Journal of accounting, auditing finance tập 27, số 02, 2012 4

To accountfor seasonal variations in quarterly cash flows, we deseasonalize our data using the X11 The economic significance of accruals’ predictive ability in our sample is most nounced

Auditing & Finance 27(2) 151–176

Ó The Author(s) 2012 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/0148558X11409149

http://jaaf.sagepub.com

The Predictive Value of

Accruals and Consequences for

on average, as opposed to CFO alone They also find that Sloan’s accrual anomaly is related

to our accrual contribution anomaly Indeed, when accruals’ contribution to future cashflow prediction is the highest, the accrual anomaly vanishes Collectively, the results suggestthat the predictive value of accruals and market participants’ ability to process it are a sig-nificant driver of accrual-based anomalies

Keywords

accruals, cash flows, cash flow predictions, anomalies

The amount of aggregate future cash flows is key to the valuation of a firm’s securities.Alternative valuation models by both academics and financial analysts have focused on theprediction of free cash flows (FCFs; Copeland, Koller, & Murrin, 1994) or residual income(Edwards & Bell, 1961; Ohlson, 1995; Preinreich, 1938) The prediction of cash flows isinvariably based on past accounting numbers One question that has occupied much of theresearchers’ attention is the extent to which the accrual component of past earnings contri-butes to the prediction of future realizations of cash flows and market participants’ expecta-tions of future cash flows

3

New York University, Stern School of Business, USA

Corresponding Author:

Seunghan Nam, Lally School of Management & Technology, Rensselaer Polytechnic Institute, 110 8th Street, Room

1120, Pittsburgh Bldg., Troy, NY 12180, USA

Email: nams2@rpi.edu

Promoting the accrual basis of accounting, the Financial Accounting Standards Board(FASB) asserts that earnings and their components are better predictors of future cashflows than current cash flow (FASB, 1978) In spite of the FASB argument, scholars andpractitioners argue that the subjectivity inherent in estimates embedded in accruals intro-duces noise that can have a negative impact on their informational value (Dechow &Dichev, 2002) Firm managers may engage in self-serving earnings manipulation by report-ing numbers based on distorted estimates, which has been shown to decrease the value rele-vance of earnings (Marquardt & Wiedman, 2004) Hence, whether they are made in goodfaith or with manipulative intent, accruals can be misleading and not representative offirm’s future performance.

We first revisit the findings on cash flow predictability by testing the Dechow, Kothari,and Watts’s (1998) theoretical predictions with a methodology that simultaneouslyaddresses the following three dimensions: (a) judgment of the superiority of the predictor

estimation of firm-specific versus cross-sectional coefficients, and (c) the level of tion of future cash flows as the predicted variable Our evidence based on these methodolo-gical choices supports the view that accruals contribute to the prediction of future cashflows and provides detailed information on cross-sectional differences in the predictive

Our sample utilizes post-SFAS 95 quarterly data from Compustat We define cash flow

as cash flow from operations (CFO) and accruals as the difference between net income andCFO, consistent with Hribar and Collins (2002) In our main analysis, we require 56 time-series observations to develop firm-specific regression estimates As a result, our holdoutsample period is from the third quarter of 2002 to the fourth quarter of 2006 To accountfor seasonal variations in quarterly cash flows, we deseasonalize our data using the X11

The economic significance of accruals’ predictive ability in our sample is most nounced when the predicted variable is current or one-quarter-ahead market value ofequity: The model including accruals as a predictor along with CFO exhibits significantlysmaller mean and median absolute prediction errors than the model using current CFOalone, by about 5% of total assets

pro-In our portfolio tests, the average hedge return adjusted for the three Fama–French tors and momentum for a 90-day holding period when going long (short) on the highest(lowest) quintile of the quarterly predicted return distributions is insignificantly differentfrom zero, with or without accruals as a predictor However, as the holding period

increases, the returns earned on the portfolio using CFO and accruals become significantlyhigher than those using CFO only as a predictor For instance, 270- and 365-day incremen-tal returns when accruals are added as a predictor are about 2% per quarter on average.

As for our tests related to the accrual anomaly, we replicate the results first documented

by Sloan (1996) using quarterly data and find that the accrual anomaly is nonexistent forstocks in the top quintile of accruals’ contribution to future cash flow predictions Thisresult supports our view that the current accruals’ ability to forecast future cashflows—rather than properties of current accruals per se, such as their sign and size—is the

Our contribution to the literature is twofold First, our study demonstrates that accruals’contribution to future cash flow predictions is most significant when predicting futuremarket capitalizations Assuming that market capitalization is a good proxy for all futurecash flows, this implies that accruals contribute to the prediction of all future cash flows.Many studies show that cash flow and accruals exhibit higher associations with future cashflows and/or stock returns than current cash flow alone (e.g., Barth, Cram, & Nelson, 2001;Dechow, 1994), but none provides such evidence in terms of out-of-sample predictions.Second, our results add to the literature on accounting-based stock anomalies By docu-menting predictable abnormal returns based on hedge portfolios that use current accountingdata as a sorting criterion, we show that market participants do not fully understand theimplications of current CFO and accruals for the present value of future cash flows In par-ticular, the contribution of accruals to future cash flow predictions does not appear to befully taken into account by investors, as accruals help improve upon CFO alone in earningabnormal returns over horizons of 6 months and more Finally, we show that our contribu-tion anomaly is related to the accrual anomaly documented by Sloan (1996)

In addition, our methodological considerations have practical implications because theyaddress issues of relevance to investors who use current accounting data for equity valuationpurposes With respect to finite cash flow predictions, finite horizon predictions are of particu-lar relevance to equity valuation techniques that consist of forecasting earnings, cash flows, ordividends over a finite period and computing a terminal value (Penman & Sougiannis, 1998).Our study is subject to caveats that apply to most studies in this field First, by usingfirm-specific regressions, we not only require time-series data that unavoidably reduce

deferrals are estimates subject to moral hazard between managers who report them andshareholders Our attempt to separate accruals based on their discretionary or unverifiablecomponents using the Jones (1991) model is subject to the usual criticism regarding discre-tionary accruals estimation error

The rest of the article is organized as follows: Section titled ‘‘Prior Literature andEmpirical Predictions’’ reviews the relevant literature Section titled ‘‘Research Design’’specifies the empirical tests, and the next section titled ‘‘Research Design’’ describes thesample selection process and presents the main results The final section titled

‘‘Conclusion’’ summarizes and concludes

Prior Literature and Empirical Predictions

Prior Literature

Our article relates to an extensive literature that investigates the valuation implications ofcomponents of accounting earnings, either indirectly through their association with future

accounting measures or directly through their association with market values of equity.Wilson (1986), who uses stock returns around earnings announcements and Form 10-K fil-ings to show that the accrual component of earnings has incremental information contentover cash flow, is one of the earliest studies in this literature strand Because this questionhas subsequently generated a vast number of studies, which generally differ by methodolo-gical choices, we provide a matrix (see Appendix A and all other appendices online athttp://jaaf.sagepub.com/supplemental) that highlights the key findings of prior researchbased on the three dimensions along which we position our study.

Determinants of Accruals’ Contribution

The extent to which current accruals contribute to more accurate predictions of future cashflows is expected to vary across firms and time periods

First, we expect the accruals’ contribution to vary with specifics of the economics of afirm, as manifested in properties of past or current accounting numbers For example, iffirms operate in an uncertain environment, their stream of cash flows is more likely to exhi-bit greater volatility As a result, past realizations of cash flows are likely to be noisy and

to be a less useful predictor of future cash flows Financial statement users are more likely

to draw inferences about the timing and amount of future cash flows by using accruals.Indeed, accruals tend to smooth out some of the variability in cash flow patterns by mitigat-ing issues arising from discrepancies between cash flows and the underlying economics interms of timing of recognition

In addition, current cash flow in firms with greater growth options could be of relativelylimited use in predicting future streams of cash flows Indeed, growth firms are more like-ly—ceteris paribus—to be in a transitory stage where past realizations of cash flows bearlittle association with future cash flows Although short-term accruals may also be uninfor-mative, long-term accruals are likely to provide incremental information For instance,amortization policies for recent investments can provide useful insight about the economiclife of the type of projects that the firm can undertake in the near future However, this ben-efit would not arise if a firm invests in R&D and other unrecognized internally developedintangibles At any rate, rather than the volatility of past cash flow, it is the expected volati-lity of future cash flows that provides a role for current accruals in terms of predictive abil-ity for growth firms

Our predictions so far rely on the assumption that management uses accruals in amanner that is not self-serving However, agency conflicts between managers and share-holders can induce management to deviate from truthful reporting to maximize their ownwealth Indeed, prior studies have shown that executives report income-increasing discre-tionary accruals in years where they sell their stock so as to increase the proceeds fromthose transactions (Bartov & Mohanram, 2004; Cheng & Warfield, 2005) If reportedaccruals are distorted by measurement bias, their informativeness vis-a`-vis future cashflows may be impaired to a point where they no longer provide incremental predictionvalue or even worsen predictions compared with those based on current cash flowsonly Consistent with this idea, the results documented by Xie (2001) suggest that inves-tors’ mispricing of accruals is driven by discretionary accruals However, managers canalso report income-decreasing accruals for their own benefit In particular, when truthfulreporting falls short of expectations by a large margin, they are better off taking a ‘‘bigbath,’’ that is, reporting large income-decreasing accruals Hence, we expect that themagnitude of discretionary accruals (to the extent they are driven by measurement bias)

exhibits a negative association with the contribution of total accruals to the prediction offuture cash flows.

