f*********e
发帖数: 13 | 1 看到一个quantitative modelers的招聘, 里面要求market risk VaR models, 请问实
际工作中market risk VaR都有哪些model？
在john hull 的书上看到计算VAR 的方法 historical simulation, model based
method (linear model, quadratic model), simulation methods。 请问, 实际中
工作中那些方法比较常用？
对VAR的stress test, 实际中是不是主要用simulation 的方法？importance sampling
, Brownian bridge techniques or other， 实际中那些方法比较常用？谢谢！ | k**k
发帖数: 61 | 2 Guess these days the only trustworthy way is through Monte-Carlo for VaR.
The idea is simple: simulate 100 portfolio returns and take the 5th worse
case as your 95% VaR. But the key is HOW to simulate your returns based on
what you've already known for the assets in your portfolio. Basically, this
is done through
1. getting the marginal distribution of individual returns
right via fitting to some fat-tail distribution such as Pareto.
2. getting the cross-dependence right via copula such as T-cop | f*********e
发帖数: 13 | 3 Thank you very much for your great inputs. I got it. I remembered we can
also calculate to use calculate VAR using GARCH.
this
【在 k**k 的大作中提到】
: Guess these days the only trustworthy way is through Monte-Carlo for VaR.
: The idea is simple: simulate 100 portfolio returns and take the 5th worse
: case as your 95% VaR. But the key is HOW to simulate your returns based on
: what you've already known for the assets in your portfolio. Basically, this
: is done through
: 1. getting the marginal distribution of individual returns
: right via fitting to some fat-tail distribution such as Pareto.
: 2. getting the cross-dependence right via copula such as T-cop
| i***6
发帖数: 442 | 4 maybe you could incorporate GARCH or EWMA into Monte Carlo simulation.
However, the most widely used method for computing VaR in London is the
simple variance/covariance method, according to Jorion. | k**k
发帖数: 61 | 5 Still follow my previous argument, if we assume individual asset return
follows the following relation
r(t) = c + sigma(t)*z(t)
where z(t) ~ N(0,1) and c is constant
and you will have different sets of c, sigma(t), z(t) for different assets
in or portfolio. With the above setup we can then use GARCH(1,1) to model
sigma(t) and T-copula to capture the cross dependence among differen z(t).
That way we can obtain a more realistic relation between the different r(t)
and hence our M
【在 f*********e 的大作中提到】
: Thank you very much for your great inputs. I got it. I remembered we can
: also calculate to use calculate VAR using GARCH.
:
: this
| p*******i
发帖数: 309 | |
|
