Profits: Mean Diverting with High Volatility

This paper seeks to characterize the behavior of the U.S. profit growth rates series over the economic cycle. We raise three fundamental questions which are; first, does the average growth in profits over time exhibit a mean-reverting behavior?  That is, does the growth in profits exhibit a tendency to return to some average value? Second, how volatile are profits and does this volatility obscure the message of profit growth? Finally, do profit growth rates vary between decades/ sub-samples?
We divide the profit growth series into decades since 1970 and then calculate the mean, standard deviation, and the stability ratio. Our efforts suggest that since 1970 the mean and standard deviation of profit growth had actually been decreasing up until 1990s. For the most recent (2000-08) periods, the profit growth shows a tick up in both the mean and the standard deviation. Yet, when we evaluate the entire period as a whole, 1970-2008, we find that the trend coefficient is statistically insignificant. In addition, the higher standard deviation, 12.43, than the mean, 8.19, is an evidence of high volatility in the profits series.
We employ the traditional unit root tests, which are the ADF, PP, and the KPSS, as well as the efficient unit root tests such as the ERS and the Ng-Perron tests on the profits series. The Perron and the Zivot-Andrews tests were also employed. In addition, we follow Hamilton (2008)’s approach and apply an ARCH approach on the profits series.
 If we sum-up our empirical analysis then the level of the U.S. profit growth rates is mean diverting and subject to a structural break. Therefore, the level of the profits series is not appropriate for the modeling and forecasting purpose because of a unit root problem. Due to the presence of a structural break in the profits series it would be better to employ only those techniques which are assuming a structural break in the data e.g., cointegration tests with structural break. In the presence of the ARCH effect OLS standard errors can be misleading, with a spurious regression possibility in which a true null hypothesis is asymptotically rejected with probability one. The ARCH effect and unit root problem have serious consequences for forecasting and that is the forecast band could be narrower then actual.