An Automatic Model Selection Technique Based on Parallel Trans-Dimensional Simulated Annealing

In econometrics, one of the difficulties is to identify the optimum model for a given application. Some traditional methods such as statistical tests and information criteria are usually used to choose the models, but when the number of competing models is large, these methods will quickly become infeasible. The simulated annealing (SA) may be a alternative technique to solve this big data problem, but SA´s implementation is frequently less efficient due to inherent intensive computing features. In this paper, therefore, we simultaneously consider three kinds of techniques, which are simulated annealing, reversible jump Markov Chain Monte Carlo and parallel strategy, propose the parallel trans-dimensional simulated annealing algorithm (PTDSA). Next, we empirically employ the PTDSA algorithm to the Shandong regional economic dataset in China, and obtain the optimum model automatically and efficiently.