International Differences in Conditional Conservatism – The Role of Unconditional Conservatism and Income Smoothing

Abstract

Prior research documents that conditional conservatism, measured as the asymmetric timeliness of earnings reflecting bad vs. good news, varies with cross-country differences in institutional regimes. In this paper, we examine the determinants of conditional conservatism and related earnings attributes internationally. First, using panel data, we investigate whether competing earnings attributes such as unconditional conservatism and income smoothing affect conditional conservatism and its international differences. We find that these attributes are predictably correlated with conditional conservatism. Second, we address the question whether income smoothing and conditional conservatism are two fundamentally different earnings attributes. We show theoretically that both attributes yield different earnings distributions and that the motivations for producing earnings which possess these attributes differ. To test these predictions empirically, we calculate firm-specific time-series measures of asymmetric timeliness, using a novel trigonometric measure based on the standard Basu (1997)-type regression. Using this cross-sectional data, we test whether conditional conservatism and income smoothing are different and find them to be only weakly correlated for a broad international sample. Also, we demonstrate that income smoothing explains international differences in conditional conservatism. Finally, we estimate simple determinant models of conditional conservatism and income smoothing, showing that both earnings attributes are driven by different explanatory firm-level factors: Conditional conservatism increases with the importance of debt financing, while income smoothing increases with the importance of dividends. Despite some important limitations, we believe our results to be meaningful because they show that cross-country differences in conditional conservatism are influenced by the effects of other accounting properties, predominantly income smoothing. Especially, legal regime appears to drive income smoothing while losing its explanatory power for conditional conservatism when firm-specific factors are controlled for.

Acknowledgement

We thank two anonymous referees, Laurence van Lent and Jim Ohlson (the editors), Juan Manuel García Lara, Ryan LaFond, Ernst Maug, Bill Rees, seminar participants at the Humboldt-Universität zu Berlin, Ruhr-Universität Bochum, Westfälische Wilhelms-Universität Münster, Universität Paderborn, Universität Zürich, and delegates at the EAA annual congress in Gothenburg, Sweden, for providing helpful comments.

Notes

1. For a variety of institutional variables that influence conditional conservatism, see, for example, Bushman and Piotroski (2006).

2. Consistent with much of the literature, we use the terms conditional conservatism and asymmetric timeliness interchangeably throughout this paper.

3. For an overview of related empirical work, refer to Watts (2003b).

4. Our focus in this paper is on inter-temporal, not classificatory income smoothing. The latter is discussed in, for example, Ronen and Sadan (1975).

5. Theoretical articles such as Lambert (1984), Trueman and Titman (1988) and Arya et al. (1998) have emphasized the benefits of income smoothing from the perspective of shareholders.

6. More literature on cross-country differences in smoothing incentives is reviewed by García Lara et al. (2005b, pp. 6–8).

7. In the words of Pope and Walker (2003, p. l2), this ‘distinction [between conditional and unconditional conservatism] could prove helpful in explaining why some GAAP regimes, such as Germany, which have historically enjoyed a reputation for conservatism, appear to score ‘badly’ according to the Basu-type measures of conservatism. See also Ball and Shivakumar (2005, pp. 89–90), who associate the German ‘vorsicht (prudence) principle’ with the unconditional notion of conservatism. Consistent with this intuition, García Lara and Mora (2004) document that, while conditional conservatism tends to be more pronounced in the UK, unconditional conservatism is higher in code-law countries.

8. The following analysis covers only one year. Biased recognition of economic events in one year will cause distortions in the following years. Bias introduced by conservatism can be expected to shift the earnings distribution to the right in the following years. Upward-biased smoothing can be interpreted as smoothing combined with unconditionally agressive accounting behavior, which will shift future earnings distributions to the left. Downward-biased smoothing can be interpreted as smoothing combined with unconditionally conservative behavior, and will shift future earnings distributions to the right. Neutral smoothing will leave future earnings distributions unaffected.

9. As an alternative specification of earnings, we use net income, finding almost identical results.

10. As a robustness check, we use 15-month returns and 12-month returns, both ending three months after fiscal year end, again finding almost identical results.

11. Apart from Basu's (1997) own paper, the ‘reverse regression’ approach to measuring the asymmetric timeliness of losses vs. gains recognition is explained in detail by Dietrich et al. (2005), who question the validity of the ‘reverse regression’ model to measure conservatism, raising a number of econometric issues. In this paper, we assume that the model is valid.

12. Since model (1) has a coefficient for the interaction term NEG (β₁), there may be an offset between the two regression lines at the ordinate. This effect is not visualized in Figure 2. It does not affect the validity of our BASU metric. Although the inclusion of β₁ in model (1) is motivated normally only for mechanical reasons: When β₁, is significant, it can be regarded as an indicator of some sort of non-linear asymmetric reaction of earnings to the nature of news.