Accruals’ Contribution and Their Mispricing

Starting with Sloan (1996), prior research has shown that investors do not fully understandthe implications of accruals for future earnings Sloan posits that investors fixate on earn-ings and fail to recognize the lower persistence of accruals compared with cash flows As aresult, one can generate a profitable trading strategy by buying low-accrual stocks and sell-ing high-accrual stocks We further explore the role of accruals in explaining future stockreturns through their predictive ability for future cash flows

To the extent that accruals do contribute to more accurate predictions of future cashflows and market values of equity, and if investors fail to act such as to cause stock prices

to fully reflect the predictive ability of accruals at the time accounting information becomespublicly available, then one may observe predictable stock returns subsequent to the release

of that information For instance, the predicted market capitalization conditional on currentaccounting data can be viewed as a proxy for fundamental value, and if current marketprices gravitate toward fundamental values, one can sort stocks based on the degree towhich their prices deviate from fundamental value so as to predict stock returns (Frankel &Lee, 1998) The role of accruals in explaining such anomaly can be judged by comparingpredicted values of future cash flows (proxies for fundamental value) with and withoutaccruals as a predictor We expect that if, indeed, investors do not fully understand theimplications of accruals for future cash flows, then a trading strategy going long (short) onhigh (low) ratios of predicted to actual market values of equity should yield higher risk-adjusted returns with accruals as a predictor than without accruals

Finally, we investigate whether accruals’ predictive value is related to the accrual aly first documented by Sloan (1996) We posit that accounting-based anomalies should beprimarily driven by investors’ incorrect expectations of future cash flows, rather than prop-erties of current accounting data per se Hence, we expect that sorting stocks on accrualsize need not be associated with predictable stock returns when current accruals are an

Research Design

Prediction Models

We use regression models to predict various measures of future cash flows out of sample

market value of equity (MKTCAP), at the beginning or at the end of the fiscal quarter, as a

predicted or used as predictors, are scaled by total assets at the end of the previous fiscalquarter Our main analysis is based on firm-specific estimations using time-series data Ourbenchmark ‘‘cash-flow only’’ model is the following:

Our accounting variables are subject to seasonality This is particularly the case forfirm-level quarterly cash flows time series, which exhibit purely seasonal characteristics, as

documented by Lorek and Willinger (1996) As we use adjacent quarters to make our dictions, we need to adjust for seasonality in our cash flow series To do so, we use theX11 method as described in Appendix B In brief, the X11 procedure, developed by theBureau of Census, decomposes monthly or quarterly data into trend, seasonal, and irregularcomponents using moving averages One can subsequently subtract the estimated seasonalcomponent to come up with a deseasonalized series.

pre-To test whether accruals contribute to reducing prediction errors, we compare Model 1

to models wherein aggregate accruals are included as an independent variable, either gated with cash flows or as a separate predictor:

ACC stands for total accruals, defined as the difference between net income beforeextraordinary items EARN (Compustat Quarterly Data Item 8) and CFO (CompustatQuarterly Data Item 108) net of extraordinary items/discontinued operations that affect

flow and accrual components of earnings are equal, whereas they are allowed to differ inModel 2 We include Model 3 to assess whether aggregate earnings improve upon currentcash flow alone in predicting cash flows

We further proceed to disaggregate total accruals into their components, based on thepremise that different subsets of accruals carry different implications for future cash flows(Barth et al., 2001), such as stemming from the horizon over which cash collectability isexpected or from differing degrees of subjectivity inherent in different subsets of accruals:

Model 4 is similar to the cross-sectional regression that Barth et al (2001) run to testthe incremental explanatory power of disaggregated earnings This model presents the high-est level of accrual disaggregation that we consider DAR, DINV, and DAP are changes inworking capital accounts: accounts receivable, inventories, and accounts payable, respec-tively DEPAMOR is depreciation and amortization.OTHER is simply the differencebetween total accruals ACC and (DAR 1 DINV 2 DAP 2 DEPAMOR When it is avail-able, we use data from the statement of cash flow for our individual accrual components;otherwise, we use changes in balance sheet accounts That is, we use changes in accountsreceivable, inventory, and accounts payable (Compustat Quarterly Data Items 103, 104,and 105, respectively) if they are available; otherwise, we use changes in Data Items 37,

38, and 46 from the previous fiscal quarter Depreciation and amortization expense isCompustat Quarterly Data Item 77 Market capitalization is the product of CompustatQuarterly Data Items 14 and 61 Finally, our deflator is total assets (Compustat QuarterlyData Item 44) as of the beginning of the quarter One major distinction among accrual com-ponents is the timing of their conversion into cash in- or outflows The changes in workingcapital variables are expected to affect future cash flows in the near term (within a year)

By contrast, DEPAMOR should exhibit a greater association with cash flows in the longerrun Indeed, depreciation and amortization expenses are intended to match costs of

investments with their benefits over the expected life of the asset that is being depreciated/amortized, typically several years Overall, although the use of individual accrual compo-nents may help improve prediction accuracy, the decrease in the number of degrees of free-dom may offset such a benefit for firm-specific estimations, for which the number ofobservations is limited.

use rolling windows so that coefficients are ‘‘updated’’ every quarter The required number

of observations represents a trade-off between sample size and the reliability and stability

of time-series estimates Alternatively, we estimate coefficients cross-sectionally, separatelyfor each fiscal quarter Once we run a regression, we use the coefficient estimates to com-

Assets t is equal to cg01cg1 CFOt

Assets t1 ,

the actual value We compute our absolute prediction errors as follows:

The subscript j indicates which model was used to compute the predicted value (1, 2, 3,

com-pare the predictive ability of our different models, we compute the mean and median diction errors across all firm-quarters in the holdout sample period

pre-Multivariate Analysis

To investigate determinants of the contribution of accruals to the prediction of future cashflows, we use the following multivariate specification:

ABS_DISC_ACC is the absolute value of discretionary accruals, which we estimateusing the firm-specific version of the modified Jones (1991) model as in Dechow, Sloan,and Sweeney (1995) Details are provided in Appendix C There are two main views in theliterature regarding managers’ motivations to use their discretion in reporting generallyaccepted accounting principles (GAAP) numbers The first one (the ‘‘opportunistic’’ view)

is that managers manipulate accounting reports to maintain the firm’s stock price at

artificially high levels and benefit from this overvaluation in terms of equity-based pensation The second one (the ‘‘informational’’ view) is that managers use their discretion

com-to signal their private information about future cash flows Badertscher, Collins, and Lys(2007) provide evidence that earnings managed with an apparently informational purposeexhibit a higher association with future cash flows than earnings managed opportunistically

do As we do not attempt to disentangle opportunistic from signaling motives behind cretionary accruals, we leave the sign of the coefficient on ABS_DISC_ACC as an empiricalquestion We also control for the magnitude of nondiscretionary accruals, that is, the differ-ence between total accruals and discretionary accruals

dis-With respect to SIGN_ACCRUALS, which is an indicator variable equal to one if totaldeseasonalized accruals are strictly positive, we test whether net positive accruals havegreater predictive ability for future cash flows than negative accruals do We predict a

reflect a smoothing/matching perspective, whereas negative accruals are more likelydriven by impairments due to fair value accounting (Dechow & Ge, 2006).SEASONALITY is the degree to which quarterly cash flows are seasonal We compute thisvariable by taking the difference between actual CFO and deseasonalized CFO.CFO_VOLATILITY is the standard deviation of firm-level CFO measured from t 2 16 to

t 2 1 We expect that accruals should be more helpful in predicting future cash flows,the more volatile current cash flows are (Dechow & Dichev, 2002) This should be

potential determinants of accruals’ contribution to cash flow predictions We expect anegative coefficient on both variables With respect to firm size, larger firms are presum-ably more mature firms with more stable cash flows, which can be predicted more easilyusing past cash flow observations As for book-to-market ratio, we argue that accrualsare likely to be informative about growth options beyond current cash flow; that is,their contribution to future cash flow prediction should be higher in firms with a lowbook-to-market ratio

The last two variables are proxies for the quality of monitoring that managerial actionsand reporting incentives are subject to BIG4 is an indicator variable equal to 1 if thefirm’s auditor is one of the big four auditing firm and 0 otherwise We expect that accrualswill be more informative if audited by one of the leading firms in the industries, which hasbeen extensively used as a proxy for auditor quality in the literature (Francis, 2004).Finally, we use the Gompers, Ishii, and Metrich (2003) GINDEX to proxy for firm-levelgovernance quality We expect that managers will be less inclined to manipulate accruals

in firms where shareholders’ rights are stronger Consequently, we expect a negative sign

on the coefficient of GINDEX because the index takes higher values when there are moreanti-takeover provisions

Portfolio Analysis

We test whether the predictive ability of current cash flow and accruals for future marketvalues of equity translates into predictable abnormal returns for portfolio allocations based

on current accounting data To do so, we use the following methodology First, using

return and compute the difference between the mean returns across observations in the top

decile/quintile of the predicted return distribution and in the bottom decile/quintile, wherereturns are cumulated from the day following each firm-quarter’s 10-K or 10-Q filing dateover a 90- to 365-day period Finally, we compare this return across different prediction

financial services (SIC 6000-6999) and regulated industries (SIC 4900-4999) To producereliable firm-specific coefficient estimates in our regressions, we must use a reasonablylarge number of observations We choose to require the availability of 56 consecutive quar-terly observations prior to a given firm-quarter to predict the latter’s cash flow (or aggre-gates of the predicted cash flow of that firm-quarter and those of following quarters) Theserequirements result in an upper bound of 16,594 predicted firm-quarters with data available

to predict one-quarter-ahead CFO Fiscal quarters 2002:3Q to 2006:4Q constitute our out period For our cross-sectional regressions, we winsorize the independent variables at1% and 99% of their quarterly distributions

hold-Descriptive Statistics

Table 1 reports summary statistics for the variables used in our analysis Consistent withprior studies, mean and median earnings and CFO are positive, whereas mean and medianaccruals are negative As explained in Barth et al (2001), this is most likely driven bydepreciation and amortization, which is much larger than other accrual components onaverage

Table 2 presents summary results for a subset of our firm-specific regression models

mean and median coefficients on CFO are positive (0.66 and 0.68, respectively) In Model

2, the mean (0.897) and median (0.867) coefficients on CFO are about 3 times as large asthe coefficients on ACC (mean 0.299 and median 0.231) The ratio is smaller when the

(25.55%) on average when aggregate accruals are added as a predictor Disaggregatingaccruals into their individual components also contributes to an increase in mean firm-

with what Barth et al (2001) document using cross-sectional regressions and annual data

Prediction Results

Mean and median absolute prediction errors across firm-specific estimates Tables 3 and 4report the comparisons of absolute prediction errors (ABSE) for future CFO and FCFacross firm-specific models with and without accruals as predictors, over horizons of one-

to eight-quarter-ahead and with market capitalization as a proxy for all future cash flows

Table 1 Descriptive Statistics

Variable definitions with Compustat Quarterly data item numbers (all variables are scaled by ASSETS, except ASSETS and MKTCAP, which are expressed in million dollars):

ASSETS: Total assets (Data44) as of the beginning of the quarter.

MKTCAP: Market value of equity as of the end of the fiscal quarter (from Center for Research in Security Prices, price 3 shares outstanding at the end of fiscal quarter).

CFO: Cash flow from operations (Data108).

EARN: Income before extraordinary items and discontinued operations (Data8).

ACC: EARN minus (CFO 2 extraordinary items/discontinued operations that affect cash flows [Data 78]).

DAR: Change in accounts receivable from previous quarter (Data103 if available, DData37 otherwise).

DINV: Change in inventories from previous quarter (Data104 if available, DData38 otherwise).

DAP: Change in accounts payable from previous quarter (Data105 if available, DData46 otherwise).

DEPAMOR: Depreciation and amortization (Data77).

OTHER: ACC 2 (DAR 1 DINV 2 DAP 2 DEPAMOR).

FCF: EARN (Data8) 2 (1 2 d) 3 (capital expenditure [Data90] 2 depreciation [Data77]) 2 (1 2 d) 3 D working capital, where working capital = (current assets [Data40] 2 current liabilities [Data49]) and d is debt (debt in cur- rent liabilities, Data45 1 long-term debt, Data51) to total assets (Data44) ratio.

ABS_DISC_ACC: Absolute value of discretionary accruals, as measured using the modified Jones (1991) model, mated on a firm-specific basis See Appendix C.

esti-ABS_NONDISC_ACC: Absolute value of nondiscretionary accruals Nondiscretionary accruals are the difference between total accruals and discretionary accruals.

SEASONALITY: Difference between total CFO and deseasonalized CFO.

CFO_VOLATILITY: Standard deviation of quarterly deseasonalized CFO from t 2 16 to t 2 1.

BOOK-TO-MARKET: Ratio of book value (Data59) of equity to market value of equity, as of the beginning of the fiscal quarter.

Table 3 reports mean and median absolute prediction errors scaled by total assets as ofthe beginning of the quarter (ABSE) for Models 1, 2, and 3 At this stage, we do not reportresults from Model 4 because we wish to focus on the results for the larger sample whereour data requirements are less constraining The results for finite measures of cash flowsare based on comparisons of predicted values with future deseasonalized cash flows Forall levels of aggregation of future cash flows, Model 2 produces lower mean and medianabsolute prediction errors than Model 1 The difference in means is statistically significant

at conventional levels except when one-quarter-ahead cash flow is predicted For example,

predic-tors is 1.30% of total assets, whereas the mean absolute prediction error from CFO alone is

two-tailed In addition, we find that aggregate earnings do not outperform CFO andaccruals as separate predictors in forecasting finite measures of cash flows When FCF isthe predicted variable, we also find that accruals help reduce absolute prediction errors at

firm-quarters show that there is no statistically significant difference for FCF predictions

Table 2 Summary of Firm-Specific Regression Results

Dependent variable: CFO t11 Dependent variable: MKTCAP t11

Model 1 (N = 16,549)

Intercept 0.0190 0.0189 0.0112 0.0191 0.0282 1.3273 1.3952 0.5846 0.9361 1.5448 CFO 0.1842 0.2691 20.0111 0.1664 0.3687 6.4588 14.3054 0.1253 2.6518 9.0134

R2 8.09% 13.54% 20.95% 2.38% 11.73% 6.89% 11.61% 20.99% 2.08% 10.06% Model 2 (N = 16,549)

Intercept 0.0174 0.0166 0.0099 0.0172 0.0260 0.9976 1.3515 0.4204 0.6955 1.2063 CFO 0.3644 0.4968 0.1021 0.3895 0.6421 23.6222 32.2929 5.7757 15.7165 31.8475 ACC 0.2664 0.4872 0.0263 0.2186 0.5101 20.8262 30.9259 4.2215 12.3207 27.7030

R2 12.69% 16.53% 0.57% 6.96% 19.61% 25.55% 20.70% 8.59% 21.74% 39.55% Model 3 (N = 16,549)

Intercept 0.0190 0.0182 0.0121 0.0199 0.0281 1.0888 1.3758 0.4650 0.7597 1.3276 EARN 0.3017 0.5014 0.0470 0.2709 0.5610 21.9380 30.9702 4.8272 13.2813 29.4849

R 2 8.94% 14.44% 20.77% 2.84% 12.90% 23.32% 20.41% 6.18% 18.55% 36.74% Model 4 (N = 12,327)

Intercept 0.0094 0.0365 20.0064 0.0094 0.0261 0.9106 1.9176 0.1726 0.7265 1.4453 CFO 0.3355 1.0693 0.0209 0.3375 0.6165 21.8475 31.5010 4.7697 13.2081 28.5958 DAR 0.1908 1.3548 20.1882 0.1341 0.5226 16.9327 36.6139 0.7616 8.3110 22.8994 DINV 0.4567 4.9156 20.0632 0.2391 0.6400 13.0609 156.3166 0.3436 8.4391 23.8046 DAP 20.4567 1.8547 20.7136 20.3526 20.0846 217.9243 31.7867 225.0108 29.9917 22.7853 DEPAMOR 0.0264 1.7794 20.5426 20.0094 0.5710 218.6138 73.7745 238.0973 213.7426 2.6167 OTHER 0.2832 1.0822 20.0104 0.2205 0.5270 19.0486 30.6056 3.0924 10.2049 24.3896

R2 19.20% 17.78% 5.67% 15.65% 29.39% 37.86% 20.87% 21.61% 37.51% 53.49% Note: This table reports summary statistics for coefficients in firm-specific regressions of one-quarter-ahead cash flow from operations (CFO) and market capitalization The coefficients are estimated using 56 consecutive quarterly observations over rolling windows, starting from the first fiscal quarter of 1987.

See Table 1 for variable definitions.

Hence, as with CFO, there is some evidence that accruals help improve FCF predictionsbut only to a limited extent.

accruals is statistically significant in terms of both mean and median For example, mean

stat of 5.31 (the corresponding p value being below 01) In general, accruals improve uponCFO by reducing the mean and median absolute prediction errors for current or next quar-ter market value of equity by an order of magnitude of 5% of total assets

the mean and median differences are different from zero The results indicate that the meancontribution of accruals to finite CFO and FCF prediction is positive and significantly sofor all levels of aggregation The highest mean contribution as a percentage of total assets

is 0.068% when six quarters of CFO are predicted (not tabulated) Accruals also cantly contribute to prediction accuracy at the median level for finite CFO and FCF Themean (about 5.6% of total assets) and median (2%) contribution of accruals at the firm-quarter level is also significantly positive at the 01 two-tailed level for current and one-quarter-ahead market values of equity

signifi-Contribution of accruals and level of aggregation of future cash flows We test whetheraccruals contribute more significantly to the prediction of higher levels of aggregation offuture cash flows, as their conversion to cash in- or outflows does not necessarily occurwithin the next quarter In Table 3, we have already compared mean and median absoluteprediction errors of one- to eight-quarter-ahead (cumulative) CFO of firm-specific regres-sions based on CFO alone, to CFO and accruals as separate predictors In Figure 1, we plot

a measure of the incremental contribution of accruals compared with current CFO alone,

with the level of aggregation of the dependent variable on the horizontal axis The graphindicates an upward trending contribution of accruals as aggregation increases The incre-mental contribution is always higher in terms of median than mean, but from one to sixcumulated CFOs of the quarters being predicted, the improvement of accruals is monotoni-cally increasing at the mean level and reaches about 4% (6% for medians) The fact thatthe mean contribution of accruals tends to level-off when the dependent variable is aggre-gated over more than six quarters suggests that the trade-off between increased noise andsignal becomes more severe beyond six quarters Also, untabulated results show that interms of median, the incremental contribution of accruals for current market capitalization

is 17.6% Overall, our results tend to show that using one-period-ahead or finite horizon measures may understate accruals’ usefulness in predicting future cash flows, espe-cially on a quarterly basis

short-Disaggregating accruals into individual components and prediction accuracy When we requiredata to be available for individual accrual components as in Model 4, the sample is smaller

To evaluate whether disaggregating accruals into individual components helps improveupon aggregated accruals in predicting cash flows, we compare absolute prediction errorsacross our models for all firm-quarters with data available for all variables in Model 4.Table 4 reports the results The results indicate that mean absolute prediction error from

aggregate accruals) when the predicted variable is finite CFO aggregated over six to eight

when the level of aggregation of predicted future CFO is four quarters or more However,the differences are not statistically significant The same holds for predictions of market

median levels Hence, in our sample and with our research design choices, we find no tistically significant improvement in prediction accuracy for future cash flows when disag-

Multivariate results The results we provide thus far are averaged across firms with ent economic and financial reporting attributes We test whether the ability of accruals tocontribute to future cash flow prediction varies with firms’ accounting and economic prop-erties as identified in Model 6

differ-Table 5 reports regression results where the dependent variable is the difference between

predicting future cash flows In the first column, the dependent variable is measured forone-quarter-ahead predictions of future CFO The positive coefficient on SIGN_ACC sug-gests that net positive accruals are more likely to improve cash flow prediction than nega-tive accruals The coefficient is statistically significant, although only at the 10 level Thisresult is consistent with the argument that positive accruals are more likely to be driven by

a matching perspective and thus to be useful in predicting future cash flows, especially inthe short run The coefficient on CFO_VOLATILITY is significantly positive This suggeststhat the more volatile cash flows are, the more accruals will improve upon current cashflows in predicting future cash flows This result is, again, consistent with the smoothingproperties of accruals mitigating the volatile nature of cash flows time series With respect

to the discretionary component of accruals, the coefficient on ABS_DISC_ACC is cantly negative This shows that the greater the magnitude of discretionary accruals, thelower the contribution of total accruals to the prediction of future cash flows Hence, itappears that, on average, discretionary accruals, as estimated through the Jones (1991)model, have a negative impact on the forecasting abilities of accruals To the extent that

our measure of discretionary accruals captures managerial discretion in financial reporting,

The significantly negative coefficient on GINDEX corroborates the idea that entrenchedmanagers are more likely to use accruals in a self-serving and less informative manner Incontrast, the absolute value of nondiscretionary accruals exhibits a positive (but not signifi-cant) association with the contribution of accruals to one-quarter-ahead cash flow predic-tions Although insignificant, the coefficient on BIG4 is positive as predicted Finally, thesignificantly negative coefficients on firm size and book-to-market ratio indicate thataccruals contribute more in improving future cash flow predictions for small growth

Table 5 Multivariate Analysis of Accruals’ Contribution to Cash Flow Predictions

Dependent variable is Accruals’

contribution for the prediction of

CFOt11Coefficients

stats

t-CFOt11,t14Coefficients

stats

t-MKTCAPCoefficients

t-tstats

***, **, *indicate significance at the 01, 05, and 10 two-tailed levels, respectively.

The independent variables are winsorized at 1% and 99% level.

FOURTH_QUARTER: Indicator variable equal to one if the dependent variable is measured over the fourth fiscal quarter and zero otherwise.

ABS_DISC_ACC: Absolute value of discretionary accruals, as measured using the modified Jones (1991) model, mated on a firm-specific basis.

esti-ABS_NONDISC_ACC: Absolute value of nondiscretionary accruals Nondiscretionary accruals are the difference between total accruals and discretionary accruals.

