Consider the k-variate random vector $Y\equiv (Y_1, Y_2, ..., Y_K)$ with cumulative distribution function (cdf) $F$.

How can I express in terms of $F$ the following probability: $$ \mathbb{P}(a_k\leq Y_k\leq b_k \text{ }\forall k) $$ ?

For example, if $K=2$, I know that

$$ \mathbb{P}(a_1\leq Y_1\leq b_1\text{, }a_2\leq Y_2\leq b_2)= F(a_2, b_2)+F(a_1, b_1)- F(a_1, b_2)-F(a_2, b_1) $$

(Sometimes this is also called rectangle formula)

Is there a way to generalise this to any $K>2$?

Thanks to the comment below, I now write the expression for $K=3$

1) I list all the $2^K=8$ vertices

$$ \begin{cases} a_1,a_2, a_3\\ b_1, a_2, a_3\\ a_1, b_2, a_3 \\ b_1, b_2, a_3\\ a_1,a_2, b_3\\ b_1, a_2, b_3\\ a_1, b_2, b_3 \\ b_1, b_2, b_3\\ \end{cases} $$

2) Algebraic sum of cdf at each vertex with +1 if the number of $a$'s is even and -1 otherwise

$$ \mathbb{P}(a_1\leq Y_1\leq b_1\text{, }a_2\leq Y_2\leq b_2\text{, }a_3\leq Y_3\leq b_3)= -F(a_1,a_2, a_3) +F(b_1, a_2, a_3) +F(a_1, b_2, a_3) -F(b_1, b_2, a_3) +F(a_1,a_2, b_3) -F(b_1, a_2, b_3) -F(a_1, b_2, b_3) +F(b_1, b_2, b_3) $$

  • 1
    $\begingroup$ Billingsley's Probability and Measure, Section 12 has a thorough discussion of this. $\endgroup$ – Kavi Rama Murthy Jul 30 '18 at 8:50
  • $\begingroup$ Thank you. It is a bit overcomplicated for me. Any easier source? $\endgroup$ – TEX Jul 30 '18 at 8:56
  • 1
    $\begingroup$ Sorry to say that the expression for the probability in $\mathbb R^{k}$ is a bit complicated and I cannot think of a better source. $\endgroup$ – Kavi Rama Murthy Jul 30 '18 at 8:59
  • $\begingroup$ @KaviRamaMurthy Thank you, following the book I have written above the expression for $K=3$. Is it correct? $\endgroup$ – TEX Jul 30 '18 at 9:20
  • 1
    $\begingroup$ I think you have got it right. $\endgroup$ – Kavi Rama Murthy Jul 30 '18 at 9:23

In general $1_{A\cap B}=1_A\times 1_B$ and $1_{(a_i,b_i]}=1_{(-\infty,b_i]}-1_{(-\infty,a_i]}$.

This leads to:$$1_{(a_1,b_1]\times\cdots\times(a_n,b_n]}=\left(1_{(-\infty,b_1]}-1_{(-\infty,a_1]}\right)\times\cdots\times\left(1_{(-\infty,b_n]}-1_{(-\infty,a_n]}\right)$$

Working out the RHS we get $2^{n-1}$ terms with sign $+$ and $2^{n-1}$ with sign $-$.

The sign is $+$ if and only if the number of $a_i$'s that appear in the term is even.

Taking the integral on both sides gives the equality:$$P(a_1<Y_1\leq b_1,\dots, a_n<Y_n\leq b_n)=\sum_{(c_1,\dots,c_n)\in\prod_{i=1}^n\{a_i,b_i\}}(-1)^{e(c_1,\dots,c_n)}F(c_1,\dots,c_n)$$where $e(c_1,\dots,c_n)$ denotes the cardinality of the set $\{i\in\{1,\dots,n\}\mid c_i=a_i\}$


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.