Stock markets in the world are linked by complicated and dynamical relationships into a temporal network. Extensive works have provided us with rich findings from the topological properties and their evolutionary trajectories, but the underlying dynamical mechanism is still not in order. In the present work, we proposed a technical scheme to reveal the dynamical law from the temporal network. The index records for the global stock markets form a multivariate time series. One separates the series into segments and calculates the information flows between the markets, resulting in a temporal market network representing the state and its evolution. Then the technique of the Koopman decomposition operator is adopted to find the law stored in the information flows. The results show that the stock market system has a high flexibility, i.e., it jumps easily between different states. The information flows mainly from high to low volatility stock markets. And the dynamical process of information flow is composed of many dynamic modes distribute homogenously in a wide range of periods from one month to several ten years, but there exist only nine modes dominating the macroscopic patterns.