We combine convolutional neural network (CNN) and gated recurrent unit (GRU) to form a new structure and embed it as a base layer in the autoencoder (AE) framework to efficiently extract features from financial time series data. To better utilize the hierarchical features, skip connection is...