SIGN_ACC: Indicator variable equal to one if total deseasonalized accruals are strictly positive, zero otherwise SEASONALITY: Difference between total CFO and deseasonalized CFO.

CFO_VOLATILITY: Standard deviation of quarterly deseasonalized CFO from t 2 16 to t 2 1.

FIRM SIZE: Natural logarithm of market capitalization as of the beginning of the fiscal quarter.

BOOK-TO-MARKET: Ratio of book value of equity to market value of equity, as of the beginning of the fiscal quarter BIG4: Indicator variable equal to 1 if the firm’s auditor is one of the big four auditing companies, and 0 otherwise GINDEX: Governance index from Gompers, Ishii, and Metrich (2003).

In the second column, the dependent variable is the contribution of accruals to tions of CFO aggregated over the next four quarters The coefficients on the independentvariables generally exhibit the same signs as for one-quarter-ahead predictions, but thecoefficients on cash flow volatility, the governance index, and book-to-market are nolonger significant.

predic-Finally, in the last column, we report regression coefficients where the dependentvariable is the contribution of accruals to forecasts of market capitalization In contrast to

SEASONALITY This suggests that the greater the seasonal component of current cashflow, the more useful are accruals in improving forecasts of market values of equity.Overall, the results indicate that managerial opportunism (captured by the magnitude ofdiscretionary accruals and low governance quality) is associated with less informativeaccruals, although accruals tend to be more informative in smaller growth firms withvolatile cash flows

Stock Return Analysis

Stock returns earned by portfolios based on predicted returns As one of our predicted ables is one-quarter-ahead market value of equity, we can test whether out-of-sample

vari-Table 6 Returns on Hedge Portfolios Based on Future Market Capitalization Predictions

Panel A: Quintiles

CFO (1)

CFO andaccruals (2) Earnings (3) (2) 2 (1)

p value for(2) 2 (1)

Panel B: Deciles

CFO (1)

CFO andaccruals (2) Earnings (3) (2) 2 (1)

p value for(2) 2 (1)

Note: This table reports mean equal-weighted abnormal stock returns for portfolios going long on the highest quintile and short on the lowest quintile of the distribution of future return prediction Abnormal returns are com- puted as the intercept from a firm-specific regression of daily returns on the three Fama–French factors (Fama & French, 1993) and momentum Portfolios are rebalanced every fiscal quarter on the filing date of the Form 10-K or 10-Q We compute predicted quarterly stock returns using contemporaneous market capitalization and out-of- sample predictions of one-quarter-ahead market capitalization (plus dividends), both divided by shares outstanding

at the end of quarter t, with three sets of predictors: cash flow from operations (CFO) only, CFO and aggregate accruals, aggregate earnings, all deseasonalized using the X11 procedure The sample includes all firm-quarters from the third quarter of 2002 to the fourth quarter of 2006, preceded by 56 consecutive observations available for CFO, accruals, and market capitalization.

Bold-faced returns are significantly different from zero at the 10 level or higher All p values are based on tailed tests.

forecasts based on current accounting data translate into predictable stock returns In ular, we test whether the contribution of accruals to future cash flow predictions translatesinto superior returns to trading strategies that do not take such a contribution into account.

partic-If those returns are in excess of what common risk factors can explain, this would suggestthat investors misprice securities by not properly adjusting their expectations of future cashflows when provided with information about current earnings and components thereof

Table 7 Accrual Anomaly and Accruals’ Contribution to Future Cash Flow Forecasts

Returns significantly different from zero at the 10 level or higher (two-tailed) are in bold font.

Table 6 reports stock returns earned on hedge portfolios formed on MKTCAPt11tions using information from current earnings and their components The returns are com-

the average 90-day return to a zero-investment portfolio long in the highest and short in thelowest quintile is 1.12% with CFO as the only predictor, 1.13% with CFO and accruals asseparate predictors, and 0.65% with aggregate earnings as the sole predictor However,

When we extend the holding period to 180, 270, and 365 days, the hedge portfolioreturns based on CFO alone decrease with the window length, whereas the returns based

on CFO and accruals increase and become significantly positive Hence, the incrementalreturns earned by using accruals in addition to CFO to rank stocks also increase with theholding period, from 1% on average when positions are held above 180 days to 2% for 365days We find that the excess returns of the accrual-based portfolios are, on average, signif-icantly positive for holding periods of 180 days and more (at the 10 level for 180 days and

at the 01 level for 270 and 365 days) Ninety-day excess returns are not significantly ferent from zero Finally, Panel B reports results based on deciles instead of quintiles of

Accruals’ contribution to future cash flow predictions and the accrual anomaly The results inTable 6 suggest that investors do not fully incorporate the implications of current CFO andaccruals for out-of-sample predictions of all future cash flows in their own expectationsand investment decisions We investigate whether this anomaly is related to the accrualanomaly first documented by Sloan (1996) In particular, we argue that the primary driver

of accounting-based anomalies must be investors’ incorrect expectations of future cashflows, rather than properties of current accounting data per se Hence, we expect that sort-ing stocks on accrual size should not necessarily be associated with predictable stockreturns when we control for accruals’ contribution to out-of-sample predictions of futurecash flows

We report the results in Table 7 The first column (row) indicates portfolio ranks interms of accrual size (accruals’ contribution to one-quarter-ahead market capitalizationforecasts) Hence, stocks in the first cell (1, 1) are for the lowest quintile for accruals andthe lowest accruals’ contribution quintile We test the accrual anomaly by forming portfo-lios every quarter that take a long position in low-accrual stocks (first row) and a short one

in high accruals (fifth row) The mean returns earned by the hedge portfolios are calculatedseparately for each quintile of accruals’ contribution Accruals’ contribution is evaluated expost, so this sorting criterion is not implementable as a trading strategy and is purelydesigned to investigate the performance of the hedge portfolios across different levels of

still holds when accruals’ contribution to predictive accuracy is high

The first panel shows mean 90-day returns across portfolios Except for the highest tile of accruals’ contribution, the mean return to the hedge portfolios based on accrual size

is positive Furthermore, the returns are significantly positive in the second and third tiles (p value below 01) and marginally significant in the first and fourth quartiles (pvalues of 15 and 11, respectively)

quin-In the second panel, the 180-day returns show that there is an accrual anomaly in thefirst four quartiles of accruals’ contribution For instance, the mean return to hedge portfo-lios based on accrual size in the lowest quintile of accruals’ contribution is 5.87%, which issignificantly different from zero at the 05 level In the second, third, and fourth quartiles,

the returns are also significantly positive In contrast, there is no anomaly in the topquintile.

For longer holding periods, the anomaly tends to disappear in most accruals’ tion quintiles However, overall, the results in Table 7 show that for the top quintile ofaccruals’ contribution, there is no evidence of an accrual anomaly, regardless of the holdingperiod

contribu-In addition, we compute the hedge return on a portfolio that takes a long position in lowaccruals and low contribution, and shorts high-accrual stocks with high contribution (again,accruals’ contribution is not known when portfolios are formed) We find that this portfoliodoes not produce significantly positive returns, regardless of the holding period (not tabu-lated) This provides additional evidence supporting the argument that the accrual anomaly

is partly related to accruals’ contribution to out-of-sample predictions of future cash flows

Additional Tests

We supplement our main analysis by running additional tests that address issues related tothe assessment of the relative forecast accuracy of our models at the mean and medianlevels, the use of X11-adjusted data, the relative prediction accuracy of firm-specificregressions compared with cross-sectional regressions, the time-series requirements of ourfirm-specific regressions, and the use of current FCF instead of CFO to predict futureFCF

Comparisons of mean and median absolute forecast errors offer only a limited view ofthe relative predictive ability of different models We assess whether the distribution of pre-diction errors reveals which model outperforms the other by running a stochastic domi-

significantly dominate CFO alone in predicting market capitalization at the beginning andthe end of fiscal quarter t 1 1 in the second degree However, there is no evidence of sto-chastic dominance of CFO and accruals over CFO alone for the prediction of finite mea-sures of CFO

The X11 method is usually performed to address seasonality in macro-level data, and toour knowledge, it has not yet been implemented in the accounting literature Prior studiesanalyzing quarterly accounting data have generally treated seasonality by using variablesfour quarters apart Accordingly, we repeat our main analysis by comparing absolute pre-diction errors from the following models:

j51

to test accruals’ contribution to cash flow predictions Under those specifications, we stillfind that accruals contribute positively to cash flow forecast accuracy However, we noticethat the X11 method produces more accurate forecasts

To validate our claim that firm-specific regression estimates produce more accurate dictions than cross-sectional ones, we compare absolute prediction errors from Models 1 to

pre-4 for a given predicted variable using different estimation procedures We estimate sion coefficients at the following levels: (a) using all firms with data available in a givenyear, (b) by industry (two-digit standard industrial classification code), and (c) by operatingcycle (estimated according to the Dechow et al., 1998 model) For each model, coefficientsare estimated separately fiscal quarter by fiscal quarter Untabulated results show that for agiven set of predictor(s) and predicted variable, firm-specific regressions, on average, out-perform any of the above cross-sectional models The differences are statistically signifi-cant Among the cross-sectional models, industry-level estimates are the most accurateones

regres-To minimize any implications our time-series requirements might have for the ability of our results, we relax our requirements from 56 down to 16 consecutive observa-tions In that case, we still find that accruals improve upon CFO in predicting future cashflows, although their incremental contribution diminishes beyond six quarters of aggrega-tion of the dependent variable Our conclusions with respect to accruals’ contribution tofuture cash flow predictions remain qualitatively unchanged from the conclusions we reachusing the main sample In addition, we still find that firm-specific predictions are signifi-cantly more accurate than cross-sectional ones

generaliz-The results in Tables 3 and 4 suggest that absolute forecast errors are greater whenFCFs are the predicted variable compared with CFO We test whether we can improveFCF forecasts by replacing current CFO with current FCF and accruals with the differencebetween net income and FCF Untabulated results show that this set of predictors improvesupon CFO and accruals in terms of forecast accuracy It also appears that the differencebetween net income and FCF contributes to greater forecast accuracy for future FCFbeyond current FCF

Conclusion

This study investigates the role of earnings components in predicting future cash flows andexplaining the mispricing of securities Although the FASB emphasizes the role of accrualaccounting in helping investors to predict future cash flows, unintentional errors in account-ing estimates and earnings manipulation can decrease the usefulness of accruals in predict-ing future cash flows Our tests are aimed at addressing this empirical issue, that is,documenting whether accruals contribute to the prediction of future cash flows incremen-tally to current cash flow alone and investigating cross-sectional determinants of accruals’contribution to cash flow predictions Subsequently, we investigate whether the extent towhich accruals help improve future cash flow forecast accuracy is associated withaccounting-based stock price anomalies

