«Leif Atle Beisland University of Agder Dissertation submitted to the Department of Accounting, Auditing and Law at the Norwegian School of Economics ...»
Lev and Zarowin (1999) find that the value relevance of financial reporting has been deteriorating over time. They attribute the fall to increased investments in intangible assets and the inability of current reporting regimes to account for such investments; see also Collins, Maydew and Weiss (1997) and Francis and Schipper (1999). This conclusion is also supported by e.g. Lev and Sougiannis (1996) and Aboody and Lev (1998), focusing on the value relevance of expensing versus capitalizing research and development expenditures, and e.g. Jennings, Robinson, Thompson and Duvall (1996) and Henning, Lewis and Shaw (2000), focusing on the value relevance of goodwill recognition. For example, Lev and Sougiannis (1996) analyze price and return regressions in which the misstatement of earnings and book value due to absent capitalization of research and development expenditures enter as explanatory variables.
The coefficients of both the misstatements are positive, but the balance sheet effect seems to be more significant than the earnings effect. Thus, a substantial number of empirical studies find that increased recognition of intangible assets in the balance sheet leads to increased value relevance of book value and earnings, though the earnings evidence is more mixed.
Prior research presents evidence that fair value accounting increases the value relevance of the balance sheet – though some mixed evidence exists; see e.g. Barth (1994), Petroni and Wahlen (1995), Barth, Beaver and Landsman (1996), Nelson (1996), Eccher, Ramesh and Thiagarajan (1996), Venkatachalam (1996), Park, Park and Ro (1999), Carroll, Linsmeier and Petroni (2003), Beaver and Venkatachalam (2003), Khurana and Kim (2003), Nissim (2003)
(1996) study fair value reporting for banks and find that fair value estimates of financial instruments such as loans, securities and long term debt provide significant explanatory power for bank share prices beyond that provided by transactional cost values. Khurana and Kim (2003) provide evidence that the relationship between stock market value and fair value is larger than the relationship between market value and transactional cost value for availablefor-sale securities, but not for loans and deposits since these assets are not actively traded and hence contain more measurement errors. Hann, Heflin and Subramanayam (2007) find that the regression coefficient of the balance sheet increases if net pension obligations are reported at fair and not smoothed value. However, in their price regression with both earnings and book value, the coefficient of the book value increases only from 0.70 when smoothed to 0.74 at fair value, which is not significant.
While the evidence regarding the information content of fair values in the balance sheet is generally positive as long as fair values are not unreliable due to illiquid valuation or managerial manipulation, evidence on the value relevance effects of periodic unrealized gains and losses in the income statement is more mixed. Barth (1994) finds that unrealized gains and losses do not explain stock returns; see Ahmed and Takeda (1995) and Park, Park and Ro (1999) for opposite results. Hann, Heflin and Subramanayam (2007) analyze fair value pension accounting and find that fair values impair the value relevance of the income statement.
In their price regression with both earnings and book value, the coefficient of earnings decreases from 3.92 when smoothed to 3.14 at fair value. The difference of 0.77 is highly sigSome evidence indicates that fair values might be manipulated, for example for banks in financial distress.
Manipulation of the recognized value might lead to transactional costs becoming the most value relevant principle of measurement; see Barth, Beaver and Landsman (1994), Beaver and Venkatachalam (2003) and Nissim (2003). The lacking increase in the response coefficient of book values due to the implementation of IFRS found by Capkun, Cazavan-Jeny, Jeanjean and Weiss (2008) and Horton and Serafeim (2008) may have similar explanations.
(2004), who studies the effect of reporting real estate investments at fair value relative to reporting them at historical cost. They conclude, like Hann et al., that as earnings become more transitory, they lose value relevance.8 If the value relevance of the balance sheet increases and the value relevance of net earnings decreases when shifting from transactional cost to fair value, the effect on the combined value relevance is ambiguous. However, Hann, Heflin and Subramanayam (2007) analyze the effect on adjusted R2 and report 57.3% when pension gains and losses are smoothed and 55.1% when they are reported at fair value. The fall in adjusted R2 of 2.3 percentage point is significant by the Vuong test (p-value = 0.000).
2.3 Hypotheses to be Tested We have concluded in subsection 2.1 that the two major differences between IFRS and NGAAP are more recognition of intangible assets and more measurement at fair values, but several minor differences exist. In subsection 2.2, the empirical evidence suggests that more recognition of intangibles as well as measurement at fair value tend to increase the value relevance of the balance sheet, i.e. equity. As for the value relevance of earnings, more recognition of intangible assets is found to have a positive effect. However, increased measurement at fair value may be detrimental to the value relevance of earnings relative to measurement at
transactional cost. These empirical findings are the basis for specifying our test hypotheses:
Hypothesis 1: The response coefficients of stock prices to equity book values are different under IFRS than under NGAAP.
Bradshaw and Sloan (2002) document that so called “Street” earnings have gained increased focus and value relevance, where earnings used by “The Street”, i.e. financial analysts, are typically normalized by removing onetime items.
than under NGAAP.
If the effect of more recognition of especially intangible assets is dominating other effects, we expect, based on previous empirical findings, both the balance sheet and the income statement under IFRS to be more relevant than under NGAAP. If the effect of more fair value is dominating other effects, we expect, again based on previous empirical findings, that the balance sheet becomes more value relevant and earnings less value relevant under IFRS than under NGAAP.9 Since the main effects predict increased value relevance of the balance sheet, Hypothesis 1 could be specified one-sided. However, since there also are some minor differences between IFRS and NGAAP identified in subsection 2.1, we choose to specify Hypothesis 1 two-sided and accordingly apply two-sided tests in the empirical analysis.
