Concluding that MLEs for exponential distribution parameters are biased and then unbiasing them

165 Views Asked by At

I have the i.i.d. exponential random variables $X_1, \dots, X_n$ with the density functions

$$f(x; \sigma, \tau)= \begin{cases} \dfrac{1}{\sigma} e^{-(x - \tau)/\sigma} &\text{if}\, x\geq \tau\\ 0 &\text{otherwise} \end{cases}$$

I want to (1) calculate the MLEs for $\sigma$ and $\tau$, (2) show that they are biased, and (3) correct the bias and in doing so find unbiased estimators.

I have that the MLEs for $\tau$ and $\sigma$ are $$\tau^{\text{MLE}} = \min\left\{x_{i}\right\}\qquad \sigma^{\text{MLE}} = \frac{\sum_{i=1}^{n}\left(x_{i}-\min\left\{x_{i}\right\}\right)}{n}$$

From this Youtube video, I found that $$E[\min{\{x_i\}}] = \dfrac{\sigma}{n}$$

I also calculated that $$E\left[ \dfrac{\sum\limits_{i = 1}^n (x_i - \min{\{x_i\} })}{n} \right] = E\left[ \dfrac{1}{n} \sum_{i = 1}^n x_i \right] - \dfrac{1}{n} E[\min{\{x_i\} }] = \bar{x} - \dfrac{\sigma}{n^2}$$

So since we have that
$$E[\min{\{x_i\}}] = \dfrac{\sigma}{n} \not= \tau$$ and $$E\left[ \dfrac{\sum\limits_{i = 1}^n (x_i - \min{\{x_i\} })}{n} \right] = E\left[ \dfrac{1}{n} \sum_{i = 1}^n x_i \right] - \dfrac{1}{n} E[\min{\{x_i\} }] = \bar{x} - \dfrac{\sigma}{n^2} \not= \sigma,$$ we can conclude that the MLEs for the parameters $\tau$ and $\sigma$ are biased, right?

Is my work here correct? If so, then how do I now 'unbias' them?

1

There are 1 best solutions below

0
On

Youtube video, I found that

I did not look at the Youtube video but I do not agree...indeed the expectation is independent of $\tau$ that is not nice...

$$F_X(x)=1-e^{-(x-\tau)/\sigma}$$

and thus

$$F_{\min}(z)=1-e^{-n(z-\tau)/\sigma}$$

and

$$f_{\min}(z)=\frac{n}{\sigma}e^{-n(z-\tau)/\sigma}$$

which means that the expectation is

$$E[X_{(1)}]=\int_{\tau}^{\infty}\frac{n}{\sigma}ze^{-n(z-\tau)/\sigma}dz=\tau+\frac{\sigma}{n}$$

concluding:

$X_{(1)}$ is biased for $\tau$ but $T=X_{(1)}-\frac{\sigma}{n}$ is unbiased