A hybrid deep learning model for cryptocurrency returns forecasting: Comparison of the performance of financial markets and impact of external variables

Abstract

This study introduces a finetuned hybrid forecasting model combining both Discrete Wavelet Transform (DWT) and Long Short-Term Memory network (LSTM) to predict dirty and clean cryptocurrency returns (Bitcoin and Ripple). The findings show that the proposed DWT-LSTM model outperforms a large set of benchmark models in terms of forecasting accuracy. We investigate a broader set of predictors involving financial markets (other cryptocurrencies and commodities) and external variables (blockchain information, Twitter economic uncertainty, and CO2 emissions). Our findings underline the comparable performance of the considered predictors, with the Twitter Economic Uncertainty index being the best predictor of Bitcoin returns and S&P GSCI Energy being the best predictor of Ripple returns. We also highlight the superior performance of the trading strategies based on our forecasting results.