I was wondering if anyone could help with choosing optimal lag length for ZA test. There have been two lag order selection methods commonly used in the literature: 1) The ZA paper recommends to run the test with maximum number of lags. Then the lag order is reduced sequentially until the longest lag is statistically significant; 2) One could also use AIC or SBC or other criteria to choose lag order. I am using annual series with 22 observations. Which of the above lag order selection procedures would be correct to apply? -- View this message in context: http://r.789695.n4.nabble.com/ZA-unit-root-test-lag-order-selection-tp4664183.html Sent from the R help mailing list archive at Nabble.com.
Un texte encapsul? et encod? dans un jeu de caract?res inconnu a ?t? nettoy?... Nom : non disponible URL : <https://stat.ethz.ch/pipermail/r-help/attachments/20130416/d2f15e72/attachment.pl>
Dear Matthieu, Many thanks for your reply. I was not sure what the best way forward in selecting lag length. Eventually I wrote a function that carries out serial correl test and AIC based lag length selections. I used "urca" package. Here is what I come up with in the end:>zamod.A=ur.za(x, model="intercept",lag=j) >AIC(eval(attributes(zamod.A)$testreg))# for lag order selection>bgtest(attributes(zamod.A)$testreg,order=3)$p.value # to test forserial correlation the final decision on the best model comes from examining the results of AIC and BG tests. Thanks, Rosh -- View this message in context: http://r.789695.n4.nabble.com/ZA-unit-root-test-lag-order-selection-tp4664183p4664350.html Sent from the R help mailing list archive at Nabble.com.