Forecasting Financial Time Series Using LSTM Networks

Abstract

In this paper, we investigate the application of Long Short-Term Memory (LSTM) recurrent neural networks for financial time series forecasting. We compare LSTM networks against traditional statistical models including ARIMA and classical machine learning approaches. Our experiments on equity and cryptocurrency markets demonstrate that LSTM networks capture temporal dependencies and non-linear patterns more effectively, yielding improved directional accuracy and reduced mean-squared error across multiple prediction horizons.