LinearRegression4

k ver 10232012 p kolm 11 large sample inference

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: that is1 E (u ) = 0 Cov(x j , u ) = 0, for j = 1, 2,..., k VER. 10/23/2012. © P. KOLM 11 Large Sample Inference (1/2) Recall: • Under the classical linear model (CLM) assumptions, the sampling distributions are normal, so we could derive t - and F -distributions for hypothesis testing • The normality was due to assuming the population error was normal • The assumption of normal errors implied that the distribution of the dependent variable, y , given the independent variables, x 1,..., x k , was normal as well VER. 10/23/2012. © P. KOLM 12 Large Sample Inference (2/2) • There are many situations in which this exact normality assumption will fail o For example: Any skewed variable (like hedge fund returns, option returns, wages, savings, etc.) cannot be normal as a normal distribution is symmetric • Note that the normality assumption is not needed to conclude OLS is BLUE o We only used it for inference (hypothesis testing) VER. 10/23/2012. © P. KOLM 13 Central Limit Theorem • Recall: The central limit theorem states that the standardized average of any population with mean μ and variance σ 2 is asymptotically ∼ N (0,1), that is Y −μ a Z= ∼ N (0,1) σ n • Asymptotic normality means that P(Z < z ) → Φ(z ) as n → ∞ (where Φ(z ) denotes a normal CDF) • Based on the central limit theorem, we can show that OLS estimators are asymptotically normal. In particular, we have the important result on the next page VER. 10/23/2012. © P. KOLM 14 Asymptotic Normality of OLS (1/2) Facts: Under the Gauss-Markov assumptions, we have that 1. For each j = 1,..., k (excluding the intercept, j=0) ˆ (β a j s ˆ where se (β j ) = 2 j SSTj (1 − R ) ˆ − β j ) se (β j ) ∼ N (0,1) (the usual OLS standard error) n 1 ˆ 2. s = ∑ ui2 is a consistent estimator of σ2 n − k − 1 i =1 2 3. ( a ˆ n (β j − β j ) ∼ N 0, Avar where Avar ˆ ( n (β − β ))) j j ˆ ( n (β − β )) is the asymptotic variance of j j ˆ n (β j − β j ) (for the ˆ slope coefficients).2 We say that β j is asymptotically normally distributed VER. 10/23/2012. © P. KOLM 15 Remarks: • Because the t -distribution approaches the normal distribution for large df , we can write a ˆ ˆ (βj − βj ) se (βj ) ∼ tn −k −1 ∼ N (0,1) • Note that while we no longer need to assume normality with a large sample, we do still need homoscedasticity (For you: Why?) • Hypothesis testing: Asymptotic normality implies that . . . o . . . the t -statistic is approximately normal if the sample size is large enough → we can conduct t-tests of single regression coefficients using the normal distribution o . . . the F -statistic is approximately F distributed if the sample is large enough3 → thus for testing exclusion restrictions or other multiple hypothesis, nothing changes from what we have done before VER. 10/23/2012. © P. KOLM 16 OLS is Asymptotically Efficient • Under the Gauss-Markov assumptions, the OLS estimators will have the smallest asymptotic variances • We...
View Full Document

This document was uploaded on 02/17/2014 for the course COURANT G63.2751.0 at NYU.

Ask a homework question - tutors are online