Short-Rate Models: Calibration, Limitations, and Market Practice
Article Quant Q&A · Author: user41699
Summary
This discussion outlines how short-rate models can support pricing for interest-rate derivatives. A modeled short rate produces zero-coupon bond prices and, in some models, closed-form prices for bond options that can be used for products such as caps, floors, and swaptions. Trees may help with early-exercise and exotic features, while Monte Carlo simulation can be more practical for models with multiple factors.
It highlights calibration choices: relative errors can prevent expensive long-dated caps from dominating a fit, and a deterministic shift can fit the yield curve while model parameters address volatility. It also distinguishes equilibrium models with constant parameters from no-arbitrage models that can be recalibrated to current market data. The answer notes that positive-rate models may not suit negative-rate settings, and that one-factor models cannot represent all yield-curve and correlation behavior. This is a brief overview rather than a comprehensive comparison; it points readers toward broader model families and texts for deeper study.
Key ideas
- Short-rate models can generate bond prices and support pricing of several vanilla interest-rate derivatives.
- Closed-form pricing is available for some models, while trees or Monte Carlo methods may be needed for other products and model structures.
- Relative-error calibration can give short-maturity instruments more influence than an absolute-error objective.
- A deterministic shift can fit the current yield curve, while other model parameters can be calibrated to volatility data.
- Model choice depends on rate constraints, product features, and whether multiple yield-curve factors are needed.
Tags
Full text
# Interest Rate Models cheat sheet - Need for advice # Interest Rate Models cheat sheet - Need for advice I'm trying to get through the litterature of interest rate models for some time now. As I don't have any experience working with them, I started looking for some kind of a cheat sheet that would compare the models in terms of their limitations : - Why can't they be used to price certain instruments - What curves they reproduce well : interest rates, volatilities etc ... - Stability of parameters when calibrated ... - Which ones are really used in the industry Also I find it really difficult to know all of them. Is it more common that people would know two or three models well and be kind of "experts" in them? The other thing I want to know is a list of practices and techniques that are really popular in the industry. I mean practices like CMS-Replication, Basket Pricing methods (Levy-like) etc ... Anyone can help with links or any resources? My purpose is really to target my readings and not be taken in interviews as someone who never worked in quantitative team Thank you ## Answer by Kevin (score 2) https://quant.stackexchange.com/a/46625 The answer to your question clearly fills entire book shelves, not just books. I can try to give you a brief overview and answer regarding short rate, the simplest interest rates to be modelled. - Typically, you model the short rates $(r_t)$ from which you obtain zero-coupon bond (ZCB) prices. Then, you can price zero-coupon bond options (ZCBO) which allow you to price many vanilla claims such as caps, floors and swaptions. - Some models Hull-White, Cox-Ingersoll-Ross have a closed form solutions for ZCB and ZCBO, others don't, e.g. Black-Karasinski. If you price exotics or have early-exercise features, you may wish to build a tree. Trinomial trees are very popular for the short rate models, see the papers from Hull White. - Calibration issue 1. When you calibrate a simple HW model to cap volatilities, use relative errors as objective function, not absolute errors. With longer maturities, the caps are much more expensive such that an absolute-error-objective function focusses completely on the long-term products and gives you a poor fit for the instruments with short maturities. - Calibration issue 2. Keep in mind that you need to re-calibrate your model which is inconsistent with the notion of a ''long-term mean'' which many short rate models incorporate. Hull & White (2001) differentiate here between an academic view which rather have constant parameters with economic meaning and the view of a trader who always wants his model to match the current market data. See also the difference between equilibrium models (e.g. Vasicek, CIR, Dothan) with constant parameters and no-arbitrage models (Ho-Lee, Hull-White, Black-Karasinski) with time-dependent parameters. - Calibration issue 3. You asked which curve is to be fitted. Volatilities or yield curve. Arguably, the latter is more important. Brigo and Mercurio (Appendix F in their fantastic book which you ought to purchase if you're interested in the interest rate literature!) describe how to construct a tree which perfectly fits any yield curve. They use a deterministic shift. i.e. Let $x_t$ be a well-known short rate model (Vasicek, CIR, etc.) and $\theta_t$ be deterministic. Then, set $r_t=x_t+\theta_t$ and use $\theta_t$ to calibrate the model towards the yield curve. The parameters of $(x_t)$ can be used to fit the volatility curve. - Limitations 1. If you work in a market with negative interest rates, then you cannot use the exponential Vasicek model and the CIR model since they impley positive rates. Instead you got to work with an extra displaced model or use a model with normal distribution (e.g. Vasicek, Ho-Lee, Hull-White) which has allows for negative short rates. - Limitations 2. The short rate models mentioned thus far are one-factor models, i.e. they are driven by a single Brownian motion. That's bad if you want to price products that depend on correlation or the entire yield curve. One can generalise all these models (and the deterministic shift $\theta_t$) to multi-factor models but building trees is then often tedious. Monte Carlo simulations will be easier the higher the dimension is. As I said, this merely touches short rate models. Further important model classes are libor market models and swap market models as well as the HJM framework. ## Answer by user34971 (score 2) https://quant.stackexchange.com/a/46637 At risk of sounding pedantic, I don't think cheat sheets are the way to learn things. May I suggest reading the following book for a relatively quick but quite robust introduction to linear IR derivatives and the current market practice: ``` J H M Darbyshire Pricing and Trading Interest Rate Derivatives: A Practical Guide to Swaps. Revised Edition, Aitch & Dee Limited (May 17, 2017) ``` http://www.tradinginterestrates.com/ Not sure if your goal is quantitative analysis or trading/portfolio management. If the former, which I think it is, then probably at some point you should read / plough through the "bible" of IR derivs which is the Andersen & Piterbarg Trilogy. Good luck :)
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.