The key methodological features of our tests are the following: (a) Judgment of thesuperiority of our different models is based on out-of-sample criteria; (b) coefficients areestimated at the firm-level, using time-series data; and (c) predicted variables are not onlyfinite measures of future cash flows but also market values of equity as a surrogate for thepresent value of all future cash flows We use post-SFAS 95 data to measure cash flowsdirectly from the statement of cash flow, and we use quarterly data to obtain a sufficientnumber of observations for our firm-specific estimates To address seasonality in quarterlyaccounting data, we use the X11 procedure developed by the U.S Bureau of Census

We find that the contribution of accruals is most significant when predicting poraneous or next quarter market value of equity as a proxy for all (expected) future cashflows Indeed, mean and median absolute prediction errors are smaller by more than 5% oftotal assets when accruals are included as a predictor These results highlight the impor-tance of assessing the ability of accruals to predict future cash flows by measuring the pre-dicted variable over a sufficiently long horizon We also identify cross-sectionaldeterminants of accruals’ predictive value In particular, we find that the magnitude of dis-cretionary accruals is negatively associated with accruals’ contribution to future cash flowpredictions.

contem-Subsequently, we document that portfolios based on out-of-sample predictions of quarter-ahead stock return can earn positive quarterly risk-adjusted returns when sortingstocks on current CFO and accruals but not CFO alone This suggests that investors do notfully incorporate the contribution of accruals to future cash flow predictions into their ownexpectations In addition, we find that Sloan’s (1996) accrual anomaly is partly related tothe out-of-sample predictive ability of accruals for future cash flows Indeed, the accrualanomaly does not hold when accruals improve the most beyond current cash flow in pre-dicting future cash flows

one-We perform various robustness checks, such as lowering our time-series requirements,addressing seasonality using more conventional methods in the accounting literature (e.g.,using data from quarter t 2 4 to predict quarter t), and estimating regression coefficientscross-sectionally (e.g., per industry) Our conclusions remain qualitatively unchanged bythese alternative research design choices

Collectively, our results may have implications for investors who use current accountingdata for equity valuation purposes Although the ability of accruals to contribute to the pre-diction of finite measures of cash flows varies with model specifications and levels ofaggregations of the dependent variable, it is robust and unequivocally significant whenmarket value of equity is predicted Our results also shed new light on accrual-basedanomalies The evidence that the predictive ability of current accruals for future cash flows

is associated with Sloan’s (1996) anomaly can be of help to researchers (investors) whowish to further understand (exploit) such phenomena

Authors’ Note

JEL Classifications: M41, G14, M43, M44

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

regression analysis shows that the predictive value of accruals is associated with documented tors that affect in-sample accrual quality.

fac-2 As a robustness check, we use two alternative models to address seasonality in our quarterlydata First, we add as independent variables indicators for fiscal quarters to account for the aver-age component of seasonality in quarterly cash flow from operation (CFO) Second, we use datafrom four quarters prior to predict future cash flows Our conclusions with respect to the contri-bution of accruals to future cash flow prediction remain unchanged However, we observe thatX11-adjusted data produce more accurate forecasts than the aforementioned alternatives, a result

of potential relevance to financial statement users

3 As our result is based on the benefit of hindsight (we compute ex-post forecast accuracy), it iscomplementary to studies that document implementable trading strategies using current accruals

4 To address this issue, we lower our time-series requirement to 16 quarters of data available andstill find that accruals contribute to improving cash flow forecasts

5 An alternative hypothesis is that because investors fixate on earnings, they fail to process able information when accruals have high predictive value In that case, one would expectgreater mispricing of stocks for which accruals add the most in terms of cash flow forecastaccuracy

valu-6 We measure free cash flow (FCF) using Damodaran’s (2004) definition of FCF to equity, which

is net income 2 (1 2 d) 3 (capital expenditure 2 depreciation) 2 (1 2d) 3 D working capital,where d is debt to total assets ratio We repeat our analysis using another well-known definition

of FCF (cash flows from operations 2 capital expenditures) and obtain similar results

7 Consistent with prior research, we measure market values of equity as of fiscal period end (seeBarth, Beaver, Hand, & Landsman, 2005; Barth, Cram, & Nelson, 2001)

8 Hribar and Collins (2000) note that Compustat reports’ year-to-date interim cash flow statementdata, so we take differences between quarters t and t 2 1 to obtain quarterly numbers

9 X11 cannot be performed if there are missing observations in the middle of the series

10 The motivation for this procedure is to avoid using information not available at the time the diction is being made, such as shares issued or repurchased during quarter t 1 1 Results are qua-litatively unchanged when predicted MKTCAPt11 is divided by the number of sharesoutstanding at the end of quarter t 1 1 Note that we predict the sum of MKTCAPt11and divi-dends, so our predictions are not based on actual dividends paid during t 1 1

pre-11 Hribar and Collins (2002) show that mergers, acquisitions, and divestitures create significanterrors in estimates of accruals measured with changes in balance sheet accounts

12 We also attempt to distinguish between long-term accruals (depreciation and amortization) andothers but find no improvement compared with aggregate accruals The evidence in Barth et al.(2005) also suggests that CFO and aggregate accruals tend to better predict market value ofequity than CFO and accrual components in terms of median ABSE, although components dobetter at the mean level They offer no explanation as to why this may be the case Finger (1994)and Lorek and Willinger (1996) argue that, although theoretically superior, models includingmore predictors do not always outperform simpler models because of reduced degrees offreedom

13 In nontabulated tests, we also add special items as an independent variable We find a cantly negative association between the magnitude of special items and the ability of totalaccruals to predict one-quarter-ahead CFO incrementally to current CFO This suggests that spe-cial items tend to have poor predictive ability for future cash flows, most likely due to their non-recurring nature

signifi-14 In untabulated tests, we also add average annual sales growth over the last 3 years as an dent variable However, we do not find any significant association between sales growth andaccruals’ contribution

15 Technically, we cannot form portfolios until the last Form 10-Q or 10-K of the quarter is filed.Alternatively, and as is conventional in the literature, we form portfolios as of the filing deadline.

We find qualitatively similar results

16 To evaluate the statistical significance of returns earned by each portfolio, we use a Fama–McBeth procedure: Each quarter, we measure the cross-sectional mean return for each portfolio(CFO alone, CFO and accruals, and aggregate earnings) We subsequently compute t statistics,which are based on the mean and standard deviation of the time series of portfolio returns com-puted fiscal quarter by fiscal quarter, that is, 18 observations from 2002 to 2006

17 Although differently motivated, the test we perform in this section is similar in design to thevalue/price ratio tests of Frankel and Lee (1998) We check that our results are robust to a value/glamour trading strategy by performing a 5 3 5 portfolio ranking where stocks are rankedamong 25 portfolios based on their predicted returns using current accounting data and theirbook-to-market ratio Untabulated tests reveal that (a) ranking stocks on book-to-market does notyield significantly positive returns in our sample (possibly because we already adjust for fourfactors, including book-to-market), (b) hedge returns based on CFO and accruals as predictorssignificantly outperform book-to-market for all horizons, and (c) the incremental returns earned

by using CFO and accruals as opposed to CFO alone as a predictor hold across book-to-marketquintiles, although not always at a statistically significant level (which we attribute to a lack ofpower, given that we have as few as 31 firms in a given cell of the 5 3 5 table.)

18 For the same reason, we are not interested in comparing stock returns going long on high-accrualcontribution and short on low accrual contribution stocks across accrual size quintiles

19 We choose this test following Tse and Zhang (2004), who conclude that it is superior to theAnderson (1996) and the Kaur, Rao, and Singh (1994) tests For the sake of brevity, we do notdescribe the test in this article, but it basically consists of performing comparisons of the distri-butions of absolute forecast errors for two models (e.g., cash flow only vs cash flow andaccruals) at various points and assessing the statistical significance of the differences Overall,Model A is said to first-degree dominate Model B if at all points of the distribution, where thetest is performed, the absolute prediction error from Model A is significantly lower than that ofModel B Second-degree dominance is achieved if the error from Model A is always lower andsignificantly so at least at one point

20 Details can be found at the U.S Census Bureau website (http://www.census.gov/ts/TSMS/KarenH/X-11BookSummary.pdf) See also Ladiray and Quenneville (2001)

of Iowa, Northwestern University

Barth, M E., Beaver, W H., Hand, J R M., & Landsman, W R (2005) Accruals, accounting-basedvaluation models, and the prediction of equity values Journal of Accounting, Auditing, andFinance, 20, 311-345

Barth, M E., Cram, D P., & Nelson, K K (2001) Accruals and the prediction of future cash flows.Accounting Review, 76, 27-58

Bartov, E., & Mohanram, P (2004) Private information, earnings manipulation and executive option exercises Accounting Review, 79, 889-920

stock-Cheng, Q., & Warfield, T (2005) Equity incentives and earnings management Accounting Review,

80, 441-477

Copeland, T., Koller, T., & Murrin, J (1994) Valuation: Measuring and managing the value of panies (2nd ed.) New York, NY: John Wiley

com-Damodaran, A (2004) Applied corporate finance: A user’s manual New York, NY: John Wiley

Davidson, R., & Duclos, J.-Y (2000) Statistical inference for stochastic dominance and for the surement of poverty and inequality Econometrica, 68, 1435-1464.

mea-Dechow, P (1994) Accounting earnings and cash flows as measures of firm performance: The role

of accounting accruals Journal of Accounting and Economics, 18, 3-42

Dechow, P., & Dichev, I (2002) The quality of accruals and earnings: The role of accrual estimationerrors Accounting Review, 77, 35-59

Dechow, P., & Ge, W (2006) The persistence of earnings and cash flows and the role of specialitems: Implications for the accrual anomaly Review of Accounting Studies, 11, 253-296

Dechow, P., Kothari, S P., & Watts, R L (1998) The relation between earnings and cash flows.Journal of Accounting and Economics, 25, 133-168

Dechow, P., Sloan, R., & Sweeney, A (1995) Detecting earnings management Accounting Review,

Frankel, R., & M C L Charles (1998) Accounting valuation, market expectation, and tional stock returns, Journal of Accounting and Economics, 9, 195-228

cross-sec-Gompers, P., Ishii, J., & Metrick, A (2003) Corporate governance and equity prices QuarterlyJournal of Economics, 118, 107-155

Greenberg, R R., Johnson, G L., & Ramesh, K (1986) Earnings versus cash flow as a predictor offuture cash flow measures Journal of Accounting, Auditing, and Finance, 1, 266-277

