The Time Series Properties of Accrual versus Cash-Based Income Variables: Empirical Evidence from Listed Finnish Firms.

This study compares the time series properties of accrual and cash-based income variables in order to determine the basic (dis)similarities in the underlying mechanisms of their behavior over time. The main motive behind this research objective lies in the observation that while the relevant literature almost unanimously suggests that annual accrual-based income follows a submartingale (random walk with or without a drift) or similar process, the explanations for such behavior are virtually nonexistent. It has been particularly unclear whether the submartingale behavior is primarily due to industrial-organizationbased or accounting-method-based factors. The comparative analysis of accrual vs. cash-based income time series performed in this study tackles the relative strengths of these competing explanations.

Theoretical analysis of serial dependences at the sales and operating income levels produced inconclusive results with respect to the relative magnitudes of autocorrelation in accrual vis-a-vis cash-based variables. Empirical analysis was needed to provide an answer to the research question.

The empirical time series analysis was performed on accrual-based sales, operating income and net income variables, and on their direct cash-based counterparts. The time series data was obtained from the financial statements of 39 listed Finnish firms representing various industries. The data covered the 34-year period 1951-84.

The empirical inquiry involved three main phases. First, the degree of randomness was analyzed with distribution-free tests and autocorrelation analysis. Second, some parsimonious univariate time series models were estimated from the data. Third, the predictive ability of the models was tested in non-overlapping hold-out prediction periods.

The main findings of the empirical time series analysis were as follows. First, the analysis confirmed with data from Finnish firms the prior results obtained in other countries that, on average, the underlying mechanism describing the behavior of annual accrual accounting income variables is a submartingale or similar process. Second, submartingale-type behavior in accrual variables was observed across all income measurement levels analyzed in this study. Third and most important, at the operating and net income levels the submartingale model did not turn out to be robust across the accounting systems : the behavior of the cash-based operating and net income variables were much better described by constant processes than by submartingales. Thus, the main conclusion of this study is that the accountinq-method-based explanation for the observed patterns in annual accrual income numbers is obviously much stronger than the competing industrial-organization-based explanation. Furthermore, another implication of the results is that insofar as market expectations can be proxied with past realizations, the information content of the most recent cash flow number should be relatively small.