A long short-term memory algorithm-based approach for univariate time series forecasting with application to GDP forecasting

Abstract

This work presents a time series forecasting method based on Long Short-Term Memory (LSTM) network, which can be utilized for macroeconomic variable forecasting, like Gross Domestic Product. LSTM is a popular method in Artificial Neural Networks and is an active research topic, however applications in forecasting are limited. The current work focuses on one-step ahead forecast, and uses Python Keras libraries for the implementation. The method is applied to forecast Greek Gross Domestic Product and the accuracy results are high and comparable to ARIMA approach. The model we present offers a competent approach for time series and GDP forecasting, with comparable accuracy to traditional statistical approaches. It demonstrates the feasibility of the approach and can be further developed in parameter tuning and application on diverse large data sets.