On the basis of time series forecast models of neural networks (NNs) such as the traditional NN, recurrent neural network (RNN), long short-term memory (LSTM), and gated recurrent unit (GRU), this study builds a time series forecast model of ensemble learning (EL) to study the performance of NN models, the EL model, and traditional time series models in stock index prediction. This study takes 16 Chinese and international stock market indexes as samples to compare the performance of the models in different forecast periods and stock markets in different countries and regions. The main conclusions of this study are as follows: First, the NN time series forecast model and the EL time series forecast model based on NNs are significantly more robust than the traditional financial time series forecast model, and the prediction performance is improved by about 35%. Second, the performance of NN models and EL models in Chinese and American stock markets is better than that of the rest of developed countries and regions.