The purpose of this study is to investigate the time-series properties of semi-annual earnings and to compare the predictive ability of expectation models derived from these time-series properties with that of the seasonal random walk model. This study uses a sample of 195 manufacturing firms for which data are continuously available in the Korea Listed Companies Association database for a period of 10 and a half years (21 semi-annual periods) from the first half of 1987 to the first half of 1997. An examination of the autocorrelation of the time series using the first 19 semi-annual earnings observations for these firms revealed seasonality, as expected. Considering the autocorrelation and seasonality of semi-annual earnings, the autocorrelation of seasonally differenced semi-annual earnings suggests that the MA(2) and IMA(1,2) models are appropriate time-series models for the semi-annual earnings process. When comparing the predictive power of these models with that of the seasonal random walk model using the last two semi-annual periods of the sample, the MA(2) model exhibited the smallest forecast errors on average; however, no statistically significant difference was found between the predictive power of the MA(2) model and that of the intuitive, simple seasonal random walk model.