Computes all five of Pedroni's tabulated panel cointegration statistics and standardises each one, following his own five-step recipe: fit the cointegrating regression and keep the residuals; regress the differenced dependent on the differenced regressors and take the long-run variance of those residuals as \(\hat L_{11i}^2\); fit \(\hat e_{i,t} = \hat\gamma_i \hat e_{i,t-1} + \hat u_{i,t}\) and form \(\hat\lambda_i = (\hat\sigma_i^2 - \hat s_i^2)/2\). Long-run variances use the Newey-West estimator.
Value
Named list with statistics and z and
p_values (each a named list over panel_v, panel_rho,
panel_t, group_rho, group_t), plus n_units,
n_regressors, case, p_value, method
and warnings.
Details
Each statistic is read against a normal only after his equation (2), \(Z = (\chi_{N,T} - \mu\sqrt{N})/\sqrt{v}\), with \(\mu\) and \(v\) from Table 2, transcribed here for all three deterministic cases and m = 2..7 regressors.
The panel variance statistic diverges to \(+\infty\) under the alternative, so it alone is read from the right tail; the other four go negative under cointegration and are read from the left.
Table 2 has no m = 1 row, so a bivariate cointegrating regression cannot be standardised from it and that case is refused rather than extrapolated.
Mirrors morie.fn.pdcoin on the Python side.