Argue that the proposed estimators πΛ and πΛ below are both consistent and asymptotically normal. Then, give their asymptotic variances π(πΛ) and π(πΛ) , and decide if one of them is always bigger than the other.
Let π1,β¦,ππβΌπ.π.π.π―ππππ(π) , for some π>0 . Let πΜ =πβ―β―β―β―β―π and πΜ =βln(πβ―β―β―β―π) , where ππ=1{ππ=0},π=1,β¦,π .
π(πΛ) =? and π(πΛ) =?
To argue that the proposed estimators πΜ and πΜ are both consistent and asymptotically normal, we need to show that they satisfy the properties of consistency and asymptotic normality.
1. Consistency:
An estimator is consistent if it converges to the true parameter value as the sample size increases. In this case, we want to show that both πΜ and πΜ converge to π as n approaches infinity.
a. πΜ = πβ―β―β―β―β―π:
The sample mean πβ―β―β―β―β―π is a consistent estimator for the population mean due to the Law of Large Numbers. For an i.i.d. sequence of random variables π1, π2, ..., ππ with a common distribution π―ππππ(π), the sample mean πβ―β―β―β―β―π converges to π as n approaches infinity. Therefore, πΜ = πβ―β―β―β―β―π is a consistent estimator of π.
b. πΜ = βln(πβ―β―β―β―π):
In this case, ππ = 1{ππ = 0} is an indicator variable that takes the value 1 if ππ = 0, and 0 otherwise. So, πβ―β―β―β―π represents the proportion of zeros in the sample. As n approaches infinity, the proportion of zeros in the sample will converge to the probability of getting a zero from the distribution π―ππππ(π). Taking the negative logarithm of this proportion, βln(πβ―β―β―β―π), will tend to π. Thus, πΜ = βln(πβ―β―β―β―π) is a consistent estimator of π.
2. Asymptotic Normality:
An estimator is asymptotically normal if, as the sample size increases, its distribution becomes approximately normal with mean π and a finite variance. To determine the asymptotic variances π(πΜ) and π(πΜ), we need to calculate the variances of the estimators.
a. π(πΜ):
For the sample mean estimator πΜ = πβ―β―β―β―β―π, the variance is given by Var(πβ―β―β―β―β―π) = Var(ππ)/π, where Var(ππ) = π(1 - π) for π.π.π. random variables from the π―ππππ distribution. Thus, π(πΜ) = π(1 - π)/π.
b. π(πΜ):
To find the variance of πΜ = βln(πβ―β―β―β―π), we need to calculate the variance of πβ―β―β―β―π. Since πβ―β―β―β―π represents the proportion of zeros in the sample, it follows a binomial distribution with parameters π and π. The variance of a binomial distribution with parameters π and π is given by Var(πβ―β―β―β―π) = ππ(1 - π). Taking the negative logarithm does not change the variance, so π(πΜ) = ππ(1 - π).
Comparing the two asymptotic variances, we can see that π(πΜ) = π(1 - π)/π, while π(πΜ) = ππ(1 - π). The ratio of the two variances is π(πΜ)/π(πΜ) = (ππ(1 - π))/ (π(1 - π)/π) = πΒ². Thus, π(πΜ) is always greater than π(πΜ), with π(πΜ) growing quadratically with n.
In conclusion, both estimators πΜ and πΜ are consistent and asymptotically normal. However, the asymptotic variance π(πΜ) is always larger than the asymptotic variance π(πΜ).