What is the difference between historical and Gaussian method of VaR estimation?
I know how they are calculated, but what are the pros and cons of each?
3
$\begingroup$
$\endgroup$
-
$\begingroup$ Have you tried searching for this? There is a lot of info out there. $\endgroup$ – AfterWorkGuinness Oct 27 '15 at 19:11
-
$\begingroup$ I have, but didn't find anything really relevant. $\endgroup$ – luka5z Oct 27 '15 at 19:13
3
$\begingroup$
$\endgroup$
Historical Simulation
Pros:
- Easy to calculate
- Doesn't make assumptions about distribution of returns (uses empirical distribution)
- Can add some enhancements onto it such as giving a higher weighting to more recent returns (prevents ghosting mentioned below) or a weighting by volatility where more volatile returns get a higher weight.
Cons:
- Assumes the past will repeat itself, doesn't consider events that it has not seen before
- If you use the most basic historical simulation approach, as your historical window shifts, large losses or returns at the edge of the window will no longer be in your data-set and can cause a significant jump in the Var (this is called ghosting) which in very undesirable
Guassian/Parametric/Delta Normal/Variance-Covariance (has many names)
Pros:
- Relatively easy to calculate (more work than historical, but less compared to monte carlo)
Cons:
- Assumes returns are normally distributed, which is often incorrect
- Assumes delta sensitivity accounts for all the risk
- Very inaccurate for non-linear positions like options (because of above point re delta)
- Need to compute an NxN covariance matrix for the portfolio.
answered Oct 27 '15 at 19:24
-
-
1$\begingroup$ No idea sorry. If I had to make an educated guess, historical as it is the least data and process intensive. $\endgroup$ – AfterWorkGuinness Oct 27 '15 at 19:32
-
$\begingroup$ Thanks anyway! Maybe some praticioner will input something $\endgroup$ – luka5z Oct 27 '15 at 20:05
-
$\begingroup$ But guys: in usual Monte Carlo approaches: what do you do? Estimate a covariance from historical data and sample Gaussians. There are more sophisticated approaches but in the basic case: what is the difference? $\endgroup$ – Ric Oct 28 '15 at 9:13
-
$\begingroup$ Monte Carlo simulation is quite different. Here are a couple of posts on the topic: quant.stackexchange.com/questions/17910/… and quant.stackexchange.com/questions/12592/… $\endgroup$ – AfterWorkGuinness Oct 28 '15 at 16:37