Henderson, R (1916) Note on graduation by adjusted average Transactions of the American Society

of Actuaries, 17, 43-48

Hribar, P., & Collins, D W (2000) Earnings-based and accrual-based market anomalies: One effect

or two? Journal of Accounting and Economics, 29, 101-124

Hribar, P., & Collins, D W (2002) Errors in estimating accruals: Implications for empirical research.Journal of Accounting Research, 40, 105-134

Jones, J (1991) Earnings management during import relief investigations Journal of AccountingResearch, 29, 193-228

Kaur, A., Rao, B L S P., & Singh, H (1994) Testing for second-order stochastic dominance of twodistributions Econometric Theory, 10, 849-866

Kim, M., & Kross, W (2005) The ability of earnings to predict future operating cash flows has beenincreasing, not decreasing Journal of Accounting Research, 43, 753-780

Klein, A (2002) Audit committee, board of director characteristics, and earnings management.Journal of Accounting and Economics, 33, 375-400

Ladiray, D., & Quenneville, B (2001) Seasonal adjustment with the X11 method (Lecture Notes inStatistics) New York, NY: Springer-Verlag

Lev, B., Siyi, L., & Sougiannis, T (2010) The Usefulness of Accounting Estimates for predictingcash flows and earnings Review of Accounting Studies, 15, 779-807

Lorek, K S., & Willinger, G L (1996) A multivariate time-series prediction model for cash-flowdata The Accounting Review, 71, 81-102

Macaulay, F (1931) The smoothing of time series New York, NY: National Bureau of EconomicResearch

Marquardt, C., & Wiedman, C (2004) The effect of earnings management on the value relevance ofaccounting information Journal of Business, Finance, and Accounting, 31, 297-332.

Ohlson, J (1995) Earnings, book values, and dividends in equity valuation ContemporaryAccounting Research, 11, 661-687

Penman, S., & Sougiannis, T (1998) A comparison of dividend, cash flow, and earnings approaches

to equity valuation Contemporary Accounting Research, 15, 343-383

Preinreich, G (1938) Annual survey of economic theory: The theory of depreciation Econometrica,

6, 219-241

Shiskin, J (1958) Decomposition of economic time series Science, 128, 1539-1546

Shiskin, J., Young, A H., & Musgrave, J C (1967) The X-11 variant of the census method II nal adjustment program (Technical Paper No 15, Revised) Washington, DC: Bureau of theCensus

seaso-Sloan, R (1996) Do stock prices fully reflect information in cash flows and accruals about futureearnings? The Accounting Review, 71, 289-315

Tse, Y K., & Zhang, X (2004) A Monte Carlo investigation of some tests for stochastic dominance.Journal of Statistical Computation and Simulation, 74, 361-378

Wallis, K F (1982) Seasonal adjustment and revision of current data: Linear filters for the X-11method Journal of the Royal Statistical Society, Series A, 145, 74-85

Wilson, G P (1986) The relative information content of accruals and cash flows: Combined dence at the earnings announcement and annual report release date Journal of AccountingResearch, 24, 165-200

evi-Xie, H (2001) The mispricing of abnormal accruals The Accounting Review, 76, 357-373

Yoder, T (2007) The incremental predictive ability of accrual models with respect to future cashflows (Working paper) Starkville: Mississippi State University

Auditing & Finance 27(2) 177–207

Ó The Author(s) 2012 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/0148558X11409150

http://jaaf.sagepub.com

Ex Ante Severance

Agreements and Timely

Disclosures of Bad News

Abstract

This study explores the puzzle of CEO severance agreements by examining the associationbetween the existence of ex ante severance agreements and the timeliness of bad news dis-closures Classifying severance agreements by type and the way boards grant them, thisarticle documents a positive association between the timeliness of bad news disclosuresand the existence of an ex ante single-trigger severance agreement, especially when it isgranted alone This association remains positive in the CEO’s last year of tenure where per-formance is poor Further analyses show that this association is stronger among CEOs with

a high-variable pay structure than among CEOs with a low-variable pay structure Theseresults suggest that an ex ante single-trigger severance agreement may play a role in form-ing timely disclosures of bad news and that paring it with a high-variable pay structureenhances the chance of its success

Keyword

severance, golden parachute, bad news disclosure, timeliness, executive compensation

This study explores the financial reporting effect of CEO severance agreements by gating the association between the existence of ex ante severance agreements and the time-liness of bad news disclosures CEO severance agreements present a quandary because theyguarantee a significant amount of money if the CEO’s employment is terminated This istrue even if the termination is based on poor performance This promise of paymentreduces the deterrent effect of employment termination against adverse selection and moralhazard that arise from the separation of ownership and management The payment itselfalso directly reduces funds available for business operations, and it sends a signal ofrewarding failure to incoming executives (Bebchuk & Fried, 2004) Why do boards signseverance agreements with CEOs? Except for the study by Rusticus (2006), which exam-ines the relationship between ex ante severance agreements and CEO turnover, empirical

Taking a step in this direction, this study tests the association between ex ante severanceagreements and the timeliness of bad news disclosures.

Analytical models suggest that ex ante severance agreements encourage CEOs to close private information early (Inderst & Mueller, 2010; Laux, 2008) Following prior lit-erature (Rau & Xu, 2008; Rusticus, 2006; Schwab & Thomas, 2006), this study classifiesseverance agreements by type: the single-trigger severance agreement (ST) and the double-trigger severance agreement (DT), which is also called a golden parachute In addition,severance agreements are classified based on the way boards grant them Boards may grant

dis-a single-trigger severdis-ance dis-agreement only (STONLY), dis-a double-trigger severdis-ance dis-agreementonly (DTONLY), or both (STDT) to their CEOs Although both types of severance agree-ments (ST = STONLY 1 STDT and DT = DTONLY 1 STDT) are triggered by the termina-tion of employment, the termination under the double-trigger agreement must be within aspecified period following a defined change in control of the firm Prior studies show thatthe DT type of severance agreements is also used as a means of takeover defense(Gompers, Ishii, & Metrick, 2003; Singh & Harianto, 1989; Stein, 1988) Its effect on badnews disclosures may be different from that of the single-trigger severance agreement.Thus, it is also important to examine severance agreements by the way boards grant them.Using the full sample, a positive association is found between the timeliness of bad newsdisclosures and the single-trigger severance agreement (ST) The result is primarily driven bythe single-trigger severance agreement when it is granted alone (STONLY) The results arerobust after controlling for the possible effects of firm size, historical accounting practice,information demand by debt holders, the legal environment of the industry, CEO tenure, andtakeover probability These findings suggest that the single-trigger severance agreement mayplay a role in forming timely disclosures of bad news The association is also tested in thecontext of the CEO’s last year of tenure, where performance is typically poorer than innormal years It is in poor performance years that timely disclosures of bad news are espe-cially valuable to the boards In this context, the positive association between the timeliness ofbad news disclosures and the single-trigger severance agreement, ST or STONLY, still exists.This association with the timeliness of bad news disclosures is not found for the sever-ance agreement type DT When how boards grant severance agreements is considered, noassociation is found for the double-trigger severance agreement alone (DTONLY) or thecombined severance agreements (STDT) This is consistent with the prior literature that thedouble-trigger severance agreement is used as a takeover defense tool (Gompers et al.,2003; Singh & Harianto, 1989; Stein, 1988) Lowering the probability of being a takeovertarget also lowers the probability of triggering the severance agreement Thus, the incentivefrom the double-trigger severance agreement to disclose bad news early is constrained.Partitioning the full sample on the degree to which company performance influences theCEO’s compensation, this study has two subsamples: a high-variable pay structure and alow-variable pay structure A high-variable pay structure means that a significant amount ofthe CEO compensation varies with the performance of the company The converse is truewith a low-variable pay structure On conducting tests within the subsamples, I find that thepositive association between the single-trigger severance agreement alone (STONLY) and thetimeliness of bad news disclosures is significant in both cases However, the positive associa-tion is stronger among CEOs with a high-variable pay structure than among CEOs with alow-variable pay structure These findings are consistent with the theoretical predictions ofLevitt and Snyder (1997), Inderst and Mueller (2010), and Van Wesep (2008)

This study makes two contributions to the literature First, it sheds light on the role exante CEO severance agreements may play in timely disclosures of bad news The effect

suggested by theoretical models (Inderst & Mueller, 2010; Laux, 2008) that ex ante ance agreements encourage early disclosures of bad news applies to the single-triggerseverance agreement type but not to the double-trigger severance agreement Second, thisstudy provides some insights into corporate governance If boards want to take advantage

sever-of the effect sever-of the single-trigger severance agreement alone on early disclosures sever-of badnews, I find that pairing the severance agreement with a high-variable CEO pay structuremay enhance the power of this mechanism

The remainder of this article includes these sections: ‘‘Severance Agreements andPayments,’’ ‘‘Hypothesis Development and Empirical Model,’’ ‘‘Sample Selection andDescriptive Statistics,’’ ‘‘Results,’’ ‘‘Additional Tests,’’ and ‘‘Conclusion.’’

