Income Classification Shifting and Mispricing of Core Earnings

Abstract

This study examines whether the market misprices core earnings (operating income before depreciation and special items) when firms use income classification shifting tactics to boost their core earnings. We find that the market’s expectation of core earnings’ persistence is higher than the actual reported earnings persistence of firms that have shifted their core earnings. We also find that core earnings are more negatively associated with future returns for shifters than for non-shifters. Overall, we find strong evidence that the market overprices shifters’ core earnings. These results are robust to controlling for earnings management and real earnings management, endogeneity and self-selection, and using alternative measures of classification shifting. Our findings are timely given the Securities and Exchange Commission’s recent concerns of firms’ income classification shifting behavior.

Keywords

classification shifting mispricing anomalies

Introduction

This article provides large sample evidence of negative economic consequences to shareholders when firms increase their core earnings by using income classification tactics. Specifically, we document that the market overvalues the core earnings of firms that engage in income classification shifting (“classification shifters”). We are motivated by studies providing large sample evidence of income classification shifting (Barua, Lin, & Sbaraglia, 2010; McVay, 2006), increasing concerns of regulators,¹ and the lack of systematic evidence on the mispricing of reported core earnings for the classification shifters. It is likely that earnings management using classification tactics is not easily detected,² and accordingly, significant resource allocation inefficiencies (e.g., mispricing of earnings) are likely to exist. Drawing on the expected core earnings model developed by McVay (2006), we use various methods to categorize our sample firms as likely“classification shifters” or “non-classification shifters” and investigate whether the market overvalues the core earnings reported by the classification shifters. Our large sample evidence that core earnings are more mispriced for classification shifters has important implications for policy makers, investors, and researchers (Chi, Pincus, & Teoh, 2014).³

The issue of income classification shifting is of high importance to the Securities and Exchange Commission (SEC) as it states, “The appropriate classification of amounts within the income statement is as important as the appropriate measurement or recognition of such amounts” (SEC, 2000b). The SEC further stresses that they are concerned about account classification because they have noted a number of improper classifications of line items in financial statements, especially the income statement. In the last decade, the SEC has been actively pursuing companies that engage in income classification shifting. For example, on November 12, 2009, the SEC charged SafeNet, Inc., with improper classification of ordinary operating expenses as non-recurring integration expenses (costs incurred to integrate acquired companies into current operations; SEC, 2009). Furthermore, on April 2, 2010, the SEC charged Symbol Technologies, Inc., with misclassifying unrelated operating expenses to a restructuring charge, which was recorded for its acquisition of Texlon Corporation and relocation of its manufacturing operations to new facilities (SEC, 2010a).

Despite the SEC’s concern of the possible negative market consequences of firms’ misclassification behavior, academic research in this area is limited, especially regarding the market implications of classification shifting.⁴ For instance, McVay (2006) proposes a method to identify the core earnings amount that is shifted to special items but does not find an association between shifted core earnings and future returns for the full sample.⁵ Using a similar approach, Haw, Ho, and Li (2011) find that classification shifting generates negative future abnormal returns for a sample of eight East Asian countries. However, Athanasakou, Strong, and Walker (2011) examine firms in the United Kingdom and do not find evidence of mispricing for firms that use classification shifting to meet analyst expectations. As U.K. and U.S. capital markets are similar in many aspects, it is likely that the results from Athanasakou et al. (2011) imply that the U.S. market does not misprice shifted core earnings. Such a conclusion requires a systematic analysis, which is the focus of our article.

To carry out our tests, we first classify firms as shifters and non-shifters. According to previous studies, a positive relation between unexpected core earnings and income-decreasing special items (McVay, 2006) and income-decreasing discontinued operations (Barua et al., 2010) indicates that classification shifting is likely to have occurred. As our main measure of shifting, we categorize a firm as a classification shifter if the firm has positive unexpected core earnings and income-decreasing transitory items (i.e., special items or discontinued operations) in any given fiscal year. We conduct both the Mishkin test and hedge return/multiple regression analyses that are adopted in previous research and contrast our findings between the shifters and the non-shifters. We use this simple definition for our main analysis; however, it is possible that this classification is crude and will bias our results. In our robustness tests (discussed in “Robustness Check and Additional Analyses” section), we use various techniques to ensure that our finding is more likely due to shifting rather than other unknown factors.

Classification shifting increases core earnings in year t but such an increase is unlikely to persist into period t+ 1 (McVay, 2006)⁶; accordingly, ceteris paribus, the core earnings persistence after shifting will be lower. If the market does not see through the shifting, investors will expect the core earnings persistence to be higher than the actual persistence of core earnings. We adopt the Mishkin test for this prediction. Consistent with our prediction, we find that the market fails to recognize the lower persistence of shifters’ artificially inflated core earnings. In the zero-investment portfolio test, we take a long position in firms in the most negative core earnings decile in year t and a short position in firms in the most positive core earnings decile in year t. We find that portfolio returns are significantly positive and higher for classification shifters than for non-classification shifters, consistent with mispricing of inflated core earnings.

We find that the mispricing of core earnings continues to hold for classification shifters after controlling for cross-sectional differences in risk and the cash flow anomaly (Desai, Rajgopal, & Venkatachalam, 2004). This suggests that the results of our previous tests are more likely due to market mispricing than risk. To ensure that classification shifting differs from accruals management and real earnings management, we add controls for these two earnings management techniques and find that classification shifting exerts adverse effects that cannot be explained by these two management tactics. As we rely on special items to define classification shifting and studies have shown that special items may be related to future returns (e.g., Burgstahler, Jiambalvo, & Shevlin, 2002; Dechow & Ge, 2006), we also control for the relation between special items and future returns in our regression.⁷ Finally, we control for idiosyncratic risks and transaction costs (Mashruwala, Rajgopal, & Shevlin, 2006) and the earnings response coefficient (ERC; Shi & Zhang, 2012), and continue to find more mispricing for shifters. We have conducted numerous additional robustness checks, including alternative definitions of classification shifting, mispricing of core earnings when special items are large and when unexpected core earnings are negative. Our inferences remain unchanged.

Kraft, Leone, and Wasley (2007) claim that the Mishkin test is misspecified because variables omitted from the forecasting and pricing equations are not rationally priced. However, Lewellen (2010) suggests that the test of market efficiency using the Mishkin test is still valid even though there are correlated omitted variables. To mitigate concerns that our results are due to relevant variables being excluded from the forecasting equation of the Mishkin test, we following Kraft et al. (2007) and re-run the Mishkin test incorporating additional explanatory variables that could potentially alter the relation between future returns and the persistence of core earnings. We find consistent results. To mitigate concerns about endogeneity and self-selection of managers intentionally choosing to engage in classification shifting, we use a propensity-score matched sample of shifters and non-shifters to re-run the Mishkin test. We estimate a probit model using a comprehensive set of firm characteristic variables, including operating cash flow, accruals, discretionary accruals, earnings persistence, number of business segments, level of industry competition, and prior sales growth. We identify matched pairs by selecting one shifter observation and one non-shifter observation with the closest propensity score on the basis of the probabilities generated from the first-stage probit model. Our results remain strong using the propensity-score matched sample.

We contribute to the literature on the mispricing of firms that engage in classification shifting behavior. Previous literature documents that unexpected core earnings are correlated with future returns, implying that the market has mispriced the core earnings. We provide a direct test of the mispricing of “core” earnings rather than “non-core” earnings. We show that investors can use core earnings to form investing portfolios to earn significant excess returns if they have suspicions of core earnings shifting. Also, we provide a more complete test to show that the mispricing is due to the market overreaction to higher reported earnings persistence and the mispricing still exists when we control for risk factors. Specifically, we report that the magnitude of mispricing is large; this large economic consequence provides strong support for the SEC’s recent investigations into and concerns about firms that are potentially engaging in classification shifting. Furthermore, we show that the mispricing of classification shifting is distinct from the mispricing of earnings management and real earnings management. This is an important finding because high unexpected core earnings can also be attributed to alternative opportunistic managerial strategies, such as accruals management and the manipulation of real activities.

The rest of this article is organized as follows: In “Background and Hypotheses” section, we provide a review of studies that are relevant to our article and develop our hypotheses. In “Research Design” section, we outline our research design. “Data and Descriptive Statistics” section describes our sample selection and descriptive statistics. “Results” section presents the results. In “Robustness Check and Additional Analyses” section, we provide additional analyses and robustness checks. We conclude in “Conclusion” section.

Background and Hypotheses

Income Classification Shifting and Market Mispricing

Compared with accruals management and real activities management, income classification shifting is a subtle yet viable tool for managing earnings. As managers are shifting items between categories, net income will not be affected. However, managers should have incentives to boost core earnings because the market has become increasingly focused on core earnings that exclude non-recurring or unusual items from generally accepted accounting principles (GAAP) earnings (Bradshaw & Sloan, 2002; Gu & Chen, 2004).⁸ The operating expenses that are removed from core earnings this year could recur next year in core earnings, resulting in lower persistence of current-period core earnings into the next-period.⁹ If investors are unable to recognize the lower persistence of core earnings of firms that have shifted expenses and revenues, they will overvalue the securities of shifters, which will create investor losses when the lower than expected core earnings are later revealed.

Income classification shifting refers to the intentional misclassification of core expenses as non-recurring items to boost core earnings.¹⁰ The literature recently identifies shifting as an important type of earnings management (Barua et al., 2010; Fan, Barua, Cready, & Thomas, 2010; McVay, 2006).¹¹ Compared with accruals management and real earnings management, income classification shifting has several advantages. First of all, income classification shifting may be more difficult to detect than accruals management. When management uses accruals to recognize revenue prematurely or delay the recognition of expenses, current- and future-period earnings are affected. However, when management allocates regular operating expenses to non-recurring expenses, the total amount of expenses and income remains the same.¹² This may draw less attention from the auditor and the public. Moreover, the standards on the classification of expenses as recurring and non-recurring can be subjective; hence, auditors may have limited ability to detect such misclassification or less incentive to debate with the managers and require adjustments.

Consequently, the use of income classification shifting has a lower cost of detection for the managers compared with accruals management.¹³ Second, income classification shifting has lower economic costs than real earnings management for the firm. With real earnings management, managers undertake or defer actual operational transactions or activities to increase revenue or net income at the expense of future benefits. Income classification shifting, however, is pure accounting manipulation that entails no real business transactions. Consequently, it imposes less cost to achieve various objectives for earnings management (McVay, 2006).

As income classification shifting is less costly to implement, yet more difficult to detect, it offers a viable alternative to other earnings management tools. Both McVay (2006) and Barua et al. (2010) establish large sample evidence that management shifts core expenses to special items or discontinued operations to inflate core earnings. According to McVay, the effect of income classification shifting on core earnings can be substantial. For instance, firms that have been identified as opportunistic shifters can, on average, increase their core earnings by half a cent per share. Firms with income-decreasing special items of at least 5% of sales can boost their core earnings by almost 3 cents per share. Not surprisingly, managers have been found to use classification shifting to meet/beat analysts’ earnings benchmarks (Barua et al., 2010; Fan et al., 2010; McVay, 2006) or to increase offering valuations around seasoned equity offerings (Siu & Faff, 2012). This misclassification of expenses, however, will only be effective if it affects investors’ perceptions of the firm’s underlying economic performance.

