Information complexity criteria for detecting influential observations in dynamic multivariate linear models using the genetic algorithm

Estimating all possible SUR models with permuted exogenous data matrices derived from a VAR process

“…The problem of estimating the SUR models with permuted exogenous data matrices arises in subset VAR modeling (Bozdogan and Bearse, 2003;Lu¨tkepohl, 1993;Maringer, 2004;Winker, 2001). The estimation of all subset VAR models is infeasible even for modest dimensions of the original model.…”

“…Thus, the best-subset VAR models could be derived without computing the whole regression tree which generates all sub-models Hand, 1981). The branch-and-bound method which employs a cutting test based on statistical criteria such as BIC, AIC or ICOMP is currently investigated (Akaike, 1969(Akaike, , 1974Bozdogan and Bearse, 2003;Schwarz, 1978).…”

“…The first is the use of heuristics and genetic algorithms (Bozdogan and Bearse, 2003;Winker, 2000). A second alternative is to search through restricted subsets of models.…”

“…A subset VAR model can be seen as a VAR model with zerorestrictions on some of the coefficients (ZR-VAR). In order to identify a ''good'' submodel, several subset VAR models are enumerated by enforcing some coefficients to be zero, and compared with respect to some criterion (Akaike, 1969(Akaike, , 1974Bozdogan and Bearse, 2003;Hannan and Quinn, 1979;Schwarz, 1978). Imposing zerorestrictions on some coefficients is equivalent to selecting the non-zero ones.…”

An efficient branch-and-bound strategy for subset vector autoregressive model selection

“…A subset VAR model can be seen as a VAR model with zero-restrictions on some of the coefficients. In order to select a ''good'' submodel, several subset VAR models are enumerated by enforcing some coefficients to be zero, and compared with respect to a given statistical criterion (Akaike, 1969;Bozdogan and Bearse, 2003;Hannan and Quinn, 1979;Schwarz, 1978). A primary difficulty in the specification of the subset VAR models is the large number of candidate submodels.…”