We know that the density function of standard Cauchy Distribution is $p(x)=\frac{\lambda}{\pi (\lambda^2+x^2)}$.

It seems that $p(x)$ can't be written as the form of Exponential Family $$f_{X}(x;\theta)=h(x)\exp\Big(\sum^{s}_{i=1}{\eta_i (\theta)T_i (x)-A(\theta)}\Big)$$

But how to prove it strictly? or how to show that Cauchy Distribution doesn't belong to the Exponential Family?

  • $\begingroup$ I'm sorry, the probability density function for Cauchy Distribution with parameter $\lambda$ should be $p(x;\lambda)=\frac{\lambda}{\pi (\lambda^2 +x^2)}$ $\endgroup$ – Victor Chen Mar 29 '16 at 16:43
  • 2
    $\begingroup$ You can use the "edit" button under the question. $\endgroup$ – zhoraster Mar 29 '16 at 16:47

Hint: If $X$ were from exponential family, it would have finite expectation (you may even express it in terms of $A(\theta)$).

  • 1
    $\begingroup$ Can you explain why $X$ must have finite expactation? $\endgroup$ – Daniel Yefimov Dec 7 '17 at 13:10
  • $\begingroup$ @Daniel Yefimov Because we assumed $X$ is from exponential family. Check Bartlett identity. $\endgroup$ – Zhanxiong Dec 7 '17 at 13:38
  • $\begingroup$ The sufficient statistic must have finite expectation. $\endgroup$ – Sextus Empiricus Jun 12 '20 at 21:28

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.