Earnings Management using Classification Shifting: Relation between Core Earnings and Special Items

Transcription

1 UPPSALA UNIVERSITY Department of Business Studies Bachelor Degree of Business Autumn Earnings Management using Classification Shifting: Relation between Core Earnings and Special Items Authors: Michael Bondegård & David La Advisor: Katerina Hellström Abstract: In this paper we examine if managers of Swedish firms listed on OMX Large Cap engage in earnings management by shifting core expenses to income decreasing special items in order to increase core earnings, with other words classification shifting. To measure classification shifting we use regression models by McVay (2006) to measure unexpected core earnings, this is the difference between the reported and expected core earnings, and unexpected change of core earnings. By examining the relation between unexpected core earnings, unexpected change of core earnings and special items we find no significance evidence of classification shifting in year We also decompose the special items into two groups, those that are unsusceptible and those that are susceptible, and test these against unexpected core earnings and unexpected change of core earnings. Based on our sample on large firms in 2004, we find no significant relation. Keywords: Earnings Management, Classification shifting, Core Earnings, Special Items

2 TABLE OF CONTENTS I. INTRODUCTION... 3 II. PRIOR RESEARCH AND HYPOTHESES... 4 III. RESEARCH DESIGN... 8 PHASE ONE... 8 PHASE TWO... 9 IV. PRESENTATION OF DATA & DESCRIPTIVE STATISTICS PRESENTATION OF DATA DESCRIPTIVE STATISTICS V. RESULTS PHASE ONE MEASURE EXPECTED EARNINGS PHASE TWO TESTING THE HYPOTHESES VI. CONCLUSION AND DISCUSSION REFERENCES LITERATURE ARTICLES WEBSITES OTHER SOURCES FIGURE APPENDIX A Bondegård & La Page 2 of 25

3 I. INTRODUCTION Evidence from prior research in the U.S. (Graham et al., 2005) and in the U.K. (Choi et al., 2006) show that meeting the expectations from the analysts is a significant earnings target. Negative earnings surprises lead to a penalty in the stock market while the positive earnings surprises lead to a market reward (Bartov et al., 2002; Kasznik and McNichols, 2002). From this point of view management might have the incentives to use their judgment over reported earnings to meet the expectations (Athanasakou et al., 2009). Since the analysts and investors focus more on core earnings (net sales or revenues less core expenses), pro forma earnings and street earnings than on bottom-line income, the management therefore takes advantage of this focus and might have the incentive to misclassify some core expenses within the income statement as special items (Bhattacharya et al., 2004; Gu and Chen, 2004). McVay (2006) develops regression models to test if management in U.S. firms engages in earnings management by classifying parts of core expenses (defined as costs of goods sold, and selling, general and administrative expenses) as special items in order to increase core earnings. By shifting parts of core expenses to income-decreasing special items, the firm experience increased core earnings while bottom-line income remains unaffected. In other words, as income-decreasing special items increase, the firm reported core earnings tend to be higher than expected core earnings and unexpected core earnings arise. She finds evidence to support her hypothesis. Athanasakou et al., 2009, use the same regression models to test if management in U.K. firms engages in classification shifting, and their findings are not consistent with the results in of McVay (2006). Since these regression models are relatively new and applied to few markets, we therefore want to test them on firms from of Sweden s OMX Large Cap. We also decompose special items into income-decreasing special items that are susceptible and those that are not and test these two subsets on unexpected core earnings and unexpected change of core earnings. The data we use to examine the regression models are from Thomson Reuters DATASTREAM Sweden OMX Large Cap, The original data contains 79 observations. After removing those firms with missing data, our final sample contains of 38 firms. We find no evidence of classification shifting in Swedish firms in Bondegård & La Page 3 of 25

4 The paper proceeds as follows. In the next section, we review in prior research and develop our hypotheses. In section III we describe in detail the measurement of unexpected core earnings and the test design. In section IV, we discuss the data, sample, and descriptive statistics and in Section V we present the results. Section VI, conclude and discuss the paper. II. PRIOR RESEARCH AND HYPOTHESES In their paper Graham et al. (2005) conclude that analyst expectations and prior earnings from firms are managers two most essential benchmarks. By reporting small negative earnings surprises can strongly affect the stock market negatively and give the investors a view that the firm is not well-managed. On the other hand reporting zero or positive earnings surprises give the investors the opposite view, the firm s underlying performance is better than predicted. By meeting the analyst expectations the firms earn a market reward and by missing the expectations the market punish the firms (Bartov et al., 2002; Kasznik and McNichols, 2002). Since analysts and investors typically focus rather on pro forma earnings 1 and street earnings 2 than the reported earnings (Bhattacharya et al. 2004), Brown and Caylor (2005) therefore, find evidence that from the mid-1990s managers rather report losses or earnings decreases than earnings surprises. They strengthen this conclusion by reporting evidence of a considerably higher reward/penalty for meeting/missing the analyst expectations than for achieving/missing the other earnings targets. The managers therefore have the incentives to use different methods to meet the expectations, and one of these is by earnings management 3 : classification shifting of core expenses to special items. 1 Pro forma earnings are a non-gaap measure of performance, and there is no official definition of this expression. This particular way of reporting earnings exploded in the late 1990s, pro forma earnings may exclude expenses that are non-recurring, non-cash, and a variety of other miscellaneous charges but also e.g. R&D costs and write-offs, restructuring charges, asset impairment charges, losses on the sale of businesses and assets, and goodwill amortization (Doyle et al., 2003; Dechow and Schrand, 2004). 2 Street earnings are the actual earnings reported by I/B/E/S. It is this earnings number that analysts are trying to forecast (e.g. earnings before depreciation and amortization) (Dechow and Schrand, 2004). 3 Earnings management is a meaningful intervention in the external financial reporting process, with the purpose of achieving some private gain (as opposed to, say, merely facilitating the neutral operation of the process (Schipper, 1989). Earnings management tends to occur when management use their opinions in financial reporting and in structuring transactions to change financial reports to either mislead some stakeholders about the firm s underlying economic performance or to affect contractual outcomes that depend on reported accounting numbers (Healy and Wahlen, 1999). Bondegård & La Page 4 of 25

5 According to former studies there are two main methods of earnings management: real activities management and accrual management (Schipper, 1989). When using real activities manipulation managers can e.g. provide price discounts to temporarily boost the sales, cut expenditures such as Research & Development, advertising and structuring costs, and other transactions (Baber et al., 1991; Bushee, 1998). Such actions have major impacts on earnings quality 4 and devastating effects on the firm s future performance (Dechow and Schrand 2004). Real activities management increase net income in the short term, but also have real costs. For example cutting R&D may result in the loss of future income related to forgone R&D opportunities. Since these activities are not GAAP 5 violation, they bear a relative lower cost of detection (Dechow and Sloan, 1991). The latter earnings method of manipulation, accrual management means that a manager can borrow earnings from future periods, through the acceleration of revenues or deceleration of expenses, in purpose to improve current earnings. Like the real activities management this process also bears a cost, since this action borrows earnings from the future period it will reduce with the same amount borrowed in the next period (Healy, 1985; Jones, 1991; McNichols and Wilson, 1988; Dechow and Schrand, 2004). McVay (2006) investigates a third form of earnings management. She tests whether U.S. firms manipulate core earnings which are defined as sales minus core expenses (costs of goods sold and selling, general and administrative expenses). To examine this statement she predicts that managers use classification shifting by classifying core expenses as special items 6 within the income statement. Since analysts and investors typically focus rather on core earnings (pro forma earnings) and street earnings than bottom-line income (reported 4 A high-quality earnings number is one that correctly replicates the firm s current operating performance, and is a good sign of future operating performance. In other words, earnings are considered to be of high quality is when the earnings number accurately annuitizes the intrinsic value of the firm and another way to think about high quality numbers are when return on equity is a good measure of the internal rate of return on the firm s current portfolio of project. Therefore when management decides to engage in earnings management then they decrease the earnings quality (Dechow and Schrand, 2004). 5 Generally Accepted Accounting Principles (GAAP) which are a widely accepted collection of regulations, principles, conventions, standards, and procedures for reporting financial information (fasb.org, 2009; investorwords.org, 2009). 6 Special items are a set of many unusual or infrequent items, which contain (-)special charge, (-)special liability accruals, (±) nonrecurring items, (-) asset write-downs, (±) changes in estimates, (-) start-up costs expensed, (±) profits and losses from asset sales, (-) restructuring costs, (±) profits and losses from discontinued operations, (±) extraordinary operating items, (±) accounting charges, (±) unrealized gains and losses on equity investments, (±) gains from share issues in subsidiaries, (±) currency gains and losses, (±) derivative gains and losses (operations) (Penman, H. S., 2007). Bondegård & La Page 5 of 25

6 earnings), because they consider them to be more value-relevant, managers might have the incentive to misclassify some core expenses within the income statement as special items (Bhattacharya et al. 2004; Gu and Chen 2004). Bradshaw and Sloan (2002) also find evidence that the market responds to street earnings more than to bottom-line earnings. McVay (2006) finds evidence to support her hypothesis and remarks that classification shifting bears a relatively low cost compared to the other two earnings management actions. There is no accrual that later reverses (illustrated in Figure 1), or forgone revenues or income by manipulating real activities. Furthermore the detection by auditors is fairly low. As the bottom-line income does not change, they might spend less energy on the identification or adjustments of these accounts. And to measure unexpected core earnings she uses a core earnings expectation model, similar to the accrual expectation model (Jones, 1991). The expected core earnings are measured by the relation between reported core earnings and a number of firm performance measures (e.g. core earnings for prior period, asset turnover ratio for the fiscal year, accruals for the fiscal and prior year, and change in sales for the fiscal year) for all other firms in the same industry. The unexpected core earnings are then calculated as the difference between the reported core earnings and the expected core earnings. If unexpected core earnings increase within overall in as the negative special items, this is a sign of classification shifting, since the management shift core expenses to negative items so that reported core earnings are higher than the expected. If management does not engage in classification shifting then the reported core earnings should be equal to the expected core earnings. McVay (2006) makes clear that the unexpected core earnings are increasing in special items in the fiscal year, t and for this increase to reverse in the next year, t+1. This is to assure that it is due to classification shifting and not to an economic improvement associated with the income-decreasing special items. Athanasakou et al., (2009) use the same methods as McVay (2006) on firms in the U.K. to examine the relation between core earnings, unexpected core earnings and special items. Their conclusion is opposite to the findings by McVay (2006). There is no significant relation between unexpected core earnings and income-decreasing special items. Since classification shifting has been examined only on U.S. Data (McVay, 2006) and U.K. data (Athanasakou et al., 2009), we find it interesting to examine if managers in of Swedish firms engage in the third form of earnings management, classification shifting. This leads to our first hypothesis: Bondegård & La Page 6 of 25

7 H1: Managers classify core expenses as special items McVay (2006) admits that by using COMPUSTAT special items, it contains many types of special items. These special items include those that are not susceptible to classification shifting such as asset write-downs, and other items that are more open to classification shifting, e.g. restructuring expenses other than asset write-downs, or merger related costs. Therefore, she hand-collects data for a sub-sample of firms, which have income-decreasing special items of at least 5% of sales in and have transitory charges. Then she divides them into two subsets of special items, those that are open to classification shifting and those that are not. She considers Property, Plant & Equipment write-offs, goodwill writeoffs, and losses on asset sales to be unsusceptible and all other are considered as acceptable for expense shifting. If the PP&E write-offs, goodwill write-offs, or losses on asset sales are not clearly broken out from susceptible charges, then she classifies the entire charge as susceptible. And by testing unsusceptible and susceptible special items with unexpected core earnings and unexpected change of core earnings she concludes that susceptibility is consistent with classification shifting. As McVay (2006), Athanasakou et al., (2009) examine these two subsets of income-decreasing special items on U.K. firms. Instead of using incomedecreasing unsusceptible and susceptible special items, they sort the special items as nonoperating exceptional items and other non-recurring items. By categorizing parts of special items differently to McVay (2006) they find no evidence of a significant correlation between unexpected core earnings and income-decreasing other non-recurring items for the entire sample. When analyzing the relation between unexpected core earnings and non-operating exceptional items, they find no significant evidence for any subset of the firms. But since our focus is on McVay (2006) we use the same categorization as McVay and apply it to Swedish firms. This lead us to our second hypothesis: H2: There is a significant relation between unexpected core earnings and susceptible special items Bondegård & La Page 7 of 25

8 III. RESEARCH DESIGN We replicate McVay (2006) two stage regression model. By following this model we create two phases, phase one to measure unexpected core earnings and phase two to examine if there is a relation between unexpected core earnings and special items. PHASE ONE Our first regression, regression (1) measures expected core earnings. The dependent variable core earnings (CE t ) is scaled by sales. Following McVay (2006), the first variable in the regression is lagged core earnings (CE t-1 ) and it is included because prior research finds that core earnings tend to be persistent 7. The asset turnover ratio (ATO t ) tends to have an opposite effect on core earnings (Nissim and Penman, 2001) and therefore is included in the model 8. Sloan (1996) suggests that accruals can be explained, if holding earnings constant, accruals can be used to predict future performance. For that reason, in line with McVay (2006), we include both lagged accruals (ACCRUALS t-1 ) and accruals (ACCRUALS t ) for current year. Following McVay (2006) we finally include sales growth (ΔSALES t ) and negative sales growth (NEG_ΔSALES t ). The regression (1) is shown below 9 : Regression (1) In regression (2) we use change in core earnings scaled by sales (ΔCE t ) as a dependent variable. We include, following McVay (2006), lagged accruals, current year accruals, sales growth, negative sales growth, change in asset turnover ratio (ΔATO t ) and add change in core earnings from prior year. (ΔCE t-1 ). The regression (2) shown is below 10 : 7 The Spearman correlation in McVay (2006) is between core earnings and lagged core earnings compared to our Spearman correlation which is in APPENDIX A. 8 The Spearman correlation in McVay (2006) is between core earnings and ATO compared to our Spearman correlation which is in APPENDIX A. 9 See Table 1 Panel B for our definition for the variables 10 See Table 1 Panel B for our definition for the variables Bondegård & La Page 8 of 25

9 Regression (2) Both regressions are estimated by fiscal year and we have ignored an industry classification since our sample is too small. We have make two tests on regression 1 and regression 2. The first test consists of data for one year and the second test is a control if our sample is valid even if we double the observations. PHASE TWO To test H1 we follow McVay (2006) unexpected core earnings for current year (UE_CE t ) and unexpected change in core earnings for the next year (UE_ΔCE t+1 ). We calculate coefficients 11 for each observation from regression (1) respectively regression (2). Further, we use reported core earnings (reported change in core earnings) and subtract obtained coefficients to calculate unexpected core earnings (unexpected change in core earnings). Next we include Special items (SI t ) and define it as income-decreasing special items 12. We compose special items in a slightly different way than McVay (2006) because of missing data 13. We obtain special items in two parts, in the first part we collect data from DATASTREAM and in the second part we collect data manually from annual reports. The data from DATASTREAM contains extraordinary charges 14, extraordinary items and gain/loss from sales of assets 15. The annual report data contains goodwill write-offs and losses from discontinued operations 16. We combine these two sets of variables to create our definition for special items. The regression (3) and regression (4) are shown below 17 : 11 We acquire unstandardized coefficients from SPSS 12 See Table 1 Panel B for our definition for the variables 13 Further discussion in the result section 14 Extraordinary charges are pretax extraordinary charge which is infrequent or unusual included in the net income of a company. For non-u.s. companies it is as defined by the company. It includes but not restricted to for example force majeure costs, expropriation of assets by forgeign government and write-down or write-off of inventories or other write-downs that are normally a part of operations if they exceed 25% of earnings. (DATASTREAM, ) 15 Extraordinary items & gain/losses of assets represents gains and losses resulting from nonrecurring or unusual events (DATASTREAM, ) 16 We manually collect goodwill write-offs and losses from discontinued operations compared to McVay (2006) which collect Goodwill write-offs, losses on asset sale and additionally PP&E (Property, plant and equipment) 17 See Table 1 Panel B for our definition for the variables Bondegård & La Page 9 of 25

10 Regression (3) Regression (4) The last step in phase two is to form two subsets of special items for testing our second hypothesis. We define special items that are susceptible to classification shifting as SHIFTABLE t and special items that are not susceptible to classification shifting as NOT_SHIFTABLE t. In line with McVay (2006) we move goodwill write-offs, PP&E (property, plant and equipment) write-offs or losses of assets sales to NOT_SHIFTABLE t if it is clearly broken out in the annual report. Since we find all goodwill write-off items in the annual reports we classify all goodwill items as NOT_SHIFTABLE t. The remaining parts of our composed special items are defined as SHIFTABLE t. The regression (5) and regression (6) is shown below 18 : Regression (5) Regression (6) IV. PRESENTATION OF DATA & DESCRIPTIVE STATISTICS PRESENTATION OF DATA We gather data for all Swedish listed companies on the OMX Large-Cap from Thomson Reuters DATASTREAM for the period from 2003 to Our original data contains 79 observations where we exclude firms which report missing data and we also remove the least traded share if the same company issued A and B shares. From this sample we also eliminate, 18 See Table 3 for our definition for the variables Bondegård & La Page 10 of 25

11 consistent with McVay (2006) all firms which report less than 7.03 MSEK 19 in net sales to avoid outliers as sales are used as a component for most of the variables. The final sample contains a total of 38 firms in different industries. We have chosen to not classify our sample into smaller subgroups such as type of industry and amount of income decreasing special items of sales due to the extremely small sample. DESCRIPTIVE STATISTICS Table 1 panel A presents descriptive statistics. The reported values for the variables in the sample are comparable to some extent to McVay (2006). The mean sales growth (ΔSales t ) is approximately 27.9% and is about 6% higher in our full sample compared to McVay (2006). The mean Core Earnings (CE t ) in our full sample is 20.3% and is also larger than McVay (2006) which is 7%. The rest of the variables seem approximately identical to our results in terms of mean is income decreasing special items (SI t ) 2.9%, Accruals (ACCRUALS t ) -7.9% and Asset Turnover Ratio (ATO t ) 2.72 compared to McVay (2006) which is SI t 2.7%, Accruals t -10.4% and ATO t The mean Change Core Earnings (ΔCE t-1, t ), Unexpected Core Earnings (UE_CE t ) and Unexpected Change Core Earnings (UE_ΔCE t+1 ) are nearly identical to our result. Table 1 panel B explains the variables and how we calculate the data. Table 1 Panel A Descriptive statistic for full sample (N=38) Variables: Mean Median Standard Deviation Percentile 25 Percentile 75 ΔSALES t 0, , , , , CE t 0,203 0,160 0,181 0,109 0,209 ΔCE t-1, t 0,011-0,002 0,061-0,008 0,012 ΔCE t, t+1 0,017 0,007 0,073-0,001 0,019 UE_CE t 0, , , , , UE_ΔCE t+1 0, , , , , SI t -0, , , , , ACCRUALS t -0,079-0,035 0,384-0,071-0, McVay (2006) excludes firms that report net sales of less than 1 million dollar. We estimate one million as 7.03 million Swedish krona based on forex exchange rate ( ) Bondegård & La Page 11 of 25

