A Technical Guide on RNN/LSTM/GRU for Stock Price Prediction

Sequential data prevail in our lives. Voice data, song data, or language data are examples of sequential data, and univariate time series data are special cases. In sequential data, data arrive sequentially, and the new data should not be abrupt from the previous data. For example, a sentence can be:

• “I am going to feed the dog then do my homework,” or
• “After dinner, they went out for a walk.”

The next word should follow certain grammar rules to its previous word. The serial connectivity is an important property of sequential data. When someone raises an unfinished sentence like “I am going to …”, the hearer expects to hear certain words that can complete the sentence and making grammar sense. For example, it would be obscure if I hear “I am going to I make a ball”, in which the underlined words should not be there. Similarly, a song is a sequential data type. It would be abhorrent if the next musical notes do not follow coherently with the previous notes.

In sequential data, we are interested in forecasting multiple periods in the future. For example, in the sentence “I am going to …”, we are interested in knowing the multiple words after the existing words “I am going to”. We are not just interested in a one-word or one-period forecast.