Resumen
Despite a highly debated subject abroad since the 1960s, mainly in the U.S., the literature about the time series properties of earnings is practically nonexistent in Brazil. Thus, this paper is designed to fill this gap by analyzing the time series behavior of quarterly earnings (operating and net) of a sample of 109 Brazilian listed companies during the period 1995 to 2008. To perform the study, we used five models of earnings forecasting that were previously estimated in the international literature, here called common-structure models, as well as using the Box and Jenkins (BJ) method to verify the possibility of finding a model appropriate to the conditions of the Brazilian market. Empirical evidence is provided showing that only operating earnings can be predicted. It is not possible to apply BJ method to predict net income, as it presents a random behavior, described by a random walk with drift. Also, it is not possible to identify any common-structure ARIMA model for all companies. A model should be identified individually for each company.