How well do Markov switching models describe actual business cycles? The case of synchronization