The Relation between Persistence Theory and the Time-Series Properties of Earnings
Posted: 27 Feb 1998
In this analytical study we question the theoretical consistency of using time-series parameters estimated from autoregressive integrated moving average [ARIMA] models as proxies for the persistence component of earnings construct. We examine Miller and Rock's (1985) theoretical arguments that the earnings announcement effect is associated with not only the magnitude of the unexpected portion of earnings but also with the persistent component of the unexpected earnings as determined by a persistence parameter. Next we examine the validity of Kormendi and Lipe's (1987) theoretical arguments suggesting that Miller and Rock's persistence construct could be proxied empirically by a moving average parameter from an ARIMA time-series model. The conclusion we draw from our analysis is that the use of an ARIMA moving average parameter is not theoretically consistent with Miller and Rock's persistence construct. Furthermore we find that the argument for use of parameters of an ARIMA model requires that Miller and Rock's persistence construct be equal to zero suggesting no earnings persistence. The most important and urgent implication of our findings is for future earnings persistence research. If the analysis is valid the persistence construct should not be proxied by using parameters from ARIMA time-series models.
JEL Classification: M41
Suggested Citation: Suggested Citation