moment generating function of a random variable dependent on a different one

251 Views Asked by At

Assuming I have a random variable $Y \sim \mathrm{U}(0,1)$, and a random variable $X$ has conditional distribution dependent on $Y$: $X\mid _{Y=p} \; \sim \mathrm{Bin}(n,p)$ (Bernoulli with success probability $p$ and $n$ experiments).
How can I find the moment generating function for $X$?

My idea was to use the fact that $\mathrm{E}(X)=\mathrm{E}(\mathrm{E}(X\mid Y))$, but I get stuck in the calculations (maybe my calculus is lacking, if so I would love to know the trick for solving it).

How would you solve and approach these type of questions?

Any help is greatly appreciated.

1

There are 1 best solutions below

0
On BEST ANSWER

Recall the definition of MGF:

$$M_X(t) = \operatorname{E}[e^{tX}].$$ So by the law of total expectation,

$$\operatorname{E}[e^{tX}] = \operatorname{E}[\operatorname{E}[e^{tX} \mid Y]].$$

Now think about what the inner expectation means: $\operatorname{E}[e^{tX} \mid Y]$ is the MGF of the conditional distribution of $X$ given $Y$, because given $Y = p$, $X$ is binomial with parameters $n$ and $p$. So $\operatorname{E}[e^{tX} \mid Y]$ is simply the MGF of a binomial distribution where $p$ has been replaced by $Y$:

$$\operatorname{E}[e^{tX} \mid Y] = (1 + (e^t - 1)Y)^n.$$

Then we take the expectation of this with respect to $Y$, which is continuous uniform on $[0,1]$:

$$M_X(t) = \operatorname{E}[(1 + (e^t - 1)Y)^n].$$ This calculation I leave as an exercise for the reader.