Transition-based approaches have shown competitive performance on constituent and dependency parsing of Chinese. State-of-the-art accuracies have been achieved by a deterministic shift-reduce parsing model on parsing the Chinese Treebank 2 data (Wang et al., 2006). In this paper, we propose a global discriminative model based on the shift-reduce parsing process, combined with a beam-search decoder, obtaining competitive accuracies on CTB2. We also report the performance of the parser on CTB5 data, obtaining the highest scores in the literature for a dependency-based evaluation.