DISCRETIONARY ACCRUALS AND MANAGEMENT FORECAST ERRORS BY TIM CAIRNEY

APPLICATION FOR DISCRETIONARY RATE RELIEF QUESTION 1 NAME
Blue_Badge_Discretionary_form_2021
CHARGE RATES 20212022 FWORDORIGINALCHARGES MILES STANDARD DISCRETIONARY 15 4200

DENVER TIER III DISCRETIONARY GRANTS INNOVATIVE COLLABORATION GRANT APPLICATION
DISCRETIONARY ACCRUALS AND MANAGEMENT FORECAST ERRORS BY TIM CAIRNEY
DISCRETIONARY TRUSTS SUMMARY 2 DISCRETIONARY POWERS A DISCRETIONARY

MANAGEMENT FORECASTS AND DISCRETIONARY ACCRUALS



DISCRETIONARY ACCRUALS AND

MANAGEMENT FORECAST ERRORS


by

TIM CAIRNEY

SCHOOL OF ACCOUNTING

FLORIDA ATLANTIC UNIVERSITY

FORT LAUDERDALE, FL, USA 33301

E-MAIL: [email protected]

PHONE: 954 762 5327

and

ALASTAIR MURDOCH

FACULTY OF MANAGEMENT

UNIVERSITY OF MANITOBA

WINNIPEG, MB, CANADA R3T 5V4

E-MAIL: [email protected]

PHONE: 204 474 8439



First draft: 96.04.30

This draft: 98.09.12


Useful comments on previous drafts of the paper were received from Greg Waymire, Bill Hopwood, John Reisch, Bob Bowen, participants at the Certified General Accountants’ Associations / University of Manitoba 1996 Accounting Research Mini-Conference, participants at the 1998 Southeast Regional AAA, Gordon Richardson, Associate Editor, and two anonymous referees. We are grateful to I/B/E/S for supplying analysts’ forecasts, the Faculty of Management at the University of Manitoba for funding the purchase of data from Compustat and CRSP, and Grace Pownall, Charles Wasley, and Greg Waymire for supplying data from their 1993 project.

DISCRETIONARY ACCRUALS AND

MANAGEMENT FORECAST ERRORS

Abstract

Positive accounting theory suggests a number of reasons why management might decide to manage earnings. This paper examines the extent to which such a decision may be a result of management having previously issued a forecast of earnings. Based on a sample of 225 such forecasts issued during the period 1986 to 1992, we find that discretionary accruals are adjusted to bring reported earnings more in line with management’s forecast. We also find that the adjustments are larger when there is greater investor diversity concerning expected earnings for the forecast year.


DISCRETIONARY ACCRUALS AND

MANAGEMENT FORECAST ERRORS


Introduction

Positive accounting theory (Watts and Zimmerman 1986; 1990) suggests a number of incentives for management to manage earnings. For example, higher earnings imply higher management bonuses because annual bonuses are either a direct function of annual earnings or are a direct function of common stock value that is in turn a function of reported earnings (Gaver, Gaver, and Austin 1995; Healy 1985). Also, higher earnings reduce the likelihood of bond covenant violations (DeFond and Jiambalvo 1994). On the other hand, lower earnings may reduce regulation and political scrutiny (Jones 1991). They may also allow higher future earnings - the “Big Bath” hypothesis (Amir and Livnat 1996; Beatty and Verrecchia 1989; Copeland and Moore 1972; Healy 1985).

This paper presents evidence of an additional reason for managing reported earnings: inaccurate management forecasts of annual earnings. We consider the situation where management makes a forecast of annual earnings and then discovers that the forecast will be inaccurate unless the accounts are adjusted prior to reporting earnings. Inaccurate forecasts may be undesirable for the following reasons. First, investors may interpret inaccurate forecasts as indicative of more serious management deficiencies (Trueman 1986). Second, investors may reward a reputation for accurate earnings forecasts with a lower cost of capital (King, Pownall, and Waymire 1990). Third, inaccurate forecasts (particularly overly optimistic forecasts) may lead to lawsuits alleging misrepresentation (Skinner 1994). Thus, there are incentives for management to minimize the difference between forecast and reported earnings. Our research question is:

Under what conditions may management use discretionary accruals to improve the apparent accuracy of its earnings forecast?

We examine this question using a sample of 225 management forecasts issued between 1986 and 1992 and find the following. First, discretionary accruals are managed to achieve a more accurate forecast. Second, more accruals management occurs when there is greater diversity of investors' expectations of earnings.

