Stock Market Prediction with LSTMs

We implement a paper that performs stock market prediction with three different neural networks using LSTM: single LSTM, stacked LSTM, and attention LSTM.

Comment 23
Share this project:

Updates

posted an update —

Project Check in #2

Introduction:

The paper we want to implement builds three different neural networks using LSTM, in order to perform stock market prediction with S&P 500 index (SPY). We decided to work on the paper because we were not only interested in solving problems in the stock market, but also interested in comparing different uses of LSTM. We are solving a prediction problem with supervised learning.

Challenges:

First challenge is deciding which interval of years to use. We are not sure how many years of data we should choose. 1 year, 5 years, 10 years?

Insights:

We have the ARIMA model that we will compare with our LSTM models, but we are yet to create deep neural networks with LSTM. The ARIMA model shows convincing results, and demonstrates that we can explore the size of S&P 500 datasets. In other words, since the ARIMA model performs well given a one-year interval of S&P 500, there is a possibility that LSTMs can likewise perform well with the one-year interval data.

The ARIMA model performs well and predicts values almost perfectly, satisfying our expectations. We are yet to complete the three remaining models .

Plan:

We are on track with this project. We need to dedicate more time into developing a model that would predict the stock price accurately. Right now, we plan to go ahead with our initial ideas and develop three different LSTM models, compare them and see which one is more reliable in predicting the prices.

Log in or sign up for Devpost to join the conversation.