Properties of Accounting Numbers: Models and Tests

Abstract

Statistical models for accounting numbers, particularly income-related numbers, have attracted considerable interest. Some reasons for this interest are: (1) the potential use of forecasts of accounting numbers as inputs to decision models; (2) the need to secure proxies for unobservable expectations in order to test economic theories; (3) the need to use such statistical models within the context of studies dealing with the predictive ability or information content of accounting numbers, subjects that have been receiving increased attention during the past few years;' (4) the growing interest in examining the forecasting success of, for example, managers and financial analysts relative to statistical models that are appropriate for the accounting number series of interest;2 and (5) the need to use accounting numbers in testing hypotheses regarding industrial organization (e.g., market concentration), profitability,3 and the growth and decline of firms.4 The interest in statistical models for accounting numbers obviously induces a need for model-formulation efforts and empirical results regarding the properties of alternative statistical models. If, in fact, a statistical model used in a given study suffers from important misspecifications or if