Managers are likely to engage in income classification shifting if investors value core earnings higher than non-core earnings. Core earnings are believed to be a better indicator of future profitability than non-core earnings. For example, the American Institute of Certified Public Accountants (AICPA) Special Committee on Financial Reporting suggests that a company’s core activities which are usual and recurring provide the best indicator of the future and should be reported separately from non-core activities that are transitory in nature. Practitioners, including financial analysts, also advocate distinguishing between recurring (core) and non-recurring (non-core) items in firm valuations. Numerous academic studies document that operating components receive higher weights than non-operating components in forecasting future profitability (Barth, Cram, & Nelson, 2001; Fairfield, Sweeney, & Yohn, 1996; Givoly & Hayn, 1992). Furthermore, as the persistence of earnings affects the valuation weight placed by the market (Kormendi & Lipe, 1987), the market places a higher valuation weight on core earnings than on non-core earnings, which has been extensively documented in previous literature (Bradshaw & Sloan, 2002; Brown & Sivakumar, 2003; Cheng, Cheung, & Gopalakrishnan, 1993; Elliott & Hanna, 1996; Francis, Hanna, & Vincent, 1996; Lipe, 1986).

These research findings imply that the market understands the implications of earnings persistence in firm valuation. Sloan (1996) also shows that the market correctly assesses earnings persistence. However, as the reported core earnings persistence of the shifters is manipulated, there remains the question of whether investors can correctly identify the persistence of core earnings of shifters and price them properly. When managers artificially boost core earnings by shifting expenses from core expenses in year t, these expenses are likely to recur as core expenses in year t+ 1. As a result, some of the current “higher” core earnings among classification shifters are unlikely to persist in the future, leading to lower persistence of core earnings in year t+ 1.¹⁴ We expect that when recurring expenses are buried in special items and other transitory items, the market will have greater difficulty untangling the effect of excluded expenses that are recurring and will overestimate the persistence of shifters’ core earnings. Hence, our hypothesis states the following:

Hypothesis 1 (H1): Investors do not fully disentangle classification shifting.

Hypothesis 1a (H1a): The market’s expected persistence of core earnings from period t to period t+ 1 is higher than the reported persistence of core earnings from period t to period t+ 1 for shifters.

Hypothesis 1b (H1b): The core earnings of shifters in period t are more negatively associated with future returns than the core earnings of non-shifters.

For H1a, we use the Mishkin test to assess whether the market’s perceived core earnings persistence from period t to period t+ 1 is greater than the reported persistence. For H1b, we use zero-investment portfolio analyses and multiple regression analyses as adopted in previous studies.

The Mishkin Test

For the mispricing tests, we follow Sloan (1996) and use both the Mishkin test and the zero-investment return method. We also use the multiple regression method to relate portfolios formed based on current information to future returns (e.g., Cheng & Thomas, 2006; Desai et al., 2004; Pincus, Rajgopal, & Venkatachal, 2007; Shi & Zhang, 2012). Many recent studies do not use the Mishkin test method because conceptually the Mishkin test and zero-investment hedge return method as well as other portfolio methods generate similar results (Sloan, 1996). In addition, the Mishkin test has been subject to criticism (Kraft et al., 2007). However, the Mishkin test builds upon whether the market’s perceived earnings persistence equals the actual reported earnings persistence. Thus, it is particularly pertinent for our research question because we predict that core earnings persistence will be lower after classification shifting and the market’s expected core earnings persistence will be higher than ex post realizations. In this section, we provide a discussion of this method and discuss the literature that has debated the appropriateness of the Mishkin test.

The Mishkin test uses two simultaneous equations. The first equation is an earnings prediction model which generates the earnings persistence coefficient relating current- to next-period earnings (we term it the reported persistence coefficient [RPC]). The second equation is a concurrent unexpected return-unexpected earnings model where the unexpected earnings are derived from the residual of the first equation and we can derive the expected persistence coefficient from the equation (we term it the expected persistence coefficient [EPC]). Mishkin (1983) posits that these two coefficients should not differ if the market is efficient. Sloan (1996) documents that these two coefficients are not significantly different when testing the mispricing of net income that is defined as operating income after depreciation, implying that either the market is efficient in assessing the persistence of earnings or the market is fixated on earnings. When Sloan decomposes earnings into the accrual and cash flow components, he finds that the market is not efficient in assessing the persistence of accruals and cash flows. Accordingly, he concludes that the market is fixated on earnings and cannot see through the differential persistence between accruals and cash flows. In the context of shifters, if the market sees through the shifting, then the EPC should be equal to the RPC.

It is important to note that we do not assume that earnings, accruals, and cash flows represent the full set of information available to investors. Under the null hypothesis of market efficiency, the coefficients in the forecasting equation should be equal to their corresponding coefficients in the valuation equation regardless of whether other correlated omitted variables can predict earnings (Lewellen, 2010). Therefore, the test of market efficiency remains valid even if there are correlated omitted variables. This argument is in contrast to Kraft et al. (2007), who argue that only if the omitted variables are rationally priced can one claim that the inferences of the Mishkin test are valid. They further show that accruals are rationally priced when one adds certain explanatory variables, such as lagged sales, the level of and change in sales, the level of and change in capital expenditures, lagged cash flows, and lagged accruals. Although Richardson, Tuna, and Wysocki (2010) acknowledge that such criticism is valid, they state that it does not invalidate the predictive ability of a given variable. They argue that the Mishkin test is “purely a predictive regression that places some priors on the forecasting so as to allow inferences to be made about the validity of that structure” (p. 429). Lewellen (2010) also explains that correlated omitted variables can affect the slope on a predictive variable, but they do not affect the overall test of market efficiency. In essence, the omission of a variable can affect the magnitude of the slope on our variable of interest but cannot make returns predictable when they are not. We believe that the Mishkin test has merits, especially because it will provide us with a clearer understanding of the persistence of core earnings. However, in our robustness tests, we also control for additional variables as suggested in Kraft et al. (2007).

Research Design

Measuring Income Classification Shifting

To categorize a firm as a classification shifter in any given year, we use a firm-specific measure of shifting derived from prior literature. McVay (2006) finds evidence of income classification shifting by showing a positive relation between unexpected core earnings and income-decreasing special items (Model 1b below):

\begin{matrix} CoreEar n_{t} = β_{0} + β_{1} CoreEar n_{t - 1} + β_{2} AT O_{t} + β_{3} ACC R_{t - 1} + β_{4} ACC R_{t} \\ + β_{5} Δ SALE S_{t} + β_{6} NEG_Δ SALE S_{t} + ε_{t}, \end{matrix}

U E_C E_{t} = α_{0} + α_{1} % S I_{t} + ε_{t},

where CoreEarn = operating income before depreciation divided by sales; ATO = asset turnover ratio measured as sales divided by average net operating assets (NOA). NOA is operating assets minus operating liabilities; ACCR = net income before extraordinary items minus cash flow operations divided by sales; ΔSALES = percent change in sales over prior-year sales; NEG_ΔSALES = percent change in sales (ΔSALES) if ΔSALES is less than 0, and 0 otherwise; UE_CE = unexpected core earnings measured as the difference between reported and predicted core earnings from Model 1a; %SI = special items times −1 divided by sales when special items are income-decreasing, and 0 otherwise.

In the expected core earnings Model 1a, McVay first regresses core earnings on lagged core earnings, asset turnover ratio, lagged accruals, current-period accruals, change in sales, and negative change in sales. In Model 1b, she shows that unexpected core earnings, estimated from the first model, are positively related to income-decreasing special items. She argues that this positive relation is evidence that firms engage in income classification shifting. We design our main measure of income classification shifting directly on the basis of the predicted relations between unexpected core earnings and income-decreasing special items and discontinued operations (Barua et al., 2010; McVay, 2006). We categorize a firm as a classification shifter if the firm-year observation has positive unexpected core earnings and either income-decreasing special items or income-decreasing discontinued operations. All other firm-year observations are categorized as non-classification shifters.

As our measure does not establish a direct link between positive unexpected core earnings and income-decreasing transitory items, our categorization of classification shifters may be imprecise. In untabulated analyses, we replicate McVay (2006) and Barua et al. (2010) and find that the coefficients on %SI and %DO are significantly higher for shifters than for non-shifters. This result helps validate our classification of shifters and non-shifters and provides assurance that our results are closely related to shifting. We employ several additional analyses, including defining shifters on the basis of the component of unexpected core earnings that is associated with transitory items, and analyzing mispricing when the transitory items are especially large and when the unexpected core earnings are negative. We also use two alternative specifications of the McVay (2006) model following Haw et al. (2011) to ensure that our results are robust. First, we estimate the model only for firms that have positive core earnings after removing accruals from the model to avoid spurious correlations. Second, we replace total accruals with working capital accruals, because total accruals include the special item accruals to which managers are shifting. More detailed discussion is provided in the robustness check section.

Data and Descriptive Statistics

We construct our sample with financial statement data from Compustat and stock return data from Center for Research in Security Prices (CRSP) over the period from 1988 to 2010. We delete observations with sales of less than US$1 million, firms with Standard Industrial Classification (SIC) codes 6000-6999, and firms with negative book values. Following McVay (2006), we require a minimum of 15 observations per industry per fiscal year to accurately estimate expected core earnings. We categorize industries according to Fama and French (1997). Our final sample includes 94,221 firm-year observations and 13,584 firms. We find that approximately 19% of the sample are categorized as shifters (17,902 observations) whereas about 81% of the sample are in the category of non-shifters (76,319 observations).

The existence of large income-decreasing transitory items increases the opportunity that managers have to “disguise” their core expenses through classification shifting. Therefore, to evaluate the pervasiveness of classification shifters, we focus on firms with large income-decreasing transitory items. McVay (2006) shows that special items greater than 5% of sales are associated with a significant increase in core earnings. Using 5% of sales as a benchmark to define large transitory items, we find that for firms with non-zero negative transitory items, about 62% of them have large transitory items. Among these large transitory item firms, approximately half of them are categorized as classification shifters. These results suggest that one in two firms would engage in classification shifting given the opportunity, a non-trivial number.

Table 1 provides comparative descriptive statistics for our main variables for the full sample of income classification shifters versus non-shifters. The table shows that shifters have a significantly higher level of core earnings (0.162) compared with non-shifters (0.068). Note that, by construction, many non-shifters have negative unexpected earnings, but we require shifters to have positive unexpected earnings. The table also reports that shifters have larger income-decreasing special items (0.042) than non-shifters (0.038). The amount of negative special items is slightly higher than that in prior studies on shifting (e.g., McVay reports 2.7%), but is consistent with Riedl and Srinivasan (2010), who find that the frequency and magnitude of negative special items have been significantly increasing over time. For the full sample, mean income-decreasing discontinued operations as a percentage of sales is also significantly larger for shifters (0.008) compared with non-shifters (0.005), consistent with Barua et al. (2010).

Table 1.

Descriptive Statistics.

