Study of instantaneous dependence among several variable is important in many of the high-dimensional sciences. Multivariate GARCH models are as a standard approach for modelling time-varying covariance matrix such phenomena. Cholesky GARCH is one of these approaches where the time-varying covariance matrix can be written parsimoniously, containing variance components through a diagonal matrix and dependency components through a unit lower triangular matrix with regression coefficients as entries. In this paper, we proposed a stochastic structure for dependency components in Cholesky GARCH model by considering linear regression model as a state-space model and using kalman filtering for estimating regression coefficients. We find that the MSE of stochastic Cholesky GARCH model is smaller than the MSE of other models also show that the stochastic Cholesky GARCH has better performance in compare to another models based on MAE and MSE criterions for the real data.