Financial and Non-Stationary Time Series Forecasting using LSTM Recurrent Neural Network for Short and Long Horizon

Abstract

Multi step ahead (long horizon) forecasting in a time series is a difficult task for various engineering applications in finance, geology, and information technology, etc. It is observed that different approaches mentioned in literature are at some disadvantage to give long term prediction for a non linear time series. Long Short-Term Memory (LSTM) network a type of Recurrent Neural Network (RNN) can be used to solve aforementioned problem and it may replace many practical implementations of the time series forecasting systems. This paper presents a novel LSTM model to give short and long horizon forecasting for a time series data. The LSTM method is preferable over other existing algorithms as LSTM network is able to learn non-linear and non-stationary nature of a time series which reduces error in forecasting. Also, long horizon forecasting is targeted in this paper which give wide range of its applicability. Comparative experiments with existing models demonstrate that our proposed LSTM ensemble method achieves state-of-the-art forecasting performance on four different real-life time series datasets publicly available.