Abstract: Anticipating the occurrence of financial bubbles holds paramount significance as it empowers investors to make judicious decisions and navigate potential losses adeptly. Additionally, the prediction and identification of bubbles play a pivotal role in achieving financial stability objectives. In light of these considerations, this research paper endeavors to address the challenge of predicting financial bubbles through a methodology that integrates the BSADF test with machine learning algorithms. The initial phase involves the identification of bubbles within the stock prices of all entities comprising the STOXX 600 index, followed by the application of a machine learning framework to forecast bubble values. The study aims to discern and incorporate all relevant features for the prediction of bubbles, employing a diverse array of neural network algorithms to formulate forecasts. Subsequently, the research evaluates the out-of-sample prediction accuracy of these algorithms.

Keywords: financial bubbles forecasting, BSADF, MLP, NBEATS, NHITS

JEL codes: G12, G17

DOI: ...