12 ATO t 2, , , , , Table 1 Panel B Variables SALES t ΔSALES t Definitions and [DATASTREAM CODES] Company net sales [X(WC01001)] The amount in percent of change in sales (Sales growth), calculated as (Sales t -Sales t-1 )/Sales t-1 CE t ΔCE t+1 UE_CE t UE_ΔCE t Core Earnings t = Core Earnings scaled by sales = (Sales t - Cost of Goods Sold t - Selling, Gen. & Adm. Costs t )/Sales t [(X(WC01001) - X(WC01051) - X(WC01101)) / X(WC01001)] Change in Core Earnings, CE t+1 - CE t Unexpected Core Earnings is calculated as the difference between Core Earnings (CE t ) and predicted value of Core Earnings. Predicted values collected from Regression of CE t where t = 2004 Unexpected change in Core Earnings is calculated as the difference between change Core Earnings (ΔCE t+1 ) and the predicted value of Core Earnings. Predicted values collected from Regression of ΔCE t+1 NEG_ ΔSALES t If ΔSales t is less than 0 then ΔSales t else 0 ACCRUALS t Accruals t = Operating Accruals = (Net Income before Extraordinary Items t - cash flow from operations t )/Sales t [(X(WC01551) - X(WC04860)) / X(WC01001)] NOA t Net Operating Assets t = Operating Assts t - Operating Liabilities t [(X(WC02999) - X(WC02001)) - (X(WC02999) - X(WC03999) - X(WC03255) + X(WC03351))] ATO t Asset Turnover Ratio t = ATO t = Sales t /((NOA t + NOA t-1 )/2) ΔATO t Change in Assets Turnover Ratio, ATO t ATO t-1 SI t Income decreasing special items scaled by sales calculated as if negative (goodwill write-off t + Extraordinary charges t + Losses Bondegård & La Page 12 of 25

13 from discontinued operations t ) * (-1) / Sales t [Goodwill write-offs collected by hand + X(WC01254) + Losses from discontinued operations collected by hand / X(WC01001)] SHIFTABLE t NOT_SHIFTABLE t Income decreasing special items that are susceptible to classification shifting scaled by sales calculated as if negative (Extraordinary charges t + Losses from discontinued operations t ) * (-1) / Sales t [X(WC01254) + Losses from discontinued operations collected by hand / X(WC01001)] Income decreasing special items that are not susceptible to classification shifting scaled by sales calculated as if negative (goodwill write-off t ) * (-1) / Sales t [Goodwill write-offs collected by hand / X(WC01001)] t represents the year; for all cases if not stated differently: t = year 2004 V. RESULTS We begin to explain our results in phase one and phase two. PHASE ONE MEASURE EXPECTED EARNINGS In Table 2 panel A we see that almost all results are consistent in term of sign prediction and significance level with McVay (2006) 20, for example CE t-1, ACCRUALS t-1 and ΔSALES t. The predicted sign of ATO t and ACCRUALS t is on the other hand not consistent with McVay (2006). ATO t however is not significant. To solve ACCRUALS t, we run a second test by increasing our sample by adding a second year shown in Table 2 panel B. We find that ACCRUALS t is not significant anymore in our sample. Our adjusted R 2 for the mean 20 McVay (2006) uses a different approach to display significance level. We run one regression on all firms in the sample, compared to McVay (2006) who runs several regressions based on industries and therefore displays a percentage of all runs when expressing significance level. 21 ATOt in our results (Table 2 panel A) is and not significant which can be compared to McVay (2006): and not significant. 22 ACCRUALS t in our result (Table 2 panel A) is -0,031 and significant which can be compared to McVay (2006): and significant. Bondegård & La Page 13 of 25

