|
saw this mentioned on the W forum:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2026350
looks promising from the point of view of matching the "true" (monte-carlo) distribution, and avoiding negative densities
the two authors work for Numerix
I know "SABR is just a tool for interpolating" blah blah blah but this one does seem to offer more potential for improvement in practice than the various other recent tweaks (Paulot etc)
any opinions ? tbh I don't think I'll be rushing to implement the complex integral but if somebody else wrote some code I'd love to play around with it.
|
|
|
|
 |
mtsm
|
|
| Total Posts: 94 |
| Joined: Dec 2010 |
| |
|
That could be nice. I would like to know how useful you find this to be. Do you plan on implementing? How much guidance do the authors give?
I tend to agree that not matching SABR MC for long expiries, long tenors and OTM strikes is a nuisance, so any solution that comes close to the real thing is a plus.
Negative densities are always bad, but.in this light unfortunately, I don't think that the conclusion of the paper's abstract is particularly relevant.
"These formulas are important in volatility surface construction and CMS products replication because they provide correct behavior for far strikes and reduced approximation error."
The main problem is that true SABR model does not provide correct behavior for far strikes and reduced approximation error. In this sense this is just another applied math
paper.
In my experience you can go very far with the Hagan (lognormal) expansion if you are prepared to put in some work on using it properly. Things get dicey really only for far OTM puts, which can be a problem in market practice to some extent. But other than this it is really quite good.
|
|
|
 |
Hydro
|
|
| Total Posts: 6 |
| Joined: Jul 2009 |
| |
|
actually, Antonov et al. paper is very good in my opinion.
it carries on previous work done by M. Forde at Kings' College London and previously great ground work on zero correlation approach done by O. Islah.
On the "another applied math paper": when you think about it, the (single) perturbation technique used by Hagan et al. is quite challenging mathematically. But eventually it was an application massively used. Surely it has its limits in today's market.
Antonov is going through the same exercise (almost) but improves it dramatically (using zero correlation, general correlation, mapping technique).
As for the numerical implementation, it's not complicated at all. there is a double integral to deal with, but this can be easily handled with a Gauss-Legendre which code is widely available.
the paper is one of the rare to actually display results. I believe the authors are preparing updates to this paper. |
|
|
|
|
mtsm
|
|
| Total Posts: 94 |
| Joined: Dec 2010 |
| |
|
Nobody is saying this is bad work, on the contrary it could be a significant technical improvement.
I am saying it is not going to improve CMS pricing. What it is going to improve is convergence to a model that is wrong. All I am saying really. |
|
|
|
Hydro
|
|
| Total Posts: 6 |
| Joined: Jul 2009 |
| |
|
good stuff,
you're right, CMS prices might not be improved.
|
|
|
|
|
