Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I am now reading a paper which has a time series model given by $$ y_t|\theta_t\sim N(0,\theta_t^{-1}),\\ \theta_t=e^{r_t}\theta_{t-1}\eta_t,\\ \theta_0|Y_0\sim \text{Gamma}(a_0,b_0),\\ \eta_t\sim \text{Beta}(wa_{t-1},(1-w)a_{t-1}), $$ where $Y_t$ is the information available at time $t$, and as I understand $r_t$ is deterministic. The Gamma distribution is defined as $$ f(x;\alpha,\beta)=\frac{x^{\alpha-1}\beta^\alpha\exp(-x\beta)}{\Gamma(\alpha)}. $$ The paper says if we combine the transition equation of $\theta_t$ with the prior, then we have that $$ \theta_1|Y_0\sim \text{Gamma}(wa_0,e^{-r_1}b_0). $$ Bayes' theorem then delivers $$ \theta_1|Y_1\sim \text{Gamma}(wa_0+\frac{1}{2},e^{-r_1}b_0+\frac{1}{2}y_1^2). $$ I am not sure how I can get the distribution of $\theta_1|Y_0$. Thanks for any help!

share|cite|improve this question
Though it is not hugely informative, you might check the wikipedia page on conjugate priors, this would be the normal known mean unknown variance case. you know that $\theta = \frac 1 {\sigma^2}$ , so the likelihood is $\propto e^{-.5 x^2 \theta}$, looking gamma-like in $\theta$.? – mike Jan 15 '14 at 14:39
Hi @mike I added more stuff from the paper to the question. What I have in mind is the first distribution might not be the result of Bayes' updating? – Mathlover Jan 15 '14 at 14:52
up vote 1 down vote accepted

The statement to prove is:

Let $c\gt0$, $S$ with distribution gamma $(a,b)$, and $U$ with distribution beta $(wa,(1-w)a)$ and independent of $S$. Then $cSU$ has distribution gamma $(wa,b/c)$.

First, the distribution of $cX$ is gamma $(v,b/c)$ if and only if the distribution of $X$ is gamma $(v,b)$ hence one can assume without loss of generality that $c=1$. Second, the distribution of $SU$ has density $$ f(z)=\int_0^1f_U(u)f_S(u^{-1}z)u^{-1}\mathrm du, $$ where $f_U$ denotes the density of $U$ and $f_S$ the density of $S$. By hypothesis, $$ f_U(u)\propto u^{wa-1}(1-u)^{(1-w)a-1},\qquad f_S(s)\propto s^{a-1}\mathrm e^{-bs}, $$ hence, after some compulsory simplifications, $$ f(z)\propto z^{a-1}\int_0^1u^{-(1-w)a-1}(1-u)^{(1-w)a-1}\mathrm e^{-bz/u}\mathrm du. $$ The change of variable $u=1/(1+x)$ yields $x$ in $(0,+\infty)$ and $\mathrm du=\mathrm dx/(1+x)^2$ hence $$ f(z)\propto z^{a-1}\mathrm e^{-bz}\int_0^\infty x^{(1-w)a-1}\mathrm e^{-bzx}\mathrm dx. $$ Finally, the change of variable $\xi=bzx$ yields $\xi$ in $(0,+\infty)$ and $\mathrm dx=\mathrm d\xi/(bz)$ hence $$ f(z)\propto z^{a-1}\mathrm e^{-bz}\int_0^\infty \xi^{(1-w)a-1}z^{1-(1-w)a}\mathrm e^{-\xi}z^{-1}\mathrm d\xi\propto z^{wa-1}\mathrm e^{-bz}, $$ which proves the claim.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.