calculating the covariance using joint pdf

189 Views Asked by At

Given the joint pdf as $f_{X,Y}(x,y)=3x$ where $0\le y\le x\le1$ , I'm trying to calculate the covariance

My attempt:

First calculate $E\left(X\right)$ , $E\left(Y\right)$ and $E\left(XY\right)$ separately

$$f_X\left(x\right)=\int_{0}^{x}3x\>dy$$

$$E\left(X\right)=\int_{0}^{1}x\cdot f_X\left(x\right)dx$$

$$f_Y\left(y\right)=\int_{y}^{1}3xdx$$

$$E\left(Y\right)=\int_{0}^{1}y\cdot f_Y\left(y\right)dy$$

$$E\left(XY\right)=\int_{0}^{1}\int_{0}^{x}xy\left(3x\right)dydx$$

then substitute the above values in the equation $Cov\left(X,Y\right)=E\left(XY\right)-E\left(X\right)E\left(Y\right)$

My main question is regarding selecting the correct upper and lower bounds when intergrating. For instance, when finding $E(XY)$, why is it wrong if we use $$E\left(XY\right)=\int_{0}^{1}\int_{0}^{1}xy\left(3x\right)dydx$$

instead of

$$E\left(XY\right)=\int_{0}^{1}\int_{0}^{x}xy\left(3x\right)dydx$$

3

There are 3 best solutions below

0
On BEST ANSWER

Always remember the support.

Your support is $0\leqslant y\leqslant x\leqslant 1$. The joint density function only equals $3x$ when $x$ and $y$ exist inside the support, elsewhere the density is $0$. That is, it is a piecewise function.$$\begin{align}f_{X,Y}(x,y) & = \begin{cases} 3x &:& 0\leqslant y\leqslant x\leqslant 1\\0&:&\textsf{otherwise}\end{cases}\\[2ex] &= 3x\,\mathbf 1_{0\leqslant y\leqslant x\leqslant 1}\end{align}$$

When finding $f_X(x)$ you "integrate out" $y$, which lives between $0$ and $x$, leaving $x$ between $0$ and $1$.

$$\begin{align}f_X(x) &= \int_{0\leqslant y\leqslant x}3x\,\mathbf 1_{0\leqslant x\leqslant 1}\,\mathrm d y\\[2ex]&= 3x\,\mathbf 1_{0\leqslant x\leqslant 1}\,\int_0^x\,\mathrm d y\\[2ex]&= 3x^2\,\mathbf 1_{0\leqslant x\leqslant 1}\end{align}$$

When finding $f_Y(y)$ you "integrate out" $x$, which lives between $y$ and $1$, leaving $y$ between $0$ and $1$.

$$\begin{align}f_Y(y) &= \int_{y\leqslant x\leqslant 1}3x\,\mathbf 1_{0\leqslant y\leqslant 1}\,\mathrm d x\\[2ex]&= 3\,\mathbf 1_{0\leqslant y\leqslant 1}\,\int_y^1 x\,\mathrm d x\\[2ex]&= \tfrac 32(1-y^2)\,\mathbf 1_{0\leqslant y\leqslant 1}\end{align}$$

When performing a double integral, you "interate out" the inner variable, then integrate out the outer. $$\begin{align}\mathsf E(g(X,Y)) &=\iint_{0\leqslant y\leqslant x\leqslant 1} 3x\,g(x,y)\,\mathrm d (x,y)\\&=\int_{0\leqslant x\leqslant 1}\int_{0\leqslant y\leqslant x} 3x\,g(x,y)\,\mathrm d y\,\mathrm dx&&=3\int_0^1 x\int_0^x g(x,y)\,\mathrm d y\,\mathrm dx\\&=\int_{0\leqslant y\leqslant 1}\int_{y\leqslant x\leqslant 1} 3x\,g(x,y)\,\mathrm d x\,\mathrm d y&&=3\int_0^1\int_y^1 x\,g(x,y)\,\mathrm d x\,\mathrm d y \end{align}$$

So $$\begin{align}\mathsf E(X) &= \int_0^1 x \int_0^x 3x\,\mathrm d y\,\mathrm d x&&=\int_0^1 x\, f_X(x)\,\mathrm d x\\[2ex]\mathsf E(Y) &= \int_0^1 y\int_y^1 3x\,\mathrm d x\,\mathrm d y &&= \int_0^1 y\,f_Y(y)\,\mathrm d y\\[3ex]\mathsf E(XY) &= \int_0^1\int_0^x xy\cdot 3x\,\mathrm d y\,\mathrm d x \\[1ex]& = \int_0^1\int_y^1 xy\cdot 3x\,\mathrm d x\,\mathrm d y \end{align}$$

And so forth.

0
On

The inner integral is about $y$, and the probability is nonzero when $0\leq y\leq x$ not $0\leq y\leq 1$. Thus the first integral is wrong.

If you want to integrate $x$ first and then $y$, the integral would become $$E(XY)=\int_0^1{\int_y^1{xy\cdot 3x\, dx}\, dy}.$$

0
On

$f_{X,Y}(x,y)=0$ for $y\gt x$, so $f_{X.Y}(x,y)\ne 3x$ there. Integral limits reflect this.