In this paper, we present a data-driven approach to forecasting stock prices in the Moroccan Stock Exchange. Our study tests three predictive models: ARIMA, LSTM, and transformers, applied to the historical stock price data of three prominent credit companies (EQD, LES, and SLF) listed on the Casablanca Stock Exchange. We carefully selected and optimized hyperparameters for each model to achieve optimal performance. Our results showed that the LSTM model achieved high accuracy, with R-squared values exceeding 0.99 for EQD and LES and surpassing 0.95 for SLF. These findings highlighted the effectiveness of LSTM in stock price forecasting. Our study offers practical insights for traders and investors in the Moroccan Stock Exchange, demonstrating how predictive modeling can aid in making informed decisions. This research contributes to advancing stock market forecasting in Morocco, providing valuable tools for navigating the Casablanca Stock Exchange.

In this study, the reliability of the Fama–French Three-Factor model (FF3F) and the Carhart Four-Factor model (C4F) is examined thoroughly. In order to determine which of the asset pricing models is the best to explain portfolio returns on the Moroccan share market, these two models are indeed evaluated in the Moroccan market. Additionally, it is worth mentioning that five years of monthly data from the firms that listed on the Casablanca Stock Exchange are used in this research, as well over the period of nine years. The results of this inquiry show that these models barely have a partial hold on the Casablanca Stock Exchange (CSE), which limits their ability to predict the cross-sections of returns. In accordance with this, the C4F model has somewhat greater explanatory power than the FF3F Model. Moreover, our research adds to the body of knowledge by inserting two learned material asset pricing theories to the proof in the market, which is still evolving, and where distinctive anomalistic traits still exist (the CSE).

This paper aims at comparing the performance of the different state-of-the-art machine learning techniques in anticipating the performance of stock prices of the energy sector. The data collected cover the period from January 2020 to February 2023 with a daily frequency for the three most imported refined petroleum products in Morocco and trained four regression machines learning (linear regression, lasso regression, ridge regression, and SVR) and four classifiers machine learning (logistic regression, decision tree, extra tree and Random Forest) so that anticipating one day ahead prices direction can take place no matter whether they are negative or positive prices. The performance of regression algorithm is then evaluated using different evaluation metrics, especially MSE, RMSE, MAE, MAPE and R2 to evaluate the performance of regression algorithm while precision, recall and F1 scores are used to evaluate the performance of classifiers algorithm. The outcomes propose that the performance of linear regression and ridge regression takes place equally and outperform other single regression that is lasso regression and SVR for-one-day predictions as a whole. In addition to that, we have come to find that in the classifiers, algorithms group all machine learning display similar predictive accuracy, this is on one hand. On the other hand, the best of them is the logistic regression. In brief, this study suggests that all performance metrics are significantly improved by ensemble learning. Therefore, this study proves that critical information affecting stock movement can be captured by utilizing historical transactions.