I have a problem and looked at this forum to find this thread: Probability with covariance matrix

However, I am looking for the solution involving covariance matrix with three variables.

Let $ X_1, X_2, X_3 $ be normal with the following covariance matrix:

$ \Sigma = \begin{pmatrix} 6 & -3 & -5\\ -3 & 7 & -2 \\ -5 & -2 & 3 \end{pmatrix} $

Find the probability that: $ P(X_1 - X_2 ≥1 \cap X_1-X_3 ≥ 1) $

I get the probability for either one of the legs the following way: $ P(X_1 - X_2 ≥1) $: $ Var(Y)=Var(X_1 )+Var(X_2 )−2Cov(X_1 ,X_2 )= 6+7-2*(-3)=19$

$ P(Y≥1)=1−P(Y<1)=1−Φ(\frac{1}{\sqrt{19}}) $

Similarly for $ P(X_1 - X3 ≥1)$. But these two events are clearly correlated and hence I cannot multiply them to get the final probability I am looking for. I thought about it and did not figure out how to incorporate the correlation between $X_1 ,X_2$ into the calculation. Does anyone know this?


1 Answer 1


If we have the joint distribution, $f_{\underline X}(\underline x)$ of the random veczor variable, $X_,X_2,X_3$ and if we have to calculate the probability that $\underline X=(X_1,X_2,X_3)\in A\subset R^3$, where $A$ is a nice (Borel) set then the solution is the following integral:

$$\iiint_Af_{\underline X}(\underline x)\ d\underline x.$$

In our case

$$f_{\underline X}(\underline x)=\frac1{\sqrt{(2\pi)^3\det \Sigma}}e^{-\frac12(\underline x-\underline m)^T\Sigma^{-1}(\underline x-\underline m)}$$ where $\underline m$ is the expectation vector, $\Sigma$ is the covariance matrix and it is only impliciteky given that the joint distribution is normal.


$$A=\{\underline x=(x_1,x_2,x_3):x_1-x_2\geq 1, x_1-x_3\geq 1\}.$$

Unfortunately $\underline m$ is not given. Even if the same was given the integral can only be evaluated by some numeric integration method, that is, an easy closed formula cannot be given.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .