(i) Types of Overrides
All the override disclosures identified in our search were carefully read and analyzed and then classified into nine major types of overrides according to the underlying accounting treatment affected by the override. We then further assign each of the override types to one of the four cost categories discussed above. Table 1 presents these override types, classified by cost category, in descending order of frequency. It also shows how many overrides were quantified in each cost category. The Appendix presents further details regarding the expected effect of each override type on the financial statements. As can be seen from Table 1, Category 1 overrides comprise the majority of the sample, and the rate of occurrence is substantially lower for Category 4 than for Category 1. Also, the total number of overrides under Categories 2 and 3 (161 observations) is lower than that under Category 1, but higher than that for Category 4. The most common override type is non-depreciation of investment properties, which is required by GAAP but disallowed by the Companies Act. Hence this type is a Category 1 override. The second most frequent override type is non-amortization of goodwill. While UK GAAP allows for non-amortization, the relevant standard (FRS 10, ASB, 1997) does not favor it and, furthermore, requires annual impairment reviews that have been perceived as costly.14 As non-amortization of goodwill also stands in contrast to the CA, it is classified as Category 2.
As Table 1 indicates, seemingly similar override types may relate to more than one category. This is because each case is individually analyzed to determine the rule that was departed from. Thus, for example, in the case of piecemeal calculation of goodwill, two cases were found to violate GAAP while the other 32 cases were consistent with GAAP, but inconsistent with the CA.
The findings in Table 1 show that Category 1 overrides are the most common, and Category 4 overrides (outright departures from UK GAAP) are much less common. However, Category 4 includes 133 firm-year observations invoked by 68 firms in our 5-year sample period, suggesting that overriding GAAP is a non-negligible phenomenon.
(ii) Firm Performance by Type of Override
In this section, we describe several analyses that provide evidence on whether firms that invoke TFV overrides are doing so to report more favorable performance (hypothesis 1). If firms invoke overrides solely because they want to increase their reported income, or avoid violation of debt covenants, we would expect to see significant differences in pre-override performance or debt-related financial indicators between override and control sample firms. On the other hand, if firms invoke overrides to increase the informativeness of their financial statements, we would not expect TFV firms to exhibit weaker indicators before the effect of the override.
First, we examine the annual effect of overrides on income and the cumulative effect on equity for the subsample of firms that disclose these amounts. The Companies Act and UK GAAP require that firms invoking a TFV override disclose the particulars of any departure, the reasons for it and its effect, where the effect is quantified if possible. As reported in Table 1, only a small number of companies actually quantify the effect of their override. In some cases, this occurs because a firm would not be able to characterize what number it would have reported, as in the case of adopting merger (pooling) accounting instead of purchase accounting in acquisition of a subsidiary. In other cases, such as non-depreciation of investment properties, quantification would require that the firm calculate the depreciation amount it would have recorded, which is arbitrary if the firm views the asset as having an indefinite life.15 We expect that the median income and equity effects of an override are significantly positive if firms invoke overrides opportunistically. If overrides are not income and equity-increasing, on average, it seems less likely that they are invoked for opportunistic reasons. However, we note that because only a fraction of firms quantify their override effects, the median for the sample as a whole may differ from the median for the sample that disclosed the magnitude of the effects on income and equity. Furthermore, our tests that the median effect is positive are less powerful than they would be if based on a larger sample.
Second, we compare the characteristics of TFV firms to a control sample of industry and size-matched firms. We test whether firms that invoke overrides are less profitable and financially weaker than firms that do not. It should be noted that to the extent an override is successful in masking poor performance or financial position, we are less likely to find evidence supporting opportunistic behavior in measures based on net income. To mitigate potential effects of overrides on profits, assets and equity, we use two measures to assess underlying performance before the effect of an override: OPINC, the ratio of operating income before depreciation and amortization to sales, and CFOTOS, the ratio of cash from operations to sales. In addition, to assess the tightness of debt covenants before the effect of overrides, we use DETOFIX, debt to gross book value of tangible fixed assets, and INTCV, the interest coverage ratio.16 Note that if contracts regularly contain clauses to undo the effect of overrides, we would not expect to find an association between debt levels and the occurrence of overrides. Thus, finding such an association would suggest either contracting parties agree not to undo overrides or did not originally anticipate them.
