2
$\begingroup$

I'm wondering about the monotonic quantile transformation in case of a t distribution. First, for explanation, lets consider a normal distribution:

\begin{align*} f(l | \mu , \sigma ^2)=\frac{1}{\sigma \sqrt{2 \pi}} e^{-\frac{(l-\mu)^2}{2 \sigma ^2}} \end{align*}

A quantile at $\alpha$ can be calculated by

$Quantile_\alpha=\mu + \sigma \Phi^{-1}(\alpha)$

so e.g. lets say $\mu = -0.0000615$ and $\sigma=0,0145$, the 0,95 quantile of the normal distribution is 1,644854 so

$Quantile_{0.95}=-0.0000615+0,0145*1,644854=0.02379$

So now for the t distribution, I use the following notation:

\begin{align*} f(l|\nu ,\mu ,\beta) = \frac{\Gamma (\frac{\nu+1}{2})}{\Gamma (\frac{\nu}{2}) \sqrt{\pi \nu} \beta} \left(1+\frac{1}{\nu}\left(\frac{l - \mu}{\beta}\right)^2 \right)^{\text{$-\frac{1+\nu}{2}$}} \end{align*}

now, I have this from a book: quantile

ignore the "VaR", it only sais, we focus on the right side of the distribution.

And now I am not really sure about the quantile computation anymore. I think my $\beta$ is the $\sigma^2$ right?

So is this correct: $Quantile_{\alpha}^{t}=\mu + \beta t_\nu^{-1}(\alpha) $

or this: $Quantile_{\alpha}^{t}=\mu + \sqrt{\beta} t_\nu^{-1}(\alpha) $

As an example, I have the following values:

$\mu=-0.0012827$

$\beta=0.009891722$

$\nu = 3.34331$

and therefore $t_{3.34331}^{-1}(0.95)=2.258039$

now, is

$Quantile_{0.95}^{t}=-0.0012827+0.009891722*2.258039=0.02105319$

or

$Quantile_{0.95}^{t}=-0.0012827+\sqrt{0.009891722}*2.258039=0.2232954$

correct?

Mh, I am not sure, since I am confused about the $\sigma^2$ they use, but 0.2232954 CANNOT be empirical correct, so it MUST be the 0.021? Where is my mistake?

the book: Alexander J. McNeil, Rüdiger Frey, Paul Embrechts: Quantitative Risk Management 2005 page 40.

$\endgroup$
3
$\begingroup$

You have to use the scale parameter (whatever you denote it, $\sigma$, $\beta$, or $\spadesuit$).

In your density $f(\cdot; \mu,\beta,\nu)$, $\beta$ is the scale parameter and its value has to be plugged into the "monotonic" transformation.

P.S. By the terminology you use, I guess you also asked Quantile of a non standardized t distribution?

$\endgroup$
  • $\begingroup$ so please tell me directly @user1 is 0.02105319 correct? or 0.223? $\endgroup$ – Stat Tistician Mar 28 '13 at 16:22
  • 1
    $\begingroup$ The correct result is $ 0.02105321$ which uses the value of the scale parameter $\beta$. You can check this by running qnst(0.95,−0.0012827,0.009891722,3.34331). $\endgroup$ – user1 Mar 28 '13 at 16:23

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.