This work aims to provide a comparative evaluation of deep learning models and traditional econometric methodologies to forecast returns at the sector level. Specifically, the work will focus on financial markets. Based on historical data from the Moroccan stock market, we used Long Short-Term Memory networks and econometric approaches to forecast future returns for the MASI Index and the sectors that are linked with it. Our research has provided us with insights into the strengths and limitations of various methodologies in terms of the accuracy of forecasts, the management of sectoral correlations, and the resilience to systemic market risks. Empirical data has demonstrated that econometric models provide improved interpretability and shed light on the interconnections between different firms. This is despite deep learning models having higher capabilities for short-term forecasting.