We are on $\mathbb{R}^d$ with Borel $\sigma$-algebra. Let $\mu_1, ..., \mu_d$ be probability measures on $\mathbb{R}$ and $\Pi(\mu_1, \mu_2, ..., \mu_d)$ be the set of probability measures on $\mathbb{R}^d$ with first marginal $\mu_1$, second marginal $\mu_2$, etc.

Given $\pi^{(1)}, \pi^{(2)} \in \Pi(\mu_1, ..., \mu_d)$ I am interested in conditions for the absolute continuity $\pi^{(1)} \ll \pi^{(2)}$. More specifically, I am looking for sufficient conditions that do not depend on the marginals $\mu_1, ..., \mu_d$, but solely on the dependence structures of $\pi^{(1)}$ and $\pi^{(2)}$.

I am first of all very interested whether these types of conditions or related questions have already been studied.

More specifically, I have the following idea for such a condition, which I could so far neither prove nor disprove:

Let $C_1, C_2 : [0, 1]^d \rightarrow [0,1]$ be two copulas. I.e., $C_1, C_2$ are CDFs corresponding to $\nu^{(1)}, \nu^{(2)}$, which are probability measures on $[0, 1]^d$ with uniform marginals.

Let $\pi^{(1)}, \pi^{(2)} \in \Pi(\mu_1, ..., \mu_d)$ be given by the dependence structures corresponding to $C_1, C_2$, respectively.

Question: If $\nu^{(1)} \ll \nu^{(2)}$, does then also hold $\pi^{(1)} \ll \pi^{(2)}$?

  • $\begingroup$ If I understand all of your terms correctly, then no...Consider $\nu^1$ which is uniform over all $(x,y)$ where $x$ and $y$ begin with the same digit in base 4, and $\nu^2$ which is similar but with base 2. Then $\nu^1$ is absolutely continuous with respect to $\nu^2$, but this says nothing about the broader dependence structures $\pi$, which may do different things in the null regions of the $\nu$’s. $\endgroup$ – Matt F. Jan 26 at 2:09
  • $\begingroup$ @MattF. Thanks, this sounds like an interesting example. Do I understand it correctly that $(0.1, 0.5)$ might be a sample of both $\nu^1$ and $\nu^2$ (since 1 mod 4 = 5 mod 4), but $(0.1, 0.3)$ could only a sample of $\nu^2$ but not $\nu^1$? If so, could you describe what kind of structure for $\pi$ you have in mind such that absolute continuity might be lost? $\endgroup$ – Steve Jan 26 at 10:01

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.