Forecasting of corporate revenues with machine learning models versus traditional methods in the digital industry

• Autor(a) principal: Pattenden, João Ribeiro dos Santos
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.14/42245
• topic: Machine learning; Sales and revenues forecasting; Digital companies; Pre-processing; Time series; Empirical comparison; Accuracy; Machine learning; Previsão de vendas e receitas; Pre-processamento; Séries temporais; Comparação; Empresas digitais; Domínio/Área Científica::Ciências Sociais::Economia e Gestão

There has been a growing interest in the applicability of machine learning models in corporate sales and revenues forecasting. Past research has found promising results in this field, which show that these models might outperform more traditional methods. In this thesis, three real-world datasets containing information about the revenues and other fea tures of Amazon, Microsoft and Netflix in the last two decades are investigated to forecast the revenues of these digital companies. Firstly, we apply different pre-processing tech niques on the data, which include seasonal differencing using logarithm transformations. Then, some more classical time-series methods including Autoregressive model or order 1 are built. Moreover, different machine learning models including Partial Least Squares and Deep Neural network are applied. Finally, an empirical comparison of the models is performed using metrics such as Mean Absolute Error and Akaike Information Criterion. The results show that Autoregressive model of order 1 outperforms all the other models in terms of revenues forecasting accuracy in all datasets. Particularly, comparing with the benchmark machine learning model in each dataset, this method is able to reduce the error by more than 12 % and up to 72 %. Although these findings require further research to ad dress any possible limitations, they provide insights on the performance of several models in revenues forecasting of digital firms, which can be a valuable tool for the decision-making process of businesses in this industry.