![]() |
From | Christopher Baum <baum@bc.edu> |
To | statalist@hsphsun2.harvard.edu |
Subject | st: Re: problem with foreach in Stata 10.1 |
Date | Tue, 18 Nov 2008 11:05:23 -0500 |
< >All good advice, but from a time series econometrics perspective, first-generation unit root tests such as -dfuller- and -pperron- are deprecated in favor of more powerful tests such as -dfgls- (disclaimer: I am one of the original coauthors (with Richard Sperling) of -dfgls- (STB57, STB58, freely available from http://ideas.repec.org/s/tsj/stbull.html) . As demonstrated empirically in the talk
http://ideas.repec.org/p/boc/asug07/7.htmlthe augmented Dickey-Fuller test has very low power compared to the DF- GLS of Elliott, Rothenberg, Stock. An additional advantage is that - dfgls- automatically runs for a number of lag orders and provides a sophisticated selection criterion, due to Ng and Perron, which is superior to AIC/BIC. For instance
webuse lutkepohl dfgls invest, maxlag(12) Kit Baum, Boston College Economics and DIW Berlin http://ideas.repec.org/e/pba1.html An Introduction to Modern Econometrics Using Stata: http://www.stata-press.com/books/imeus.html On Nov 18, 2008, at 2:33 AM, Brian wrote:
- -dfuller- is an r-class command, but prior to Stata 10 it had been leavingbehind regression results in the e-class macros. In general, r-class commands should not affect any estimation results the user has lyingaround, so in Stata 10 we modified -dfuller- to back up and restore theuser's prior estimation results. In Joao Ricardo's example using Stata 10.1, his -estat ic- results arealways showing the results for the first call to -dfuller-; presumably hedid not do any estimation before calling -dfuller-, since in that case- -dfuller- did not see any estimation results that it needed to back up and restore. In fact, -dfuller- should not be saving e-class results in thatcase, either; we will address that in an ado-file update. Using -estat ic- in conjunction with -dfuller- to choose the optimal number of lags to include might at first blush seem like a good idea.However, looking at his results using Stata 9 reveal a problem: the samplesize used in each case depends on the number of lags specified with- -dfuller-. When using an information criterion to select the lag order, you should restrict the estimation sample to the subset of observationsthat can be used with the maximum number of lags you wish to consider.
* * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/
© Copyright 1996–2025 StataCorp LLC | Terms of use | Privacy | Contact us | What's new | Site index |