Deep Learning and Statistical-Based Daily Stock Price Forecasting and Monitoring

Abstract

Designing a machine learning algorithm to predict stock prices is a subject of interest for economists and machine learning practitioners. Financial modeling is a challenging task, not only from an analytical perspective but also from a psychological perspective. After 2008 financial crisis, many financial companies and investors shifted their interest toward predicting future trends. Most of the existing methods for stock price forecasting are modeled using nonlinear methods and evaluated on specific datasets. These models are not able to generalize for diverse datasets. Financial time series data is highly dynamic in nature and makes it difficult to analyze through statistical methods. Recurrent neural network (RNN)-based long short-term memory (LSTM) networks were able to capture the patterns of the sequences data meanwhile statistical methods tried to generalize by memorizing data instead of recognizing patterns. In this book chapter, we examined the performance of LSTM model and statistical models over stock prices of different companies to generalize the model. The experimental results of this study show that, LSTM networks outperformed traditional statistical methods like ARIMA, MA, and AR models. Furthermore, we have noticed that, LSTM networks were able to perform consistently on different datasets while statistical methods showed varied performance. Through this project, we addressed the gaps in current models of stock price prediction in both economic and machine learning perspective.