14 coefficient is 97.5% for Table 2 panel A and 93% for Table 2 panel B compared with McVay (2006) which is 75.5%. Table 2 Panel A Dependent Variable: CE t Observations: 38 Year t = 2004 Mean Coefficent t Sig. Collinearity Statistics Toleranc VIF e (Constant) 0,012 1,043 0,305 Independent Variable CE t-1 0,897 31,955 0,000 0,612 1,634 ATO t 0,001 0,409 0,686 0,709 1,411 ACCRUALS t-1-0,108-9,595 0,000 0,694 1,441 ACCRUALS t -0,031-2,167 0,038 0,725 1,380 ΔSALES t 0,041 6,903 0,000 0,805 1,242 NEG_ ΔSALES t 0,260 0,864 0,394 0,914 1,094 Adjusted R 2 97,50% Table 2 Panel B Dependent Variable: CE t Observations: 76 (Pooled) Year t = 2004/2005 Mean Coefficent t Sig. Collinearity Statistics Tolerance VIF (Constant) 0,019 1,322 0,190 CE t-1 0,900 26,967 0,000 0,671 1,491 Independent Variable ATO t 0,000-0,228 0,820 0,784 1,276 ACCRUALS t-1-0,061-4,311 0,000 0,735 1,360 ACCRUALS t -0,019-1,027 0,308 0,707 1,414 ΔSALES t 0,042 5,520 0,000 0,863 1,159 NEG_ ΔSALES t 0,064,559 0,578 0,890 1,124 Dummy Variable 0,001,074 0,941 0,949 1,053 Adjusted R 2 93,00% In Table 3 Panel A we find the results to match McVay (2006), both sign prediction and significance level except of ACCRUALS t. We perform a second test to find out if the Bondegård & La Page 14 of 25

15 regression is valid for two years shown in Table 3 Panel B. We find no change to prediction sign values and significance compared to McVay (2006). Our adjusted R 2 for the mean coefficient is still high for Table 3 Panel A but decreases in Table 3 Panel B. Our adjusted R 2 for Table 3 Panel A is 85.4% and for Table 3 Panel B is 51% compared to McVay which is 51.7%. Table 3 Panel A Dependent Variable: ΔCE t Observations: 38 Year t = 2004 Mean Coefficent t Sig. Collinearity Statistics Tolerance VIF (Constant) 0,008 0,912 0,369 CE t-1-0,092-3,356 0,002 0,620 1,614 Independent Variable ΔCE t-1-0,026-0,282 0,780 0,696 1,337 ΔATO t 0,012 1,569 0,127 0,767 1,303 ACCRUALS t-1-0,108-9,686 0,000 0,680 1,470 ACCRUALS t -0,034-2,443 0,021 0,761 1,313 ΔSALES t 0,041 7,453 0,000 0,889 1,125 NEG_ ΔSALES t 0,372 1,229 0,229 0,869 1,151 Adjusted R 2 85,40% Table 3 Panel B Dependent Variable: ΔCE t Observations: 76 (Pooled) Year t = 2004/2005 Mean Coefficent t Sig. Collinearity Statistics Tolerance VIF (Constant) 0,015 1,364 0,177 Independent Variable CE t-1-0,095 3,138 0,003 0,789 1,268 ΔCE t-1-0,148 1,663 0,101 0,800 1,250 ΔATO t 0,003-0,383 0,703 0,877 1,140 ACCRUALS t-1-0,059 4,298 0,000 0,767 1,304 ACCRUALS t -0,023-1,261 0,212 0,700 1,428 ΔSALES t 0,047 6,111 0,000 0,835 1,198 NEG_ ΔSALES t -0,001-0,010 0,578 0,776 1,288 Bondegård & La Page 15 of 25

16 Dummy Variable 0,003 0,248 0,941 0,892 1,121 Adjusted R 2 51,00% We compare the results in Table 2 Panel A and Panel B respectively Table 3 Panel A and Panel B and we can conclude that there are no or small differences between one year data and pooled data. We also find that our dummy variable in both cases display no significance, for this reason we decide to continue our further phase two analysis based on one year data since there should be no impact. PHASE TWO TESTING THE HYPOTHESES We report results from unexpected core earnings on special items and unexpected change in core earnings on special items, shown in Table 4 and Table 5. We see that our result for regression (3) in Table 4 differs from McVay (2006). We find a positive sign but no evidence of a significant relationship between UE_CE t and SI t. McVay (2006) finds on the contrary a significant relationship between UE_CE t and SI 23 t. However, our results are consistent with Athanasakou et al. (2009) who find no evidence of significance between unexpected core earnings and special items 24. Our adjusted R 2 is 1.00% compared to McVay (2006) which is 0.03% and compared to Athanasakou et al., (2009) which is 0.01%. Next, our result for regression (4) is shown in Table 5 and illustrates a negative predicted sign in line with McVay (2006) but there is still no significance between UE_ΔCE t+1 and SI t. Table 4 Dependent Variable: UE_CE t Observations: 38 Year t = 2004 Mean Coefficent t Sig. Independent Variable (Constant) 0,004 0,679 0,502 SI t 0,120-1,164 0,252 Adjusted R 2 1,00% 23 McVay (2006) t-statistics is The t-statistics in Athanasakou et al., (2009) is 1.08 Bondegård & La Page 16 of 25

