The influence of accounting changes on financial analysts' forecast accuracy and forecasting superiority: Evidence from the Netherlands

Abstract

This study assesses the influence of discretionary accounting changes on financial analysts' individual forecast errors in the Netherlands from 1988 to 1999. It contributes to previous research by examining whether accounting changes (1) influence analysts' earnings forecast accuracy; and (2) change analysts' forecasting superiority relative to a mechanical earnings prediction model because of the change in the time series and composition of earnings. The empirical results indicate that changes in accounting procedures can significantly affect analysts' forecast accuracy and forecasting advantage, conditional on the change-year effect, prior disclosure and the type of change. Specifically, this study finds that in the year that firms adopt accounting changes with a material effect on earnings before extraordinary items, analysts' forecast accuracy significantly worsens if the changes have not been previously disclosed. Further, in the earliest years after the adoption of changes from current cost accounting to historical cost accounting and changes from expensing to capitalization analysts' forecast accuracy and forecast superiority significantly improves, whereas analysts' forecast accuracy and superiority significantly worsens after the adoption of other changes.

Keywords:

• Accounting changes
• financial analysts
• forecast accuracy
• earnings predictability

Acknowledgements

The author gratefully acknowledges the contribution of I/B/E/S International Inc. for providing earnings per share forecast data, available through the Institutional Brokers Estimate System. These data have been provided as part of a broad academic programme to encourage earnings expectations research. He appreciates helpful comments and suggestions from Ivo Blij, Thomas Buchman, Willem Buijink, Kees Camfferman, Marc De Ceuster, Henri Dekker, Jan Klaassen, Frederick Lindahl, Frank Moers, Stephen Ryan, Ruud Vergoossen and accounting seminar participants at Maastricht University and the Vrije Universiteit Amsterdam.

Notes

¹In this paper, when I use the term mechanical model, I refer to those earnings prediction models that have most prominently been studied in the accounting literature, i.e. time-series-based models (see, e.g., Brown et al., 1987a), share-price-based models (see, e.g., Elgers and Murray, 1992), and financial-statement-based models (see, e.g., Stober, 1992). In the empirical analysis, I focus on share-price-based earnings prediction models.

²In this paper, analysts' data fixation means that analysts fail to recognize that accounting numbers that are prepared under different accounting systems have different economic meanings.

³Two distinctive characteristics of the Dutch requirements concerning the adoption of accounting changes may constitute an important determinant of the influence of accounting changes on analysts' forecast accuracy. First, Dutch firms are allowed to present the effect of a change in accounting principles on the financial figures of the preceding year instead of the financial figures of the current year. Second, contrary to disclosure requirements in the United States, Dutch firms are not obliged to restate figures for prior years in the historical summary.

⁴My study focuses on earnings forecasts of which the forecast horizons are less than one year (short-term forecasts). It is therefore reasonable to assume that these forecasts incorporate transitory earnings components, such as the earnings effect of accounting changes.

⁵A reason why accounting changes are relatively costly tools for earnings management is that an accounting change not only affects present figures but also has a fairly unpredictable effect on future accounting earnings. For instance, the effect of a change from current cost accounting to historical cost accounting (or vice versa) on future earnings crucially depends on future movements in market prices. Further, the adopted accounting change should probably be widely accepted, or it would receive too much attention to be effective in manipulating earnings.

⁶The explanations for accounting changes presented in this study, i.e. improving the financial statements and earnings management, are not mutually exclusive. For instance, an accounting change can improve the informativeness of financial statements in the years after the change, but at the same time be used opportunistically by adopting the change in a year in which the earnings effect of the change increases management's personal wealth.

⁷Likewise, Baldwin (1984) and Swaminathan (1991) show that after mandated segment disclosures analysts' forecast dispersion decreases and analysts' forecast accuracy increases.

⁸I realize that accounting changes can be disclosed in reports other than prior annual reports. However, excluding interim disclosure of changes ensures that all individual one-year-horizon earnings forecasts can be assumed to be affected by prior disclosure. Using only this type of disclosure may understate the impact of prior disclosure, i.e. reduce the significance of the empirical results.

⁹The 5% rule is a materiality criterion that is often used in Dutch guidelines for financial reporting. I classify accounting changes with a material change-year effect on earnings before extraordinary items and a material change-year effect or cumulative effect on net profit as material changes.