This paper contributes to the literature in the following ways. Agency theory predicts that agents will manage reporting on their performance because of their inherent self-interest. We find evidence of the use of income adjusting discretionary accruals and, thus, confirm the agency theory prediction as it relates to financial statement reporting, despite such constraints as GAAP. We also contribute to the earnings management literature by demonstrating an additional circumstance when earnings may be managed, namely, after management publicly discloses a forecast of annual earnings. In concurrent research, Kasznik (1996) also provides evidence that discretionary accruals are used to manage earnings after management forecast disclosures. The two papers provide robust results given that they use different methods of estimating discretionary accruals (we use time series methods; Kasznik uses cross-sectional methods), they incorporate different proxies for testing (for example, Kasznik uses reported earnings as a benchmark for pre-managed earnings while we use the last analyst forecast), and they use different statistical tests (we use regression methodology; Kasznik uses t-tests on means). Finally, we contribute to the literature on discretionary accruals by including a performance measure in the regression test for earnings management, as suggested by Dechow et al. (1995).

The rest of the paper is organized as follows. We develop the hypotheses and describe the data and variable selection in the next section. The following two sections present the results and sensitivity tests. We offer conclusions in the final section.


Hypotheses and variable selection

Hypotheses

Prior research has noted the possibility of earnings management by firms issuing management forecasts. Several studies (Hassell and Jennings 1986; Imhoff and Pare 1982; Jaggi 1980; Waymire 1986) have shown that management forecasts are more accurate than preceding analysts’ forecasts. McNichols (1989) finds that management forecasts are generally accurate and unbiased estimates of earnings. However, she uses reported earnings numbers and observes (page 19),

The nature and extent of earnings management by forecasting firms remain as interesting issues for future research.”

We hypothesize:


H1. Management uses discretionary accruals to reduce the difference between reported earnings and management’s forecast of earnings.


We assume that discretionary accruals cannot be easily detected by investors. More specifically, we assume that investors' earnings expectations are diverse, so that each firm has a range of plausible earnings amounts, within which range earnings management is difficult to detect. Thus, within this range, any expected reputation costs associated with detection of earnings management are dwarfed by the expected reputation costs associated with forecast errors, which are highly visible. For instance, if the dispersion of expected earnings is small, then management is less likely to use discretionary accruals to correct forecast errors because the risk of detection is greater. A sufficiently large dispersion, on the other hand, may permit the use of accruals to completely eliminate the forecast error. Consequently, we hypothesize:


H2. Management’s use of discretionary accruals to reduce the difference between reported earnings and management’s forecast of earnings is greater when the diversity in investors' expectations of earnings is greater.


Variables Selection

Testing the hypotheses requires proxies for the following constructs:

Proxy Construct

FE forecast errors.

DA discretionary accruals.

DISP diversity of investor beliefs about expected earnings.

We discuss proxies for these constructs in order.

Our hypotheses propose that management uses discretionary accruals to produce an earnings amount close to the amount forecast. Therefore whether discretionary accruals are needed for this purpose depends on what the earnings amount was prior to the use of the accruals. If this “pre-accrual” earnings amount is close to the earnings forecast, then there is little incentive to use accruals. On the other hand, if there is a large difference between the “pre-accrual” earnings amount and the forecast, then we predict that management will use accruals to reduce this difference. FE measures the difference between management’s forecast of earnings and our proxy for “pre-accrual” earnings. We discuss next how we obtained management forecasts of earnings. Then we discuss possible proxies for “pre-accrual” earnings.

Our sample of point and range management forecasts includes (1) those reported in the Wall Street Journal, for the period January, 1986 to December, 1992, and (2) the point and range annual forecasts from Pownall, Wasley, and Waymire (1993). Qualitative forecasts, being less precise, would not have the level of management commitment offered by the more precise point and range forecasts. Only the first forecast in each fiscal year was selected because later forecasts may be only repeats of the earlier forecast and because the first forecast by management sets the stage for investors to evaluate management's ability to control the firm (Trueman, 1986). The sample was further reduced to control for concurrent news releases (identified through the Wall Street Journal Index). This step was taken to focus our study on the incentives provided by the release of management forecasts only. If a management forecast was issued concurrently with, for example, a takeover announcement, then the information set would be more complex and the forecast may implicitly depend on timeliness of the completion of the takeover, as well as the performance of the new entity. In summary, the incentives to achieve a stand-alone forecast may differ compared with those associated with complex and conditional information packages. These steps reduced our original sample of 1196 forecasts to 745. Concurrent announcements include earnings releases (18%), dividend announcements (7%), investment changes (5%), ownership/management changes (4%), and other announcements (4%).

