5

How would I simulate observations from a bivariate Poisson distribution such that they have a nonzero covariance? The hint I was given is that I need to use the fact that the sum of two Poisson random variables is also Poisson.

Glen_b
  • 257,508
  • 32
  • 553
  • 939
user69502
  • 51
  • 1
  • 2
  • 1
    See http://stats.stackexchange.com/questions/108705. Both answers there show how values can be simulated by means of three independent Poisson variates (although they do not explicitly point that out). – whuber Feb 21 '15 at 20:25
  • 2
    Please see the `self-study` [tag wiki](http://stats.stackexchange.com/tags/self-study/info). The hint you already have is a complete giveaway; I'm not sure there's much of anything between the hint you already have and telling you exactly which Poissons to add ... i.e. a complete solution. Oh, wait, there's one -- start with more than two independent Poissons. – Glen_b Feb 22 '15 at 09:53

1 Answers1

9

Since you do not impose any constraint on the joint distribution, any copula structure gives you a solution. For instance,

  1. take a bivariate normal $$(x_1,x_2)\sim\mathcal{N}_2\left((0,0),\left[\matrix{1 &\rho\\\rho &1}\right]\right)$$generation;
  2. turn $(x_1,x_2)$ in correlated uniforms as $$(u_1,u_2)=(\Phi(x_1),\Phi(x_2))$$where $\Phi(\cdot)$ is the normal CDF;
  3. derive two Poisson variates $(n_1,n_2)$ with parameters $\lambda_1$ and $\lambda_2$ from $(u_1,u_2)$ by inverting the Poisson CDF.
AWB
  • 103
  • 2
Xi'an
  • 90,397
  • 9
  • 157
  • 575
  • 1
    +1 but could you maybe also add an introductory reference to the copula link? It's a wikipedia stub at present. – conjugateprior Feb 22 '15 at 10:22
  • @conjugateprior: thank you, the link was damaged and I just corrected it to the entire Wikipedia page. – Xi'an Feb 22 '15 at 10:42