To give a comprehensive overview of the results of these tests, we begin by presenting the findings for the entire sample of firm-years invoking any category of override. Table 2 presents descriptive statistics on several variables for both the TFV and the control samples.17 We calculate the Wilcoxon matched pair signed rank Z-statistic for the median difference in the firm-specific means of the TFV and control samples, and report the results under the Z heading. The advantage of this approach is that it is less susceptible to cross-sectional dependence in the TFV sample. It also mitigates the influence of outliers.
Comparison of Override, Descriptive, Performance, Contracting and Market-based Variables for the Full Sample of TFV Overrides 1998–2002 and Size and Industry-matched Control Sample, Excluding Real Estate Firms
|% EQUITY||44||0.059||0.012||0.137|| || || || || 2.31|
|% INCOME||47||0.096||0.023||0.182|| || || || ||0.90|
|BM4M||523||0.767||0.607||0.613||522||0.719||0.503|| 523|| 1.86|
Consistent with the theoretical considerations discussed earlier in the paper, we group the variables into five categories: override effects, descriptive measures, performance measures, debt contracting and market-related variables.18 The first category includes the override effects on equity and income, on a percentage basis, for the subsample of firm-years in which the amount of the effect was disclosed.19 The mean (median) increase in equity is 5.9% (1.2%) and the mean (median) effect on income is 9.6% (2.3%). However, only the median effect for equity is significantly positive with probability value less than 0.001. These findings thus provide some support for the notion that firms invoking overrides tend to adopt income- and equity-increasing accounting choices.
The second category includes descriptive measures, and indicates that TFV firms report SALES (total turnover), INCOME (net income), ASSETS (total assets) and EQUITY (total shareholders' funds) that are not significantly greater than those of the control group. Turning to performance measures, we observe that the net profit margin, NETPRO, is not significantly different for the TFV and control samples. However, TFV firms have significantly lower depreciation to sales, DTOS, which is not surprising given the large frequency of overrides that result in non-depreciation of investment properties. The difference in the two measures of pre-override profitability, CFOTOS and OPINC, are insignificant.
Under the hypothesis that TFV firms are motivated to override accounting rules to avoid violating a debt covenant, one might expect DETOFIX to be higher for TFV firms. We find that the median debt to gross fixed assets is indeed greater for TFV firms, suggesting that for the sample as a whole, concern over leverage ratios may have motivated them to override asset-reducing standards. Because we measure DETOFIX using gross fixed assets, this measure is not affected by non-depreciation overrides. Alternatively, INTCV likely reflects the effects of any income-increasing overrides and here we do not find a significant difference between TFV and control firms for the entire sample.
Market-based variables comprise the final category of variables in Table 2. The ratio of book to market value as measured four months after fiscal year-end, BM4M, is significantly greater for override firms. The ratio of earnings to price, again measured four months after fiscal year-end, EP4M, is not significantly different. These findings suggest that the market discounts the equity of TFV firms relative to that of control firms and therefore that investors at least partially adjust for the financial statement effects of the override.20
In summary, we do not find a significant difference between the pre-override profitability of TFV firms and those of a size- and industry-matched sample. This may be due, at least in part, to the fact that most of the overrides in Table 2 are mechanical. Nonetheless, we do find that TFV firms have higher debt to fixed assets, suggesting that a motive for override may be related to debt contracts.
It is important to recognize that the findings presented in Table 2 reflect the aggregation of diverse override types. We therefore examine three subsamples, with an aim to understanding whether differences in performance and debt contracting are observed where the overrides involve the greatest discretion and are most costly.
The first subsample we examine is that of firms invoking an override to avoid amortizing goodwill, which we also refer to as the nonamortization of goodwill (NOG) subsample. This override is conjectured to be costly because it is not the preferred method in FRS 10 and its implementation involves real costs to review for impairment. As such, it is a Category 2 override, implying that we expect differences in financial performance or position to be more pronounced than the overall sample. Put differently, firms overriding the principle of amortizing goodwill may be exercising greater discretion than firms who override the depreciation of investment property, as FRS 10 indicates that goodwill may have an indefinite useful life in exceptional cases.