The second proxy required for the measure of forecast error is “pre-accrual” earnings. One proxy is reported primary earnings per share before extra-ordinary items. This proxy has the advantage of being a reliably reported item that does not require additional judgment on our part. However, it may not be the most relevant proxy since reported earnings include the effects of any earnings management. As a consequence, we seek a more relevant proxy for “pre-accrual” earnings. DeFond and Park (1997) note that analysts’ forecasts of earnings are unlikely to include discretionary adjustments because earnings before discretionary adjustments are easier to forecast. The mean of the last analysts' forecasts made during the last fiscal month is, therefore, likely to include the most information available to make an unbiased statement of the yearend earnings.1

Our second construct, discretionary accruals, is measured using the "modified Jones model" as recommended in Dechow et al. (1995, p. 223). Variables necessary to perform our time-series estimation of discretionary accruals reduced the sample from 745 observations to 225 observations.2 Even though the final sample is much smaller than the original sample, we believe the firm-specific controls offered by the time-series procedure provide clearer inferences than a cross-sectional procedure.3

The third construct, diversity in expected earnings, is proxied by the dispersion of analyst forecasts (Ajinkya, Atiase, and Gift 1991; Morse, Stephan, and Stice 1991). Specifically, we use the standard deviation of the last analyst forecasts in the last fiscal month of the forecasting year, based on the I/B/E/S Summary File.4


Control Variables

Our empirical analysis uses two control variables:

EPS Firm performance.

YEAR Fiscal year of forecast.

Dechow et al. (1995) note that researchers using the discretionary accruals estimation model should control for firm performance. If accruals are available to increase or decrease earnings on a discretionary basis, then their use may be restricted by the magnitude of earnings. To control for firm performance, reported primary earnings per share before extraordinary items (EPS) is included as an independent variable.5

We include a control variable (YEAR) for the fiscal year of the management forecast release for the following reasons. Dechow et al. (1996) report a wide annual variation in the number of firms over their sample period in which the SEC believes earnings manipulation took place. Also, time related differences that relate to macroeconomic and industry wide effects have been observed to impact accruals related ratios (Kane, Graybeal, and Richardson 1996).


Testing Model

Our tests for a linear relation of discretionary accruals with forecast error (hypothesis 1) and with earnings uncertainty (hypothesis 2) employ the following regression models (respectively):

Hyp.1: DA = b0 + b1(FE) + b2(EPS) + Σb3-8(YEAR),

Hyp.2: DA = b0 + b1(FE) + b2(FE*HDISP) + b3(FE*LDISP)+ b4(EPS) + b5-10Σ(YEAR),

where:

DA = discretionary accruals computed using the modified Jones model.

FE = forecast error computed as management’s forecast of earnings

less the mean of the analysts' forecasts during the last fiscal month of the forecasting year.

HDISP = a dummy variable equal to one for observations with high levels

of the standard deviations in the analysts' forecasts in the last fiscal month of the fiscal year.

LDISP = a dummy variable equal to one for observations with low levels

of the standard deviations in the analysts' forecasts in the last fiscal

month of the fiscal year.

EPS = reported primary earnings per share before extraordinary items.

YEAR = discrete variable (1,0) representing the calendar year of the sample.

Hypothesis 1 predicts a positive coefficient for FE, and hypothesis 2 predicts a positive coefficient for FE*HDISP and a negative coefficient for FE*LDISP.


Results

Table 1 presents full sample characteristics and univariate comparisons of optimistic and pessimistic forecasts. Optimistic (pessimistic) forecasts are defined as forecast errors greater than (less than) 0. Variable measures from the "full sample" column 1 indicate that most of our variables are significantly skewed, with means being larger than medians in absolute value. Discretionary accruals are significant and negative at the mean, indicating that our sample of firms use income decreasing adjustments, on average. This mean measure is consistent with that reported in Becker et al. (1998).6 On the other hand, discretionary accruals scaled by the absolute value of the last analyst forecast is positive but insignificant.7

An explanation for the mean negative discretionary accruals may involve subsequent revisions in market expectations. Although we have eliminated firms with subsequent management forecast releases, the interaction between analysts and management in the forecasting environment (Brown, Foster, and Noreen 1985) implies that revisions in analyst forecasts may be used as a mechanism to update earnings expectations. The mean analyst revision is negative (-0.167), with 62% of the sample having downward revisions. The downward revisions would be consistent with lower expectations by the market, less pressure for management to increase earnings, and less need for positive discretionary accruals.

Table 1 also presents means of the optimistic and pessimistic forecasts (positive and negative forecast errors, respectively). While mean discretionary accruals remain negative for both forecast error types (and not significantly different) the mean scaled forecast error is positive for the optimistic forecasts. The optimistic mean scaled forecast error is also significantly greater (at p<0.08) than the pessimistic error, consistent with hypothesis 1. Additionally, analyst forecast revisions are significantly smaller (greater) for the optimistic sample, indicating analysts lowered their expectations through the year. As discussed above, this could explain why even for the optimistic forecasts the mean (negative) discretionary accruals are negative (for the optimistic sample).

