Nonlinearities And Volatility Patterns In Corporate Earnings

Abstract

This paper explores the stochastic properties of the quarterly earnings per share series for industrials, railroads and utilities since 1935. Evidence of stochastic dependency suggests modeling them as a conditionally heteroskedastic process. Changes in earnings tentatively reject the random walk hypothesis since future returns can be predicted from past information. The conditional variance is found to be sensitive to market advances and positively correlated with the conditional mean since the risk premium is statistically significant. Finally, volatility persistence appears to be high in all series.