0
$\begingroup$

Suppose we have two random variables $X$ and $Y$ both distributed as gamma random variables with parameters $\alpha_X,\beta_{X},\alpha_{Y},\beta_Y$, with characteristic functions given by:

$$\phi_{X}(t) = \left(1 - \frac{it}{\beta_X}\right)^{-\alpha_X} \;\;\text{ and }\;\;\phi_{Y}(t) = \left(1 - \frac{it}{\beta_Y}\right)^{-\alpha_Y}$$

Furthermore, suppose that $X$ and $Y$ are comonotone : the joint c.d.f is then given, via the upper frechet-hoffding bound, by :

$$F_{X,Y}(x,y) = \min(F_X(x),F_Y(y))$$

Note that this random vector has no density.

Can we obtain a formulation for the joint characteristic function ? Same question for more than $2$ comonotone gammas, say $X_i$ with parameters $\alpha_i,\beta_i$ for $i \in \{1,...,n\}$.

$\endgroup$

1 Answer 1

1
$\begingroup$

If you are asking for a closed-form expression, this is quite impossible, at least with the current state of the art. Indeed, for any cdf $F$ and its generalized inverse defined by $$F^{-1}(u):=\inf\{x\in\mathbb R\colon F(x)\ge u\} =\min\{x\in\mathbb R\colon F(x)\ge u\}$$ for $u\in(0,1)$, we have $$x\ge F^{-1}(u)\iff F(x)\ge u$$ for all $x\in\mathbb R$ and $u\in(0,1)$.

Letting now $F:=F_X$ and $G:=F_Y$, we see that the joint distribution of comonotone $X$ and $Y$ with cdf's $F$ and $G$ is the same as the joint distribution of $F^{-1}(U)$ and $G^{-1}(U)$, where $U$ is a random variable uniformly distributed on the interval $(0,1)$.

Therefore, the joint characteristic function (c.f.) $f$ of $X$ and $Y$ is given by $$f(s,t)=\int_0^1\exp\{i(s F^{-1}(u)+t G^{-1}(u))\}\,du$$ for real $s,t$. This integral cannot be taken in closed form when $X$ and $Y$ have arbitrary gamma distributions.

This cannot be done even in the simplest case when $\alpha_X=2$, $\beta_X=1$, $\alpha_Y=1$, $\beta_Y=1$. Indeed, then $F(x)=1-(x+1)e^{-x}$ and $G(x)=1-e^{-x}$ for $x>0$, whence for real $s,t$ $$f(s,t)=\int_0^\infty\exp\{i(s x+t G^{-1}(F(x)))\}\,dF(x) \\ =\int_0^\infty \exp\{i(s+t)x-x\}\,\frac{x\,dx}{(x+1)^{it}}.$$ Mathematica cannot do anything with this integral, returning just a trivially identical expression:

enter image description here

$\endgroup$
5
  • $\begingroup$ Thanks for all the work; Except you made a mistake at the end : the $i$ is in factor of everything, not just (s+t)x, so you obtain an exponent of $i(s+t-1)x$ if i'm right. Can mathematica solve it with this correction? $\endgroup$
    – lrnv
    May 5, 2020 at 14:54
  • $\begingroup$ @lrnv : I think everything is correct as written. $-x$ in the exponent comes from $dF(x)=e^{-x}\,dx$. $\endgroup$ May 5, 2020 at 15:00
  • $\begingroup$ Ho yeah you are right. Here comes down all my hopes ;( Thanks anyway ! $\endgroup$
    – lrnv
    May 5, 2020 at 15:08
  • $\begingroup$ @lrnv : Oops! I did confuse $F$ and $G$ at some point ($G$ must be the simpler one, as compared to $F$, to make $G^{-1}$ explicit). This is now fixed, and the integral has become only slightly more complicated, with the integrand now containing the extra factor $x$. $\endgroup$ May 5, 2020 at 15:13
  • $\begingroup$ Hahaha it got worse :) $\endgroup$
    – lrnv
    May 5, 2020 at 15:37

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.