|
Aggregation within Firms, Long Memory,
and the Prediction of Earnings
Richard M Frankel, Washington University in St. Louis
Bjorn N Jorgensen, Columbia University
Hans Ole Ae Mikkelsen, Bank of America Securities
ABSTRACT. Accountants produce earnings by aggregating realizations from underlying processes within the firm. We propose a framework for investigating the effect of aggregation on the time-series properties of earnings. We predict that earnings exhibit a specific form of persistence called long memory when earnings are produced by aggregating heterogeneous auto-regressive processes. That is, current earnings help predict earnings in the distant future. We provide empirical evidence consistent with long memory in earnings. Further, we document that taking into account long memory reduces the errors in out-of-sample earnings forecasts. In addition, we find a relation between earnings persistence and economic firm characteristics, such as product-type, size, and leverage is stronger for ARFIMA-based earnings persistence measures than for AR(1)-based earnings persistence measures.
Full-Text is no longer available online. Please contact the author(s) for more information about this manuscript.
Back to Session Listing
|