Variable	Shifters			Non-shifters
	M	Median	SD	M	Median	SD	Difference (M)	t
CE	0.162	0.148	0.080	0.068	0.104	0.179	0.094***	72.24
UE_CE	0.031	0.020	0.037	−0.008	−0.007	0.049	0.039***	92.34
Negative SI	0.042	0.011	0.113	0.038	0.000	0.387	0.004**	2.52
Negative DO	0.008	0.000	0.051	0.005	0.000	0.243	0.003***	2.98
EARN	−0.053	0.013	0.215	0.002	0.039	0.177	−0.055***	35.70
NI	−0.062	0.010	0.258	0.004	0.039	0.209	−0.066***	35.80
CFO	0.064	0.086	0.157	0.026	0.068	0.213	0.039***	17.15
Total ACC	−0.086	−0.054	0.394	−0.062	−0.043	0.231	−0.024***	10.92
DACC	−0.044	−0.032	0.139	0.012	0.002	0.145	−0.056***	−34.54
APROD	−0.020	−0.008	0.251	0.006	0.006	0.310	−0.026***	−7.06
ADISX	0.037	0.040	0.407	−0.010	0.019	1.130	0.047***	12.14
REM	0.027	0.033	0.502	0.022	0.028	0.739	0.005	−0.93
SIZE	5.602	5.938	2.747	4.882	5.077	2.645	0.720***	24.02
SALES	6.041	6.143	2.060	5.323	5.370	2.252	0.719***	29.04
SEG	2.762	2.000	2.089	2.416	1.000	1.884	0.346***	15.92
BM	0.051	0.446	2.292	0.059	0.467	2.336	−0.008	−0.30
LEV	0.278	0.228	0.285	0.247	0.190	0.279	0.031***	9.83
HERF	0.032	0.030	0.029	0.033	0.028	0.049	−0.001**	−2.48
SG	0.143	0.060	1.028	0.317	0.086	6.605	−0.174***	−2.70

Note. There are 94,221 firm-year observations estimated over 1988-2010. Shifters (non-shifters) have 17,902 (76,319) observations. Core earnings, CE, is operating income before depreciation divided by average total assets (oibdp / ata). Unexpected core earnings, UE_CE, is the difference between reported and predicted core earnings, as determined by Equation 1 following McVay (2006). Special items, SI, are negative values divided by sales (spi *−1) / sale and 0 otherwise. Discontinued operations, DO, are negative values divided by sales (do *− 1) / sale and 0 otherwise. Earnings, EARN, is income before extraordinary items divided by average total assets (ib / ata). NI is net income divided by average total assets (ni / ata). CFO is operating cash flows divided by average total assets (oancf − xidoc) / ata. ACC, accruals, is income before extraordinary items minus CFO divided by average total assets [ib − (oancf − xidoc)] / ata. Discretionary accruals, DACC, are estimated using the modified Jones model within each two-digit SIC code each year. Abnormal Production Costs, APROD, is the residual from the following regression: $PRO D_{t} / A_{t - 1} = α_{0} + α_{1} (1 / A_{t - 1}) + α_{2} (Δ S_{t} / A_{t - 1}) + α_{3} (Δ S_{t} / A_{t - 1}) + α_{4} (Δ S_{t - 1} / A_{t - 1}) + ε_{t},$ where PROD is the sum of costs of goods sold in year t and the change in inventory from t−1 to t, A_t₋₁ is total assets in year t− 1, S_t is net sales in year t, and ΔS_t is the change in net sales from year t− 1 to t. Abnormal discretionary expenditures, ADISX, is negative 1 times the residual from the following regression: $DIS X_{t} / A_{t - 1} = α_{0} + α_{1} (1 / A_{t - 1}) + α_{2} (Δ S_{t - 1} / A_{t - 1}) + ε_{t},$ where DISX_t is the sum of research and development, advertising, and selling, general and administrative expenditures in year t. Real earnings management, REM, is measured as the sum of Abnormal Production Costs and Abnormal Discretionary. Expenditures following Roychowdhury (2006). SIZE is the natural log of price times shares outstanding (prcc_f × csho). SALES is the natural log of sales. SEG is the number of business segments. BM, book-to-market, is book equity divided by market value [ceq / (prcc_f × csho)]. LEV is the sum of long-term and current debt divided by total assets (dltt + dlc) / at. HERF is the Herfindahl index of industry concentration computed with net sales. Sales growth, SG, is the average of sales growth over most recent 3 years. All variables are winsorized at 1% and 99%. SIC = Standard Industrial Classification.

, **, and *** are significant at the 1%, 5%, and 10% level, respectively.

Shifters also have more negative total accruals (−0.086) compared with non-shifters (–0.062).¹⁵ The average size of shifting firms is significantly larger (5.602) than that of non-shifting firms (4.882). Furthermore, firms that shift also tend to have more sales (6.041) and more segments (2.762) than firms that do not (5.323 and 2.416, respectively). However, shifters appear to have significantly weaker sales growth on average (14.3%) compared with non-shifters (31.7%). As the characteristics of our categorization of shifters (positive unexpected core earnings, income-decreasing special items, and income-decreasing discontinued operations) are likely to be correlated with the size and performance of the firm, we do not directly attribute the difference in firm characteristics to classification shifting.

Table 2 reports the Pearson and Spearman correlation coefficients for our main variables for the combined sample of classification shifters and non-shifters. Pearson correlation coefficients are reported on the upper-right corner and Spearman correlation coefficients are reported on the lower-left corner. The Pearson correlation between accruals (ACC) and future returns (AR) of −0.022 is negative and significant, which is consistent with prior research on accruals overpricing. The Pearson correlation between operating cash flows (CFO) and future returns (AR) of 0.093 is positive and significant, which is also expected according to earlier findings of cash flow underpricing. The Pearson correlation between earnings (EARN) and future returns (AR) of 0.207 is positive and significant. The majority of the correlations among the independent variables are statistically significant but their magnitudes are not large. This suggests that multicollinearity should not be a concern. We investigate the issue of multicollinearity further by calculating the Variance Inflation Factor (VIF). All variables have a VIF of less than the recommended cut-off value of 10 (Kennedy, 1992).

Table 2.

Pearson/Spearman Correlation Coefficients.

	AR_t ₊₁	ACC_t	CFO_t	EARN_t	BM_t	SG_t	CE_t	NCE_t
AR_t ₊₁		−.047	.060	.265	.006^a	−.038	.034	.006^a
ACC_t	−.022		−.329	.379	−.049	.152	.091	.384
CFO_t	.093	−.026		.630	.005^b	−.063	.805	.063
EARN_t	.207	.564	.723		−.266	.146	.837	−.013
BM_t	.137	−.011	−.064	.000^b		.001^b	.000^b	−.002^b
SG_t	−.084	−.021	.043	−.084	−.250		−.025	−.025
CE_t	.047	.334	.704	.827	−.171	.245		.027
NCE_t	−.000^b	.383	−.201	.292	.197	−.049	−.343

Note. There are 94,221 firm-year observations estimated over 1988-2010. The upper-right represents Pearson correlation coefficients. The lower-left represents Spearman correlation coefficients. AR are measured as size-adjusted returns over the 12-month period beginning 3 months after the fiscal year-end. ACC, accruals, is calculated as income before extraordinary items minus cash from operations divided by average total assets [ib − (oancf − xidoc)]. EARN is income before extraordinary items divided by average total assets. BM, book-to-market, is book value divided by market value [ceq / (prcc_f × csho)]. SG, sales growth, is the average of annual growth in sales over the most recent 3 years. CE, core earnings, is operating income before depreciation divided by average total assets (oibdp). NCE, non-core earnings, is net income minus core earnings, divided by average total assets.

All correlations are significant at the 1% level except those marked with “^b” which are not significant and those with “^a” which are significant at the 10% level.

In Table 3, we present the mean regression results of the earnings expectation Model 1a. The mean adjusted R² is approximately 79%, slightly higher than McVay’s adjusted R² of 76%. All of the mean coefficients have the same sign and similar magnitude as reported by McVay (2006), which supports the consistency of her model over time. The coefficient on prior-year core earnings (CoreEarn_t_–1) is 0.777, which confirms that core earnings are highly persistent. Lagged accruals (ACC_t_–1) have a coefficient of −0.164, which shows that higher accruals contribute to lower core earnings persistence. The coefficient of current-period accruals (ACC_t) is 0.211 as expected, because extreme performance is significantly related to changes in accrual levels (DeAngelo, DeAngelo, & Skinner, 1994). In categorizing a firm as an income classification shifter, we require positive unexpected core earnings as calculated from the residual of this model as well as income-decreasing special items or income-decreasing discontinued operations.

Table 3.

Model of Expected Core Earnings.

C o r e E a r n_{t} = β_{0} + β_{1} C o r e E a r n_{t - 1} + β_{2} A T O_{t} + β_{3} A C C R_{t - 1} + β_{4} A C C R_{t} + β_{5} Δ S A L E S_{t} + β_{6} N E G_Δ S A L E S_{t} + ε_{t} .

Parameter	M coefficient	SD	t statistic	p value
Intercept	0.033***	0.066	14.85	<.0001
CoreEarn_t ₋₁	0.777***	0.219	105.26	<.0001
ATO_t	−0.002**	0.023	−2.14	0.033
ACC_t ₋₁	−0.164***	0.234	−20.72	<.0001
ACC_t	0.211***	0.309	20.21	<.0001
ΔSALES_t	0.065***	0.124	15.41	<.0001
NEG_ΔSALES_t	0.362***	0.778	13.78	<.0001
Adjusted R²	78.9%

Note. There are 109,074 firm-year observations and 876 industry-year regressions. Regressions are estimated by industry and fiscal year., core earnings, is defined as operating income before depreciation divided by sales (oibdp / sale). ATO, asset turnover ratio, is sales divided by average net operating assets (NOA). NOA is operating assets minus operating liabilities. ACCR, total accruals, is income before extraordinary items minus cash flow operations divided by sales [(ib − (oancf − xidoc)] / sale. ΔSALES is the percent change in sales over prior-year sales. NEG_ΔSALES is percent change in sales (ΔSALES) if ΔSALES is less than 0, and 0 otherwise. UE_CE, unexpected core earnings, is the difference between reported and predicted core earnings from Model 1a. All variables are winsorized at 1% and 99%.

, **, *** are significant at the 1%, 5%, and 10% level, respectively.

Results

The Mishkin Test

Our first test, the Mishkin test (1983), is used to examine whether investors correctly price core earnings for classification shifters relative to non-shifters. The Mishkin test assumes that markets are efficient, which implies that the expectation of abnormal returns equals zero:

E (R_{t + 1} - E [R_{t + 1}]) = 0 .

On the basis of the condition of market efficiency, we can infer that the market will only react to unexpected core earnings but not to the portion of core earnings that is anticipated:

E (R_{t + 1} - E [R_{t + 1}]) = β (Actual C E_{t + 1} - Forecasted C E_{t + 1}) + ε_{t + 1} .

Given these assumptions, we employ the Mishkin framework by estimating a system of equations for core earnings following Sloan (1996) for the full sample¹⁶:

C E_{t + 1} = γ_{0} + γ_{1} C E_{t} + υ_{t + 1},

A R_{t + 1} = β (C E_{t + 1} - γ *_{0} - γ *_{1} C E_{t}) + ε_{t + 1} .

Kraft et al. (2007) show that when relevant variables are not included in the forecasting and pricing equations, an omitted variable problem can lead to incorrect inferences about the rational pricing of accounting variables. In the robustness check section, we re-run our Mishkin test controlling for variables that have significant explanatory power for core earnings.

In Panel A of Table 4, we show the reported persistence coefficient (RPC) and the expected persistence coefficient (EPC) for core earnings from Equations 3 and 4 using the Mishkin test. The RPC (γ₁ = 0.755) is not significantly different from the EPC (γ*₁ = 0.756) as evidenced by the likelihood ratio statistic of 0.060. This shows that Sloan’s results hold in our sample period between 1988 and 2010. However, we predict that core earnings of shifters will have a higher EPC than RPC. The average results may be due to a large number of non-shifters in our sample. Indeed, we find about 19% of our sample firms are classified as shifters.

Table 4.

The Market Pricing of Core Earnings.

Panel A: Test of Rational Pricing of Core Earnings (Full Sample).

C E_{t + 1} = γ_{0} + γ_{1} C E_{t} + ν_{t + 1} .

A R_{t - 1} = β (C E_{t + 1} - γ *_{0} - γ *_{1} C E_{t}) + ε_{t + 1} .

Reported persistence			Expected persistence			Test of rational pricing
Parameter	Estimate	SE	Parameter	Estimate	SE	Difference	LR
γ₁(CE)	0.755***	0.002	$γ_{1}^{*}$	0.756***	0.007	0.001	0.060

Panel B: Tests of Rational Pricing of Core Earnings for Classification Shifters Versus Non-Shifters.

C E_{t + 1} = γ_{0} + γ_{1} SHIF T_{t} + γ_{2} C E_{t} + γ_{3} C E_{t} \times SHIF T_{t} + ν_{t + 1} .

A R_{t - 1} = β (C E_{t + 1} - γ_{0}^{*} - γ_{1}^{*} SHIF T_{t} - γ_{2}^{*} C E_{t} - γ_{3}^{*} C E_{t} \times SHIF T_{t}) + ε_{t + 1} .

Reported persistence			Expected persistence			Test of rational pricing
Parameter	Estimate	SE	Parameter	Estimate	SE	Difference	LR
γ₁(SHIFT)	0.008**	0.004	$γ_{1}^{*}$	−0.055***	0.012
γ₂(CE)	0.760***	0.002	$γ_{2}^{*}$	0.749***	0.008	−0.011	1.620
γ₃(CE×SHIFT)	−0.025***	0.006	$γ_{3}^{*}$	0.052***	0.019	0.077***	14.400
γ₂+γ₃	0.735***	0.004	$γ_{2}^{}$ + $γ_{3}^{}$	0.801***	0.012	0.066***	12.830

Note. Equations are jointly estimated using an iterated generalized nonlinear least squares estimation procedure. There are 94,221 observations over the time period 1988-2010. See Table 1 for variable descriptions. LR = Likelihood Ratio

, **, *** are significant at the 1%, 5%, and 10% level, respectively.

Next, we expand the Mishkin test for the benchmark core earnings model by assessing whether the market rationally prices core earnings for firms identified as income classification shifters by estimating the following system of equations:

C E_{t + 1} = γ_{0} + γ_{1} SHIF T_{t} + γ_{2} C E_{t} + γ_{3} CE \times SHIF T_{t} + υ_{t + 1},

A R_{t +1} = β (C E_{t +1} - {γ *}_{0} - {γ *}_{1} S H I F T_{t} - γ *_{2} C E_{t} - γ *_{3} C E \times S H I F T_{t}) + ε_{t +1} .

In Panel B of Table 4, the EPC (γ*₂ = 0.749) for non-shifters is not significantly different from the RPC (γ₂ = 0.760). This implies that the market does not misprice the core earnings of non-shifters. However, we find that for the interaction term CE×SHIFT, the EPC (γ*₃) is 0.052 and RPC (γ₃) is −0.025. This indicates that classification shifters have expected earnings persistence of 0.801 (0.749 + 0.052), which is significantly greater than their reported earnings persistence of 0.735 (0.760 − 0.025). These results confirm our prediction in H1a that the market’s expected persistence of core earnings is higher than the firm’s reported persistence. Alternatively speaking, the market does not see through shifting because it does not anticipate the lower persistence associated with core earnings that have been artificially inflated by shifting recurring expenses to either the special item or discontinued operation.

Zero-Investment Portfolio Results

Our second test calculates the difference in returns on a zero-investment portfolio between classification shifters and non-shifters. Our trading strategy is to take a long position in firms in the most negative decile of core earnings in year t and a short position in firms in the most positive decile of core earnings in year t after separating the firms into shifters and non-shifters. The portfolio returns are calculated as follows:

Shifter Portfolio Retur n_{t + 1} = Low Decile Retur n_{t + 1} - High Decile Retur n_{t + 1},

Non - Shifter Portfolio Retur n_{t + 1} = Low Decile Retur n_{t + 1} - High Decile Retur n_{t + 1},

Incremental Retur n_{t + 1} = Shifter Portfolio Retur n_{t + 1} - Non - Shifter Portfolio Retur n_{t + 1} .

In Table 5, we examine whether there is a difference in the zero-investment portfolio returns between the shifter portfolio and the non-shifter portfolio. In Panel A, we group firms into portfolio deciles each year on the basis of their ranking of core earnings. We form zero-investment portfolios by establishing a long position in the most negative core earnings decile and a short position in the most positive core earnings decile for both shifters and non-shifters. We report the average of the 23 annual abnormal size-adjusted returns for the lowest and highest core earnings deciles over the period between 1988 and 2010, as well as the zero-investment return for each group.

Table 5.

Relation Between Core Earnings and Future Abnormal Returns for Classification Shifters Versus Non-Shifters.

Panel A: Univariate Analysis Between Core Earnings and Future Returns.

Shifter Portfolio Retur n_{t + 1} = Low Decile Retur n_{t + 1} - High Decile Retur n_{t + 1} .

Non - Shifter Portfolio Retur n_{t + 1} = Low Decile Retur n_{t + 1} - High Decile Retur n_{t + 1} .

Incremental Retur n_{t + 1} = Shifter Portfolio Retur n_{t + 1} - Non - Shifter Portfolio Retur n_{t + 1} .

	Shifters		Non-shifters		Difference
	M	SE	M	SE	M
Core earnings
Lowest decile	0.062***	0.073	−0.056***	0.011	0.118***
Highest decile	−0.086***	0.009	−0.078***	0.005	−0.008
Zero-investment return	0.148***	0.051	0.022**	0.011	0.126***

Panel B: Multivariate Analysis Between Core Earnings and Future Returns.

A R_{t + 1} = ρ_{0} + ρ_{1} SHIF T_{t} + ρ_{2} {(- CE)}_{t} + ρ_{3} {(- CE \times SHIFT)}_{t} + ρ_{4} B M_{t} + ρ_{5} SG + ρ_{6} \frac{CFO}{P_{t}} + ε_{t} .

	Coefficient	SE	Coefficient	SE
Intercept	−0.057***	0.004	−0.144***	0.005
SHIFT	0.064***	0.011	0.026**	0.011
CE	−0.001	0.006	−0.013*	0.008
CE×SHIFT	0.092***	0.016	0.034**	0.017
BM			0.189***	0.007
SG			−0.358***	0.057
CFO / P			−0.009	0.008

Note. 1988-2010. There are 94,221 observations in Panel A. The average n in Panel B is 3,377. SHIFT is 1 if positive unexpected core earnings and negative special items or negative discontinued operations, 0 otherwise. CE, core earnings, is operating income before depreciation divided by average total assets (oibdp / ata). BM is book value of equity divided by market value of equity [ceq / (prcc_f × csho)]. SG, sales growth, is the average of annual growth in sales over most recent 3 years. CFO / P is operating cash flows divided by market value of equity [oancf − xidoc / (prcc_f × csho)]. Zero-investment portfolio returns in year t+ 1 are calculated by subtracting the average size-adjusted return of firms in the highest decile (i.e., short position) from the average size-adjusted return of firms in the lowest decile (i.e., long position). All independent variables are decile ranked from 0 to 9 and scaled by 9. p values are based on the time series standard errors of the coefficient estimates. Standard errors are corrected for autocorrelation using the Newey–West procedure.

, **, and *** are significant at 1%, 5%, and 10% level, respectively.

In support of our H1b, we find that the zero-investment portfolio return of 14.8% for shifters is positive and significant at the 1% level, which is consistent with the overpricing of shifters’ core earnings from the Mishkin test.¹⁷ The zero-investment return of 2.2% for non-shifters is also positive and significant, albeit at the 5% level. However, this return is not likely to be economically significant after transaction costs.¹⁸ The difference in portfolio returns between shifters and non-shifters is 12.6% indicating that the mispricing of shifters’ core earnings is likely to be more severe than that of non-shifters. These univariate results are consistent with our findings from the Mishkin test on the mispricing of shifters’ core earnings.

Multivariate Regression Results

Our third test uses a multivariate regression to analyze the mispricing of classification shifting using future stock returns while controlling for cross-sectional differences in risk-related anomalies. If there is a correlation between the core earnings of classification shifters and cross-sectional differences in risk, then it is possible that our results from the Mishkin test and the zero-investment portfolio analysis represent underlying changes in risk and not actual mispricing. We use the following multivariate regression model:

A R_{t + 1} = ρ_{0} + ρ_{1} SHIF T_{t} + ρ_{2} (- C E_{t}) + ρ_{3} (- CE \times SHIF T_{t}) + ρ_{4} B M_{t} + ρ_{5} S G_{t} + ρ_{6} \frac{CFO}{P_{t}} + ε_{t} .

In Panel B of Table 5, we use standardized decile ranking and report the mean coefficient across 23 annual cross-sectional regressions over the period from 1988 to 2010, using the Fama and MacBeth (1973) t statistics corrected for autocorrelation with the Newey–West procedure. We also use the Desai et al. (2004) framework and control for book-to-market ratio, sales growth, and operating cash flows-to-price.¹⁹ Our main variable of interest is the interaction term of core earnings and shifters (CE×SHIFT). Consistent with our H1b, we find that the coefficient on the interaction term is positive and significant at the 1% level (0.034) after including control variables. This signifies that investors are surprised that the shifters’ core earnings in year t+ 1 are lower than expected, resulting in a negative abnormal return. In contrast, the coefficient for non-shifters’ core earnings is negative and significant at the 10% level (−0.013), suggesting that the market does not overprice the core earnings of non-shifters.

Accruals Management, Real Earnings Management, and Special Items

In our main analyses, we classify a firm as a shifter if it has positive unexpected core earnings and either income-decreasing special items or discontinued operations. However, firms could have abnormally high core earnings due to other earnings management tactics. In addition to classification shifting, it is likely that abnormally high core earnings can be due to accruals management, real earnings management, or a combination of the three strategies. To mitigate the concern that our main results are not attributable to shifting but rather to firms engaging in accruals management and real earnings management, we add controls for these two management tactics in our mispricing tests of shifters. Moreover, previous studies (Burgstahler et al., 2002; Doyle, Lundholm, & Soliman, 2003) suggest a relation between income-decreasing special items and future abnormal returns; as classification shifters are special item firms, for completeness, we also add controls for special items.²⁰ We therefore augment our multiple regression model as follows:

\begin{matrix} A R_{t + 1} = ρ_{0} + ρ_{1} SHIF T_{t} + ρ_{2} {(- CE)}_{t} + ρ_{3} {(- CE \times SHIFT)}_{t} + ρ_{4} B M_{t} \\ + ρ_{5} S G_{t} + ρ_{6} \frac{CFO}{P_{t}} + ρ_{7} {(- DACC)}_{t} + ρ_{8} {(- DACC \times SHIFT)}_{t} \\ + ρ_{9} {(- REM)}_{t} + ρ_{10} {(- REM \times SHIFT)}_{t} + ρ_{11} S I_{t} + ε_{t + 1} . \end{matrix}

In columns (1) and (2) of Table 6, we find that the coefficient on the interaction term (CE×SHIFT) is still positive and significant when controlling for accruals management (DACC), 0.042, and real earnings management (REM), 0.064.²¹ These results suggest that the mispricing of the core earnings of shifters is not subsumed by any mispricing of accruals management and real earnings management. In column (3) of Table 6, we find that controlling for special items increases the coefficient on CE×SHIFT to 0.059 (from 0.034 in Table 5), implying that for some observations, special items have a positive association with future abnormal returns. This is consistent with Dechow and Ge (2006) who find that special items for the lowest decile of accruals have a positive relationship with future abnormal returns. In column (4) of Table 6, we show that our results continue to be robust after controlling for the combination of accruals management, real earnings management, and special items.²²

Table 6.

Multivariate Regressions of Future Returns on Core Earnings Controlling for Accruals Management, Real Earnings Management, and Special Items.

\begin{matrix} A R_{t + 1} = ρ_{0} + ρ_{1} SHIF T_{t} + ρ_{2} {(- CE)}_{t} + ρ_{3} {(- CE \times SHIFT)}_{t} + ρ_{4} B M_{t} + ρ_{5} S G_{t} + ρ_{6} \frac{CFO}{P_{t}} + ρ_{7} {(- DACC)}_{t} \\ + ρ_{8} {(- DACC \times SHIFT)}_{t} + ρ_{9} {(- REM)}_{t} + ρ_{10} {(REM \times SHIFT)}_{t} + ρ_{11} S I_{t} + ε_{t + 1} . \end{matrix}

	(1)	(2)	(3)	(4)
	Coefficient	Coefficient	Coefficient	Coefficient
Intercept	0.002	0.012	−0.019	0.023
SHIFT	0.065***	0.101***	0.082***	0.114***
CE	0.188***	0.185***	0.172***	0.183***
CE×SHIFT	0.042**	0.064**	0.059**	0.061**
BM	−0.006***	−0.065***	−0.051***	−0.006***
SG	0.150***	0.157***	0.148*	0.139***
CFO / P	0.054*	0.051*	0.042	0.041
DACC	0.018			0.016
DACC×SHIFT	0.002			0.021
REM		0.037*		0.034
REM×SHIFT		0.044**		0.040**
SI			0.051	0.086

Note. The maximum number of observations is 74,975 over 1988-2010. Flexible sample is used for each regression. SHIFT is 1 if positive unexpected core earnings and negative special items or negative discontinued operations, 0 otherwise. CE, core earnings, is operating income before depreciation divided by average total assets. BM is book value of equity divided by market value of equity. SG, sales growth, is the average of annual growth in sales over most recent 3 years. CFO / P is operating cash flows divided by market value of equity. REM, real earnings management, is measured as the sum of the abnormal level of production costs and the abnormal level of discretionary expenditures (Roychowdhury, 2006). DACC, discretionary accruals, are measured using the cross-sectional modified Jones model (Dechow, Sloan, & Sweeney, 1995) within each two-digit SIC code each year. SI is negative special items scaled by sales, where income-increasing items have been set to 0. All independent variables are decile ranked from 0 to 9 and scaled by 9. SIC = Standard Industrial Classification.

, **, and *** are significant at 1%, 5%, and 10% level, respectively.

Robustness Check and Additional Analyses

Kraft et al. (2007) show that the Mishkin test can bias tests of market efficiency when pertinent variables are omitted from the forecasting equation. They state that the exclusion of such variables is irrelevant only if the omitted variables are rationally priced. Specifically, they show that when additional explanatory variables are added to the forecasting and pricing equations, the accrual anomaly disappears. Therefore, we test the robustness of our results by adding similar explanatory variables to the Mishkin framework below to ascertain whether our results hold:

\begin{matrix} C E_{t + 1} = σ_{0} + σ_{1} SHIF T_{t} + σ_{2} C E_{t} + σ_{3} CE \times SHIF T_{t} + σ_{4} A R_{t} + σ_{5} SALE S_{t} \\ + σ_{6} Δ SALE S_{t} + σ_{7} CAPE X_{t} + σ_{8} Δ CAPE X_{t} + σ_{9} NO A_{t} + υ_{t}, \end{matrix}

A R_{t + 1} = β (\begin{matrix} C E_{t + 1} - σ_{0}^{*} + σ_{1}^{*} SHIF T_{t} + σ_{2}^{*} C E_{t} + σ_{3}^{*} CE * SHIF T_{t} + σ_{4}^{*} A R_{t} + \\ σ_{5}^{*} SALE S_{t} + σ_{6}^{*} Δ SALE S_{t} + σ_{7}^{*} CAPE X_{t} + σ_{8}^{*} Δ CAPE X_{t} + σ_{9}^{*} NO A_{t} \end{matrix}) + ε_{t} .

In Table 7, we show that our results continue to hold even after controlling for current-period stock returns, sales, sales growth, capital expenditures, change in capital expenditures, and net operating assets. We find that classification shifters have expected earnings persistence of 0.660 (0.588 + 0.072), which is significantly greater than their reported earnings persistence of 0.594 (0.614 − 0.020). For non-shifters, we find that the market underestimates the persistence of core earnings.²³ This provides further confirmation that the market does not see through shifting because it does not anticipate the lower persistence associated with core earnings.

Table 7.

The Market Pricing of Core Earnings for Classification Shifters With Additional Explanatory Variables.

\begin{matrix} C E_{t + 1} = σ_{0} + σ_{1} SHIF T_{t} + σ_{2} C E_{t} + σ_{3} CE \times SHIF T_{t} + σ_{4} A R_{t} + σ_{5} SALE S_{t} + σ_{6} Δ SALE S_{t} \\ + σ_{7} CAPE X_{t} + σ_{8} Δ CAPE X_{t} + σ_{9} NO A_{t} + υ_{t} . \end{matrix}

A R_{t + 1} = β (\begin{matrix} C E_{t + 1} - σ_{0}^{*} + σ_{1}^{*} SHIF T_{t} + σ_{2}^{*} C E_{t} + σ_{3}^{*} CE \times SHIF T_{t} + σ_{4}^{*} A R_{t} + σ_{5}^{*} SALE S_{t} + \\ σ_{6}^{*} Δ SALE S_{t} + σ_{7}^{*} CAPE X_{t} + σ_{8}^{*} Δ CAPE X_{t} + + σ_{9}^{*} NO A_{t} \end{matrix}) + ε_{t} .

Reported persistence			Expected persistence			Test of rational pricing
Parameter	Estimate	SE	Parameter	Estimate	SE	Difference	LR
σ₁(SHIFT)	−0.015**	0.006	$σ_{1}^{*}$	−0.069	0.017	−0.054
σ₂(CE)	0.614***	0.002	$σ_{2}^{*}$	0.588***	0.008	−0.026	9.75
σ₃(CE×SHIFT)	−0.020**	0.008	$σ_{3}^{*}$	0.072***	0.018	0.092***	10.99
σ₄(AR)	0.095***	0.002	$σ_{4}^{*}$	0.109***	0.008	0.014***	8.63
σ₅(SALES)	0.005***	0.001	$σ_{5}^{*}$	−0.009***	0.008	−0.014***	261.61
σ₆(ΔSALES)	0.003***	0.001	$σ_{6}^{*}$	0.017***	0.008	0.014***	251.08
σ₇(CAPEX)	0.007***	0.001	$σ_{7}^{*}$	0.007***	0.008	0.000	0.20
σ₈(ΔCAPEX)	−0.003***	0.001	$σ_{8}^{*}$	0.007***	0.008	0.010***	139.60
σ₉(NOA)	−0.010***	0.001	$σ_{9}^{*}$	0.025***	0.008	0.035***	302.38
σ₂+σ₃	0.594***	0.004	$σ_{2}^{}$ + $σ_{3}^{}$	0.660***	0.015	0.066**	5.91

Note. Equations are jointly estimated using an iterated generalized nonlinear least squares estimation procedure based on 75,015 observations from 1988 to 2010. All independent variables are decile ranked from 0 to 9, scaled by 9. AR is measured as size-adjusted returns over the 12-month period beginning 3 months after the fiscal year-end. SHIFT is defined as firms with positive unexpected core earnings and negative special items or negative discontinued operations. CE, core earnings, is operating income before depreciation (oibdp) divided by average total assets. Sales and Change in Sales are scaled by average total assets. Capital expenditures and Change in Capital Expenditures (capx) are scaled by average total assets. NOA is operating assets minus operating liabilities scaled by average total assets.

, **, *** are significant at the 1%, 5%, and 10% level, respectively.

Propensity-Score Matched Model

As managers intentionally choose to engage in classification shifting, it is possible that our main results are biased due to self-selection and endogeneity. We address the concerns about these endogeneity issues by using a propensity-score matched analysis to test the mispricing of core earnings for classification shifters. In Panel A of Table 8, we estimate propensity scores using a first-stage probit model for firms that are categorized as classification shifters regressed on a set of firm characteristic variables. The propensity-score matching procedure produces a matched sample of 7,860 control observations leading to a combined sample of 15,720 observations.

Table 8.

Propensity-Score Matched Model.

Panel A: First-Stage Probit Model for Income Classification Shifting.

\begin{matrix} SHIF T_{t} = λ_{0} + λ_{1} \frac{CFO}{P_{t}} + λ_{2} AC C_{t} + λ_{3} AAC C_{t} + λ_{4} PERSISTENC E_{t - 1} + λ_{5} SIZ E_{t} + λ_{6} SALE S_{t} \\ + λ_{7} SEGMENT S_{t} + λ_{8} B M_{t} + λ_{9} LEVERAG E_{t} + λ_{10} HER F_{t} + λ_{11} RO A_{t} + λ_{12} S G_{t} + ε_{t} . \end{matrix}

	Coefficient	SE	p value
Intercept	−2.3133***	0.0423	<.0001
CFO / P	2.0091***	0.1453	<.0001
ACC	0.1026***	0.0291	0.0004
AACC	−1.5914***	0.1191	<.0001
PERSISTENCE	0.1169***	0.0278	<.0001
SIZE	0.0731***	0.0096	<.0001
SALES	0.0902***	0.0113	<.0001
SEGMENTS	0.0281***	0.0063	<.0001
BM	0.0000	0.0000	0.6896
LEVERAGE	0.2874***	0.0486	<.0001
HERF	−0.5625**	0.2736	0.0398
ROA	−1.8927***	0.1204	<.0001
SG	−0.1195***	0.0273	<.0001
R ²	8.67%

Note. There are 40,103 firm-year observations estimated over 1988-2010. SHIFT equals 1 if firm has positive unexpected core earnings and negative special items or negative discontinued operations, and 0 otherwise. CFO / P is operating cash flows divided by market value of equity [(oancf 2 xidoc) / (prcc_f 3 csho]. ACC, equals income before extraordinary items less CFO [ib 2 (oancf 2 xidoc)]. AACC, abnormal accruals, are estimated using the modified Jones model within each two-digit SIC code each year. PERSISTENCE equals 1 for high earnings persistence and 0 for low earnings persistence in the year prior to shifting. Earnings persistence is calculated from the coefficient on core earnings (CEt–1) estimated for each firm over rolling 10-year windows. Firms with earnings persistence greater (lower) than the median are coded as high (low) persistence for year t 2?1. SIZE is the natural logarithm of price times shares outstanding (prcc_f 3 csho). SALES is the natural log of sales. SEGMENTS is the number of business segments. BM, book-to-market, is stockholders’ book equity divided by market value [ceq / (prcc_f 3 csho)]. LEVERAGE is the sum of long-term and current debt, divided by total assets [(dltt + dlc) / at]. HERF is the Herfindahl index of industry concentration computed with net sales. ROA is income before extraordinary items scaled by average total assets. SG is the average of annual growth in sales over most recent 3 years. All continuous variables are winsorized at 1% and 99%. SIC = Standard Industrial Classification.

