Some Informational Aspects of Conservatism

Abstract

When two value estimates are about equally likely, conservatism dictates reporting the less optimistic one (e.g. Lower of Cost or Market). We use an analytical model to investigate informational implications of this dictum, and identify types of environments where the conservative accounting treatment is more informative than a predetermined choice. The bias induced by the conservative choice is found to be adequately moderate, never excessive. It benefits users of the financial statements that take the reported figures at face value whenever upside errors are more costly (possibly only slightly more costly) than similar downside errors. Sophisticated users, who know how to give the reports the best possible interpretation, benefit from the lower variability, not from the bias. These latter benefits are least ambiguous when upside errors and downside errors are about equally costly.

Acknowledgements

The authors acknowledge useful comments on earlier versions of this paper from Eli Amir, Michael Bromwich, Eti Einhorn, Haim Falk, Nils Hakansson, Baruch Lev, Ulf Schiller, Jeroen Suijs, Alfred Wagenhofer, and from participants of the Accounting and Economics Workshop (Frankfurt, 2004) and participants in the London Business School seminar. The most helpful remarks of James Ohlson and Laurence van Lent (Joint Guest Editors) and two anonymous referees are especially appreciated. As always, sole responsibility is ours.

Notes

¹The Framework for the Preparation and Presentation of Financial Statements adopted by IASB in 2001 requires reliable information contained in financial statements to be free from bias (p. 36).

²Basu (1997) and Watts (2003a,b) give extensive reviews.

³Fischer and Verrecchia (2000), Gigler and Hemmer (2001), Gjesdal and Antle (2001), Kwon et al. (2001) and Christensen and Demski (2003) model various aspects of conservatism using a principal–agent framework. Zhang (2000) examines links between accounting data and firm value under conservatism.

⁴See, for example, Devine (1963), Levitt (1998) and Watts (2003a).

⁵An early implicit treatment of error losses in accounting appears in Ijiri and Jaedicke (1966). Later and more explicit examples are Scott (1975) and Antle and Lambert (1988).

⁶Sophisticated users have superior abilities (Beaver, 1998) and can learn better from experience (Bonner et al., 2003).

⁷In their early study of accounting reliability, Ijiri and Jaedicke (1966) relate to users who interpret the data according to some idea about the relationships between the reported value and the underlying value (in which they are interested). Such users can be regarded as ‘sophisticated users’. Ijiri and Jaedicke's concept of ‘alleged value’ relates to the value that, if reported, would give rise to an interpretation that coincides with the true value.

⁸We believe that similar results hold if the two estimation errors are not independent, but only less than perfectly correlated.

⁹Under this latter assumption, adjusting the reported value by a constant (upward or downward) shift D, independent of the reported value r, is indeed an optimal interpretation of the report. When e _i is not independent of V, the optimal interpretation may involve adjustments that depend on the reported value.