Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I am asked to explain why the following function is not a legitimate (multivariate) distribution function.

$$ F(x,y) = 1 - e^{-x-y}, x,y \geq 0 $$

I am tempted to reason as follows: The function has right continuity on each of its variables, so its not that. Monotonicity doesnt pose a problem either. The function also varies between $0$ and $1$, so basically what remains out of the basic properties is the limit one, which says:

$$ \lim_{x_{1},...,x_{n}\rightarrow+\infty}F(x_{1},...,x_{n})=1 $$

Maybe I am missing something on this multivariate limit, but I suppose it works as well for the given cumulative function. What I thought I could argue is that we know the distribution function by FTC would satisfy:

$$ \frac{\partial^2F}{\partial x\, \partial y} = f_{X,Y}(x,y) $$

However:

$$ \frac{\partial^2F}{\partial x\, \partial y} = - e^{-x-y} \leq 0 $$

Since the joint density cannot assume negative values. Is that it? Would there be another reason for it not being legitimate?

share|improve this question

1 Answer 1

up vote 1 down vote accepted

Suppose $F$ is the joint cumulative distribution function of the random variables $X$ and $Y$, so that $P(X \leq x, Y \leq y) = F(x,y)$.

Fix $x = 0$ and consider $$\lim_{y\rightarrow \infty} F(0,y) = \lim_{y\rightarrow \infty}1 - e^{-y} = 1$$

For $x_{0} < 0$

$$\lim_{y\rightarrow \infty}F(x_{0}, y) = 0$$

follows from $F(x,y)$ being equal to zero outside the first quadrant.

It follows that $P(X = 0) = 1$.

A similar argument shows that $P(Y = 0) = 1$.

But then the random vector $(X,Y)$ can only take the value $(0,0)$, which is inconsistent with our alleged cumulative distribution function. Hence $F$ can't really be a cumulative distribution function after all.

Another way to see it is that all bivariate joint cumulative distribution functions must satisfy the property

For all $x_1 < x_2 \in \mathbb{R}$ and $y_1 < y_2 \in \mathbb{R}$

$$F(x_2,y_2) - F(x_2,y_1) - F(x_1, y_2) + F(x_1,y_1) = P(x_1 < X \leq x_2, y_2 < Y \leq y_2) \geq 0$$

To see why the left hand side gives the probability in the middle first consider the region $\left\{(x,y) : x \leq x_2, y \leq y_2\right\}$. $F(x_2, y_2)$ gives the probability $(X,Y)$ lies in this region. $F(x_1, y_2) = P(X \leq x_1, y \leq y_2)$, while $F(x_2, y_1) = P(X \leq x_2, y \leq y_1)$. If we subtract by $F(x_1, y_2)$ and $F(x_2, y_1)$, we almost have $P(x_1 < X \leq x_2, y_1 < Y \leq Y_2)$, except we've subtracted out the probability for the region $\left\{(x,y) : x \leq x_1, y \leq y_1\right\}$ twice. The probability $(X,Y)$ lies in this last region is $F(x_1, y_1)$, adding $F(x_1, y_1)$ back in to make up for the double subtraction we did gives us

$$P(x_1 < X \leq x_2, y_1 < Y \leq y_2) = F(x_2, y_2) - F(x_1, y_2) - F(x_2, y_1) + F(x_1, y_1)$$

It can easily be seen that $F(x,y) = 1 - e^{-x-y}$ doesn't satisfy this property by choosing $x_1 = y_1 = 0$ and $x_2 = y_2 = \infty$.

share|improve this answer
    
You mean the random vector $(X,Y)$ would only take the value $(0,0)$, right? –  user191919 Jan 16 at 18:35
    
Yes. I've fixed it. –  Albert Steppi Jan 16 at 18:44
    
Thanks. I can't see why the left hand side is equal to the right hand side of the new equation you put in... –  user191919 Jan 16 at 19:04

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.