The second subsample we examine is Category 4 overrides, firms that override UK GAAP, which we refer to as the true GAAP override sample. We expect these to be the most costly overrides and to involve the greatest discretion. The third subsample we examine, the S19 subsample, includes Category 1 overrides of the Companies Act requirement that all fixed assets of finite life be depreciated, in order to follow the UK GAAP (SSAP 19) requirement that investment assets be carried at fair value. Because the override is invoked to follow a more authoritative standard, we would not expect to see differences in performance, debt contracts or governance between firms invoking this override and the control sample.
The results for the non-amortization of goodwill subsample, reported in Table 3, indicate that firms quantifying the magnitude of their override significantly increase reported income and shareholders' equity by not amortizing goodwill. They also exhibit weaker performance than that of industry- and size-matched control sample firms. Specifically, OPINC, the ratio of operating income before depreciation and amortization to sales, and CFOTOS, the ratio of cash from operations to sales are significantly lower for non-amortizing TFV firms than for control firms. However, reported net income, net margin, return on equity and return on assets are not significantly lower, consistent with the income-increasing effect of the override. Second, these TFV firms have recognized significantly more goodwill than control firms, and due to the override, report lower amortization as a percent of sales than control firms. This is clearly reflected in the ratio of ending balance of goodwill and intangible assets to total assets, GWTOASS, and in new goodwill capitalized during the year, PGWTOASS.21 The picture that emerges for firms not amortizing goodwill is that they experience profitability before the override that is lower than the control sample, and which may further deteriorate in the future should they amortize the large amount of newly recognized goodwill. Third, these firms have a significantly lower interest coverage ratio. Bearing in mind the lower cash generation, this suggests that potential concern for violating debt covenants may have motivated the choice not to amortize goodwill. Finally, investors do not appear to discount the book value or earnings of these firms relative to the control sample, suggesting that investors view the reported numbers as comparable to those of firms that did not invoke an override.
Comparison of Override, Descriptive, Performance, Contracting and Market-based Variables for the Subsample of Non-amortization of Goodwill Overrides, 1998–2002, and Industry and Size-matched Control Firms
|% EQUITY||6||0.017||0.013||0.011|| || || || || 2.53|
|% INCOME||21||0.178||0.098||0.213|| || || || || 3.31|
Our third analysis focuses on overrides of GAAP (Category 4), which we hypothesize involve the highest cost. We expect to see significant differences in performance relative to control firms if the higher cost of these departures is offset by greater benefits from reporting the alternative financial statement amounts. As noted in Table 1, during 1998–2002 we find 133 firm-year observations of GAAP overrides. Given the considerable concern that the SEC and various commentators have expressed regarding granting managers the ability to depart from GAAP as well as the lack of sufficient evidence on such departures, we augmented our search to encompass the nine-year period of 1994–2002. This procedure gives us a larger sample and longer time series to examine factors associated with GAAP overrides: the resulting subsample has 203 firm-years.22
Table 4 provides the comparison that pertains to the augmented subsample of GAAP overrides. Due to limits on data availability, the number of observations for variables other than override effects varies from 91 to 151 depending on the specific variable. For the subset of firms that quantified the effect of their override, we find that the equity and income effects are not significantly different from zero. Firms overriding GAAP are significantly less profitable after the override than their respective control firms, as reflected in lower net margin and return on equity, though cash from operations to sales and operating income to sales are not significantly lower. Consistent with the lower level of reported profitability, interest coverage is also significantly lower. However, we do not find a higher level of debt to fixed assets for TFV firms. These findings provide support for the hypothesis that firms overriding GAAP are experiencing weaker performance or have a greater incentive to raise earnings for debt contracts.
