Can we predict crashes? The case of the Brazilian stock market

In this study we analyze Brazilian stock prices to detect the development of bubbles and crashes in individual stocks using a log-periodic equation. We implement a genetic algorithm to calibrate the parameters of the model and we test the methodology for the most liquid stocks traded on the Brazilian Stock Market (Bovespa). In order to evaluate whether this approach is useful we employ nonparametric statistics and test whether returns after the predicted crash are negative and lower than returns before the crash. Empirical results are consistent with the prediction hypothesis, e.g., the method applied can be used to forecast the end of asset bubbles or large corrections in stock prices.