeling simplification you may still be
left with a very expensive problem to solve numerically. How can you
leverage
technology such as distributed computation on graphic cards (e.g. Cuda) to
speed things up using farms of hardwares? If your underlying pricing modules
are not fast enough, will you be willing to rewrite everything in a faster-
execution language (e.g. C++).
- How can you align physical measures and risk-neutral measures for
different asset classes if market is not complete and you don't have a
unique risk measure?
- How can you validate your implementation for human errors before it's too
late? You may be able to validate a swap CVA by a swaption pricer, but what
about other instruments?
- What about path-dependent and credit-dependent instruments for which you
lose your Markovian assumptions? What about instruments without any semi-
analytic solutions for which you will need to run simulations of simulations
? What if collateral discounting is significantly different than market
implied discounting and you are forced to double your curve generation?
The list goes on and on, and at the end of day, traders may not have clear
directions and just charge a rich risk premium and warehouse some of the
risks. The regulators understand the problem even more poorly and sometimes
issue guidelines that trade off efficiency and performance for political
benefits.
For OP's role, they may pay OK if it's regulation driven hence demand is
high but most people will not be able to move forward much so the upside is
small (relatively). It should be good for work-life balance though, but so
are many other middle-office roles.
【在 L*******t 的大作中提到】
: 没人说，这是看steve zhu的文章有感而发。
|
L*******t
发帖数: 2385 | 12 赞！学习了！
once
【在 s*******h 的大作中提到】
: Steve Zhu's paper is actually the easy part. The complexity can explode once
: you try to implement it in real world. You are about to solve one of the
: hardest yet mandatory cross-asset-class problems that each bank faces.
: Smaller firms may not be able to set up the problem even if they want to.
: - How can you generate scenarios across different asset classes and
: instruments in each netting set and across all netting sets? Are simple
: difussion processes enough to model individual assets, and are simple
: copulas enough to capture the essense of tail risks? Will it be different
: for the gold price and dollar spot in a rising rate or falling rate
: environment?
|
M*******e
发帖数: 546 | 13 听起来你没做过。各大银行都有国人在做CVA。是个无聊的事倒是真的
once
【在 s*******h 的大作中提到】
: Steve Zhu's paper is actually the easy part. The complexity can explode once
: you try to implement it in real world. You are about to solve one of the
: hardest yet mandatory cross-asset-class problems that each bank faces.
: Smaller firms may not be able to set up the problem even if they want to.
: - How can you generate scenarios across different asset classes and
: instruments in each netting set and across all netting sets? Are simple
: difussion processes enough to model individual assets, and are simple
: copulas enough to capture the essense of tail risks? Will it be different
: for the gold price and dollar spot in a rising rate or falling rate
: environment?
|
s*******h
发帖数: 1361 | 14 此版人气不行，本想借贴帮忙攒攒。我做与不做又有何干。你哪怕回帖有一丝干货也不
枉我笔墨。
此地虽常年有相关从业人员关注，可惜被各路杂碎把持，人气低迷，几句以内必然转移
到攻击，猜疑，自卑，你妈同性别新港小留身上。
叹,偌大北美没有一个像样的华人论坛可以讨论相关话题。也好，本就僧多粥少,老人各
赚各钱，新人自求多福.无限传承，发扬光大。
【在 M*******e 的大作中提到】
: 听起来你没做过。各大银行都有国人在做CVA。是个无聊的事倒是真的
:
: once
|
L*******t
发帖数: 2385 | 15 又看了一遍你的帖子，又有了新的感受。
once
【在 s*******h 的大作中提到】
: Steve Zhu's paper is actually the easy part. The complexity can explode once
: you try to implement it in real world. You are about to solve one of the
: hardest yet mandatory cross-asset-class problems that each bank faces.
: Smaller firms may not be able to set up the problem even if they want to.
: - How can you generate scenarios across different asset classes and
: instruments in each netting set and across all netting sets? Are simple
: difussion processes enough to model individual assets, and are simple
: copulas enough to capture the essense of tail risks? Will it be different
: for the gold price and dollar spot in a rising rate or falling rate
: environment?
|
L*******t
发帖数: 2385 | 16 大家可能现实中都是明白人，可惜到了网上，就是另一个样子。
【在 s*******h 的大作中提到】
: 此版人气不行，本想借贴帮忙攒攒。我做与不做又有何干。你哪怕回帖有一丝干货也不
: 枉我笔墨。
: 此地虽常年有相关从业人员关注，可惜被各路杂碎把持，人气低迷，几句以内必然转移
: 到攻击，猜疑，自卑，你妈同性别新港小留身上。
: 叹,偌大北美没有一个像样的华人论坛可以讨论相关话题。也好，本就僧多粥少,老人各
: 赚各钱，新人自求多福.无限传承，发扬光大。
|
