For portfolio value-at-risk analysis, a novel approach is proposed based on Independent Component Analysis (ICA) and Mixture Density Network (MDN). Specifically, the original data is first transformed into separate signals which are independent from each other through ICA. Then using MDN their conditional density functions are fitted, from which the joint distribution function of the multivariate time series could be derived. Finally VaR estimates are calculated based on Monte Carlo simulation. This method successfully circumvents the difficult correlation issue within multivariate time analysis and it achieves superior performance compared to traditional EWMA and MVGARCH techniques in empirical study.