13. An alternative, non-market-related approach to measuring conditional conservatism has been introduced by Ball and Shivakumar (2005), who developed a returns-independent, accruals-based metric of asymmetrically timely loss recognition. Although this metric is undistorted by international differences in market efficiency, we refrain from using it because those differences are not the focus of our paper.

14. For more references, see Penman (1996, pp. 235–236).

15. In this respect, we are in line with Roychowdhury and Watts (2006, p. 15), who exclude rents from their concept of conservatism. We refrain from using the price–earnings ratio as a proxy for conservatism (e.g. Givoly and Hayn, 2002, pp. 68–71) due to its analogy to the MTB measure (e.g. Fama and French, 1992; Penman, 1996).

16. Literature provides other alternatives to MTB not applied in this paper, for example, the ‘C-score’ metric (of balance sheet-oriented conservatism) introduced by Penman and Zhang (2002) and Easton and Pae's (2004) new cash investments and book values of lagged operating assets (and changes herein). Roychowdhury and Watts (2006, p. 8) point out that the Basu metric ‘is a better measure of conservatism (total understatement of net assets) when estimated cumulatively over several periods.’

17. Another, similar metric used by Leuz et al. (2003, p. 510) is the contemporaneous correlation between changes in accruals and changes in cash flow from operations.

18. As a robustness test, we also prepare within-country analyses (not tabulated) and shorten our time period to the pre-1998 period to exclude the growing influence of accounting internationalization. Also, we delete as a robustness test all firm-years for which Worldscope indicates that non-local/non-EU accounting standards are followed. These sample modifications yield qualitatively very similar results.

19. Since we use the negative of this ratio, higher levels of SMOOTH represent smoother earnings.

20. All these differences are significant at conventional levels (t-test, Wilcoxon).

21. It is interesting to note that, for the common-law sample, β₅ is significantly positive, indicating that smoothing appears to increase the timeliness of good news recognition in earnings.

22. These disadvantages have also been discussed in a recent paper by Penalva et al. (2006), who conclude that Francis et al.'s (2004) measure does not capture conditional conservatism and propose a so-called ‘portfolio approach’ similar to that used in Section 3. We try to address the most prominent problems of firm-specific time-series measures of conservatism, documenting that the modified method yields results consistent with the portfolio approach. Thus, we see both methods as complementary.

23. The difference of SK_NI_TA between legal regimes is only marginally significant (0.12, t-test, two-sided). All other reported differences are significant at conventional levels.

24. Recall that SK_NI_TA is negative when earnings are left-skewed, so negative SK_NI_TA represents high conditional conservatism, while the opposite is true for the BASU metric.

25. Recall that higher values of TCTOTACC represent lower unconditional conservatism.

26. A different approach to modeling the determinants of income smoothing is to investigate its role as an information device. See Goel and Thakor (2003) for a literature review and a recent model. Since the literature on conditional conservatism explicitly or implicitly assumes that earnings carry no information, we have to assume that earnings in general are uninformative to the capital market in order to consistently model determinants for our two earnings attributes. Whether this assumption is valid is outside the focus of our study. Thus, the only role which earnings can fulfill in our theoretical framework is that of a contracting device.

27. Dechow and Dichev (2002) use the frequency of negative earnings as the proxy for negative shocks in the firm's operating environment. We use cash flows instead because the time series of earnings is mechanically related to our dependent variables. In addition, we do not include the accruals-related variables of the Dechow and Dichev model since they are also mechanically related to our dependent variables and are specific to the purpose of the original model, namely, to assess accrual quality.

28. All reported differences are significant at conventional levels.

29. We exclude SD_CFO from the models to verify that this mis-specification does not influence our results. We also include other controls (a dummy indicating whether dividends are paid, market capitalization instead of total assets and industry dummies) without qualitative changes.

30. Gu and Zhao (2006) show that firms with smoother earnings experience higher bond rating on average, lending some support to the assumption that public debt holders demand smooth earning streams.

31. We replicate our research design with proxies for different dimensions of the institutional infrastructure as used by Bushman and Piotroski (2006) (not tabulated). For most of these proxies, the connection to income smoothing seems much more robust than the connection to conditional conservatism when we control for firm-specific factors.

32. We investigate this point by incorporating the standard deviation of returns as a risk proxy in our model (7.2). We find (not tabulated) that, while risk measured by the standard deviation of returns is positively related to conditional conservatism, it does not render the negative relation between MTB and conditional conservatism insignificant. Since we are not aware of an undisputed proxy for risk, this (and probably no other) test can rule out this competing explanation for our findings presented above.