Comparison of Override, Descriptive, Performance, Contracting and Market-based Variables for the Subsample of True GAAP Overrides, 1994–2002, and Industry and Size-matched Control Firms
|% EQUITY||18||0.038||0.019||0.066|| || || || ||1.36|
|% INCOME||8||−0.053||0.055||0.253|| || || || ||−0.92|
In our next subsample analysis, we examine overrides that can be regarded as ‘mechanical’ in that they represent an accounting treatment that is consistent with GAAP but is in contrast to the requirements of the Companies Act (i.e., Category 1). This subsample comprises non-depreciation of investment properties (by non real-estate firms) and is presented in Table 5. We are interested in this subsample as a benchmark to the analysis of the more costly overrides, since we do not expect to find significant differences for mechanical overrides. The median income effect of this override is significantly positive for the subset of firms that quantified their effects, though none of these overriding firms quantified the equity effect. However, the net profit margin for TFV firms is not greater than that of control firms. Consistent with the override, TFV firms report significantly lower depreciation to sales relative to control firms. Adjusting for depreciation however, the profitability of TFV and control firms seems to be similar, as OPINC and CFOTOS are not significantly different between TFV and control firms. The interest coverage ratio is insignificantly different for TFV firms and the debt to fixed assets ratio, DETOFIX, is significantly lower, suggesting that debt covenants are not tighter for TFV firms. Lastly, earnings to price, EP4M, is significantly higher for the override sample than the control sample, indicating that the market at least partially adjusts for the difference in depreciation and its implications for reported income and book value of equity. The overall evidence thus seems to support the claim that non-depreciation of investment properties is mechanical in nature and not an opportunistic accounting choice, consistent with our prediction for a Category 1 override.23
Comparison of Override, Descriptive, Performance, Contracting and Market Variables for the Sample of S19 (non-depreciation of investment properties) 1998–2002, and Control Firms
|% EQUITY||0|| || || || || || || || |
|% INCOME||9||0.075||0.060||0.053|| || || || || 2.76|
The previously reported tests are based on a comparison of the TFV firms to industry and size-matched control firms. The next set of tests examines the change in performance of TFV firms in the year they first adopted an override relative to the prior year. To the extent that overrides are undertaken to increase income or equity, we would expect to find first-time adoptions associated with worsening financial conditions. We therefore use the TFV firm in the year prior to the first override as the control. We present results on the descriptive and performance-related variables for the sample as a whole, and for the subsample of GAAP overrides.
The left side of Table 6 presents the findings for the overrides for which we could identify the first year of the override from the firm's financial statements and collect relevant data for the prior year. The findings indicate that the changes in sales, assets and equity, ΔSALES, ΔASSETS and ΔEQUITY, respectively, are all significantly positive, whereas the change in income, ΔINCOME, is not significantly positive. Consistent with this, the change in scaled profitability measures, ΔNETPRO, ΔNETROE and ΔTURNOVER are all significantly negative. However, the change in our measures of profitability before the effect of the override, ΔCFOTOS and ΔOPINC, are not significantly negative. The debt contracting variables indicate a marginally significant increase in debt to fixed assets and a statistically insignificant decline in interest coverage. Finally, the change in the book-to-market and earnings to price ratios are significantly positive, consistent with investors discounting book value of equity and earnings, respectively, in the year of the override, relative to the prior year.
Analysis of Differences Between First Year of Override and the Prior Year for Descriptive, Performance, Contracting, and Market Variables. Positive (negative) value implies year of override higher (lower) than prior year
|ΔSALES (£M)||224||185023.12||6539.50||1299122.72|| 3.83||46||225275.81||9||2513000||0.03|
|ΔASSETS (£M))||225||1476257.49||14212.00||16431363.20|| 3.96||46||4720054.87||4682||35601056|| 2.04|
|ΔEQUITY (£M))||225||−15166.39||4458.00||2049770.24|| 4.25||46||−75410.83||4359.5||3902808||0.73|
|ΔDTOS||213||0.042||0.001||0.429|| 3.30||41||0.027||0.01||0.067|| 3.70|
The right side of Table 6 presents the findings for the GAAP override subsample. Similar to the sample as a whole, the change in scaled profitability measures, ΔNETPRO, and ΔTURNOVER are significantly negative. In contrast to the entire sample, however, ΔCFOTOS and ΔOPINC are negative, though ΔCFOTOS is marginally significant. The changes in the debt contracting variables and the market-based variables are not significantly different from zero. Taken along with the findings in Table 4, the findings suggest that eroding performance may have contributed to the decision to override GAAP.