¹⁰A third large category of changes consists of changes in principles for revenue recognition. However, a large proportion of these changes stems from an amendment to the Dutch Investment Incentives Act (WIR), which induced firms to instantly recognize formerly deferred investment grants or accelerate amortization of these investment grants. Since these changes are part of a once-only, country-specific event, I do not separately examine changes in principles for revenue recognition.

¹¹Philips' accounting change from current cost accounting to historical cost accounting is included in this study's sample.

¹²In addition to the information advantage, analysts also have a timing advantage over time-series prediction models (see Brown et al., 1987b). The timing advantage exists since analysts are able to make use of information published after the initiation date of time-series models.

¹³Due to limitations of the dataset individual changes in forecast errors are calculated using two forecasts of one broker, which in theory does not necessarily coincide with two forecasts of one analyst. In my dataset analyst codes are available for 73.6% of the observations. Examinations of this subset reveal that 62.1% of all individual one-year changes in forecast errors, 36.3% of all individual two-year changes in forecast errors, and 20.3% of all individual three-year changes in forecast errors stem from the same analyst. However, analyst turnover is not significantly different between the sample of firms adopting accounting changes and the sample of non-changing firms and is therefore not likely to systematically affect the results.

¹⁴The assumption underlying the use of decile-ranked changes in stock return volatility as a proxy for changes in information asymmetry is that this variable especially filters out the effect of extreme changes in information asymmetry on changes in forecast accuracy. This is a desirable property of the proxy, since I do not wish to remove the impact of less extreme variations in information asymmetry. The reason for this is that these less extreme variations in information asymmetry can be caused by accounting changes, which are the subject of my investigation.

¹⁵My motivation for using a fixed-effects model is to avoid assuming that the relationship between the determinants of forecast accuracy and forecast errors is linear.

¹⁶For instance, the Shapiro–Wilk statistic of abnormal changes in forecast errors from the prior year to the change year equals 0.72 (p < 0.01).

¹⁷The Wilcoxon tests require that the differences in forecast errors be symmetrically distributed. Descriptive statistics indicate that in some instances abnormal changes in forecast errors are skewed (see Table 1). However, I recalculated the Wilcoxon test statistics when using a bootstrapping method. This indicated that the bootstrap means are not significantly different from the reported test statistics. Hence, I base all z-statistics of Wilcoxon tests presented in this paper on the rank sums and signed ranks that are obtained by using the original sample.

¹⁸To test the effect of the accounting changes on the forecast accuracy of the share-price-based earnings prediction model, I have also compared the model's forecast errors after the accounting change with the errors before the change. This test indicates that none of the three types of changes (CCA to HCA, expensing to capitalization and other changes) affect the forecast accuracy of the share-price-based prediction model significantly in the first or second year after the changes.

¹⁹Individual forecast errors for different firms by one particular analyst (or broker) may be correlated, since forecasting ability may vary among analysts. This type of correlation is likely to be reduced when analysing first differences of analysts' forecast errors, as I do in my study.

²⁰This finding is explained by the fact that in my sample, the average cross-product of the error terms (within firm-years) from equation (4) is small and negative, suggesting that abnormal changes in analysts' forecast errors are modestly negatively correlated.

²¹The GMM t-statistic on the indicator variable for prior disclosed material changes in the change year weakly suggests that prior disclosed material accounting changes can also negatively affect analysts' forecast accuracy (t = 1.71, p < 0.10). However, when using the generalized method of moments, the influence of prior disclosed changes on analysts' forecast errors remains significantly smaller (p < 0.10) than the influence of not prior disclosed material changes. This finding confirms my earlier conclusion that prior disclosure mitigates the influence of material accounting changes on forecast accuracy.

²²Two factors can explain the difference in effect of material and immaterial accounting changes. First, since manipulation towards analysts' expectations likely requires smaller amounts of manipulation, the greater size of the earnings effect of material accounting changes makes these accounting changes less useful for this particular type of earnings management. Second, Dutch disclosure requirements are less strict for immaterial accounting changes than for material accounting changes. For instance, it is not required to disclose the size of the effect, or to report pro forma adjustments of prior year's figures when the accounting change is immaterial. This makes it easier and less visible to misuse immaterial changes for the purpose of earnings management.