Changes in the measurement of fair value: Implications for accounting earnings

Abstract

With the FASB's issue of staff position papers in 2009 and the relaxation of how fair value standards are applied, there has been a change in the practice of how fair value is measured. Since the FASB staff position papers in 2009, fair value measurement by financial institutions has increasingly relied on managerial assumptions. This study examines the impact of this change on the quality of earnings. Consistent with attribute substitution theory that emphasises reliability over relevance, we find that an apparent increase in managerial discretion in fair value measurement is associated with a higher probability of earnings management and lower earnings informativeness. The results indicate that allowing more managerial discretion in fair value measurement adversely affected the quality of financial reporting. Our study highlights the issue of reliable measurement in the debate among academics and practitioners of increasing the use of fair value accounting.

Keywords:

• Fair value measurement
• Earnings quality
• Reliability

Notes

1 See the debate between Dechow, Myers, et al. (2010) and Barth and Taylor (2010) on whether fair value estimates of securitisation gains are manipulated by managers. Barth and Taylor (2010) show that the evidence on this issue is inconclusive and further investigation is needed.

2 A Wall Street Journal article estimates that the change in fair value accounting boosted banks’ earnings by 7% on average in Q2 2009 (Pulliam & McGinty, 2009).

3 For example, the CEO of Goldman Sachs, Lloyd Blankfein, wrote in the Financial Times: “At Goldman Sachs, we calculate the fair value of our positions every day, because we would not know how to assess or manage risk if market prices were not reflected on our books. This approach provides an essential early warning system that is critical for risk managers and regulators” (‘To avoid crises, we need more transparency’, FT.com, October 2009).

4 A representative example of an acknowledgement by management of the flexibility in measurement afforded by the choice of assumptions is: “The methods to estimate fair value may produce a fair value calculation that may not be indicative of net realisable value or reflective of future fair values.” JP Morgan Chase 2010 10-K, page 157.

5 In the case of level 2 estimates, managers were given the discretion to adjust the prices indicated by market indices after the relaxation of fair value rules. In the case of level 3 estimates, managers are given more discretion to determine the extent of market illiquidity and hence the extent as well as the magnitude of internal model inputs.

6 It is assumed that there is greater management flexibility when valuing level 2 and 3 securities relative to level 1 where an observable market price is typically available. While this assumption has been used in prior research it must be acknowledged that the types of securities held also vary between levels 1, 2 and 3 and, as with previous research, the results must be interpreted with respect to this potential confounding of type of securities held and the method for estimating the discretionary component of the fair value of the securities held.

7 For example, level 3 fair value assets of some bank holding companies, such as Merrill Lynch, increased by as much as 70 percent compared with the pre-crisis balance. In this way, banks were able to limit the negative effect of declines in fair value on net income or owners’ equity (Laux & Leuz, 2009). Marking-to-model at level 3 allows declines in current value to be considered to be temporary and helps avoid recognising losses.

8 The tests in Table 1 include all bank quarter observations with non-missing data for disclosures of fair value assets (Level23), level 1 fair value assets (Level1), net income before extraordinary items (Inc), natural log of total assets (Size), market value of equity (Mve), total liabilities to total assets (Lev), and tangible common equity (Tce) from Bank Compustat. The variables Level23, Level1, Inc and Tce are deflated by beginning-of-period total assets.

9 Most banks experienced negative earnings growth in the sample period due to the impact of the financial crisis. Matsunaga and Park (2001) report 8.3 per cent of companies meeting or beating forecasts for firms with negative growth in earnings.

10 We also tested the correlation between the levels of these two variables. Specifically, we tested the correlation between the amount in cents of beating or missing analyst forecasts (AEM) and the amount of discretionary fair value assets of a bank (DFV). The correlation is at 0.014 and insignificant.

11 The higher value of Spearman correlation between ULLP and POST than the Pearson correlation indicates the correlations between ULLP and POST can be better described by a monotonic function using rank scales than parametric values.

12 The VIF values for all the variables used in Eq. (1)(1) $E{M}_q={\gamma }_0+{\gamma }_{1}POST+{\gamma }_2\Delta DF{V}_q+{\gamma }_3\Delta DF{V}_q*POST+{\gamma }_{4}ULL{P}_q+{\gamma }_5\Delta CF{O}_q+{\gamma }_{6}SIZ{E}_{q-1}+{\gamma }_{7}B{M}_{q-1}+{\gamma }_{8}DISPER{S}_q+{\gamma }_{9}NUMES{T}_q+{\gamma }_{10}CHGIN{C}_q+{\gamma }_{11}PERSIS{T}_q+{\gamma }_{12}CR{1}_q+{\gamma }_{13}C{R}_q+firmfixedeffect+yearfixedeffect+{\epsilon }_q$(1) are below 3.5, indicating that multi-collinearity is not a concern in this model.

13 Note that the fixed effect Logit model includes fewer observations. The firms which do not demonstrate enough time variation in the dependent variable are dropped in the statistical calculations when estimating the coefficients for the fixed effect Logit models.

14 We also ran the ERC regression separately for each year from 2007 to 2010. The coefficient on the change in discretionary fair value assets (DFV) is negative and significant in 2008 and 2010. The coefficient on the interaction of DFV and Earnings is negative and significant in 2009 and 2010.