A Hybrid Approach of Wavelet Transform, Convolutional Neural Networks and Gated Recurrent Units for Stock Liquidity Forecasting

Abstract

Stock liquidity forecasting is critical for investors, issuers, and financial market regulators. The object of this study is to propose a method capable of accurately predicting the liquidity of stocks. The few studies on stock liquidity forecasting have focused on single models such as Seasonal Auto-Regressive Integrated Moving Average with eXogenous factors, the nonlinear autoregressive network with exogenous input, and Deep Learning. A new trend in forecasting which attempts to combine several approaches is emerging at the moment. Inspired by this new trend, we propose a hybrid approach of Wavelet Transform, Convolutional Neural Networks, and Gated Recurrent Units to predict stock liquidity. Our model is tested on daily data of companies listed on the Casablanca Stock Exchange from 2000 to 2021. Its forecasting performances are evaluated based on the Mean Absolute Error, the Root Mean Square Error, the Mean Absolute Percentage Error, Theil's U statistic, and the correlation coefficient. Finally, the outperformance of the proposed model is confirmed by comparison with other reference forecasting models. This study contributes to the enrichment of the field of prediction of financial risks and can constitute a framework of analysis allowing to help the stakeholders of the financial markets to forecast the liquidity of the actions.