It is very important for investors, market regulators, and policy makers to possess a trustworthy ex-ante tool capable of anticipating price exuberance events. This paper proposes a new statistical mechanism to predict speculative bubbles by inferring a significant probability of exuberance at least one step ahead of a bubble peak period. Contrary to other approaches, we combine asset pricing modeling and non-stationarity statistical analysis and use both in the context of adaptive learning to build a dynamic model specification. Monte Carlo simulations show that the ex-ante prediction is improved enormously by adding the estimated abnormal returns into the model. In some cases our mechanism predicts 100% of the last bubbles of the sample up to five periods before the peak. Furthermore, the mechanism is able to successfully anticipate the technological bubble observed in the 1990’s by estimating a probability greater than 90%, one month before the bubble peak. Thus, this new mechanism provides an advantage for investors interested in performing a very profitable “bubble surfing” strategy and for market regulators whose responsibility is to maintain market efficiency.

--

Speculative bubbles,Asset pricing,Non-stationarity,Adaptive learning,Dynamic models

• Economics and Business - Social Sciences