In the recent past, with the widespread expansion of artificial intelligence, there has also been an increase in the application of machine learning methods for the purpose of forecasting company revenues. The present thesis proposes a comparative analysis of traditional econometric and machine learning approaches to forecasting revenues for Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla. The aim of this study is to determine the most accurate method based on a comparative analysis. The study employs a range of traditional statistical methods, including simple moving averages, decomposition, Holt-Winters exponential smoothing, and ARIMA models, in addition to more sophisticated approaches such as Facebook's Prophet and Artificial Neural Networks. The performance of the models is evaluated by means of standard error metrics, including mean squared error (MSE), root mean squared error (RMSE), and mean absolute percentage error (MAPE). The study ascertains whether the incorporation of holiday effects in Prophet enhances the accuracy of the model, and whether the automation of machine learning model tuning via hyperparameter grid search yields superior performance in comparison to the manual specification of parameters as its report time. It is also presented the architecture of the ANN in a transparent manner and elucidate the adjustments that facilitate the execution and reproducibility of forecasts. The findings indicate that while traditional econometric models furnish useful baselines for the development of more sophisticated methods, Prophet and Artificial Neural Networks consistently demonstrate superiority in terms of forecast accuracy.

--

Séries temporais,Time series,Revenue forecasting,Métodos econométricos,Econometric methods,Machine learning,Redes neuronais,Neural networks,Accuracy