Deep learning framework for financial time series prediction using technical indicators and price data

Abstract

Financial Time Series prediction is a challenging task due to its dependency in many socio-economic factors. It depends on both quantitative and qualitative factors in a Financial Market. Quantitative factors can be mathematically modeled but qualitative factors are harder to model. Market behavior depends on both micro-economic as well as macro-economic behavior which includes quantitative and qualitative factors on both of them. Therefore modeling and predicting a financial time series has become a challenging task in Big Data Analytics world.

Deep Neural Networks can be identified as a main tool in Big Data Analytics which could solve the above challenge. Long Short Term Memory Units and Gated Recurrent Units in deep neural networks can accommodate memory cells which can store an accumulated memory. This helped to accurately capture the dependencies of the current data point by previous data points. Financial Time Series heavily depends on their predecessors and these concepts managed to capture such relationships.

This research use a combination of LSTM and GRU Units to accurately predict the Index Close Price of Tadawul All Share Index (TASI) and Stock Close Price of five highly tradable stocks in Tadawul Stock Exchange. Open, High, Low and Close Prices as well as Standard Technical Indicators of Stocks and Indices are primarily used to create the model. Principal Component Analysis is used to reduce the dimensionality. OHLC and Technical Indicator Values are fed to the network based on four different topologies creating four Evaluator Models.