I have two independent variables with distributions: $X \sim \text{beta}(\alpha_1,\beta_1)$ and $Y \sim \text{beta}(\alpha_2,\beta_2).$

I would like to estimate $P(X Y \geq \text{val})$, and I am looking for a simple method, possibly approximation, so that I won't need to work with product of distributions.

What are the possible (simple) approaches to solve this problem? I can probably simplify the question to accepting/rejecting the assumption of whether $X Y=\text{val}$ with 95% confidence but I am not sure if it helps.

  • $\begingroup$ For integral values of $\alpha_1,\beta_1,\alpha_2,\beta_2$ this probability is a linear function of $\log(\text{val})$ with coefficients that are polynomials in $\text{val}$. Approximation is not necessary, though, because numerical integration can obtain the values as accurately as you like. $\endgroup$
    – whuber
    Apr 3, 2014 at 15:40
  • $\begingroup$ When you say $X*Y$ and write "product of distributions" do you mean convolution of the densities (which is suggested by "$*$"), or do you mean the distribution of the product of the random variables ($X\,Y$)? Neither would be the "product of distributions" ($F_X . F_Y$). $\endgroup$
    – Glen_b
    Apr 4, 2014 at 4:10
  • 1
    $\begingroup$ I meant distribution of the product of the random variables. Sorry, I was not sure what to use for product operator. $\endgroup$
    – Roman
    Apr 4, 2014 at 15:17
  • $\begingroup$ I understand that simulation method can be used. For example, I could draw k pairs$(x,y)$ based on their distributions, multiple them and count how many of them are less than $val$. I wonder if there are simpler and computationally more efficient methods. α1,β1,α2,β2 are indeed integral. $\endgroup$
    – Roman
    Apr 4, 2014 at 15:21
  • 1
    $\begingroup$ See stats.stackexchange.com/questions/166737/… $\endgroup$ Dec 9, 2022 at 3:53


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.