Validating Interest Rate Lattices Through Calibration, Pricing, and Hedging
Summary
The document lays out a practical validation checklist for a one-factor mean-reverting interest rate lattice. Calibration checks include whether the model can fit coterminal instruments and caplets, represent skew where needed, handle zero-coupon swaptions, and maintain stable calibration across changing market conditions. It also asks whether mean reversion is calibrated or fixed and how that choice is made.
Pricing and sensitivity checks focus on out-of-the-money strikes, numerical convergence as integration and time-grid settings change, implied rate distributions and tails, and the forward volatility and skew the model produces. For Greeks, the document recommends checking stability through volatile periods and sensitivity to numerical settings, especially around discontinuous payoffs. A historical backtesting framework is presented as the way to assess hedging performance, but the author notes that building one takes substantial time and effort; the checklist is expressly not exhaustive.
Key ideas
- Check whether calibration can fit the instruments and volatility features required by the intended products.
- Test calibration stability across historical market conditions.
- Vary numerical integration and time-grid settings to assess pricing convergence and Greek stability.
- Inspect implied rate distributions, forward volatility, and skew for plausible behavior.
- Use historical backtesting to assess hedging performance, while recognizing the implementation effort.
Tags
Full text
# What's the best way to test/validate an interest rate lattice model # What's the best way to test/validate an interest rate lattice model I have some implementations of interest rate lattice models. I would like to verify their performance. What would be the best approaches? Currently I compare pricings of some interest rate dependent derivatives, but is there anything more direct? I'm more interested in accuracy and possibly hedging performance. ## Answer by mepuzza (score 6, accepted) https://quant.stackexchange.com/a/3501 OK, so you need to validate a one factor model with mean reversion, here are the questions I would ask myself For calibration Is the model capable of calibrating to both coterminals and caplets? If not how do I intend to calibrate it when pricing Callable Cap Floaters? Is the model capable of calibrating to skew? If not how do I intend to calibrate it when pricing Callable Range Accruals and Collars? Am I going to price zero coupon swaptions? What kind of calibration am I going to use? Am I going to calibrate mean reversion? If so how? If not what number do I use? Is the calibration stable? I would try running through the end of 2008 and see how many failures I get. For pricing Is the integration routine accurate? I would try pricing structures with OTM strikes and see when it breaks down. In particular, can I achieve convergence with a reasonable CPU time? Changing integration parameters (grid spacing, time steps) should give me an idea of the issues. (Connected to previous) What is the implied distribution of Libor rates? Does it have fat tails? If so, does the integration routine cover a sufficient range (number of standard deviations). Is the calibration appropriate? I would try calibrating an OTM bermudan to both ATM coterminals and at the strike coterminals. Do I get different prices? Why? What is the appropriate strike to be used? What kind of forward vols/skews does the model imply? Does it make sense? For greeks Are the greeks stable? Again, I would try backtesting the model through turbulent periods. And again, I would try changing the integration parameters and see if I get oscillations. How does the integration routine cope with disontinuities in the payoff? Is anything sophisticated used for avoiding spurious oscillations? Here is a paper with a standard technique used for eliminating noise http://www.risk.net/data/risk/pdf/technical/risk_1106_Wackertapp.pdf is anything like this employed in the integration routine? How do I intend to calculate greeks? Am I planning to use model greeks or smile greeks (i.e. adjusted for movements in the vol surface) ? Which ones are better? Not being a model validator myself I must have missed some important points. For what concerns the hedging performance, I've spent some time in the past implementing a backesting framework for one factor interest rate models and I think that's the only way to check hedging performance. Plus it allows you to find answers to a lot of the questions above. Beware however, it takes a lot of time and effort to implement one.
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.