Taken as a whole, the findings thus far suggest that firms invoking the more costly overrides are experiencing weaker profitability. Firms invoking an override to avoid amortizing goodwill have significantly lower pre-override profitability. Firms invoking an override of UK GAAP have lower post-override profitability and experienced a significant decline in profitability in the year of adoption.24 Such firms may also be concerned about violating debt covenants, such as those based on interest coverage. An interpretation of these results is that in the UK's environment, which is characterized by principles-based rules coupled with significant deterrents to abuse, some companies would nevertheless be motivated to take advantage of the TFV requirement. However, this in itself does not imply that with more rigid requirements (i.e., a rules-based environment) the frequency and magnitude of accounting manipulations in the UK would have been lower.
(iii) Implications of Overrides for Earnings Quality and Informativeness
Our next analysis examines whether investors view TFV firms' financial statements as less informative than those of control firms (hypothesis 2). In the first set of tests, we focus on one dimension of informativeness, the explanatory power of reported book value and earnings per share for share prices. Following Joos and Lang (1994), we assess informativeness by the adjusted R2 from the regression of share price on reported book value and earnings per share.25
Our tests use Ohlson's (1995) model relating stock prices to earnings and book values.26 He notes that his model can be interpreted as a weighted average of earnings and book value-based valuation models. In his model, lower persistence of earnings will be captured by a lower coefficient on income and a higher coefficient on book value of equity. We separately examine this relation for positive and negative earnings, given the prior evidence by Hayn (1995) that negative earnings are of lower persistence.
The estimation equation is:
where P is the stock price per share measured 4 months after fiscal yearend; BV is the book value of equity per share; BV− is the book value of equity per share times an indicator variable equal to 1 (0) if net earnings per share are negative (non-negative), NI is reported net earnings per share, and NI− is reported net earnings per share times an indicator variable equal to 1 (0) if net earnings per share are negative (non-negative). We include BV− and NI− because the coefficient on net income is likely lower for reported losses, and the coefficient on book value of equity is likely higher.27 We expect that β1 > 0, β2 > 0, β3 > 0 and β4 < 0. We estimate equation (1) for the override and control samples separately, and test for differences in the R2 of the two subsamples.28
We examine whether more costly overrides (NOG and GAAP) involve a loss of informativeness. We compare these results to those of the mechanical override subsample, for which we would not expect less informative financial statements. The bottom two rows of Table 7 show the R2 for the override and control subsamples. For the full sample of overrides from 1998–2002, the R2 is 0.65 for the TFV firms, in contrast to 0.47 for the control firms. These findings indicate that earnings and book value do not have less explanatory power for the share prices of override firms than for those of control firms. In addition, the untabulated coefficient of variation is higher for the control sample, so the higher R2 is not due to scale. For the NOG subsample, we find that both the TFV and control sample regressions have R2's of 0.67. Because the coefficient of variation for the control sample is higher, these findings suggest that the explanatory power of book value of equity and income for share prices of the override sample is not less than for the control sample.
Pooled Regression of Stock Price on Book Value of Equity per Share and Net Earnings per Share with Varying Coefficients for TFV and Control Firms. Coefficient Estimates with t-statistics Immediately Below (2)
|BV * TFV||−0.24||0.07||−0.08||−0.16|
|BV− * TFV||−0.04||−0.55||−1.17||0.25|
|NI * TFV||2.81||−4.58||0.25||3.81|
|NI− * TFV||−4.60||8.92||−3.21||−6.58|
The results of the market-based tests for the GAAP overrides subsample are in contrast to the previous findings. They indicate that the explanatory power of the book value of equity and income for variation in GAAP override firm share prices is lower than for the control sample. The coefficient of variation is higher for the override sample however, so the scale effect is not inducing this finding. Given the findings in Tables 4 and 6, this is also consistent with the view that GAAP overrides are invoked to mask deteriorating performance in the financial statements.
Finally, we estimate equation (1) for the subsample of firms that do not depreciate investment properties, a treatment that is consistent with GAAP but inconsistent with the Companies Act. To the extent that these are merely mechanical overrides, we do not expect to find similar results to GAAP overrides. Indeed, for this sample, the R2 comparison suggests that the book value of equity and income reported by override firms have greater explanatory power for share prices than those reported by control firms. The coefficient of variation is greater for the control sample, again indicating that scale bias is not inducing this result. This finding may be explained, at least in part, by the fact that firms invoking S19 also annually revalue their investment properties. Prior literature (e.g., Aboody et al., 1999) suggests that revaluations are informative, implying that our results may reflect the greater informativeness of revalued properties.
Our second set of tests of H2 focus on the coefficients on earnings and book values, to examine whether TFV firms' earnings are perceived by investors as less persistent than those of control firms. This test pools the control and TFV samples, but allows the coefficients to vary between control and TFV firms. Specifically, we estimate an augmented version of equation (1):
where TFV is an indicator variable assuming the value of one if the observation is for a TFV firm, and zero otherwise.29 We test whether the persistence of earnings, as reflected in stock prices, differs between TFV firms and control firms. Specifically, if the earnings of TFV firms are artificially increased by the override and therefore are less persistent, we would expect β7 < 0 and β5 > 0. By introducing the interaction variable NI− * TFV in (2), we can also examine whether investors perceive differences in the persistence of negative income, and whether investors perceive differences in such persistence between TFV and control firms.
The results of this analysis are reported in Table 7. For the entire sample, the evidence indicates differential weights on income and book value of equity between profit and loss firms. Consistent with the prediction that losses are less persistent, the incremental coefficient on NI− is negative and statistically significant (β4=−6.38, t=−6.74) and the incremental coefficient on BVE−, book value of equity for loss firms, is positive and statistically significant (β2= 0.43, t= 3.36). This finding also largely holds for the coefficients on the three subsamples: the coefficient on negative net income is negative and significant for all three subsamples and the coefficient on book value for negative income firms is positive and significant for the NOG and GAAP override samples, though not for the S19 overrides.
As for the persistence of TFV firm earnings, we find that for profitable firms, the incremental coefficient on TFV earnings, β7, is positive and statistically significant (β7= 2.81, t= 4.01). This finding indicates that investors perceive these firms to have greater earnings quality than profitable control firms. Consistent with this, the incremental coefficient on book value of equity of TFV firms, β5, is negative and significant (β5=−0.24, t=−3.19), indicating a lower weight on equity in the valuation of TFV firms. For loss firms, the findings indicate an insignificant incremental coefficient on the book value of equity of TFV firms, (β5=−0.04, t=−0.25) and a significantly lower coefficient on the earnings of TFV loss firms, consistent with these TFV firms having less earnings persistence. A substantially similar pattern is observed for the S19 subsample as well.
In the GAAP override subsample, the findings do not indicate any difference in the persistence of profit between control and TFV firms. However, β6, the coefficient on book value for loss firms, BV−, is significantly smaller for override firms than control firms, consistent with their financial statements being less informative overall. In addition, the coefficient on earnings for loss firms is marginally significantly negative, suggesting less persistent earnings for loss firms. These findings suggest that the lesser informativeness of GAAP override firms observed in the R2 test is due to the firms earning losses.
In contrast to the findings for the GAAP override sample, the NOG subsample exhibits offsetting effects on the persistence of profit and loss override firms, consistent with the finding of similar R2's overall for both the NOG and control samples. For the S19 sample, as mentioned above, we obtain qualitatively similar findings to those for the sample as a whole.
To summarize, the stock price-based tests indicate that the financial statements of firms invoking the most costly overrides, the GAAP overrides, have lower explanatory power than those of their control sample. This finding is in contrast to the findings for firms invoking the NOG or S19 overrides, and for the override sample as a whole. The coefficient estimates indicate that the weaker explanatory power of GAAP override firms is due to the market discounting both the book value and earnings of GAAP override firms with losses. Our findings therefore suggest investors at least partially discount earnings and book value of the override firms where strategic use of overrides is most likely. In contrast, for the S19 sample, consistent with our finding little support that firms invoke these opportunistically, investors view their earnings as more persistent.