[Gretl-users] Re: Full Information Maximum Likelihood (VECM)

Am 11.07.2019 um 14:49 schrieb macrocs(a)juno.com: Hi Cesar, I think the best way would be something that PcFiml/PcGive/Oxmetrics used to do in the old days: estimate the cointegration relations first and add the estimated level ECTs to a system of differences. In this second system you can then impose arbitrary restrictions. (In a 'vecm' in gretl you can only restrict beta and alpha AFAIK.) See also ch. 31 of the guide (multivariate models). So roughly: lagorder = 4 vecm lagorder 1 x y series ec = $ec # need a loop for rank > 1 diff x y list dxlags = lags(lagorder-1, d_x) list dylags = lags(lagorder-1, d_y) vahidengle <- system equation d_x const ec(-1) dxlags dylags equation d_y const ec(-1) dxlags dylags end system restrict vahidengle ... # concrete coeff restrictions end restrict estimate vahidengle method=fiml However, I don't remember the type of the restrictions needed in this case. If they also involve the levels/EC terms, it could be that some auxiliary identities must also be defined, "explaining" to gretl the relationship between the EC term and the (lagged) differences. (Using 'identity' lines in the system block as mentioned in the guide.) hth, sven