Severance Agreements and Payments

Severance agreements are roughly classified as either single-trigger or double-trigger ance agreements The difference between the two types is that under the double-triggerseverance agreement, the employment termination must be within a specified period oftime following a defined change in control over the firm However, both types of severanceagreements promise CEOs pay and benefits if they lose their position These severanceagreements qualify CEOs for the pay and benefits if their employment is terminated with-out cause or they resign for a good reason What is considered ‘‘cause’’ varies acrossfirms The most commonly specified causes are ‘‘willful misconduct,’’ ‘‘moral turpitude,’’and ‘‘failure to perform duties’’ (Schwab & Thomas, 2006) CEOs are considered to haveresigned for a ‘‘good reason’’ if their firm altered their duties, failed to compensate them

sever-as promised, or relocated them (Schwab & Thomsever-as, 2006) Leaving for a position atanother firm is not a good reason for resigning from the present employment Terminationwith cause or resignation without a good reason does not trigger the severance pay underany type of severance agreement

An ex ante severance agreement typically provides twice of the CEO’s annual salaryand bonus; it allows the CEO to accelerate the vesting of his/her stock options andrestricted stock, and it offers the CEO a supplemental executive retirement plan, insurance,and other perquisites (Rau & Xu, 2008; Rusticus, 2006; Schwab & Thomas, 2006;Yermack, 2006) Under a severance agreement, termination without cause and resignationfor a good reason award similar amounts to the departing CEO (Schwab & Thomas, 2006).However, in these same situations, the amount specified in a double-trigger severanceagreement is usually larger than the amount in a single-trigger severance agreement This

is in part because it is the acquiring firm not the currently hiring firm that will make thepayment later (Schwab & Thomas, 2006) In ex post settlements, CEOs typically receive amore generous payout than is specified in the ex ante severance agreement; CEOs removedfrom office usually receive a higher separation pay than CEOs who voluntarily retire(Bebchuk & Fried, 2003, 2004; Yermack, 2006)

Hypothesis Development and Empirical Model

Hypotheses

The modern corporate structure features the separation of ownership and management Thiscreates a situation where the CEO has an information advantage over the board From dailymanagement, a CEO gains insight into issues such as whether his/her human capital fits theneeds of the firm, what likelihood the current corporate strategy has of succeeding, and

whether the investment projects will make profits The separation of ownership and agement also suggests that the CEO’s objective and the shareholders’ may not be congru-ent When things go wrong, the CEO worries about his/her private benefits and may beunwilling to reveal negative information promptly as releasing this bad news may have anadverse impact on the CEO’s private benefits In addition, stock markets react negatively

man-to bad news (e.g., Skinner, 1994) When the sman-tock price falls, the CEO’s current wealthshrinks and the award of stock options or restricted stock is reduced With releases of badnews, the CEO may also face a termination of employment, loss of directorship, and diffi-culty in finding another CEO position Given these concerns, some managers admit thatthey delay disclosures of bad news, hoping that good news will come out in the subsequentperiod and then investors will not notice the changes (Graham, Harvey, & Rajgopal, 2005).These intentional delays of bad news and gradual release of good news lead to asymmetricmarket reactions to bad news and good news (Kothari, Shu, & Wysocki, 2009)

Lack of timely private information from managers prevents boards from taking sary actions to minimize shareholder loss For example, managers may engage in ‘‘empirebuilding,’’ keeping negative present value projects going Without timely information onprojects, boards are unable to promptly identify loss creating projects and terminate thoseprojects (Watts, 2003) Regardless of their productivity, mangers may also keep resources

neces-to themselves Without information on individual productivity, boards cannot reassignresources for their best use (Eisfeldt & Rampini, 2008) Without knowing managers well,boards may have unrealistic expectations of management ability If managers believe thatthey cannot meet the board’s expectations even given their best effort, they may undertakeprojects with excessive risks that the company cannot afford (Van Wesep, 2008) Overall,being able to access management’s private information early is important for boards tomaximize shareholder value

Based on the preceding discussion, early disclosures of private information by managers

do not occur naturally Incentives are needed to achieve that Levitt and Snyder (1997)model the information flow from the agent to the principal Theoretically, the principal canobtain private information early from the agent without cost If the principal does not inter-vene in business operations, the disclosures on their own will not affect the agent’s privatebenefits In this case, the agent is indifferent to revealing or hiding private information.However, if the principal does take action based on the revealed information, then theagent will be unwilling to disclose private information because the agent’s private benefitsmay suffer Therefore, the principal has to reward early disclosures of private information

if the principal wishes to avoid future nondisclosure or delays in disclosure Applying thisline of reasoning to the CEO (the agent) and the board (the principal), if the board does nottake action using the CEO’s private information, then the board does not need to provideincentives for the CEO to disclose the information However, if the board relies on theCEO’s private information to take action, then the board has to reward early revelation ofthe CEO’s private information In practice, it is easy for boards not to take action givengood news but difficult not to do so given bad news An extreme example is that 24% ofthe CEOs in Yermack’s (2006) sample were forced to leave by their boards

An ex ante severance agreement provides an incentive for the CEO to disclose badnews early Inderst and Mueller (2010) argue that the CEO evaluates the costs and bene-fits of revealing bad news If disclosures eventually cost the CEO job and there is nothing

in return, then the CEO will try to hide bad news If the CEO receives a severance paywhile losing the job, then the CEO is better motivated to disclose bad news Similarly,Van Wesep (2008) argues that severance pay helps the board to distinguish high-quality

CEOs from low-quality CEOs because severance pay induces low-quality CEOs to close bad news early and then leave The value of bad news increases in proportion to theeagerness that the board has for information Bad news appears to be more valuable if theboard has an aggressive policy of replacing the incumbent CEO (Laux, 2008) These the-oretical studies suggest that the severance agreement would lessen the anxiety of theCEO over the negative consequences of disclosing bad news Even in the worst-case sce-nario of disclosures leading to a termination of employment, the severance agreementwill compensate the CEO Therefore, the promise of severance pay should encourage theCEO to disclose bad news early This leads to the first hypothesis stated in the alternateform:

dis-Hypothesis 1: The existence of ex ante severance agreements and the timeliness ofbad news disclosures are positively associated

A properly designed compensation package improves the chance of a successful transfer

of information from the CEO to the board of directors To make the information transferwork effectively, the CEO’s compensation should include stock options or other forms ofperformance-based pay (Inderst & Mueller, 2010; Van Wesep, 2008) This part of compen-sation varies with company performance As a consequence, the annual pay may be lowdue to poor performance Lacking a severance agreement, the CEO will prefer the lowannual pay to being fired and receiving nothing if disclosures of bad news eventually pro-voke the termination of employment In this case, the CEO is financially better off hidingbad news However, with a severance agreement in place, the CEO is better off disclosingbad news because the severance pay is typically twice the annual salary and target bonus,and it allows accelerated vesting of options (Rau & Xu, 2008; Rusticus, 2006; Schwab &Thomas, 2006; Yermack, 2006) This pay from severance is greater than the performance-based compensation computed on the poor performance, and the pay will be realized in ashort period of time

The effect of the severance agreement is expected to be more salient with a variable pay structure Compared with a low-variable pay structure, a high-variable paystructure features a relatively low salary but a relatively high amount of target bonus andoptions In the case of poor performance, a high-variable pay structure substantially lowersthe CEO’s expected annual pay because the performance-based pay is small This widensthe difference between the annual pay and the severance pay, making the severance pay rel-atively more attractive Moreover, it is to the CEO’s great advantage to accelerate the vest-ing of a large number of stock options before the stock price plummets The incentive forthe CEO with a low-variable pay structure to disclose bad news early is weaker becausethe difference between the annual pay and the severance pay is smaller, and the amount ofoutstanding options is also lower Thus, a high-variable pay structure provides a strongerincentive for the CEO to communicate bad news earlier than a low-variable pay structure.Thus, the second hypothesis stated in the alternate form is as follows:

high-Hypothesis 2: The association between the existence of ex ante severance agreementsand the timeliness of bad news disclosures is stronger among CEOs with a high-variable pay structure than among CEOs with a low-variable pay structure

Empirical Model

This study examines disclosures that take the form of recognizing bad news in accountingearnings Basu’s (1997) model separates good news and bad news, and captures the timeli-ness of these disclosures In his model, good news is indicated by positive stock returnsand bad news by negative stock returns The relationship between the stock returns andaccounting earnings indicates the timeliness of news disclosures The Basu (1997) model isspecified as follows:

where EARN is earnings before extraordinary items, RET is the buy-and-hold return, andNEG is a dummy variable that equals 1 for a negative RET and 0 otherwise The coeffi-cients on the variable RET and the interaction term NEG 3 RET capture the timeliness ofgood news disclosures and the incremental timeliness of bad news disclosures,respectively

The Basu (1997) measures may be contaminated by the accounting treatment of nomic events in the past (Roychowdhury & Watts, 2007) To address this concern, themarket-to-book (MTB) ratio is included as a control variable in this study’s empiricalmodel because the composition of equity value at the beginning of the year reflects thecumulative effect of past accounting practice (Roychowdhury & Watts, 2007) This studyalso controls for other variables that have a potential impact on timely disclosures of badnews, including firm size, leverage, litigation risk, CEO tenure, and takeover probability.They are discussed below

eco-The full empirical model testing the association between ex ante severance agreementsand the timeliness of bad news disclosures is as follows:

EARNi;t5b01b1NEGi;t1b2RETi;t1b3NEGi;t3RETi;t1b4SAi;t1

1b5NEGi;t3SAi;t11b6RETi;t3SAi;t11b7NEGi;t3RETi;t3SAi;t1

1b8SIZEi;t11b9NEGi;t3SIZEi;t11b10RETi;t3SIZEi;t1

1b11NEGi;t3RETi;t3SIZEi;t11b12MTBi;t11b13NEGi;t3MTBi;t1

1b14RETi;t3MTBi;t11b15NEGi;t3RETi;t3MTBi;t1

1b21NEGi;t3LITi;t11b22RETi;t3LITi;t11b23NEGi;t3RETi;t

3LITi;t11b24TENUREi;t11b25NEGi;t3TENUREi;t11b26RETi;t

ð2Þ

where for firm i in year t, EARN = earnings before extraordinary items of the year scaled

by the market value of equity at the beginning of the year; RET = the buy-and-hold stockreturn over the fiscal year; NEG = 1 if RET is negative and 0 otherwise; SIZE = the yearlydecile rank value of the market value of equity at the beginning of the year, scaled by nine;

SA = 1 if an ex ante severance agreement exists and 0 otherwise; MTB = the yearly decilerank value of the ratio of the market value of equity to the book value of equity at the

beginning of the year, scaled by nine; LEVERAGE = the yearly decile rank value of theratio of total debt to total assets at the beginning of the year, scaled by nine; LIT = 1 if thestandard industry classification (SIC) code of the firm falls in one of these ranges, 2833-

2836, 3570-3577, 3600-3674, 5200-5961, and 7370-7374, and 0 otherwise; TENURE = thedecile rank value of the number of years that a CEO is in office with the company at thebeginning of the year; and TAKEOVER = the probability of being a takeover target at thebeginning of the year

In this model, NEG 3 RET 3 SA is the variable of interest, the coefficient of whichsuggests whether the existence of ex ante severance agreements and early disclosures ofbad news are positively associated SA shows the existence of the ex ante severance agree-ment by type—the single-trigger severance agreement (ST) and the double-trigger sever-ance agreement (DT) or by the way boards grant them—a single-trigger severanceagreement alone (STONLY), a double-trigger severance agreement alone (DTONLY), orboth types of severance agreements to the same CEO (STDT) The coefficient on NEG 3RET 3 SA is expected to be positive

