Sparse Principal Component Analysis for High Dimensional Vector Autoregressive Models