This paper adopts the value-at-risk (VaR) method as the risk measurement framework, and establishes a series of VaR-GARCH models to measure the dynamic risk of short-term international capital random fluctuations. Then, the accuracy of the empirical results is measured by the Backing-Test. The results show that the measurements of VaR-GARCH(1,1) and VaR-GARCH-M models are relatively conservative; the measurement results of VaR-EGARCH model are closest to the true value of short-term international capital flows.