LinearRegression4

10232012 p kolm 49 standard errors through

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: parameters not affected • Large sample size in this example justifies use of robust standard errors • Robust standard errors can be larger or smaller than usual ones so it can make a difference to the inferences we draw VER. 10/23/2012. © P. KOLM 49 Standard Errors through Bootstrapping See the separate handout “Computationally Intensive Econometric Techniques: The Bootstrap” for discussion of the bootstrap We repeat part (i) of the previous example, calculating the standard errors using bootstrapping For the results, see the Matlab code VER. 10/23/2012. © P. KOLM 50 Important Questions Why not always use robust inference? • Researchers often do in cross-sectional applications when n is large • In small samples robust standard errors may be far out What about the efficiency of OLS under heteroscedasticity? • OLS not BLUE in presence of heteroscedasticity so a better estimator is potentially available • GLS, WLS, or Feasible Least Squares can provide us with a more efficient estimator How do we know that we are dealing with heteroscedasticity? Can we test for it? • Often, by visually inspecting the residuals we can often get some intuition whether we are dealing with heteroscedasticity • Yes, we can formally test for the presence of heteroscedasticity VER. 10/23/2012. © P. KOLM 51 Testing for Heteroscedasticity: The Basic Idea We test the null hypothesis of homoscedasticity, that is H 0 : Var (u | x 1, x 2,…, x k ) = σ 2 From the definition of variance and since E (u | x ) = 0 , this is equivalent to ( ) () H 0 : E u 2 | x 1, x 2,…, x k = E u 2 = σ 2 VER. 10/23/2012. © P. KOLM 52 Assuming the relationship between u 2 and x j is linear, we estimate the model u 2 = δ0 + δ1x 1 +… + δk x k + v and test H 0 : δ1 = δ2 = … = δk = 0 H1 :H 0 not true using the F-statistic F= 2 Ru 2 / k ˆ 2 (1 − Ru 2 ) / (n − k − 1) ˆ ∼ Fk ,n −k −1 • As usual we reject if the observed value of the test statistic exceeds an appropriate critical value • Rejection implies heteroscedasticity is present VER. 10/23/2012. © P. KOLM 53 Remarks ˆ • We don’t observe the error u, but the residuals from the OLS regression, u , provide an estimate for it • We estimate a regression of these residuals squared on a set of “suspect regressors” and an intercept and calculate the R 2 from this regression 2 (denoted by Ru 2 ) ˆ • The suspect regressors can be all (k), or just some of the regressors in the original model (s < k ) • Statistical packages typically provide tests for heteroscedasticity: o Breusch-Pagan test for heteroscedasticity (as above) o White test for heteroscedasticity (more general form of heteroscedasticity, squares and cross products of independent variables) VER. 10/23/2012. © P. KOLM 54 Generalized Least Squares (GLS) • It is always possible to estimate robust standard errors for OLS • However, if we know something about the specific form of the heteroscedasticity, we can obtain more efficient estimators than OLS • Basic idea: Transform the heteroscedastic regression model into one that has homoscedastic errors an...
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