Short-Term Forecasting of Stock Prices Using Long Short Term Memory

Abstract

Predicting stock market is not an easy task as it is a chaotic system i.e. whose dynamics are sensitive to arbitrarily small differences in initial conditions. Any small changes in the system can produce compound errors in predicting the future behavior of the system. Over the last few years, many machine learning algorithms have been used in an attempt to forecast stock prices. This paper evaluates the effectiveness of a type of Recurrent Neural Network known as Long Short Term Memory (LSTM) to implement technical analysis for making predictions about stock prices of AAPL ticker from NASDAQ exchange. Performance with three popular output activation layers is tested with Adam optimizer as back-propagation algorithm. The performance is compared using Root Mean Square Deviation. The model had an average RMSE value of 12.483 with linear output activation scaled to range (0,1) and 3.258 for the same scaled to a range of (-1,1), 21.769 with sigmoid output activation scaled to range (0,1) and 21.738 with tanh output activation scaled to a range of (-1,1).