Chapter 2: Statistical properties of the OLS estimator (part III) Florian Pelgrin HEC September-December, 2010 Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 1 / 18 1 What is an estimator? 2 What constitutes a good estimator? Unbiased estimator Best unbiased estimator in a parametric model Best invariant unbiased estimators 3 Summary Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 2 / 18 What is an estimator? 1. What is an estimator? Definition A point estimator is any function T (Y1 , Y2 , · · · , Yn ) of a sample. Any statistic is a point estimator. Examples: Assume that Y1 ,· · · ,Yn are i.i.d. N (m, σ 2 ) random variables. 1 The sample mean n Ȳn = 1X Yi n i=1 is a point estimator (or an estimator) of m. 2 The sample variance n Sn2 = 2 1 X Yi − Ȳn n−1 i=1 2 is a point estimator of σ . Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 3 / 18 What is an estimator? Remarks: 1 There is no correspondence between the estimator and the parameter to estimate. 2 In the previous definition, there is no mention regarding the range of the statistic T (Y1 , · · · , Yn ): the range of the statistic can be different from the one of the parameter. 3 An estimator is a function of the sample ⇒ it is a random variable (or vector). 4 An estimate is the realized value of an estimator (i.e. a number) that is obtained when a sample is actually taken. For instance, ȳn is an estimate of Ȳn and is given by: n ȳn = 1X yi . n i=1 Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 4 / 18 What constitutes a good estimator? 2. What constitutes a good estimator? A good estimator is one that: 1 ...is unbiased, i.e. E(T (Y )) = θ or E(T (Y )) = g(θ) where g is known; 2 ...satisfies some asymptotic properties (when the sample size is large); 3 ...is efficient, i.e. has the minimum variance among all estimators of the quantity of interest; 4 ...is the best estimator in a restricted class of estimators that satisfies some desirable properties (search within a subclass): the class of unbiased estimators, etc. 5 ...is the best estimator, which has some appropriate properties, by maximizing or minimizing a criterion (or objective function); 6 ... Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 5 / 18 What constitutes a good estimator? Unbiased estimator 2.1. Unbiased estimator... Definition An estimator T (Y ) is unbiased for θ if Eθ [T (Y )] = θ for all θ ∈ Θ. Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 6 / 18 What constitutes a good estimator? Unbiased estimator Examples: 1. Let Y1 ,· · · ,Yn be a random sampling from a Bernoulli distribution. An unbiased estimator of p is p 1X T (Y ) = Yi . n i=1 2. Let Y1 ,· · · ,Yn be a random sample from the uniform distribution U[0,θ] . An unbiased estimator of θ is n T (Y ) = 2X Yi . n i=1 Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 7 / 18 What constitutes a good estimator? Unbiased estimator Examples (cont’d): 3. The multiple linear regression model. Consider the model Y = X β0 + u where Y ∈ Rn , X ∈ Mn×k is nonrandom, E(u) = 0n×1 , and V(u) = σ 2 In . The OLS estimator −1 t T (Y ) = X t X X Y is an unbiased estimator of β 0 . 4. The generalized multiple linear regression model. Consider the model Y = X β0 + u where Y ∈ Rn , X ∈ Mn×k is nonrandom, E(u) = 0n×1 , and V(u) = σ 2 Ω. The Ω matrix is known. The GLS estimator −1 t −1 T (Y ) = X t Ω−1 X X Ω Y is an unbiased estimator of β 0 . Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 8 / 18 What constitutes a good estimator? Unbiased estimator Some comments...: 1. The unbiasedness condition must hold for every possible value of the parameter and not only for some of these values. 2. In general, the property of unbiasedness is not conserved after a nonlinear transformation of the estimator. 3. Asymptotically unbiased estimators: Definition The sequence of estimators θ̂n ≡ Tn (Y ) (with n ∈ N) is asymptotically unbiased if lim Eθ (Tn (Y )) = θ for all θ ∈ Θ n→∞ where Eθ is defined with respect to Pθ,n . Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 9 / 18 What constitutes a good estimator? Unbiased estimator Unbiasedness is interesting per se but not so much! (a) The absence of bias is not a sufficient criterion to discriminate among competitive estimators. (b) It may exist many unbiased estimators for the same parameter (vector) of interest. (c) This is also true when one requires that the estimator is asymptotically unbiased. Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 10 / 18 What constitutes a good estimator? Best unbiased estimator in a parametric model 2.2 Best unbiased estimator in a parametric model When an estimator is unbiased (restrict the class of estimators), its (matrix) quadratic risk function is given by Rθ (T (Y ), θ) = Eθ (T (Y ) − θ)(T (Y ) − θ)t and thus reduced to its variance-covariance matrix Vθ (T (Y )). Comparing two (or more) unbiased estimates becomes equivalent to comparing their variance-covariance matrices. Definition Suppose that T1 (Y ) and T2 (Y ) are two unbiased estimators. T1 (Y ) dominates T2 (Y ) if and only if Vθ (T2 (Y )) Vθ (T1 (Y )) i.e. the matrix Vθ (T2 (Y )) − Vθ (T1 (Y )) is a positive semi-definite matrix for all θ ∈ Θ. Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 11 / 18 What constitutes a good estimator? Best unbiased estimator in a parametric model Definition Under suitable regularity conditions, a (regular) unbiased estimator of θ is efficient if its variance-covariance matrix equals the Frechet-Darmois-Cramer-Rao lower bound: Vθ (T (Y )) = I(θ)−1 where I(θ) is the Fisher information matrix ∂logL(Y ; θ) ∂logL(Y ; θ) I(θ) = Eθ ∂θ ∂θt with L the likelihood function. Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 12 / 18 What constitutes a good estimator? Best invariant unbiased estimators 2.3. Best invariant unbiased estimators The previous results can only be used to find best unbiased estimators in parametric models. These results no longer apply in the case of semi-parametric models. The class of estimators must be again restricted: impose invariance conditions best linear unbiased estimator best quadratic unbiased estimator Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 13 / 18 What constitutes a good estimator? Best invariant unbiased estimators Best linear unbiased estimators: The Gauss-Markov theorem Theorem Consider the (conditional static) linear regression model: Y = X β0 + u with E(ui | X ) = 0 and V(ui | X ) = σ02 In , E(ui3 | X ) = 0, E(ui4 | X ) = 3σ04 , Y is an n-dimensional vector and X is an n × k matrix, with P(rk (X ) = k) = 1 or rk(X ) = k . The ordinary least squares estimator of β 0 defined by β̂OLS = (X t X )−1 X t Y is the best estimator in the class of linear (in Y ) unbiased estimators of β0 . Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 14 / 18 What constitutes a good estimator? Best invariant unbiased estimators β̂j,OLS is the BLUE of βj , j = 1, · · · , k : Best = smallest variance Linear (in Y or yi ) = β̂j = n P wij yi i=1 Unbiased = E(β̂j ) = βj , ∀β ∈ Θ Estimator = β̂j = f (data). Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 15 / 18 What constitutes a good estimator? Best invariant unbiased estimators Best quadratic unbiased estimators Theorem Consider the (conditional static) linear regression model: Y = X β0 + u with E(ui | X ) = 0 and V(ui | X ) = σ02 In , E(ui3 | X ) = 0, E(ui4 | X ) = 3σ04 , Y is an n-dimensional vector and X is an n × k matrix of rank k (or P(rk(X ) = k) = 1). The estimator of σ02 defined by s2 = 1 û 0 û Y 0 MX Y = n−k n−k is the best quadratic unbiased estimator of σ02 . Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 16 / 18 What constitutes a good estimator? Best invariant unbiased estimators To sum up, parametric parametric (Gaussian) semi-parametric Florian Pelgrin (HEC) Table 1: Efficiency of estimators Fixed regressors Stochastic regressors - Best unbiased estimator - Best unbiased estimator - FDCR bound - FDCR bound - Y ∼ N(X β0 , σ02 In ) - Y | X ∼ N(X β0 , σ02 In ) - BUE and FDCR bound - BUE and FDCR bound - BLUE - BLUE - Gauss-Markov - Gauss-Markov Ordinary least squares estimator September-December, 2010 17 / 18 Summary Key concepts What is an estimator? What is an unbiased estimator? is the unbiasedness property sufficient to distinguish among competing estimators? What is an (asymptotically) efficient estimator? In which sense? How can we characterize efficiency? Florian Pelgrin (HEC) Ordinary least squares estimator September-December, 2010 18 / 18