In the SVI-JW parametrization, we have

$$ w(k; a, b, \rho, m, \sigma) = a + b \left [ \rho(k-m) + \sqrt{(k-m)^{2} + \sigma^{2}} \right ] $$

Which gives us

$$ \begin{align*} \sigma_{BS}(k) &= \frac{1}{\sqrt{t}}\sqrt{a + b \left [ \rho(k-m) + \sqrt{(k-m)^{2} + \sigma^{2}} \right ]} \\ \\ \\ \frac{\partial \sigma_{BS}}{\partial k} &= \frac{b\left [\rho + \frac{(k - m)}{\sqrt{(k-m)^{2} + v^2}}\right ]}{2\sqrt{t}\sqrt{a + b \left [ \rho(k-m) + \sqrt{(k-m)^{2} + \sigma^{2}} \right ]}} \end{align*} $$

We can evaluate ATM variance $v_{t}$ by setting $k=0$ in $w(k; a, b, \rho, m, \sigma)$, we can evaluate ATM skew $\psi_{t}$ by evaluating $\frac{\partial \sigma_{BS}}{\partial k}|_{k=0}$ and we can evaluate minimum implied variance $\tilde{v}_{t}$ by setting $\frac{\partial \sigma_{BS}}{\partial k} = 0$ and plugging $k$ into $w(k)$.

How can we find the put and call slopes $p_{t}$ and $c_{t}$? I assumed it would be the limit of the variance $\frac{w(k)}{t}$ when $k \rightarrow \pm \infty$ but this gives me $\frac{b}{\sqrt{t}}(\rho \pm 1)$ which does not match Gatheral's results in his original paper.

Link to original paper Arbitrage-free SVI volatility surfaces by Jim Gatheral, Antoine Jacquier here.


$(p_t,c_t)$ are respectively related to the put/call slopes of the total implied variance, not variance $$ w(k,t)=\sigma^2(k,t) t $$

Under SVI $$ w(k) = a + b \left(\rho(k-m) + \sqrt{(k-m)^2 + \sigma^2} \right) $$ such that $$ \frac{\partial w}{\partial k}(k) = b \left( \rho + \frac{k-m}{\sqrt{(k-m)^2+\sigma^2}} \right) $$ and $$ \lim_{k \to \pm \infty} \frac{\partial w}{\partial k}(k) = b \left( \rho \pm 1 \right) $$ (see also here end of p.5)

Now, remembering should you define: $$ p_t := \frac{1}{\sqrt{w_t}} b (1-\rho) $$ $$ c_t := \frac{1}{\sqrt{w_t}} b (1+\rho) $$ with $w_t$ the ATMF total implied variance ($w_t = v_t t$ in the JW space) then you have that indeed:

  • $p_t$ is proportional to the opposite of the put slope of total implied variance and is expected to be positive (because $w_t$ and $b$ are positive)
  • $c_t$ is proportional to the call slope of total implied variance and is expected to be positive (because $w_t$ and $b$ are positive)
  • $\begingroup$ However the original paper has definitions for ($c_{t}$, $p_{t}$) of $\frac{b}{\sqrt{w_{t}}} (1 \pm \rho)$ where $w_{t} = w(0, t)$. Where is the $\frac{1}{\sqrt{w_{t}}}$ term coming from? And why is the sign different for $p_{t}$? $\endgroup$ – Ruse Sep 28 '20 at 12:09
  • $\begingroup$ Hi, I've edited my answer to make it clearer. The term $1/\sqrt{w_t}$ is basically a scaling term. And indeed, $p_t \propto -\text{put slope}$ and $c_t \propto +\text{call slope}$. $\endgroup$ – Quantuple Sep 28 '20 at 12:29

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.