Find unconditional distribution from conditional density functions of gamma and normal distributions

29 Views Asked by At

Let Y ∼ Gamma(a; λ) , a, λ > 0: that means random variable Y has continuous density:

$$f_Y(y) = \begin{cases} \frac{\lambda^a y^{a-1} e^{-\lambda y}}{\Gamma(a)}, & y > 0, \\ 0, & \text{elsewhere.} \end{cases}$$

Let random variable X, conditional to Y have normal distribution with expected value 0 and variance 1/y, therefore

$$X|(Y=y) ∼ N(0; 1/y). $$

Find unconditional distribution for random variable X.

I know density function for X|Y looks like $$ f_{X|Y=y}(x) = \frac{\sqrt{y}}{\sqrt{2\pi}} e^{-\frac{1}{2} y x^2} $$

I belive I can caluclate joint density function as $f_{X,Y}(x,y)$ = $f_Y(y)* f_{X|Y}(x,y)$,but it doesn't look nice. Any ideas?

1

There are 1 best solutions below

0
On BEST ANSWER

Perform the integration with respect to $y$ of the kernel of the joint density, discarding any constant multiplicative factors:

$$f_{X \mid Y}(x,y) f_Y(y) \propto y^{1/2} e^{-x^2 y/2} \cdot y^{a-1} e^{-\lambda y} = y^{a - 1/2} e^{-(x^2/2 + \lambda)y}.$$

This is the kernel of a gamma density with shape $a^* = a + 1/2$, and rate $\lambda^* = \lambda + \frac{x^2}{2}$. So

$$f_X(x) = \int_{y=0}^\infty f_{X \mid Y}(x,y) f_Y(y) \, dy \propto \frac{\Gamma(a+1/2)}{(\lambda + x^2/2)^{a+1/2}}.$$

I leave it as an exercise to show that this is proportional to a scale-transformed Student's $t$-distribution with $\nu = 2a$ degrees of freedom and scale parameter $1/\sqrt{a\lambda}$.