Table 2 presents the main results for the study. The evidence in Table 2 is consistent with earnings management by firms issuing forecasts during the fiscal year. When forecast error is the only independent variable in the regression, its regression coefficient is marginally significant and positive (column 1), but its significance increases (to p<0.05) after the control variables are added (columns 2 and 3). The positive association is as hypothesized and indicates that greater forecast errors are associated with greater discretionary accruals.8

The fourth column in Table 2 presents the results for Hypothesis 2, which examines the affect that the dispersion in investor expectations has on discretionary accruals. A larger dispersion of expectations would be required to reduce the probability of discovery of the adjustment needed to sufficiently reduce the forecast error. To examine this hypothesis, the sample is divided into seven approximately equal groups. Table 3 provides means of the dispersions for each of the sample's seven groups. The table illustrates a dramatic rise in dispersion in the higher groups. Categorical variables representing the two extreme sevenths were included in interaction terms of Table 2 to identify highest dispersion (HDISP) and the lowest dispersion (LDISP) of forecast errors. Evidence presented in the fourth column indicates a positive and significant (p<.05) coefficient for the interaction FE*HDISP.9 As a check, we compared the mean forecast error the top seventh partition versus the rest of the sample. The means of the two groups are not significantly different (t=0.62). Thus, the results are not driven by differences the forecast error associated with this partition.

In Table 2, the coefficient for FE*HDISP is positive and significant, while the coefficients for FE and for FE*LDISP are insignificant. The insignificant coefficient on the low dispersion interaction term implies no systematic association between the earnings management and the narrow dispersion forecast errors. When the diversity of investor beliefs is low, the costs associated with the possible detection of earnings management exceed any costs that might result from forecast inaccuracy. The positive coefficient for the high dispersion interaction term suggests that when there is sufficient dispersion in investor beliefs about expected earnings, management will use discretionary accruals to correct forecast errors. In general, the evidence refines our previous conclusions regarding hypothesis 1, in that at greater levels of investor uncertainty, greater discretionary accruals are associated with greater forecast error.10 Thus, the results in column four supports hypothesis 2 and extends prior research.

As reported in all the columns in Table 2, the significant and positive coefficient on the performance measure, EPS, is consistent with larger discretionary accruals per share being associated with larger earning per share. This result supports the notion that firm performance impacts the ability to use discretionary accruals. The coefficient for the performance variable alone (results are not shown) is positive and significant (p<0.01) and confirms the conclusion of Dechow et al. (1995) that firm performance must be controlled for when analyzing discretionary accruals.

In summary, the evidence in the Table 2 is consistent with the hypothesis that greater forecast errors are associated with greater discretionary accruals. Further, the evidence indicates that the use of discretionary accruals is most closely associated with forecast errors when there is greater dispersion in earnings expectations and, therefore, a lower risk of detecting the earnings management.


Sensitivity Tests

In this section, we (1) correct for potential errors in the estimation of discretionary accruals and (2) examine an alternative measure for the forecast error. The Jones model is poor at estimating discretionary accruals for some firms, so the tests were repeated for a reduced sample that has eliminated observations where the discretionary accruals measure appears to be a poor estimate, based on (i) low R-squares (below 5%) and (ii) low numbers of observations (less than 2 degrees of freedom) in the estimating model (1) described in the Appendix. Tests with the reduced sample of 200 observations are reported in Table 4. Column 1 presents results of a replication of tests for hypothesis 1. As can be seen the forecast error remains positive and the coefficient's significance increases from Table 2. Columns 2 and 3 present replications of tests of hypothesis 2 with differing partitions of the forecast dispersion variable. The second column employs seven partitions, as in Table 2, and again the interaction coefficient's significance is greater than that in Table 2. The third column presents partitions in thirds, but is representative of other high-low partitions of the dispersion variable. The significance of the interaction term (p<.05) is similar to other testing partitions (e.g., quartiles and quintiles) and supports the above conclusion that sufficient dispersion in earnings expectations is required by managers in order to reduce the risk of detection of earnings management.

Tests were also rerun for an alternative measure for forecast error. The alternative measure of forecast error is the difference between the management forecast and the reported earnings (Kasznik, 1996) and is labeled FE1 in Table 5. Results of tests with this alternative variable are reported in Table 5. The first column replicates tests for hypothesis 1, and the positive and significant coefficient for FE1 indicates a positive relationship between discretionary accruals and forecast error and is consistent with Kasznik's (1996) results. The second column replicates tests for hypothesis 2, and the positive and significant coefficient for FE1*HDISP (p<.01) indicates that the discretionary accruals are associated with forecast errors of firms with greater earnings uncertainty. This is consistent with the results reported in Table 2. Other variables in Table 5 exhibit coefficients similar to those in Table 2.


