Consider the following setup with known standard deviation $\sigma$: \begin{align} X|\theta &\sim N(\theta,\sigma^2)\\ \theta &\sim N(\mu_0,\sigma_0^2) \end{align}

and $(\mu_n,\sigma_n^2)$ is the belief after $n$ samples.

Can we say that the posterior mean $\mu_n$ is always finite?

The updating equation seems to suggest that it can be infinite only if the average of samples taken is $\infty$, which has the probability of zero under normal distribution. So, it should be finite. Does it sound right?

Or should I say $\mu_n$ is finite almost surely (since as $x$ goes to infinity, the probability goes to zero, but we can not say the probability of infinity is zero)?

  • 2
    $\begingroup$ For any finite $n$, the posterior mean is finite with probability 1. $\endgroup$
    – Alex R.
    May 19 '17 at 21:39

Since $$\mathbb{E}[\theta]=\mu=\mathbb{E}^{X_{1:N}}\left\{\mathbb{E}[\theta|X_{1:N}]\right\}$$ the quantity $$\mathbb{E}[\theta|X_{1:N}]$$ is almost surely finite. Which means that for almost any realisation of the random vector $X_{1:N}$ the conditional expectation is finite. Or, equivalently, that$$\mathbb{P}(|\mathbb{E}[\theta|X_{1:N}]|=+\infty)=0$$


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.