The independent variables commonly used to predict the stock price index usually contain data sampled at different frequencies, and simultaneously, there exist multiple outputs. However, most current researches ignore different frequencies among independent variables and multi-output issues. This paper proposes a multiple output support vector machine unrestricted mixed data sampling (MSVM-UMIDAS) approach – which can achieve multiple results for sequential points simultaneously by applying mixed frequency independent variables. We test the in-sample and out-of-sample performances of MSVM-UMIDAS for stock forecasting in terms of (t−1), (t−2) and (t−3) and then compare the performances of the proposed model with those of other models. The results indicate that our model performs better when assessed by four different measurements. Thus, our proposed model is more realistic in practice and an appropriate tool for multi-output and mixed frequency issues for stock price forecasting.