Panel B: 2nd Stage Model for the Market Pricing of Core Earnings for Classification Shifters.

C E_{t + 1} = θ_{0} + θ_{1} SHIF T_{t} + θ_{2} C E_{t} + θ_{3} CE \times SHIF T_{t} + υ_{t + 1} .

A R_{t + 1} = β (C E_{t + 1} - θ_{0}^{*} - θ_{1}^{*} SHIF T_{t} - θ_{2}^{*} C E_{t} - θ_{3}^{*} C E_{t} \times SHIF T_{t}) + ε_{t + 1} .

Reported persistence			Expected persistence			Test of rational pricing
Parameter	Estimate	SE	Parameter	Estimate	SE	Difference	LR
θ₁(SHIFT)	0.122*	0.063	$θ_{1}^{+}$	−0.596**	0.239
θ₂(CE)	0.794***	0.006	$θ_{2}^{+}$	0.795***	0.021	0.002	0.010
θ₃(CE×SHIFT)	−0.012	0.011	$θ_{3}^{+}$	0.091**	0.040	0.103**	6.200
θ₂+θ₃	0.782***	0.009	$θ_{2}^{+}$ + $θ_{3}^{+}$	0.886***	0.032	0.104***	8.850

Note. Equations are jointly estimated using an iterated generalized nonlinear least squares estimation procedure based on 15,720 observations from 1988 to 2010. All independent variables are decile ranked from 0 to 9, scaled by 9. CE, core earnings, is operating income before depreciation (oibdp) divided by average total assets. AR is measured as size-adjusted returns over the 12-month period beginning 3 months after the fiscal year-end. SHIFT is 1 if firms have positive unexpected core earnings and negative special items or negative discontinued operations in year t but not year t− 1 or year t+ 1, 0 otherwise.

, **, and *** are significant at the 1%, 5%, and 10% level, respectively.

In Panel B of Table 8, we evaluate the second-stage model after matching on propensity scores. We find that for the interaction term CE×SHIFT, the EPC $(θ_{3}^{*})$ is 0.091 and RPC (θ₃) is −0.012. This indicates that classification shifters have expected earnings persistence of 0.886 (0.795 + 0.091), which is significantly greater than their reported earnings persistence of 0.782 (0.794 − 0.012). These results suggest that it is not likely that underlying firm characteristics are driving investors’ mispricing of core earnings for firms engaging in income classification shifting.

Shifting Only in Year t, Not in Year t− 1 or Year t+ 1

If a firm shifts expenses for more than 1 year, it is more difficult to detect the shifting behavior. For example, if there is shifting in year t and year t + 1, the core earnings will continue to be inflated in year t+ 1 and will not reverse until year t + 2. Therefore, our main measure of shifting will contain noise. We address this issue by isolating the shifting behavior and analyzing the mispricing of core earnings when a firm shifts in year t but not in year t− 1 or year t+ 1.

In Table 9, we show that for the interaction term CE×SHIFT, the EPC (ϕ^*₃) is 0.023 and RPC (ϕ₃) is −0.042. This indicates that classification shifters have expected earnings persistence of 0.782 (0.759 + 0.023), which is significantly greater than their reported earnings persistence of 0.719 (0.761 − 0.042), suggesting the overpricing of shifters’ core earnings. Therefore, our more refined measure of shifting leads to results that are consistent with our earlier evidence of the market’s mispricing of the core earnings of shifters.

Table 9.

The Market Pricing of Core Earnings for Classification Shifters (No shifting in year t– 1 and year t+ 1).

C E_{t + 1} = φ_{0} + φ_{1} SHIF T_{t} + φ_{2} C E_{t} + φ_{3} CE \times SHIF T_{t} + υ_{t + 1} .

A R_{t + 1} = β (C E_{t + 1} - φ_{0}^{*} - φ_{1}^{*} SHIF T_{t} - φ_{2}^{*} C E_{t} - φ_{3}^{*} C E_{t} \times SHIF T_{t}) + ε_{t + 1} .

Reported persistence			Expected persistence			Test of rational pricing
Parameter	Estimate	SE	Parameter	Estimate	SE	Difference	LR
ϕ₁(SHIFT)	0.038***	0.004	$φ_{1}^{*}$	0.014	0.013
ϕ₂(CE)	0.761***	0.002	$φ_{2}^{*}$	0.759***	0.010	−0.002	0.060
ϕ₃(CE×SHIFT)	−0.042***	0.007	$φ_{3}^{*}$	0.023	0.022	0.065**	6.540
ϕ₂+ϕ₃	0.719***	0.005	$φ_{2}^{}$ + $φ_{3}^{}$	0.782***	0.019	0.063***	6.780

Note. Equations are jointly estimated using an iterated generalized nonlinear least squares estimation procedure based on 54,039 observations from 1988 to 2010. All independent variables are decile ranked from 0 to 9, scaled by 9. CE, core earnings, is operating income before depreciation (oibdp) divided by average total assets. AR is measured as size-adjusted returns over the 12-month period beginning 3 months after the fiscal year-end. SHIFT is 1 if firms have positive unexpected core earnings and negative special items or negative discontinued operations in year t but not year t− 1 or year t+ 1, 0 otherwise.

, **, and *** are significant at the 1%, 5%, and 10% level, respectively.

Validity of the McVay (2006) Model

McVay (2006) acknowledges that there are limitations to the model, because it includes contemporaneous accruals to control for current performance in Model 1a. Current-period accruals are included because special item firms are more likely to have extreme negative performance. She notes that the positive relation between unexpected core earnings and negative special items in Model 1b could be biased because some of the negative special item accruals are included in current-period accruals that are used to estimate unexpected core earnings. To address this concern, she removes accruals from the model and re-estimates the model for a subset of firms having positive core earnings, and shows that her results hold. She concludes that her evidence is more consistent with classification shifting than with model bias. Therefore, we use the McVay model to estimate unexpected core earnings, consistent with other recent studies (Barua et al., 2010; Haw et al., 2011). We follow Haw et al. (2011) in estimating two variations of the McVay model to ascertain whether our main results hold.

First, we remove current-period accruals and lagged accruals from the core earnings expectation Model 1a. We re-estimate the model for only those firms that have positive core earnings. This is because McVay shows that her results of classification shifting are robust to an expectation model without accruals for firms with only positive core earnings. This is important because the reason that accruals are included in the expectations model is to control for performance as special item firms tend to be underperformers. Second, we replace accruals with working capital accruals²⁴ (current and lagged) in the core earnings expectation model. Even though working capital accruals include write-offs of inventory and accounts receivable, they will not include write-offs of goodwill, write-offs of long-term assets, and losses on sales of long-term assets. The analysis (untabulated) confirms our main mispricing results for income classification shifters and suggests that our findings are not sensitive to the specification of the core earnings expectation model.

Firms With Large Transitory Items

Using regression models such as McVay (2006) or other extended models, we may misclassify non-shifters into shifters (or vice versa). For classification shifting to be effective, the transitory items (i.e., special items and discontinued operations) are likely to be large; hence, large special items are more likely to be associated with income shifting. For example, McVay reports that the estimated mean shifted amount per firm-year is US$1.66 million or 3 cents per share for firms with negative special items of at least 5% of sales. To ascertain that the mispricing of core earnings that we document is likely to be driven by classification shifting, we investigate whether our results are stronger for firms with large amounts of transitory items using the following model:

\begin{matrix} A R_{t + 1} = ρ_{0} + ρ_{1} SHIF T_{t} + ρ_{2} {(- CE)}_{t} + ρ_{3} {(- CE \times SHIFT)}_{t} + ρ_{4} B M_{t} + ρ_{5} S G_{t} \\ + ρ_{6} \frac{CFO}{P_{t}} + ρ_{7} {(- DACC)}_{t} + ρ_{8} {(- DACC \times SHIFT)}_{t} + ρ_{9} {(- REM)}_{t} \\ + ρ_{10} {(- REM \times SHIFT)}_{t} + ρ_{11} S I_{t} + ρ_{12} LG_N C_{t} \\ ρ_{13} {(- CE \times SHIFT \times LG_NC)}_{t} + ε_{t + 1}, \end{matrix}

where LG_NC equals 1 if the sum of income-decreasing special items and discontinued operations is greater than 5% of sales, and 0 otherwise.

In column (1) of Table 10, we find that the coefficient (0.168) on the interaction term (CE×SHIFT×LG_NC) is positive and significant at the 1% level, consistent with our expectation that larger special items are more likely to be related to shifting, hence, more mispriced.

Table 10.

Validity Tests of the Classification Shifting Measure.

\begin{matrix} A R_{t + 1} = ρ_{0} + ρ_{1} SHIF T_{t} + ρ_{2} {(- CE)}_{t} + ρ_{3} {(- CE \times SHIFT)}_{t} + ρ_{4} B M_{t} + ρ_{5} S G_{t} + ρ_{6} \frac{CFO}{P_{t}} + ρ_{7} {(- DACC)}_{t} + ρ_{8} {(- DACC \times SHIFT)}_{t} \\ + ρ_{9} {(- REM)}_{t} + ρ_{10} {(- REM \times SHIFT)}_{t} + ρ_{11} S I_{t} + ρ_{12} LG_N C_{t} + ρ_{13} {(- CE \times SHIFT \times LG_NC)}_{t} \\ + ρ_{14} {(- CE \times LGS I_{t} \times SM_NEG_UECE)}_{t} + ρ_{15} {(- CE \times LGS I_{t} \times LG_NEG_UECE)}_{t} + ε_{t + 1} . \end{matrix}

	Original measure of shifting			Alternative measure of shifting
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Coefficient	Coefficient	Coefficient	Coefficient	Coefficient	Coefficient	Coefficient	Coefficient
Intercept	0.027	0.024	0.026	−0.094**	0.025	0.026	0.026	0.026
SHIFT	0.110***	0.113***	0.108***	0.092***	0.132***	0.129***	0.132***	0.128***
CE	0.184***	0.184***	0.184***	0.076	0.186***	0.185***	0.187***	0.185***
CE×SHIFT	0.077***	0.060**	0.075***	0.043**	0.079***	0.087***	0.079***	0.085***
BM	−0.056***	−0.056***	−0.056***	0.221***	−0.065***	−0.066***	−0.065***	−0.065***
SG	0.135***	0.139***	0.135***	−0.058***	0.133***	0.130***	0.133***	0.131***
CFO / P	0.041*	0.041	0.041*	0.008	0.047	0.047*	0.047	0.047*
DACC	0.016	0.015	0.016		0.017	0.017	0.017	0.017
DACC×SHIFT	0.020	0.021	0.020		−0.002	−0.003	−0.002	−0.003
REM	0.034	0.034	0.034		0.033	0.033	0.033	0.033
REM×SHIFT	0.033**	0.040**	0.032*		0.023	0.018	0.023	0.018
SI	0.085	0.085	0.085		0.100	0.108	0.100	0.108
LG_NC	−0.021		−0.018			−0.008		−0.004
CE×SHIFT×LG_NC	0.168***		0.164***			0.086*		0.088*
CE×LGSI×SM_NEG_UECE		−0.103	−0.048				−0.138	−0.104
CE×LGSI×LG_NEG_UECE		−0.138	−0.092				−0.371	−0.342

Note. The maximum number of observations is 74,975 over 1988-2010. Flexible sample is used for each regression. SHIFT is 1 if positive unexpected core earnings and negative special items or negative discontinued operations, 0 otherwise. CE, core earnings, is operating income before depreciation divided by average total assets. LG_NC equals 1 if the sum of negative special items and negative discontinued operations is greater than 5% of sales, and 0 otherwise. BM is book value of equity divided by market value of equity. SG, sales growth, is the average of annual growth in sales over most recent 3 years. CFO / P is operating cash flows divided by market value of equity. DACC, discretionary accruals, are measured using the cross-sectional modified Jones model (Dechow, Sloan, & Sweeney, 1995) within each two-digit SIC code each year. REM, real earnings management, is measured as the sum of abnormal level of production costs and abnormal level of discretionary expenditures (Roychowdhury, 2006). SI is negative special items scaled by sales, where income-increasing items have been set to 0. LG (SM)_NEG_UECE equals 1 if the unexpected core earnings is (less than) greater than 5% of sale, and 0 otherwise. All independent variables are decile ranked from 0 to 9 and scaled by 9. SIC = Standard Industrial Classification.

