8. Non-stationary Time series

# Unit root test subtract is if dickey fuller from both

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: e is unit root or not? Unit Root Test:- Subtract Is If Dickey- Fuller from both side =1? =1 then we have unit root DF can not use t or z since under they do not esxist. There is a distribution that works to do test and tabulated in Book ( Table A6, A7) Stata Methods for lags 1) Sort data Sort trend Tsset trend (it tell compute what right variable is) trend To do test estimate If ̂ ̂ Then fail to accept at 5% level (Non Stationary) If have more lag values than 1 then use Augmented Dickey Fuller. CAN have more lags Then if not stationary use augumented D.F.Test. Cointegration (you get faster result of converges) Good thing because you have got two variables with unit root, but there difference is stationary. V) Cointegration :- A linear combination of two or more time series will be non- stationary if one or more of them is non- stationary But if two variables are non stationary a linear combination may be i.e. the errors are stationary. If so, the series are cointegrated( if not cointegrated error will explode but if you cointegrate then errors are stationary). Test: Take difference of time series and does augmented dickey fuller test on difference. Take away point: If series are cointegrated it means least square is “super” consistent. This means parameter estimates converges faster than with cross – sectional or stationary time series. Non- Stationary to Stationary 1) Detrend the data 2) Take the difference 3) Vector error correction...
View Full Document

## This note was uploaded on 10/18/2013 for the course EC 421 taught by Professor Gaus during the Fall '08 term at University of Oregon.

Ask a homework question - tutors are online