You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Split SmithWilson into a base curve and a correction #64
I'm looking at the downstream use of SmithWilson, more specifically the calibration of alpha which is also governed by Solvency II. It strikes me that a SmithWilson discount factor could be reinterpreted more generally as the product of a base yield curve (presently == Constant(Rate(ufr, Continuous()) and a correction term independent of ufr.
The Solvency II calibration of alpha is equivalent to searching for an alpha where the continuous 'forward rate' of the correction (at a given maturity) is some small value, so it would be pretty neat to have the correction separated.
I'll probably go make a pr for this in the near future, but maintainers are now warned 😃, and early thoughts are welcome.
That does sound interesting and potentially useful? A link I shared on the original PR decomposed the curve into components, is this getting at the same thing?
As I read it, the linked article does something slightly different, pulling out an additive term exp(-ufr * t). But you can go further and pull it out as a factor in the discount function, and the remaining factor will be independent of ufr.
Thanks for pointing out the Linked In post. I think she misinterpreted JuliaActuary as a personal project of mine. I worked as a freelance consultant a few years back, and something like this could have been a product on my shelf, so she's excused 🙂.
There would be some change to the API (or maybe just an extension, it might be possible to do it in a non-breaking way). Adding as a milestone item would be nice, but I don't think it's important enough to hold back v1.
I may have to think about it a bit more, and that might finishing up a pr, so maybe the problem will solve itself...
Review update against FinanceModels v7.0.0 (6cb71714), 2026-10-11.
Partly implemented internally; clarify the independence claim. Smith-Wilson already evaluates
D(t) = exp(-ufr*t) * (1 + s(t)),
with the correction in __smith_wilson_s; see implementation.
The kernel can be written without an explicit UFR when its fitted coefficients are held fixed, but those coefficients depend on UFR during calibration. Re-fitting identical quotes with UFR 3% versus 4% changes the correction. For a price of 0.94 at t=2, the correction factors are approximately 0.99812635 and 1.01828984 respectively.
An exposed decomposition or generalized base curve remains a low-priority extension. Specify whether coefficients are held fixed or re-calibrated; do not assume the calibrated correction can remain unchanged when replacing the base curve.
Activity
alecloudenback commentedon Nov 20, 2021
That does sound interesting and potentially useful? A link I shared on the original PR decomposed the curve into components, is this getting at the same thing?
Also, did you see this question on LinkedIn?
kasperrisager commentedon Nov 21, 2021
As I read it, the linked article does something slightly different, pulling out an additive term
exp(-ufr * t). But you can go further and pull it out as a factor in the discount function, and the remaining factor will be independent ofufr.Thanks for pointing out the Linked In post. I think she misinterpreted JuliaActuary as a personal project of mine. I worked as a freelance consultant a few years back, and something like this could have been a product on my shelf, so she's excused 🙂.
alecloudenback commentedon Feb 6, 2022
Do you envision that this change woudl affect the API or would be an internal change? (ie should I add to v1 milestone?)
kasperrisager commentedon Feb 7, 2022
There would be some change to the API (or maybe just an extension, it might be possible to do it in a non-breaking way). Adding as a milestone item would be nice, but I don't think it's important enough to hold back v1.
I may have to think about it a bit more, and that might finishing up a pr, so maybe the problem will solve itself...
alecloudenback commentedon Oct 11, 2026
Review update against FinanceModels v7.0.0 (
6cb71714), 2026-10-11.Partly implemented internally; clarify the independence claim. Smith-Wilson already evaluates
D(t) = exp(-ufr*t) * (1 + s(t)),with the correction in
__smith_wilson_s; see implementation.The kernel can be written without an explicit UFR when its fitted coefficients are held fixed, but those coefficients depend on UFR during calibration. Re-fitting identical quotes with UFR 3% versus 4% changes the correction. For a price of 0.94 at t=2, the correction factors are approximately 0.99812635 and 1.01828984 respectively.
An exposed decomposition or generalized base curve remains a low-priority extension. Specify whether coefficients are held fixed or re-calibrated; do not assume the calibrated correction can remain unchanged when replacing the base curve.