LEVINLIN Procedure

@LEVINLIN implements tests for unit roots in panel data from Levin, Lin and Chu(2002). The null is a unit root (for all individuals) vs an alternative of a common stationary root. The asymptotics are large N with N/T converging to zero. Alternative tests for unit roots in panel data are @Breitung, @Hadri and @IPShin.

@LEVINLIN( options ) series start end

Parameters

series	series to analyze
start, end	range of series to use. By default, the defined range of series.

Options

DET=NONE/[CONSTANT]/TREND

Individual-specific deterministic components to remove. Note NONE is an error and is provided simply to keep the form of the option similar to related procedures.

SMPL=standard SMPL option [not used]

LAGS=(maximum) number of additional lags of the differenced series to include [Schwert's suggestion]

CRIT=FIXED/[GTOS]/AIC/BIC/HQ

Criterion used in selecting the individual specific lags. FIXED uses the value of the LAGS option for each. GTOS is General TO Specific, starting at the value of the LAGS option and dropping lags as long as the t-stat on the final one has a significance level below the SLSTAY option. AIC, BIC and HQ are the Akaike, Schwarz-Bayesian and Hannan-Quinn, respectively.

SLSTAY=significance level for keeping the marginal lag in CRIT=GTOS [.10]

LWINDOW=NEWEY/BARTLETT/FLAT/PARZEN/QUADRATIC

BANDWIDTH=# of lags for the lag window (or bandwidth for LWINDOW=QUADRATIC) [Schwert's]

RECOLOR/[NORECOLOR]

These choose the method used for computing the ratio of long- to short-run variances. RECOLOR does this by using inverting the autoregressive representation. By default, this is done using a lag window calculation applied to the residuals from the differenced series on the deterministics with lag window type and bandwidth determined by the LWINDOW and BANDWIDTH options.

TITLE="title for output" ["Levin-Lin Unit Root Test: Series ..."]

[PRINT]/NOPRINT

Variables Defined

%CDSTAT	the Levin-Lin unit root test statistic (REAL)
%SIGNIF	the significance level of %CDSTAT as a one-tailed N(0,1) (REAL)
%NGROUP	number of individuals (INTEGER)

Example

open data pedroni_ppp.xls

calendar(panelobs=246,m) 1973:6

data(format=xls,org=columns) 1//1973:06 20//1993:11 ae rf wpi cpi uswpi uscpi country entry year month

set logratio = log(cpi/uscpi)

set logexrate = log(ae)

@levinlin(det=trend,crit=gtos) logratio

Sample Output

Levin-Lin Unit Root Test: Series LOGRATIO

Test has large N, N/T-->0

Null is Unit Root. Alternative is Common Stationary Root.

Individual Specific Components: Constant+Trend

With average lags 13.10 chosen from 15

Long-run variances by Bartlett(Newey-West)(5)

N 20

T 246

t-unadjusted -10.3992

t-adjusted -3.6333

Signif 0.0001