The paper considers the problem of using convolutional neural networks for predicting time series. As time series were chosen price series of two financial indices with data for the same period of time are considered. When prediction used binary classifier. Two classes were considered: growth and lack of growth. Two-dimensional maps are constructed along the time series segments of the same length. Two convolutional neural networks with a different number of tunable weights and different architecture were trained to predict time series. According to the results of a computational experiment, a conclusion was made about the advantage of a neural network with a large number of tunable weights.