Definition (Extremum Estimator)

An estimator \(\theta\) is an extremum estimator if there is an objective function \(\hat Q_n(\theta)\) such that

\[ \hat\theta \mbox{ maximizes } \hat Q_n(\theta) \mbox{ subject to } \theta \in \Theta, \tag{1} \]

where \(\Theta\) is the set of possible parameter values. In the notation, dependence of \(\hat\theta\) on \(n\) and of \(\hat\theta\) and \(\hat Q_n(\theta)\) on the data is suppressed for convenience.




앞으로 우리는 아래의 네 가지의 예제를 주로 살펴보고 이 예제 안에서의 extreme estimator들의 점근 분포에 대해 보일 것이다.

Example (MLE)

Let the data \(z_1,\ldots,z_n\) be i.i.d with p.d.f \(f(z;\theta_0)\) equal to some number of a family of p.d.f’s \(f(z;\theta)\).The MLE satisfies eq. (1) with \[ \hat Q_n(\theta)=n^{-1} \sum_{i=1}^n \log f(z_i;\theta). \tag{2} \] Here, \(\hat Q_n(\theta)\) is the normalized log-likelihood.



Example (Nonlinear Least Squares)

Assume that we have the data \(z_i=(y_i,x_i)\in \mathbb{R}^2\) for \(i=1,\ldots,n\), with \(E(Y_i|X_i)=h(x_i;\theta_0)\). The estimator solves eq (1) with \[ \hat Q_n(\theta) = -n^{-1} \sum_{i=1}^n (y_i-h(x_i,\theta))^2. \tag{3} \]



매우 중요하다

Example (General Method of Moments - GMM)

Suppose that there is a “moment function” vector \(g(z,\theta)\) such that the population moments satisfy \(E(g(Z,\theta_0))=0\). A GMM estimator is one that minimizes a squared Euclidean distance of sample moments. Let \(\hat W\) be a positive semi-definite matrix. A GMM estimator is one that solves eq (1) with \[ \hat Q_n(\theta) = - (n^{-1}\sum_{i=1}^n g(z_i,\theta))' \hat W (n^{-1}\sum_{i=1}^n g(z_i,\theta)). \tag{4} \]



Example (Classical Minimum Distance Estimation - CMD)

Suppose that there is a vector of estimators \(\hat\pi \stackrel{P}{\rightarrow} \pi_0\), and a vector of functions \(h(\theta)\) with \(\pi_0=h(\theta_0)\). An estimator of \(\theta\) can be constructed by solving eq (1) with \[ \hat Q_n(\theta) = - (\hat\pi-h(\theta))' \hat W (\hat\pi-h(\theta)). \tag{5} \]

  • 즉, \(\pi\)가 reduced form parameter, \(\theta\)가 structural parameter라고 생각하고, \(h\)를 어떠한 linear mapping s.t \(h:\theta\rightarrow\pi\)라고 생각해보자. 그리고 우리는 \(\hat\pi\)를 data로 갖고 있고 이를 maximizing하는 \(\theta\)를 찾는 것이다. 바꿔 말하면 \(\hat\pi\)와 가장 가까운 \(h(\theta)\)를 만족시키는 \(\theta\)를 찾는 것이다.



back