Optimized multivariate lag structure selection

摘要

Choosing the relevant variables is a crucial task in econometric modelling as estimation results and inference depend on this choice. The selection of lag structures in multivariate autoregressive models serves as an example. Instead of using some ad hoc procedure, the selection is performed using an explicit optimization approach for some given information criteria. A complete enumeration of all possible lag structures being infeasible already for moderate dimensions of the model, the refined local munch global optimization heuristic threshold accepting is implemented for this purpose. The results highlight the practicability of this approach.