Conclusions

This paper presents evidence supporting the association between management forecast disclosure and the level of discretionary accruals. We find that the use of discretionary accruals is consistent with the reduction in the forecast error. We also find that the use of discretionary accruals appears to be more closely associated with the forecast errors of firms with less certain expectations about their earnings. This is consistent with managers trading off costs associated with forecast errors and costs associated with earnings management detection.

This study has identified alternative measures of forecast error. The main results are based on the measure, management forecasts less the mean of the analyst forecasts from the last fiscal month of the forecast year. An alternative measure, management forecasts less reported earnings (Kasznik, 1996), provides consistent results. The alternative proxy, however, may be a less accurate measure of forecast error since reported earnings include the income effects of any discretionary accruals. In any case, our conclusions appear not to be driven by our proxies.

Our overall conclusion is that accruals' management is more common than previously thought. One direction for future research may be the earnings quality after a management forecast. For example, if stock issues often follow management earnings forecasts (Ruland, Tung, and George 1990), then the earnings used to price the stock may be of lower quality. Second, the relative accuracy of analyst versus management forecasts could be re-examined by controlling the ability of management to use discretionary accruals. Third, apparent anomalies such as the strong price reaction to the management forecast vis-à-vis the yearend announcement reported by Pownall and Waymire (1989) may be able to be explained if investors are able to see through the earnings management.




TABLE 1

Means (with p-values based on means) and medians

of selected variables broken into positive and negative values of forecast error (FE)



Full Sample

n=225

FE>0

n=114

FE<0

n=111

Significance

Test on Meansa


Mean

Median

p-value

Mean

Mean

p-value

Discretionary accruals

-0.823

-0.114

0.01

-0.397

-1.259

0.19

Scaled discretionary accruals

0.102

-0.058

0.91

1.56

-1.359

0.08

Management forecast

2.629

2.200

0.00

2.715

2.541

0.58

Last analyst forecast

2.658

2.16

0.00

2.291

3.035

0.01

Forecast error

-0.029

0.000

0.68

0.424

-0.495

0.58

Analyst revision

-0.167

-0.012

0.00

-0.358

0.028

0.00

Total assets

3423

1405

0.00

3590

352

0.68

Reported earnings

2.613

2.130

0.00

2.197

3.041

0.00

Dispersion in last analyst forecast


0.128


0.070


0.00


0.121


0.135


0.59








a Significance levels compare the mean values for the positive forecast error sample (FE>0) with the negative forecast error sample (FE<0), and are based on two tail tests.

Discretionary accruals = discretionary accruals as defined by the Jones Model.

Scaled discretionary accruals = discretionary accruals divided by the absolute value of the last

analyst forecast.

Management forecast = annual point or range earnings forecast.

Last analyst forecast = mean forecast reported during the last fiscal month of the

management forecast year from I/B/E/S Summary File.

Forecast error (FE) = Management forecast less last analyst forecast.

Analyst revision = Last analyst forecast (as defined above) less the mean of the analyst

forecasts, released in the month of the management forecast.

Total assets = Total assets at the end of the fiscal year of the management forecast.

Reported earnings = Primary earnings per share before extraordinary items for the fiscal

year of the management forecast.

Dispersion in last analyst forecast = Standard deviation of the last analyst forecast of the last fiscal month.




TABLE 2

REGRESSION RESULTS FOR DISCRETIONARY ACCRUALS AND FORECAST ERROR


Hypothesis 1: DA = B0 + B1 (FE) + B2 (EPS) + B3-8 Σ (YEAR)


Hypothesis 2: DA = B0 + B1 (FE) + B2 (FE*HDISP) + B3 (FE*LDISP)+ B4 (EPS) + B5-10 Σ(YEAR)




Hypothesis 1



Hypothesis 2


Column 1

Column 2

Column 3

Column 4


Independent


Variables





Coeff




t




Coeff




t




Coeff




t




Coeff




t


Intercept



-0.80


-2.43**


-2.03


-4.17***


-1.99


-2.38**


-1.99


-2.38**


FE



0.51


1.65*


0.66


2.17**


0.63


2.05**


-0.20


-0.42


FE*HDISP









1.39


2.22**


FE*LDISP









0.93


0.63


EPS





0.47


3.36***


0.51


3.59***


0.50


3.57***


YEAR







various


none


various


none


Adjusted R2




0.01



0.05



0.07



0.09


F-Statistic




2.74*



7.09***



3.26***



3.13***


Sample Size



225



225



225



224

Significance Levels (two-tail): *** < .01, ** < .05, * < .10.