A possible explanation for the expected increase in the response coefficient of the balance sheet due to increased recognition and measurement at fair value could be based on an informational perspective, in which the book value of equity is considered a signal about ‘underlying value’. The response coefficient of the book value of equity will increase if it becomes more correlated with the underlying value and if the precision of the book value as an informational signal about value increases. When reporting according to fair value with full recognition, the coefficient of the balance sheet will be one. Current earnings are contained in the book value at the end of the year – and will therefore indirectly have a response coefficient equal to one. When measured at transactional historic cost, current earnings are also a potential signal of future earnings and thereby an additional signal of current value compared to the Nissim and Penman (2008) explicitly state, when theoretically comparing fair value with historical cost, that “…while fair value accounting reports a balance sheet that is informative about value, it renders an income value that is uninformative about that value” (Nissim and Penman, 2008, p. i).
thereby current value, the earnings response coefficient will increase beyond one.10 But the fact that there are two signals, book value and earnings, means that they are also weighted as informational contributors relative to their correlation with value and precision as signals. As the weight of earnings increases, the weight of the book value tends to fall below one; see Penman (1998) and Zhang (2000).11 This view coincides with response coefficients being projections of the dependent variable, i.e. the stock price, into the information space of accounting information represented by book value and earnings per share; see e.g. Green (2008, pp. 25-26).
Prior research has also provided us with an important insight which explains why earnings could be an inferior signal of underlying value under fair value accounting; see e.g. Beaver (1998, pp. 72-76) or Kothari (2001, pp. 123-124). If current earnings are becoming more transitory, containing relatively more non-recurring items, the relationship between earnings and the stock price will be reduced. Thus, non-recurring gains and losses in earnings due to fair value revaluation could decrease the value relevance of earnings, as measured by the earnings “…if earnings’ time-series properties are such that earnings innovations are permanent, then assuming a oneto-one relation between earnings innovations and net cash flow innovations, the earnings response coefficient is the present value of the perpetuity of the earnings innovation calculated by discounting the perpetuity at the risk adjusted rate of return on equity. The present value of a $1 permanent innovation in annual earnings is (1+1/r) where r is the annual risk-adjusted discount rate for equity” (Kothari, 2001, p. 123-124). In contrast, under fair value accounting (Nissim and Penman, 2008, p. 13) ”…earnings are uninformative about future earnings and about value; earnings are changes in value and as such do not predict future value changes, nor do they inform about value (value follows a random walk, as it is said).” Zhang (2000, Proposition 2) shows that conservative accounting methods on average yield a weight on the book value less than zero and a weight on earnings, capitalized with the normal earnings multiplier, greater than one. Thus, more recognition of intangible assets and measurement at fair values tend to make accounting less conservative or more unbiased, which will increase the weight on the balance sheet toward one and decrease the weight on earnings toward one. In empirical studies using price regressions with book values and earnings, there are intercept terms and no restrictions on the sum of the coefficients, meaning that the measured coefficients of book value and earnings could deviate somewhat from the ones predicted by the models of Penman (1998) and Zhang (2000). In relation to our data, the valuation weight of earnings, capitalized with the estimated normal earnings multiplier, is estimated at -0.136 in the IFRS sample and at 0.325 in the NGAAP sample, suggesting weights of the balance sheet of 1.136 and 0.675, respectively. Pooled, the earnings weight is estimated at 0.235 with a normal earnings multiplier of 9.819. Thus, the earnings response coefficient is 2.037 and the coefficient of the balance sheet is 0.765. In comparison, Penman (1998) estimates the median weight in his U.S. sample to be 0.420, suggesting an earnings response coefficient of approximately 4.620 and a book response of 0.580.
about current performance but also about future performance and therefore about the current stock price, which depends on the book value of equity, including current earnings, and future abnormal earnings. If the link between current earnings and future abnormal earnings is reduced, e.g. by more fair value accounting, the informational value of current earnings is reduced and so is the earnings response coefficient; see Penman (2007). The balance sheet is, however, weighted upward. When reporting according to fair value with full recognition, the earnings response coefficient is reduced to one, as current earnings are contained in the book value and there is no correlation with future earnings. The response coefficient of the balance sheet will also be one.
2.4 Test Methodology We now discuss how the two outlined hypotheses are to be tested empirically. Hypothesis 1
could be tested by the price regression:
where PRICE is the stock price for firm i at time t, BOOK is the book value of equity per share, and IFRS is a dummy or indicator variable which equals 1 if firm i reports according to IFRS during period t, or equals 0 if the firm reports according to NGAAP during the same period. Finally, ε is the error term of the regression and α0, α1, α2 and α4 are the regression coefficients.12 The constant term α0 might be replaced by a constant term α0 · IND for each industry so to adjust for fixed industry effects, where IND is a vector of dummy variables for each industry. Of course, the model could also be extended to adjust for other fixed effects, but not fixed time effects because the shift from NGAAP to IFRS is to a large extent time dependent.
The book response coefficient BRC equals the ‘core’ coefficient α1, in our case the coefficient of NGAAP, plus a term, α4 · IFRS, depending on or moderated by the reporting regime IFRS.13 Thus, the indicator variable IFRS has a significant impact on the balance sheet’s net response coefficient BRC if α4 is significantly different from zero.14 We can restate our Hypothesis 1 as α4 ≠ 0, meaning that the BRC is moderated by IFRS. We expect that the BRC is larger under IFRS than under NGAAP – but the opposite cannot be ruled out; see subsection
2.3. This simple way of testing for structural breaks in the coefficient structure is standard and equivalent to the test approach originated from Chow (1960). The test with interaction effects is more flexible than simple Chow test statistics, e.g. in correcting for possible heteroskedasticity between the two samples.