This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: ECON 140, Fall 2008  11/6 Alex Rothenberg Practice Problems: Serial Correlation and Panel Data Problem 1 Suppose we have the following autoregressive model for the error term (the unobservables) in a regression equation: t = t 1 + u t (1) u t iid N (0 , 2 u ) (2) This question asks you to investigate some of the properties of this process. 1. Fixing the value of = 0, write out the first few terms of the sequence { t } in terms of the u t s. (Do this for t = 0 , 1 , 2 , 3 , 4) 2. Assuming E [ ] = 0, what is the expected value of each term? 3. Assuming that Var( t ) = 2 for all t (homoskedasticity), express 2 as a function of 2 u . 4. What is the covariance between 1 and 3 ? Between 1 and 2 ? Problem 2 Say we want to run the following timeseries regression using ordinary least squares: Y t = + X t + t However, the error term in this equation is actually serially correlated: t = t 1 + u t u t iid N (0 , 2 u ) 1. Suppose we ignore the serial correlation and just use ordinary least squares. Is still unbiased? Under what conditions will this be the case? (HINT: Remember the omitted bias formula from last time.) 2. Try to compute Var( ). What happens now that we have serial correlation?...
View
Full
Document
This document was uploaded on 02/07/2010.
 Spring '09

Click to edit the document details