Bayesian Model Averaging techniques are used to analyse how robustly it is possible to identify factors that may lead to the bursting of asset price bubbles in OECD economies. A large set of variables put forward in the literature is assessed, as well as interactions of these variables with estimates of asset price misalignments to evaluate the importance of the different channels postulated by theory. The results indicate that asset price misalignments are not robust determinants of house price reversals unless their interaction with other characteristics of the economy (credit growth, population growth and interest rate developments) is taken into account. On the other hand, stock price reversals are affected by misalignments, as well as other real and monetary variables. Out-of-sample prediction exercises provide evidence that dealing explicitly with model uncertainty using Bayesian model averaging techniques leads to better forecasts of reversals in asset prices than relying on model selection. Conclusions regarding the importance of dealing quantitatively with model uncertainty are drawn to improve the anticipation of asset price reversals.