[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: re: tsset with non-integers

From   Kit Baum <>
Subject   st: re: tsset with non-integers
Date   Thu, 7 May 2009 13:23:42 -0400

Frank said he can't tsset his data, and that "SPSS offers users the post regression estimation option to test the independence of residuals terms using the Durbin-Watson test so that users can check the independence assumption. "

Calculating a D-W statistic on, e.g., a cross-section is asinine, and a program that lets you do it is a pretty lousy piece of software. As the sort order of a cross section is arbitrary, I can change the "serial correlation" between successive observations by just sorting the data on some other key.

If you're saying that there are dates associated with your observations, but they are not evenly spaced in calendar time, that's a different issue (and one that sometimes causes people to say that they can't tsset their data without creating unwanted gaps). Business- daily data (e.g. stock market quotes) are not evenly spaced in calendar time, but you can still put trading-day data on a time series calendar.

As another poster recommended, Stata's cluster-robust covariance matrix allows you to consider that the errors within each cluster are arbitrarily correlated. You can calculate a cluster-robust VCE for cross-sectional data (e.g. firms nested within industry clusters), but that does not allow you to test for independence; it just gives you standard errors that take account of such departures from i.i.d. But without some structure you can't test for E e_i e_j = 0 for all i != j, because you have only one observation on each covariance in your sample. Thus to test that the error covariance matrix is, for instance, diagonal, you need to impose some additional structure on the manner in which the off-diagonal elements are generated.

Kit Baum   |   Boston College Economics & DIW Berlin   |
An Introduction to Stata Programming |
   An Introduction to Modern Econometrics Using Stata  |

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index