Optimal Lag Structure Selection in VEC-Models

For modelling economic and financial time series, multivariate linear and nonlinear systems of equations have become a standard tool. These models can also be applied to non-stationary processes. However, the resulting finite-sample estimates may depend strongly on the specification of the model dynamics. We propose a method for automatic identification of the dynamic part of VEC-models. Model selection is based on a modified information criterion. The lag structure of the model is selected according to this objective function allowing for "holes". The resulting complex discrete optimization problem is tackled using a hybrid heuristic combining ideas from threshold accepting and memetic algorithms. We present the algorithm and the results of a simulation study showing the method's performance both with regard to the dynamic structure and the rank selection in the VEC-model. The results indicate that the selection of the cointregation rank might depend strongly on the specification of the dynamic part of the VEC-model