DA = Discretionary accruals per share.

FE = Forecast error equal to management forecast less mean last analyst forecast.

FE*(H,L)DISP = Interaction between forecast error and categorical variables for highest (H), and

lowest (L) seventh of the sample's standard deviation of last analysts' forecasts.

EPS = Reported primary earnings per share.

YEAR = Indicator variable for each fiscal year of disclosure, less one (1992). If no year is

significant then "none" is noted.



TABLE 3

MEAN DISPERSION IN LAST ANALYST FORECASTS PER DISPERSION GROUP


LOWEST

SECOND

THIRD

FOURTH

FIFTH

SIXTH

HIGHEST

mean

0.012

.030

0.044

0.064

0.093

0.155

0.455

n

34

18

44

27

33

34

34





TABLE 4

REPLICATION REGRESSION RESULTS FOR DISCRETIONARY ACCRUALS AND FORECAST ERROR USING A REDUCED SAMPLE SIZE


Hypothesis 1: DA = B0 + B1 (FE) + B2 (EPS) + B3-8 Σ (YEAR)


Hypothesis 2: DA = B0 + B1 (FE) + B2 (FE*HDISP) + B3 (FE*LDISP)+ B4 (EPS) + B5-10 Σ(YEAR)



Hypothesis 1

Hypothesis 2


Column 1


Column 2: seven partitions


Column 3: three partitions


Independent


Variables





Coeff




t





Coeff




t





Coeff




t



Intercept



-0.21


-0.36



-0.18


-0.31



-0.33


-0.55



FE



0.58


2.68***



-.31


-0.95



-0.13


-0.34



FE*HDISP






1.63



3.80***



1.07


2.23**



FE*LDISP






0.54


0.55



0.53


0.62



EPS



0.09


0.81



0.09


0.87



0.10


0.91



YEAR



1990


-2.22**



1990


-2.25**



1990


-2.23**



Adjusted R2




0.03




0.09




0.10



F-Statistic




1.88*




3.05***




2.02**



Sample Size



200




199




199


Significance Levels (two-tail): *** < .01, ** < .05, * < .10.

DA = Discretionary accruals per share.

FE = Forecast error equal to management forecast less mean last analyst forecast.

FE*(H,L)DISP = Interaction between forecast error and categorical variables for highest (H), and

lowest (L) seventh (third) of the sample's standard deviation of last analysts'

forecasts.

EPS = Reported primary earnings per share.

YEAR = Indicator variable for each fiscal year of disclosure, less one (1992). If no year is

significant then "none" is noted.




TABLE 5

REGRESSION RESULTS FOR DISCRETIONARY ACCRUALS AND

ALTERNATIVE MEASURE FOR FORECAST ERROR


DA = B0 + B1 (FE1) + B2 (EPS) + B3-8 Σ (YEAR)


DA = B0 + B1 (FE1) + B2 (FE1*HDISP) + B3 (FE1*LDISP)+ B4 (EPS) + B5-10 Σ(YEAR)







Replication of Hypothesis 1



Replication of Hypothesis 2


Independent


Variables





Coeff





t





Coeff





t


Intercept


-0.21


-2.49**


-2.07


-2.50**


FE1


0.55


1.99**


-0.40



-0.98


FE1*HDISP






1.67


3.07***


FE1*LDISP






1.05


0.82


EPS


0.53


3.69***


0.54


3.78***


YEAR


1990


-1.68*



1990


-1.71*



Adjusted R2




0.07




0.10


F-Statistic




3.22***




3.60***


Sample Size




225




224


Significance Levels (two-tail): *** < .01, ** < .05, * < .10.

DA = Discretionary accruals per share.

FE1 = Forecast error equal to management forecast less reported earnings per share.

FE1*(H,L)DISP = Interaction between forecast error (FE1)and categorical variables for highest

(H), and lowest (L) seventh of the sample's standard deviation of last analysts'

forecasts.

EPS = Reported primary earnings per share for the fiscal year of management forecast.

YEAR = Indicator variable for each fiscal year of disclosure, less one (1992). The year

that the categorical variable is significant and related t-statistic is reported.


APPENDIX

A description of the discretionary accruals estimation procedure follows. First, parameters for each firm are estimated from the following regression model of total accruals:

Ai /TAi =B0+ B1REVi /TAi + B2PPEi /TAi + B3 /TAi + i (1)

where, A= Current Assets - Current Liabilities - Cash + Debt in Current

Liabilities - Depreciation Expense.

