This study implemented the classic time series RNNs (Recurrent neural network) models on S&P 500 ETF dataset.
There are 3 round of study in total. First round experiments applied SMA(simple moving average) as baseline model and compared with classic RNN, LSTM(long short- term memory networks) and GRU(Gate Recurrent Unit) models and the results of different models were compared.
The second round experiments based on GRU model, by tuning different hyperparameters, tries to find the effect on different sliding window, and different hidden layers. The third round of experiments is to test the adaptability of the model(using GRU model as sample) to different data.
Detailed discussion cann be found in RNNs_StockPrediction.pdf.