17 Table 5 Dependent Variable: UE_ΔCE t+1 Observations: 38 Year t = 2004 Mean Coefficent t Sig. Independent Variable (Constant) 0,007 0,862 0,394 SI t -0,238-1,479 0,148 Adjusted R 2 3,1% We can conclude that in our case we find no evidence that H1 can be true, large Swedish firms in our sample are probably not using classification shifting. To test hypothesis 2 we perform regression (5) and regression (6). The results of regression (5) are shown in Table 6. Our results show in line with McVay (2006) in both cases that SHIFTABLE t are more significant than NOT_SHIFTABLE t but overall both our results are still insignificant while McVay (2006) have significant t-statistics for SHIFTABLE 25 t. The results of regression (6) are provided in Table 7. Our results show no significant values compared to McVay where both values are significant. If we compare Table 6 and Table 7 with the results in McVay (2006), we see that despite the fact that we received no significant values they are following the same pattern. Table 6 Dependent Variable: UE_CE t+1 Observations: 38 Year t = 2004 Mean Coefficent t Sig. Independent Variable (Constant) 0,004 0,687 0,496 SHIFTABLE 0,199 1,126 0,268 NOT_SHIFTABLE 0,071 0,605 0,549 R 2 4,10% 25 McVay (2006) t-statistics for SHIFTABLE t is 4.43 Bondegård & La Page 17 of 25

18 Table 7 Dependent Variable: UE_ΔCE t+1 Observations: 38 Year t = 2004 Mean Coefficent t Sig. Independent Variable (Constant) -0,007-0,913 0,367 SHIFTABLE -0,350-1,280 0,209 NOT_SHIFTABLE -0,188-1,035 0,308 Adjusted R 2 1,20% We conclude, based on the none-existing significance on both variables in relationship to both depended variables that there is no evidence of a relation between unexpected core earnings and special items that are susceptible to classification shifting in large Swedish firms. VI. CONCLUSION AND DISCUSSION Based on our sample on large firms in Sweden, we find no evidence of classification shifting. These results must be taken with caution since it is a very small sample and hence we ignored a classification into industries and also a classification into income decreasing special items subgroups. The small sample could also influence the results in phase one, as we see both regression 1 and regression 2 containing no significant accruals for current period which is opposed to McVay (2006) findings. The next issue is the missing data to build a whole income decreasing special items (SI t ). We collect only a part of the income decreasing special items available. Given that, we recieve a different definition of income decreasing special items than McVay (2006). McVay (2006) include fully composed income decreasing special items from COMPUSTAT and hand collect three items: goodwill write-offs, PP&E write-offs and losses on asset sale while we combine parts of special items from DATASTREAM and goodwill write-offs from annual reports collected by hand. This makes our result somehow misleading, since the obtained result is just one part of decreasing special items. Despite this, we conclude that there is no evidence of classification shifting using our definition of income decreasing special items. Bondegård & La Page 18 of 25

19 However since our attempt has shown opposite results to McVay (2006) we welcome future research on this subject. Future research could consist of a larger sample of Swedish firms together with better data on special items to receive more reliable results. Another possible idea would be to examine relation between for example classification shifting and firms with a high book to market ratio or the 25% most traded firms. Bondegård & La Page 19 of 25

20 REFERENCES LITERATURE Penman, H. S., 2007, Financial Statement Analysis and Security Valuation (4 th ed.), New York: McGraw-Hill Pallant, J., 2001, SPSS Survival Manual, A step by step guide to data analysis using SPSS for Windows (Version 10), Buckingham: SPSS Inc., also Available [online] < > [ ] ARTICLES Athanasakou, V. E., Strong, N. C., and Walker, M., 2009, Earnings management or forecast guidance to meet analyst expectations? Accounting and Business Research, 39 (1): 3-35 Baber, R. W., Fairfield, M. P. Haggard, A. J., 1991, The Effect of Concern about Reported Income on Discretionary Spending Decisions: The Case of Research and Development. Accounting Review, 66 (4): Bartov, E., Givoly, D. and Hayn, C., 2002, The Rewards to Meeting or Beating Earnings Expectations. Journal of Accounting and Economics, 33: Bhattacharya, N., Black, L. E., Christensen, E. T. and Mergenthaler, D. R., 2004, Empirical Evidence on Recent Trends in Pro Forma Reporting. Accounting Horizons, 18 (1): Bradshaw, T. M. and Sloan, G. R., 2002, GAAP Versus the Street: An Empirical Assessment of Two Alternative Definitions of Earnings. Journal of Accounting Research, 41 (1): Brown, L. and Caylor, M., 2005, A Temporal Analysis of Quarterly Earnings Thresholds: Propensities and Valuation Consequences. The Accounting Review, 80 (2): Bondegård & La Page 20 of 25

