About Method of Moments and Maximum Likelihood

107 Views Asked by At

\begin{aligned} &X_{1}, X_{2}, \ldots, X_{n} \text { are n samples from a population with density }\\ &f(x)=\left\{\begin{array}{ll} (\theta+1) x^{\theta} & 0<x<1 ; \quad \theta>0 \\ 0 \text { otherwise } \end{array}\right. \end{aligned}

I have found that method of moment for this is $\frac{2 \sum_{i=1}^{n} x_{i}-n}{n-\sum_{i=1}^{n} x_{i}}$ and MLE is $\frac{-\left(n+\sum_{i=1}^{n} \ln \left(x_{i}\right)\right)}{\sum_{i=1}^{n} \ln \left(x_{i}\right)}$ .

Now I have this question ; Suppose $\mathrm{n}=2,$ and $X_{1}=.125, X_{2}=.375,$ what is your (moment) esti- mated value of $\theta$ ?

The value turns out to be negative but parmeter $\theta > 0$ so what is method of moment estimate in this case ?

1

There are 1 best solutions below

2
On BEST ANSWER

The fact that the MoM estimate can give an estimation that is not acceptable, out of the parameter support, is one of the reason why it is not always a good estimator, moreover if you compare it with the ML one.

Usually your MoM estimator can be expressed in a more suitable way

$$\hat{\theta}_{MM}=\frac{2\overline{X}_n-1}{1-\overline{X}_n}$$


EDIT: in this case, both the estimators are evidently out of the specified parameter support.

Actually your $f(x)$ is a beta distribution, $Beta(\theta+1;1)$ thus it is well defined for any $\theta>-1$.

Your exercise states $\theta>0$ and states a particular pair of observation to get you in trouble... thus the solution is the following

$$\hat{\theta}_{MM}=\max\left\{\frac{2\overline{X}_n-1}{1-\overline{X}_n};0^+\right\}$$

Actually, setting $\theta>0$ or $\theta\geq 0$ the only change is that if $\theta=0$ $X\sim U(0;1)$ thus I will write

$$\hat{\theta}_{MM}=\max\left\{\frac{2\overline{X}_n-1}{1-\overline{X}_n};0\right\}$$

Same story for MLE