0
$\begingroup$

In "Full implications of CMS convexity (Cedervall and Piterbarg, 2012)", a.k.a. "CMS: covering all bases (idem)", the authors develop a CMS model equipped with an annuity mapping function which captures normal volatilities $\sigma_i$ and correlations $\rho_{i,j}$ for a set of swap rates $S_1,\dots,S_n$ of tenors $T_1,\dots,T_n$ coming from a swaption smile model. We're interested in a derivative written on the swap rate $S_n$.

Letting $S_i(T)$ be the swap rate of tenor $T_i$ at time $T$, the authors work with a one-factor lognormal model:

$$1+\delta_iS_i(T)=(1+\delta_iS_i(0))e^{\mu_i+\nu_i\sqrt{T}X}$$

where $X\sim\mathcal{N}(0,1)$ and $\delta_i$ is the float leg accrual factor of the swap of tenor $T_i$. Then the article states that:

The parameters $\nu_i$ in [the equation above] can be linked to normal volatilities and correlations by projecting the rates [i.e. computing the expectation of one rate conditional on the other rate]:"

$$\nu_i\triangleq\frac{\color{blue}{\delta_i}}{1+\delta_iS_i(0)}\sigma_i\rho_{n,i}$$

Now, I understand if we assume a two-factor Gaussian model such that:

$$\begin{align} \text{d}S_n(t)&=\sigma_n\text{d}W_t \\[3pt] \text{d}S_i(t)&=\sigma_i\left(\rho_{n,i}\text{d}W_t+\sqrt{1-\rho_{n,i}^2}\text{d}\tilde{W}_t\right) \end{align}$$

where $W_t$ and $\tilde{W}_t$ are two independent Brownian Motions, we get:

$$E\left(S_i(T)|W_T\right)=S_i(0)+\sigma_i\rho_{n,i}W_T$$

If we now assume a single risk factor, then the normal volatility of $S_i(T)$ is $\sigma_i\rho_{n,i}\sqrt{T}$ $-$ this is basically a trick to capture correlation information from the swaption smile model $-$ and using the fact that:

$$\sigma_{\text{Normal}}\approx\sigma_{\text{Log Normal}}\times \text{Spot}$$

We set:

$$\nu_i\triangleq\frac{1}{1+\delta_iS_i(0)}\sigma_i\rho_{n,i}$$

But I don't see where the accrual factor comes from. Does anybody understand how it ends up there?

$\endgroup$
  • $\begingroup$ If the accrual factor tends to zero then the volatility of (1+accrual factor *rate) must tend to zero- this wouldn't be the case if it weren't there I think? $\endgroup$ – dm63 Jun 14 '18 at 1:50
2
$\begingroup$

Let $X_i(t) = 1+\delta_i S_i(t)$. Then $\nu_i$ is the log normal volatility for $X_i(t)$, and because $dX_i(t) = \color{blue}{\delta_i} dS_i(t)$ we get $\nu_i X_i(t) dW_t = \color{blue}{\delta_i} \sigma_i \rho_{n,i} dW_t$ and $\nu_i \approx (\color{blue}{\delta_i} \sigma_i \rho_{n,i} ) / X_i(0)$.

$\endgroup$
  • $\begingroup$ Exactly, I was messing up $1+\delta_iS_i(t)$ and $S_i(t)$. Cheers. $\endgroup$ – Daneel Olivaw Jun 14 '18 at 8:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.