21 Bushee, J. B., 1998, The Influence of Institutional Investors on Myopic R&D Investment Behavior. The Accounting Review, 73 (3): Choi, Y. S., Walker, M., and Young, S., 2006, Earnings Reporting and Analysts Earnings Forecasts: the Perceptions of UK Analysts and Financial Managers. Working paper, Lancaster University Dechow, P. M and Schrand, M. C., 2004, Earnings Quality, The Research Foundation of CFA Institute, Dechow, P. and Sloan, R., 1991, Executive Incentives and the Horizon Problem: An Empirical Investigation. Journal of Accounting and Economics, 14: Doyle, T. J., Lundholm, J. R. and Soliman, T. M., 2003, The Predictive Value of Expenses Excluded from Pro Forma Earnings. Review of Accounting Studies, 8: Graham, J. R., Harvey, C. R. and Rajgopal, S., 2005, The Economic Implications of Corporate Financial Reporting. Journal of Accounting and Economics, 40: 3-73 Gu, Z. and Chen, T., 2004, Analysts Treatment of Nonrecurring Items in Street Earnings. Journal of Accounting and Economics, 38: Healy, M. P. and Wahlen, M. J., 1999, A Review of the Earnings Management Literature and Its Implications for Standard Setting, Accounting Horizons, 13 (4, December): Healy, P., 1985, The Effect of Bonus Schemes on Accounting Decisions. Journal of Accounting and Economics, 7: Jones, J. J., 1991, Earnings Management During Import Relief Investigations. Journal of Accounting Research, 29 (2): Bondegård & La Page 21 of 25

22 Kasznik, R. and McNichols, M., 2002, Does Meeting Earnings Expectations Matter? Evidence from Analyst Forecast Revisions and Share Prices. Journal of Accounting Research, 40 (3): McNichols, M. and Wilson, G. P., 1988, Evidence of Earnings Management from the Provision for Bad Debts. Journal of Accounting Research, 26 (Supplement): 1-32 McVay, E. S., 2004, The Use of Special Items to Inflate Core Earnings. Dissertation, University of Michigan, 1-52 McVay, E. S., 2006, Earnings Management Using Classification Shifting: An Examination of Core Earnings and Special Items, Accounting Review, 81 (3): Nissim, D. and Penman, H. S., 2001, Ratio analysis and equity valuation: From research to practice. Review of Accounting Studies 6 (1): Schipper, K., 1989, Earnings Management. Accounting Horizons, 3 (4, December): Sloan, R., 1996, Do Stock Prices Fully Reflect Information in Accruals and Cash Flows about Future Earnings? The Accounting Review, 71 (3): WEBSITES Standard & Poors, 2002, Available [online] < -v6.pdf> [ ] Investopedia, 2009, Available [online] < [ ] Forex, 2009, Available [online] < [ ] Bondegård & La Page 22 of 25

23 OTHER SOURCES DATASTREAM, ANNUAL REPORTS Bondegård & La Page 23 of 25

24 FIGURE 1 A simple View of Accrual Management versus Expense Shifting Accrual Management Year 1 Year 2 Pre-managed Core Earnings Earnings Management 25 0 Earnings Management Reversal 0-25 Managed Core Earnings Expense-Shifting Year 1 Year 2 Pre-managed Core Earnings Earnings Management 25 0 Earnings Management Reversal 0 0 Managed Core Earnings Source: McVay (2004) Bondegård & La Page 24 of 25

25 APPENDIX A Spearman Correlation matrix Spearman Correlatio n matrix SALE S t ΔSAL ES t CE t CE t+1 ΔCE t UE_C E t UE_ ΔCE t SI t ACC RUA LS t ATO t SALES t 1-0,09-0,18-0,3 0,178 0,25-0,08 0,106-0,04 0,504 ΔSALES t -0,09 1 0,142 0,207 0,533 0,234 0,213 0,227-0,01-0,23 CE t -0,18 0, ,936-0,23 0,238 0,424 0,174 0,006-0,4 CE t+1-0,3 0,207 0, ,17 0,187 0,492 0,175 0,026-0,46 ΔCE t 0,178 0,533-0,23-0,17 1 0,638 0,188 0,38 0,051-0,04 UE_CE t 0,25 0,234 0,238 0,187 0, ,183 0,462 0,059-0,01 UE_ ΔCE t -0,08 0,213 0,424 0,492 0,188 0, ,09 0,045-0,21 SI t 0,106 0,227 0,174 0,175 0,38 0,462-0,09 1 0,332 0,142 ACCRUA -0,04-0,01 0,006 0,026 0,051 0,059 0,045 0, ,142 LS t ATO t 0,504-0,23-0,40-0,46-0,04-0,01-0,21 0,142 0,142 1 No Correlation: 0 to 0,09 (-0,09) Small Correlation: 0.10 (-0.10) to 0.29 (- Medium Correlation: 0.30(-0.30) to 0.49(- 0.49) 0.29) Strong Correlation: 050 (-0.50) to 1.00 (- 1.00) Definitions of correlation significance gathered from Pallant, J. (2001) Bondegård & La Page 25 of 25