This paper proposes and motivates a dynamical model of the Chinese stock market based on linear regression in a dual state-space connected to the original state-space of correlations between the volume-at-price buckets by a Fourier transform. We apply our model to the price migration of orders executed by Chinese brokerages in 2009–2010. We use our brokerage tapes to conduct a natural experiment assuming that tapes correspond to randomly assigned, informed, and uninformed traders. Our analysis demonstrates that customers’ orders were tightly correlated—in the highly nonlinear sense of prediction by the neural networks—with Chinese market sentiment, significantly correlated with the returns of the Chinese stock market, and exhibited no correlations with the yield of the bellwether bond of the Bank of China. We did not notice any spike of illiquidity transmitting from the US Flash Crash in May 2010 to trading in China.