How OIS Discounting Changes Relative-Value Curve Analysis
Summary
The document considers whether a Dynamic Nelson–Siegel relative-value model can use pre-crisis curves discounted on a Libor basis alongside later curves discounted using overnight indexed swaps. One response describes continued use of Libor discounting for simplicity, while noting that post-crisis curve relationships changed: the five-year point, for example, behaved differently relative to the two- and ten-year points. This can complicate the interpretation of model outputs even if the model remains usable.
A second response argues that Libor-discounted and OIS-discounted relative-value analysis are not directly interchangeable, particularly when the basis has a term structure. It suggests that a cash strategy’s funding and bond package matter, and that analysis should reflect the relevant investment behavior. The discussion offers practitioner views rather than systematic evidence or a tested modeling procedure; it does not settle how to adjust a DNS calibration across regimes.
Key ideas
- Pre-crisis Libor-discounted and later OIS-discounted curves may not be directly comparable for relative-value analysis.
- Post-crisis changes in curve relationships can make model outputs harder to interpret.
- A term structure in the Libor-OIS basis can affect relative-value conclusions.
- Discounting assumptions should reflect the funding and investment behavior relevant to the strategy.
Tags
Full text
# OIS discounting pre and post crises # OIS discounting pre and post crises I have a Dynamic Nelson Siegel (DNS) based rv model. I want to know if I can use pre and post-crises curves interchangeably in my calibration and out of sample testing. I.e. those without OIS discounting pre-crises and those with OIS discounting that we observe today? I had hoped that even though they are derived differently the market rates themselves will not have changed in terms of their relative value (if 2-10 looks flat then it still looks flat post crises) same for butterfly trades (and their dynamics should also be unchanged). The differences are in the PnL of the host institutions who can no longer fund at Libor but for my purposes i.e. relative value (I am hoping) it is not necessary to introduce this complication am I right. I can just calculate the PnL in terms of the basis point changes observed (I effectively assume that I am an institution that can fund at Libor throughout)? Or is it like comparing apples and oranges? Baz ## Answer by Helin (score 0, accepted) https://quant.stackexchange.com/a/12762 For RV purposes, I have actually continued to use libor discounting for simplicity; otherwise, you'd have to model multiple curves, which become very difficult to work with... That being said, the curve has been trading very differently after the crises. For example, 5y typically didn't deviate that much from 2y and 10y on relative value basis historically, but after the crisis, 5s traded RIDICULOUSLY rich relative to 2s and 10s ALWAYS (until recently). Doesn't mean your model doesn't work any more, but it does make interpreting model output a bit more challenging. ## Answer by rrg (score 0) https://quant.stackexchange.com/a/39993 The RV analysis of discounting at Libor will differ to that at OIS. It is like comparing apples to oranges. The main driver for this would be a term structure to the basis, but Libor RV is surely only a strategy when holding the Libor-bond package? An ideal cash strategy would use both, or calibrate according to the investment behaviours of peers.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.