Following the study by LaFond and Roychowdhury (2008), variables SIZE, MTB, andLEVERAGE take the scaled decile rank values of their respective underlying measures.Each of these variables is the yearly decile rank of its underlying measure from 0 to 9 andthen scaled by nine to make the values of these variables fall between 0 and 1 The under-lying measure of SIZE is the market value of equity (MV), that of MTB is the ratio of themarket value of equity to the book value of equity (MV/BV), and that of LEVERAGE is theratio of total debt to total assets (DEBT/ASSETS) TENURE is the scaled decile rank value

of CEOTENURE, which is the number of years that a CEO is with the company

The predicted sign of the coefficient on NEG 3 RET 3 SIZE is negative, as large firmsare less likely to report bad news in a timely manner (Givoly, Hayn, & Natarajan, 2007;LaFond & Watts, 2008) The coefficient on NEG 3 RET 3 MTB is also predicted to benegative because a high level of unrecorded good news, captured by a high value of MTB,lowers the need to record bad news in the subsequent period (Roychowdhury & Watts,2007) The expected sign of the coefficient on NEG 3 RET 3 LEVERAGE is positivebecause debt holders exhibit a strong demand for timely disclosures of bad news (Ball,Robin, & Sadka, 2008; Beatty, Weber, & Yu, 2008; Frankel & Roychowdhury, 2007;Wittenberg-Moerman, 2008; Zhang, 2008) Firms in high litigation risk industries tend todisclose bad news early, so the coefficient on NEG 3 RET 3 LIT is predicted to be posi-tive (Basu, 1997; Watts, 2003) The dummy variable LIT equals 1 if the SIC code of thecompany falls in one of these ranges, 2833-2836, 3570-3577, 3600-3674, 5200-5961, and7370-7374 (Francis, Philbrick, & Schipper, 1994), and 0 otherwise The predicted sign ofthe coefficient on NEG 3 RET 3 TENURE is positive Prior studies find that CEOs withlong tenure are very powerful and their compensation is very well protected (Adut, Cready,

& Lopez, 2003; Faulkender & Yang, 2010) Veteran CEOs are less afraid to disclose badnews early

The TAKEOVER variable is the probability of being taken over at the beginning of theyear This probability is calculated using the Billet and Xue’s (2007) probit model:

ð3Þ

where TARGET is 1 if the firm is a takeover target in a given year; ROAIA is the operatingincome before depreciation divided by total assets, minus the median ratio for all firmswithin the same two-digit SIC code; SIZEEQ is the natural log of common equity; LEVBIA

is total debt divided by total assets minus the median ratio for all firms within the sametwo-digit SIC code; MV/BV is the market value of common stock divided by the bookvalue of common equity; SALEGR is the natural log of the ratio of sales of the target yearover the sales of the previous year; NPPE is net plant, property, and equipment scaled bytotal assets; and ITODUM is a dummy variable equal to 1 if a company in the same four-digit SIC code was a target in the previous year, and 0 otherwise The takeover informationwas obtained from Thomson One (1992–2007) and the financial information is fromCompustat (1992-2006) The coefficient on NEG 3 RET 3 TAKEOVER is expected to benegative as firms subject to a takeover are very defensive (Field & Karpoff, 2002; Stein,1988), and when faced with high takeover probabilities, they are less likely to disclose badnews early

Sample Selection and Descriptive Statistics

RiskMetrics (formerly, Investor Responsibility Research Center [IRRC]) provides theseverance agreement information The main sample starts with 12,531 observations fromRiskMetrics’ governance biannual data (1993-2006) New governance data became avail-able in July 1993, July 1995, February 1998, February 2000, February 2002, January 2004,and January 2006 The RiskMetrics’ data are most applicable to the firm fiscal year endingafter the RiskMetrics information was updated So the severance agreement data and the com-pany characteristic data are matched based on the relative relationship between the RiskMetrics

10,904 observations have corresponding stock return information from the Center for Research

in Security Prices (CRSP) The CEO compensation information is retrieved from Standard &Poor’s ExecuComp data set, which begins with fiscal year 1992 After deleting the interimCEOs who were in office for less than a year and the CEOs who left office due to death, thesample is reduced to 8,665 observations Another 31 observations are lost because they have

no SIC information Firms in the utility and financial industries (SIC codes 4900-4999 and6000-6900) are excluded because these firms are subject to different legal constraints Thiscauses the sample size to drop to 6,884 observations To test Hypothesis 2, this study requirescompensation data to calculate the volatility of CEO’s performance-based pay This require-ment further reduces the sample size to 6,764 observations Without data to calculate thetenure of the CEO, another 373 observations are dropped out of the sample After excludingobservations without required data to calculate the takeover probability at the beginning of theyear, the final full sample includes 6,140 observations

Table 1 presents descriptive statistics on the sample In Panel A, the statistics show thatthe mean of ST is 7.3% and that of DT is 64.7% Approximately 6.3% of CEOs have thesingle-trigger severance agreement alone (STONLY), 63.8% of CEOs have the double-trigger severance agreement alone (DTONLY), and 0.9% of CEOs have both types of sever-ance agreements (STDT) This is consistent with other studies of severance agreements,which also conclude that more double-trigger severance agreements are granted thansingle-trigger severance agreements (Rau & Xu, 2008; Rusticus, 2006) On average, thesample has a positive income before extraordinary items scaled by the beginning marketvalue of equity (EARN) The sample also shows an average of 13.3% of 1-year buy-and-hold return (RET), but 40.0% of the firms experience a negative return (NEG) The market

value of equity (MV) varies dramatically across sample firms, with an average value

of US$6,341 million The average market-to-book ratio (MV/BV) is 3.371 Approximately21.4% of corporate capital comes from debt (DEBT/ASSETS) and 31.5% of thesample firms are in industries with a high litigation risk (LIT) The mean value ofCEOTENURE is 7.449 years The average takeover probability (TAKEOVER) of thesefirms is 9.9%

Panel B reports the correlations between the variables The Pearson product–momentcorrelations are reported above the diagonal and the Spearman rank-order correlations aregiven below the diagonal The correlation between ST and DT is significantly negative,with a correlation coefficient of 2.30 ST and STONLY exhibit a positive correlation withLIT DT and DTONLY show a positive correlation with DEBT/ASSETS and a negative cor-relation with LIT, CEOTENURE, or TAKEOVER The correlation between EARN and RET

is significantly positive and the correlation between EARN and NEG is significantly tive, suggesting that firms’ accounting systems and the stock market capture similar eco-nomic events It is not surprising that the correlation coefficient of ST and STONLY andthat of DT and DTONLY are highly positive, and that of NEG and RET is highly negative.The Spearman correlation between MV and MV/BV is 50 No other correlation coefficients

Results

In this section, Hypothesis 1 is tested using the full sample And then, Hypothesis 1 istested in the context of the CEO’s last year in office (a smaller sample) It is expected thatperformance in the CEO’s last year of tenure is worse than that in other years Thus, morebad news will be present in this year than in other years, making the private informationfrom the CEO more valuable to the board This context provides a natural experiment totest the association between the ex ante severance agreements and the timeliness of badnews disclosures Following that, the full sample is partitioned based on the proxy for thepay structure Model (2) is then estimated with each subsample, that is, the high-variablepay structure group and the low-variable pay structure group, to test Hypothesis 2

Testing Hypothesis 1

Full sample Table 2 reports the test results for the association between the existence of

ex ante severance agreements and the timeliness of bad news disclosures The association

is first examined by the type of severance agreement and then by the way boards grantthem to the CEO Column I presents the results for the examination of the single-triggerseverance agreement ST equals 1 if the single-trigger severance agreement is granted alone(STONLY) or together with a double-trigger severance agreement (STDT), and 0 otherwise.The coefficient on NEG 3 RET 3 ST (b = 0.078, t = 1.37) is significantly positive (p \.10) This suggests that bad news is disclosed in a more timely manner by CEOs with

a single-trigger severance agreement than by CEOs without such a severance agreement.This finding supports Hypothesis 1 The coefficients on the control variables are allconsistent with the prior literature The coefficient on NEG 3 RET 3 SIZE is negative,supporting the inverse relationship between firm size and timely disclosures of bad news(b = 20.129, t = 21.82, p \ 05) The coefficient on NEG 3 RET 3 MTB is alsonegative (b = 20.321, t = 24.51, p \ 01) The positive coefficient on NEG 3 RET 3LEVERAGE (b = 0.104, t = 1.98, p \ 05) indicates that firms with a high level of debt dis-close bad news early The coefficient on NEG 3 RET 3 LIT (b = 0.084, t = 2.47, p \ 01)

Table 2 Tests on the Association Between Severance Agreements and Timely Disclosures of BadNews (Full Sample)

suggests that timely disclosures of bad news are more common in industries with a highlitigation risk The coefficient on NEG 3 RET 3 TENURE (b = 0.067, t = 1.29, p \ 10)

is positive suggesting that veteran CEOs are more likely to disclose bad news early,whereas the coefficient on NEG 3 RET 3 TAKEOVER (b = 20.810, t = 23.72, p \ 01)

is negative indicating that firms with high takeover probabilities at the beginning of theyear are less likely to disclose bad news early

Column II reports the results for the examination of the double-trigger severance ment DT equals 1 if the double-trigger severance agreement is granted alone (DTONLY) ortogether with a single-trigger severance agreement (STDT), and 0 otherwise In general,CEOs with the double-trigger severance agreement do not disclose bad news in a moretimely manner than CEOs without such an agreement, as the coefficient on NEG 3 RET 3

agree-DT is not significant (b = 20.018, t = 20.60, p = 27) The results on the control variablesare very similar to those in Column I; therefore, they are not repeated here

of the market value of equity at the beginning of the year, scaled by nine; MTB = the yearly decile rank value of the ratio of the market value of equity to the book value of equity at the beginning of the year, scaled by nine; LEVERAGE = the yearly decile rank value of the ratio of total debt to total assets at the beginning of the year, scaled by nine; LIT = 1 if the standard industry classification (SIC) code of the firm falls in one of these ranges, 2833–2836, 3570–3577, 3600–3674, 5200–5961, and 7370–7374, and 0 otherwise; TENURE = the decile rank value

of the number of years that a CEO is in office with the company; TAKEOVER = the probability of being a takeover target at the beginning of the year It is the predicted value from the probit model below (Billet & Xue, 2007):

where TARGET is 1 if the firm is a takeover target in a given year; ROAIA is the operating income before tion divided by total assets, minus the median ratio for all firms within the same two-digit SIC code; SIZEEQ is the natural log of common equity; LEVBIA is total debt divided by total assets minus the median ratio for all firms within the same two-digit SIC code; MTB is the market value of common stock divided by the book value of common equity; SALEGR is natural log of the ratio of sales over the sales of the previous year; NPPE is net plant, property, and equipment scaled by total assets; and ITODUM is a dummy variable equal to 1 if a company in the same four-digit SIC code was a target in the previous year, and 0 otherwise The takeover information is obtained from Thomson One (1992–2007) and financial information is from Compustat (1992–2006).