Levin and Lin Panel Unit Root Test
res = Levin_Lin1(Y,t_order,LagOrders,pmax,bandwidth,kern,signif)
* Y = matrix (T,N) of observations
The data matrix may be unbalanced.
Missing Values must be specified as %nan
* t_order = -1 : no individual effect
0 : individual effects (Default)
1 : individual effects and time trends
* LagOrders =
- %nan if the user wants the program to determine the optimal lag order for each country (pmax=12 or T/4)
- a vector (N x 1) or (1 x N) of optimal lags for all the individuals of the panel
* pmax = Maximum of the lag order authorized
* bandwidth = 'C' (Default) common lag troncature for the Bartlett kernel (Levin and Lin 2003)
'N' for the Newey-West (1994)'s non parametric bandwidth parameter
'A' for the Andrews (1991) automatic bandwidth parametr selection with AR(1) structure
* kern = 'B' for Bartlett (Default)
'QS' for Quadratic Spectral (not possible when bandwidth = 'C')
* signif = significance level for the individual ADF tests (0.05, 0.01 or 0.1)
res = a results tlist with:
- res('meth') = 'Levin-Lin'
- res('y') = (T x k) matrix of data
- res('tstat') = (T x k) matrix of data
- res('critical') = Critical Values at 1%, 5% and 10% for the N(0,1) distribution
- res('pvalue') = Pvalue for the corrected statistic
- res('rho') = Estimate of the common autoRegresive parameter
- res('var_rho') = Variance of the estimator of the common autoRegresive parameter
- res('mu_star') = Mean Adjustement Factor
- res('sig_star') = Variance Adjustement Factor
- res('bandwidth') = Method to fix the bandwidth parameter
- res('kernel') = kernel function
- res('h') = Bandwicth parameter
- res('omega') = Individual Long run variances of first differences
- res('si') = Individual Ratios of long run standard deviation to the innovation SE
- res('pi') = Individual lag order in ADF regressions
- res('pmax') = Maximum lag order for ADF individual regressions
- res('Ti1') = Auxiliary Regression 1: Adjusted individual size, starting date and ending date
- res('Ti2') = Auxiliary Regression 1: Adjusted individual size, starting date and ending date
- res('t_order') = the trend order (-1, 0 or 1)
- res('t_orderlit') = the trend order in plain english
- res('tstat_nc') = Non corrected t-statistic (if t_order == 1 only)
- res('pvalue_nc') = Pvalue for non corrected t-statistic (if t_order == 1 only)
load(GROCERDIR+'/data/gdpan_oecd.dat'); Y=explone(dblist(GROCERDIR+'/data/gdpan_oecd.dat')); // extract matrix of data present in database res=Levin_Lin1(log(Y),0,%nan,5,'C','B',0.05) // Provides the Levin and Lin test for all Y columns, with a constant but not trend (arg # 2 = 0), // the Lag order to be determined by the program (arg # 3 = %nan), // the maximum number of lags set to 5 (arg # 4 = 5). res = Levin_Lin1(Y,1,2*ones(size(Y,2),1),%nan,'N','B',0.05) // xample 2 is provides the Levin and Lin test for all Y columns, with a constant and a trend (arg # 2 = 1), // the Lag order set to 2 for all variables, the maximum number of lags set to %nan because it is irrelevant (arg # 4 = %nan), // with Newey-West bandwidth (arg 5 set to 'N') and Barlett kernel. | ![]() | ![]() |