REV = Sales - Accounts Receivable.

PPE = Property Plant and Equipment.

TA = Beginning of fiscal year Total Assets.

i = sample firm.

If there are less than 5 years of data, then the firm is eliminated from the sample. Firm forecast years (identified from the sample period only) were eliminated from the estimation regressions.

The estimated parameters and the independent variables reported in the fiscal year of the forecast are used to calculate the nondiscretionary accruals for each firm:

NDAi,j=0+ 1 (REVi,j) + 2(PPEi,j) + 3(TAi,j-1), (2)

where, j= fiscal year of the forecast, and 1,2,3 = parameter estimates from equation (1). The discretionary accruals (DA) on a per share basis are then calculated as:

DAi,j= [(Ai,j - NDAi,j)/ TAi,j ]/ NOSHi,j

where NOSH is common shares used to calculate primary earnings per share.


References

Ajinkya, Bipin B., Rowland K. Atiase, and Michael J. Gift. 1991. Volume trading and the dispersion in financial analysts' earnings forecasts. The Accounting Review 66 (2):389-401.

Amir, Eli, and Joshua Livnat. 1996. Multi-period analysis of adoption motives: the case of SFAS No. 106. The Accounting Review 71 (4):539-553.

Beatty, Randolph P., and Robert E. Verrecchia. 1989. The effect of a mandated accounting change on the capitalization process. Contemporary Accounting Research 5 (2):472-493.

Becker, Connie L, Mark L DeFond, James Jiambalvo, and K R Subramanyam. 1998. The Effect of Audit Quality on Earnings Management. Contemporary Accounting Research 15 (1).

Brown, P., G. Foster, and Eric Noreen. 1985. Security analyst multi-year earnings forecasts and the capital market, AAA Studies in Accounting Research #21: American Accounting Association.

Copeland, R., and M. Moore. 1972. The financial bath: is it common? MSU Business Topics 20:63-69.

Dechow, Patricia M., Richard G. Sloan, and Amy P. Sweeney. 1995. Detecting earnings management. Accounting Review 70 (2):193-225.

Dechow, Patricia M., Richard G. Sloan, and Amy P. Sweeney. 1996. Causes and consequences of earnings manipulation: An analysis of firms subject to enforcement actions by the SEC. Contemporary Accounting Research 13 (1):1-36.

DeFond, Mark L., and James Jiambalvo. 1994. Debt covenant violation and manipulation of accruals. Journal of Accounting and Economics 17 (1):145-176.

DeFond, Mark L. , and Chul W. Park. 1997. Smoothing income in anticipation of future earnings. Journal of Accounting & Economics 23 (2):115-139.

Gaver, Jennifer J., Kenneth M. Gaver, and Jeffrey R. Austin. 1995. Additional evidence on bonus plans and income management. Journal of Accounting & Economics 19 (1):3-28.

Hassell, J., and R. Jennings. 1986. Relative forecast accuracy and the timing of earnings forecast announcements. Accounting Review 51 (1):58-75.

Healy, P.M. 1985. The effect of bonus schemes on accounting decisions. Journal of Accounting and Economics:85-108.

Imhoff, E., and P. Pare. 1982. Analysis and comparison of earnings forecast agents. Journal of Accounting Research 20:429-439.

Jaggi, B. 1980. Further evidence on the accuracy of management forecasts vis-a-vis analysts' forecasts. Accounting Review 55 (1):96-101.

Jones, Jennifer J. 1991. Earnings Management During Import Relief Investigations. Journal of Accounting Research 29 (2):193-228.

Kane, Gregory D, Patricia A Graybeal, and Frederick Richardson. 1996. Recession-Induced Stress and the Prediction of Corporate Failure. Contemporary Accounting Research 13 (2).

Kasznik, Ron. 1996. On the association between voluntary disclosure and earnings management. Working paper (August), Stanford University, Stanford, U.S.A.

King, Ronald, Grace Pownall, and Gregory Waymire. 1990. Expectations adjustment via timely management forecasts: Review, synthesis, and suggestions for future research. Journal of Accounting Literature 9:113-144.

McNichols, Maureen. 1989. Evidence of Informational Asymmetries from Management Earnings Forecasts and Stock Returns. Accounting Review 64 (1):1-27.

Morse, Dale, Jens Stephan, and Earl K. Stice. 1991. Earnings announcements and the convergence (or divergence) of beliefs. The Accounting Review 66 (2): 376-388.

Pownall, Grace, Charles Wasley, and Gregory Waymire. 1993. The effects of management forecast precision on equity pricing and on the assessment of earnings uncertainty. The Accounting Review 68 (4): 896-912.

