Google services have shifted the use of technology in our daily lives, enhancing our communication, collaboration and access information. Given this pervasive influence, the aim of this analysis is to compare the predictions of Alphabet’s stock price using various datasets and machine learning models and understand which models perform better, not only in terms of predictive accuracy, but also in terms of explainability and robustness.
To this aim, we have built three different database, considering the existing economic literature and trying to integrate features deriving from the analysis of the context in which the company works and, in particular, R&D costs from Alphabet’s annual financial reports.
We have applied to the database different state of the art machine learning models, ranging from statistical learning models (linear regression), improved with Ridge regularisation, to classic machine learning models such as Gradient Boosting and artificial neural networks, to more recent deep learning models such as recurrent neural networks.
Additionally, the models have been compared in terms of the recently proposed S.A.F.E. AI model, which includes metrics that can assess the Sustainability, Accuracy, Fairness and Explainability of AI application in a unified manner, with a metrics that is related to the Lorenz curve and the Area Under the ROC Curve.
Our empirical findings show that the choice of the best model to employ to predict Google stock prices depends on the desired objective. If it accuracy, the recurrent neural network is the best model. If it is robustness, the Ridge linear model is the most resilient to changes. If it is explainability, the Gradient Boosting model is the best choice.
