Comparing recurrent neural network and long short term memory for forecasting stock price of Moderna Inc

Vanilla Recurrent Neural Network (RNN) have feedback loop that do not throw away the past information in the learning process. Vanishing gradient often happen in the learning process of Vanilla RNN effect failed keep long term dependencies and decrease accuration. Long Short-Term Memory (LSTM) solved its problem. The two methods used for forecasting Moderna Inc stock price which is currently producing a covid vaccine. The daily closed stock price used here were collected over the december 10th, 2018 until March 31th, 2022 which divided 80% training data and 20% testing data. Comparison between two methods using Mean Absolute Error (MAE) in testing data. Input layer, based on difference and partial autocorrelation functions, are lag 1, 3, 4, 5 and 6. The best method is LSTM with one hidden layer consist of 50 units and 91 epochs, yielded MEA 0.0076. Forecasting the next 167 days starting from April 1st, 2022, has an uptrend until 239.85.