, **, and *** are significant at 1%, 5%, and 10% level, respectively.

Firms With Negative Unexpected Core Earnings

It is also possible that shifters may have negative unexpected core earnings based on our regression model; hence, our measure could misclassify shifters into non-shifters. This could happen if these firms have very poor performance and the model that we use does not fully control for performance. For robustness, we need to show that there is no mispricing for firms with negative unexpected core earnings. Accordingly, we also examine whether the market misprices core earnings for the sample of firms with negative unexpected earnings. As the misclassification is likely to occur when the negative unexpected core earnings are small in magnitude, we create a dummy variable (SM_NEG_UECE) indicating that negative unexpected core earnings are in the lowest decile and interact this variable with core earnings (CE) and large special items (LGSI = 1 if special items are greater than 5% of sales). In addition, if mispricing exists for negative unexpected earnings, it is more likely to occur when the negative unexpected earnings are large. Therefore, we also create a dummy variable (LG_NEG_UECE) indicating that the negative unexpected core earnings are in the highest decile and interact this variable with core earnings (CE) and large special items (LGSI). Specifically, we estimate the following model:

\begin{matrix} A R_{t + 1} = ρ_{0} + ρ_{1} SHIF T_{t} + ρ_{2} {(- CE)}_{t} + ρ_{3} {(- CE \times SHIFT)}_{t} + ρ_{4} B M_{t} + ρ_{5} S G_{t} \\ + ρ_{6} \frac{CFO}{P_{t}} + ρ_{7} {(- DACC)}_{t} + ρ_{8} {(- DACC \times SHIFT)}_{t} + ρ_{9} {(- REM)}_{t} \\ + ρ_{10} {(- REM \times SHIFT)}_{t} + ρ_{11} {(SI)}_{t} + ρ_{12} LG_N C_{t} \\ + ρ_{13} {(- CE \times SHIFT \times LG_NC)}_{t} + ρ_{14} {(- CE \times LGS I_{t} \times SM_NEG_UECE)}_{t} \\ + ρ_{15} {(- CE \times LGS I_{t} \times LG_NEG_UECE)}_{t} + ε_{t + 1} . \end{matrix}

In column (2) of Table 10, we find that the interaction terms (CE×LGSI×SM_NEG_UECE) and (CE×LGSI×LG_NEG_UECE) are both insignificant, indicating that there is no mispricing for firms with small or large negative unexpected core earnings and with large special items. When we combine our analyses for both large transitory items and negative unexpected core earnings as reported in column (3) of Table 10, we continue to find stronger mispricing results for firms with large transitory items and no mispricing for firms with negative unexpected core earnings. Taken together, we find no evidence of mispricing for firms with either small or large negative unexpected core earnings, suggesting that any misclassification of shifters as non-shifters should not significantly change our inferences.

Use Relevant Component of Unexpected Core Earnings to Define Shifters

To ensure that our definition of shifters based on unexpected positive core earnings is closely related to the special items or discontinued operations that managers use for shifting, we redefine shifters using a two-step procedure. First, we obtain the unexpected core earnings from the residuals of the McVay (2006) core earnings expectation model. Second, we modify the second-stage regression model as in Barua et al. (2010) and use it as an unexpected core earnings model. We classify a firm as a shifter if the firm has positive predicted unexpected core earnings and income-decreasing special items or income-decreasing discontinued operations using the following models:

\begin{matrix} CoreEar n_{t} = β_{0} + β_{1} CoreEar n_{t - 1} + β_{2} AT O_{t} + β_{3} ACC R_{t - 1} + β_{4} ACC R_{t} \\ + β_{5} Δ SALE S_{t} + β_{6} NEG_Δ SALE S_{t} + ε_{t}, \end{matrix}

U E_C E_{t} = α_{0} + α_{1} % S I_{t} + α_{2} % D O_{t} + α_{3} S I Z E_{t} + α_{4} B M_{t} + α_{5} A C C R_{t} + α_{6} O C F_{t} + α_{7} R O A_{t} + ε_{t} .

We find that all of the results for main analyses and the robustness tests are consistent with those of using the original measure as reported in columns (4) to (8). These results suggest that our main finding of the incremental mispricing of core earnings for classification shifters is not likely to be driven by mismeasurement of shifters.

Conclusion

In this study, we examine whether the market misprices core earnings for classification shifters. We categorize firms as classification shifters if they have, in any given year, positive unexpected core earnings and income-decreasing special items or income-decreasing discontinued operations. Firms that engage in income classification shifting tend to shift recurring operating expenses downward on the income statement to non-recurring accounts to boost the level of core earnings. As core expenses are temporarily being shifted out, the persistence of core earnings will be lower compared with non-shifters as these core expenses will recur in the subsequent period. Therefore, we expect a greater level of mispricing of core earnings for shifters relative to non-shifters.

We use the Mishkin test as one of our main tests to investigate whether the market overprices the core earnings of income classification shifters. We find that investors do not see through the lower persistence of core earnings of classification shifters and tend to overprice their core earnings more relative to non-shifters. These results are robust to an expanded Mishkin test where we include other relevant explanatory variable to ensure the validity of the findings from the Mishkin test (Kraft et al., 2007). Our results hold using alternative measures of classification shifters, different variations of the core earnings expectation model, and a propensity-score matching model to control for endogeneity. They are also consistent when using the zero-investment portfolio method and a multivariate regression that controls for cross-sectional differences in risk.

We show that our results hold after controlling for accruals management, real earnings management, and special items. We also find greater mispricing for firms with large transitory items and no evidence of mispricing for firms with negative unexpected core earnings. These findings signify that our measure captures classification shifting behavior as opposed to alternative explanations for positive unexpected core earnings. Therefore, we conclude that investors overprice the core earnings of classification shifters and do not see through managers’ opportunistic strategy of shifting normal and recurring operating expenses to non-recurring categories. Our study therefore provides direct large sample evidence that justifies the SEC’s concerns of income statement misclassification and its adverse impact on investors and market participants.

We contribute to the literature on income classification shifting and the mispricing of core earnings in several important ways. First, we extend the classification shifting and mispricing literature (Athanasakou et al., 2011; Bartov & Mohanram, 2013; Haw et al., 2011; McVay, 2006) by using various research methodologies available and provide strong evidence on the mispricing of core earnings for firms that are likely to adopt classification shifting tactics to manage earnings. Our findings suggest that investors “fixate” on core earnings and fail to recognize the lower persistence of core earnings inflated through classification shifting. Second, we document that the core earnings of classification shifters are mispriced even after we control for earnings management and real earnings management. Therefore, we provide important evidence that income classification shifting can have significant valuation implications beyond other earnings management techniques. Finally, the strong mispricing evidence that we document using the classification scheme of shifters and non-shifters suggests that our method can be adopted for future studies to examine various issues related to income classification shifting.

Recent studies have documented that the frequency and magnitude of negative special items have been increasing over time (Riedl & Srinivasan, 2010) and that Regulation G has led to an increasing use of income classification shifting by management to “camouflage” the shifted expenses through special items (Kolev, Marquardt, & McVay, 2008). We also find that one in two firms with large negative transitory items is likely to have engaged in classification shifting. Both the literature and our own evidence of the pervasiveness of classification shifting further establish the significance and timeliness of our study.

Footnotes

Appendix

Actual Cases of Classification Shifting.

Company	AAER^a	Date of AAER	Classification shifting charges
1. Dell, Inc.	AAER #3209	November 5, 2010	Dell shifted unrelated operating expenses to a restructuring charge and used material misrepresentations during conference calls to mislead investors and meet or exceed analyst consensus forecasts.
2. Symbol Technologies	AAER #3124	April 2, 2010	Symbol misclassified unrelated operating expenses to a restructuring charge of US$245 million for its acquisition of Telxon Corporation or relocation of manufacturing operations to new facilities.
3. Safenet, Inc.	AAER #3068	November 12, 2009	SafeNet improperly classified ordinary operating expenses as non-recurring integration expenses (costs incurred to integrate acquired companies into current operations).
4. HBO & Co.	AAER #2815	April 28, 2008	HBO recorded excessive acquisition charges of US$16 million by shifting current-period operating expenses, which were unrelated to the acquisition.
5. Fischer Imaging Corp.	AAER #2134	November 15, 2004	Fischer Imaging incorrectly classified labor and overhead expenses associated with its service business as other operating expenses rather than cost of sales materially overstating gross profit.
6. Kimberly-Clark Corp.	AAER #1533	March 27, 2002	Kimberly-Clark improperly recorded operating expenses in connection with a US$1.44 billion charge for restructuring and other unusual charges in relation to its merger with Scott Paper Corporation.
7. Bankers Trust Corp.	AAER #1383	April 18, 2001	Bankers Trust transferred operating expenses to unrelated reserve accounts, which was part of a bank-wide expense reduction plan.
8. CUC International, Inc.	AAER #1274	June 14, 2000	CUC intentionally overstated merger and purchase reserves and reversed those reserves directly into operating expenses which artificially increased operating income.
9. W.R. Grace & Co.	AAER #1140	June 30, 1999	W.R. Grace manipulated excess reserves by recording normal operating expenses to bring the Health Care Group segment’s results of operations in line with Grace’s targets.

Note. AAER = Accounting and Auditing Enforcement Releases.

AAER’s are obtained from the Security and Exchange Commission’s (SEC) website at www.sec.gov (SEC, 1999, 2000a, 2000b, 2001, 2002, 2004, 2008, 2009, 2010a, 2010b). AAER’s are financial reporting related enforcement actions concerning civil lawsuits brought by the SEC in federal court (www.sec.gov).

Acknowledgements

We appreciate the helpful comments and suggestions from Jeff Callen, Yue Li, Scott Liao, Hai Lu, Partha Mohanram, Scott Richardson, Dushyant Vyas, Aida Wahid, and participants in workshops at the University of Toronto, the Hong Kong Polytechnic University, and Xiamen University. We would also like to thank our discussant Lian Fen Lee at the 2013 AAA Annual Meeting and Jiajia Fu for her research assistance.

Author’ Note

All remaining errors are the responsibility of the authors.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Athanasakou

Strong

Walker

(2011). The market reward for achieving analysts’ earnings expectations: Does managing expectations or earnings matter?Journal of Business Finance & Accounting, 38, 58-94.

Barnea

Ronen

Sadan

(1976). Classificatory smoothing of income with extraordinary items. The Accounting Review, 51, 110-122.