Pownall, Grace, and Gregory Waymire. 1989. Voluntary disclosure credibility and securities prices: Evidence from management earnings forecasts: 1969-73. Journal of Accounting Research 27:85-105.

Ruland, William, Samuel Tung, and Nashwa E. George. 1990. Factors associated with the disclosure of manager's forecasts. The Accounting Review 65 (3): 710-721.

Skinner, Douglas J. 1994. Why firms voluntarily disclose bad news. Journal of Accounting Research 32 (1):38-60.

Subramanyam, K.R. 1996. The pricing of discretionary accruals. Journal of Accounting and Economics 22: 249-281.

Trueman, B. 1986. Why do managers voluntarily release earning forecasts? Journal of Accounting and Economics 8:53-72.

Watts, Ross L., and Jerold L. Zimmerman. 1986. Positive Accounting Theory. Englewood Cliffs: Prentice-Hall, Inc.

Watts, Ross L., and Jerold L. Zimmerman. 1990. Positive Accounting Theory: A Ten Year Perspective. Accounting Review 65 (1):131-156.

Waymire, Gregory. 1986. Additional evidence on the accuracy of analysts' forecasts before and after management earnings forecasts. Accounting Review 61 (1):129-142.

1 We are grateful to Gordon Richardson for suggesting this choice.

2 This step also eliminated firms with more than one forecast in a fiscal year. Also, forecasting firm-years were eliminated from the accruals estimation period.

3 Subramanyan (1996) compares the time-series and the cross-sectional estimation procedures for the Jones model. His conclusion to use the cross-sectional procedure is based on (1) greater sample sizes, (2) assumed nonstationarity of coefficients due to the lengthy estimation period, and (3) overlapping estimation and treatment periods. He also notes, however, that the cross-sectional model also has "a number of implausible coefficients" (footnote 5). Further, we believe the firm-specific controls offered by a time series procedure offer benefits not provided by the cross-sectional model. Our estimation period is six years, compared to his ten year estimation period, thereby reducing the effects of nonstationarity.

4 Tests that require a measure for the standard deviation of analysts forecasts have a lower sample size (224) because one firm had only one analyst following the firm.

5 Two other control variables, total assets and revenues, were also included as controls but were, in some cases, insignificant and in other cases produced results with lower significance. We settled for reported earnings as more direct proxy for the ability to utilize discretionary accruals because larger assets or revenues do not necessarily correspond to more income, on a per share basis.

6However, the mean discretionary accruals are inconsistent with Kasznik (1996), which may be due to the difference in estimation methods (45% of our sample have positive discretionary accruals, while 59% of Kasznik's are positive).

7 An alternative measure of discretionary accruals is the difference between reported earnings per share and the last analyst forecast prior to the yearend, assuming that the last analyst forecast is a well informed and unbiased estimate. The mean of this measure is positive (0.044) and significant (p<0.10). The mean of the scaled alternative measure (scaled by the absolute value of the last analyst forecast) is also positive and significant (p<0.01). The correlation between this (unscaled) alternative measure and our measure (DA) is highly significant (p<0.0001), which provides evidence of the validity of our measure.

8 As discussed above, the communication between analysts and management may imply that analyst forecasts reflect management expectations. To control for this, we also included the change in analyst forecasts from the time of the management forecast to yearend. After adding these measures to those in Table 2, the forecast error is still positive and significant at p<0.05.

9 Other partitions were attempted. In some cases, equivalent subsample sizes were difficult to obtain, and in all cases the results of tests were similar but of lower significance. For other partitions (i.e. thirds, quartiles, and quintiles), results indicate similarly signed coefficients, but the t-statistics on the positive coefficient for the higher dispersion group interaction term were at slightly less than normal levels of significance.

10 Tests were also run using a continuous variable for analyst dispersion in the model: DA=B0 + B1(FE) + B2(FE*DISPC) +B3(DISPC) + B4(EPS) + ΣB5-10 (YR), where DISPC is a continuous measure of the standard deviation in the analyst forecast for the month prior to year-end. Results indicate that B2 is positive and significant at p<0.05, coefficients for the intercept and EPS are directionally consistent with the results in Table 2, and the R-square rises to .11. While the interaction term may be more difficult to interpret, the results confirm those in Table 2.


MANDATORY VS DISCRETIONARY DISMISSAL RULES MANDATORY VS DISCRETIONARY DISMISSAL
NHS LOTHIAN DISCRETIONARY POINTS EQUAL OPPORTUNITIES QUESTIONNAIRE (FOR WORK
PLANNEDDISCRETIONARY ABSENCE FORM (SEE ATTACHED LISTING) NAME  


Tags: accruals and, discretionary accruals, management, forecast, errors, discretionary, accruals, cairney