Barth

M. E.

Cram

D. P.

Nelson

K. K.

(2001). Accruals and the prediction of future cash flows. The Accounting Review, 76, 27-58.

Bartov

Cohen

(2009). The “numbers game” in the pre- and post-Sarbanes-Oxley eras. Journal of Accounting, Auditing & Finance, 24, 505-534.

Bartov

Mohanram

(2013). Does income statement placement matter to investors? The case of gains/losses from early debt extinguishment. The Accounting Review, 89, 2021-2055.

Barua

Lin

Sbaraglia

(2010). Earnings management using discontinued operations. The Accounting Review, 85, 1485-1509.

Bar-Yosef

Prencipe

(2013). The impact of corporate governance and earnings management on stock market liquidity in a highly concentrated ownership capital market. Journal of Accounting, Auditing & Finance, 28, 292-316.

Bradshaw

M. T.

Sloan

(2002). GAAP versus the street: An empirical assessment of two alternative definitions of earnings. Journal of Accounting Research, 40, 41-66.

Brown

Sivakumar

(2003). Comparing the value relevance of two operating income measures. Review of Accounting Studies, 8, 561-572.

10.

Burgstahler

Jiambalvo

Shevlin

(2002). Do stock prices fully reflect the implications of special items for future earnings?Journal of Accounting Research, 40, 585-612.

11.

Chen

C. Y.

(2010). Do analysts and investors fully understand the persistence of the items excluded from street earnings?Review of Accounting Studies, 15, 32-69.

12.

Cheng

C. S.

Cheung

J. K.

Gopalakrishnan

(1993). On the usefulness of operating income, net income and comprehensive income in explaining security returns. Accounting and Business Research, 23, 195-203.

13.

Cheng

C. S.

Hopwood

McKeown

(1992). Nonlinearity and specification problems in unexpected earnings response regression model. The Accounting Review, 67, 579-598.

14.

Cheng

C. S.

Liu

Thomas

(2012). Abnormal accrual estimates and evidence of mispricing. Journal of Business Finance & Accounting, 39, 1-34.

15.

Cheng

C. S.

Thomas

(2006). Evidence of the abnormal accrual anomaly incremental to operating cash flows. The Accounting Review, 81, 1151-1167.

16.

Chi

Pincus

Teoh

(2014). Mispricing of book-tax differences and trading behavior of short sellers and insiders. The Accounting Review, 89, 511-543.

17.

Cohen

Dey

Lys

(2008). Real and accrual-based earnings management in the pre- and post-Sarbanes-Oxley periods. The Accounting Review, 83, 757-878.

18.

DeAngelo

Skinner

(1994). Accounting choice in troubled companies. Journal of Accounting & Economics, 17, 113-143.

19.

Dechow

P. M.

(2006). The persistence of earnings and cash flows and the role of special items: Implications for the accrual anomaly. Review of Accounting Studies, 11, 253-296.

20.

Dechow

P. M.

Sloan

Sweeney

(1995). Detecting earnings management. The Accounting Review, 70, 193-225.

21.

Desai

Rajgopal

Venkatachalam

(2004). Value-glamour and accruals mispricing: One anomaly or two?The Accounting Review, 79, 355-385.

22.

Doyle

Lundholm

Soliman

(2003). The predictive value of expenses excluded from pro forma earnings. Review of Accounting Studies, 8, 145-174.

23.

Elliott

Hanna

(1996). Repeated accounting write-offs and the information content of earnings. Journal of Accounting Research, 34, 135-155.

24.

Fairfield

P. M.

Kitching

Tang

(2009). Are special items informative about future profit margins?Review of Accounting Studies, 14, 204-236.

25.

Fairfield

P. M.

Sweeney

Yohn

(1996). Accounting classification and the predictive content of earnings. The Accounting Review, 71, 337-355.

26.

Fama

E. F.

French

K. R.

(1997). Industry costs of equity. Journal of Financial Economics, 43, 153-193.

27.

Fama

E. F.

MacBeth

J. D.

(1973). Risk, return and equilibrium: Empirical tests. Journal of Political Economy, 81, 607-636.

28.

Fan

Barua

Cready

Thomas

(2010). Managing earnings using classification shifting: Evidence from quarterly special items. The Accounting Review, 85, 1303-1323.

29.

Francis

Hanna

Vincent

(1996). Causes and effects of discretionary asset write-offs. Journal of Accounting Research, 34, 117-134.

30.

Givoly

Hayn

(1992). The valuation of the deferred tax liability: Evidence from the stock market. The Accounting Review, 67, 394-410.

31.

Chen

(2004). Analysts’ treatment of nonrecurring items in street earnings. Journal of Accounting & Economics, 38, 129-170.

32.

Gunny

K. A.

(2010). The relation between earnings management using real activities manipulation and future performance: Evidence from meeting earnings benchmarks. Contemporary Accounting Research, 27, 855-888.

33.

Haw

I.-M.

S. S. M.

A. Y.

(2011). Corporate governance and earnings management by classification shifting. Contemporary Accounting Research, 28, 517-553.

34.

Healy

(1985). The effect of bonus schemes on accounting decisions. Journal of Accounting and Economics, 7, 85-107.

35.

Hirshleifer

Teoh

(2011). Short arbitrage, return asymmetry, and the accrual anomaly. The Review of Financial Studies, 24, 2429-2461.

36.

Hossain

Mitra

Rezaee

Sarath

(2011). Corporate governance and earnings management in the pre- and post-Sarbanes-Oxley Act regimes: Evidence from implicated option backdating firms. Journal of Accounting, Auditing & Finance, 26, 279-315.

37.

Jones

(1991). Earnings management during import relief investigations. Journal of Accounting Research, 29, 193-228.

38.

Kennedy

(1992). A guide to econometrics. Cambridge, MA: MIT Press.

39.

Kolev

Marquardt

McVay

(2008). SEC scrutiny and the evolution of non-GAAP reporting. The Accounting Review, 83, 157-184.

40.

Kormendi

Lipe

(1987). Earnings innovations, earnings persistence, and stock returns. The Journal of Business, 60, 323-345.

41.

Kraft

Leone

Wasley

(2006). An analysis of the theories and explanations offered for the mispricing of accruals and accrual components. Journal of Accounting Research, 44, 297-339.

42.

Kraft

Leone

Wasley

(2007). Regression-based tests of the market pricing of accounting numbers: The Mishkin test and ordinary least squares. Journal of Accounting Research, 45, 1081-1114.

43.

Lewellen

(2010). Accounting anomalies and fundamental analysis: An alternative view. Journal of Accounting & Economics, 50, 455-466.

44.

Lipe

R. C.

(1986). The information contained in the components of earnings. Journal of Accounting Research, 24, 37-64.

45.

Marquardt

Wiedman

(2005). Earnings management through transaction structuring: Contingent convertible debt and diluted earnings per share. Journal of Accounting Research, 43, 205-243.

46.

Mashruwala

Rajgopal

Shevlin

(2006). Why is the accrual anomaly not arbitraged away? The role of idiosyncratic risk and transaction costs. Journal of Accounting & Economics, 42, 3-33.

47.

Matsumoto

D. A.

(2002). Management’s incentives to avoid negative earnings surprises. The Accounting Review, 77, 483-514.

48.

McVay

S. E.

(2006). Earnings management using classification shifting: An examination of core earnings and special items. The Accounting Review, 81, 501-531.

49.

Mishkin

(1983). A rational expectations approach to macroeconometrics: Testing policy effectiveness and efficient markets models. Chicago, IL: University of Chicago Press for the National Bureau of Economic Research.

50.

Myers

Skinner

(2007). Earnings momentum and earnings management. Journal of Accounting, Auditing & Finance, 2, 249-284.

51.

Phillips

Pincus

Rego

(2003). Earnings management: New evidence on deferred tax expense. The Accounting Review, 78, 491-521.

52.

Pincus

Rajgopal

Venkatachal

(2007). The accrual anomaly: International evidence. The Accounting Review, 82, 169-203.

53.

Prakash

Sinha

(2013). Deferred revenues and the matching of revenues and expenses. Contemporary Accounting Research, 30, 517-548.

54.

Richardson

Tuna

Wysocki

(2010). Accounting anomalies and fundamental analysis: A review of recent research advances. Journal of Accounting & Economics, 50, 410-454.

55.

Riedl

Srinivasan

(2010). Signaling firm performance through financial statement presentation: An analysis using special items. Contemporary Accounting Research, 27, 289-332.

56.

Ronen

Sadan

(1975). Classificatory smoothing: Alternative income models. Journal of Accounting Research, 13, 133-149.

57.

Roychowdhury

(2006). Earnings management through real activities manipulation. Journal of Accounting & Economics, 42, 335-370.

58.

Securities and Exchange Commission. (1999). Accounting and Auditing Enforcement Release No. 1140. Retrieved from http://sec.gov/litigation/litreleases/lr16008.txt

59.

Securities and Exchange Commission. (2000a). Accounting and Auditing Enforcement Release No. 1274. Retrieved from http://www.sec.gov/litigation/admin/34-42935.htm

60.

Securities and Exchange Commission. (2000b). Audit risk alert to the American Institute of Certified Public Accountants. Washington, DC: Author.

61.

Securities and Exchange Commission. (2001). Accounting and Auditing Enforcement Release No. 1383. Retrieved from http://www.sec.gov/litigation/admin/34-44193.htm

62.

Securities and Exchange Commission. (2002). Accounting and Auditing Enforcement Release No. 1533. Retrieved from http://edgar.sec.gov/litigation/admin/34-45653.htm

63.

Securities and Exchange Commission. (2004). Accounting and Auditing Enforcement Release No. 2134. Retrieved from http://www.sec.gov/litigation/admin/33-8503.htm

64.

Securities and Exchange Commission. (2008). Accounting and Auditing Enforcement Release No. 2815. Retrieved from http://www.sec.gov/litigation/admin/2008/33-8912.pdf

65.

Securities and Exchange Commission. (2009). Accounting and Auditing Enforcement Release No. 3068. Retrieved from http://www.sec.gov/litigation/litreleases/2009/lr21290.htm

66.

Securities and Exchange Commission. (2010a). Accounting and Auditing Enforcement Release No. 3124. Retrieved from http://www.sec.gov/litigation/admin/2010/34-61833.pdf

67.

Securities and Exchange Commission. (2010b). Accounting and Auditing Enforcement Release No. 3209. Retrieved from http://www.sec.gov/litigation/admin/2010/34-63263.pdf

68.

Shi

Zhang

(2012). Can the earnings fixation hypothesis explain the accrual anomaly?Review of Accounting Studies, 17, 1-21.

69.

Siu

Faff

(2012). Management of core earnings using classification shifting around seasoned equity offerings (Working paper). St Lucia, Australia: University of Queensland.

70.

Sloan

R. G.

(1996). Do stock prices fully reflect information in accruals and cash flows about future earnings?The Accounting Review, 71, 289-315.

71.

Xie

(2001). The mispricing of abnormal accruals. The Accounting Review, 76, 357-373.

72.

Zang

(2012). Evidence on the trade-off between real activities manipulation and accrual-based earnings management. The Accounting Review, 87, 675-703.

73.

Zhang

Zheng

(2011). The valuation impact of reconciling pro forma earnings to GAAP earnings. Journal of Accounting & Economics, 51, 186-202.

74.

Zhang

(2010). The relationship between audit quality and classification shifting (Working paper). Philadelphia, PA: Temple University.