The effect of information complexity on analysts' use of that information.

I. INTRODUCTION

Prior research finds that analysts fail to incorporate all available information into their earnings forecasts (e.g., Biddle and Ricks 1988; Bernard and Thomas 1990; Kim and Schroeder 1990; Abarbanell and Bernard 1992; Ali et al. 1992; Abarbanell and Bushee 1997, 1998). McEwen

and Hunton (1999) and Chen et al. (2002) conjecture that information complexity reduces analysts' assimilation of the information, but these studies do not directly measure the relative complexity of the information to support this conjecture. In this study I test whether the complexity of information reduces analysts' use of that information. Specifically, I examine the extent to which analysts' effective tax rate (ETR) forecasts incorporate information related to six tax-law changes of varying complexity.

I measure the relative complexity of these six tax-law changes (enacted by the Tax Reform Act of 1986) based on tax professionals' responses to the American Institute of Certified Public Accountants' tax-complexity index questionnaire (AICPA 1993). I find that the analysts' revisions of their forecasts of ETRs appear to impound the effects of less complex tax-law changes but not the effects of more complex tax-law changes. In contrast, the magnitude of analysts' ETR forecast errors increases with the effects of more complex tax-law changes, but is unrelated to the effects of the less complex tax-law changes, as if analysts fully impound the less complex information into their ETR forecasts. Taken together, these results suggest that analysts impound less complex information more fully than they do more complex information, consistent with information complexity reducing their use of the tax-law change information.

Empirical evidence that the complexity of information reduces analysts' use of that information is relevant to standard setters and researchers. For example, the Financial Accounting Standards Board Mission Statement charges the Board to "promulgate standards only when the expected benefits exceed perceived costs" (FASB 2002, 2). The evidence presented in this study suggests that more complex information imposes a cost on even expert users, which in turn suggests that as the complexity of a standard increases, the expected benefit of that standard must also increase to justify promulgation. (1) In addition, the Securities and Exchange Commission (SEC) adopted the "plain English rule," which requires issuers to:

   use plain English principles in writing the front and
   back cover pages, summary and risk factor sections of
   prospectuses; revise current requirements for highly
   technical information in the front of prospectuses;
   and revise the rule on the preparation of prospectuses
   to provide companies with more specific guidance on the
   clarity required in the entire document. (SEC 1997, 1)

The goal of the "plain English rule" is to increase investors' understanding and help them make informed investment decisions. My evidence that the complexity of information reduces even expert analysts' use of information supports the SEC's adoption of the plain English rule because less sophisticated users are, by definition, less able than expert analysts to incorporate complex information into their investment decisions. Finally, researchers investigating the relevance of disclosed information should consider the complexity of that information when interpreting their results. For example, evidence that there is no significant relation between specific information and stock prices or other measures of information use does not necessarily mean that the information is irrelevant; it could also mean that the information is too complex to be used cost-effectively.

In the context of tax laws examined in this study, information complexity is a function of both rule complexity (understanding and applying a tax law to a static set of facts) and strategic complexity (understanding firms' tax-planning responses to new tax laws). Because my empirical measure of complexity ranks tax-law changes in terms of their rule complexity only, the validity of my inferences rests on whether this ranking reflects the overall information complexity of these tax laws. Section III provides evidence supporting the validity of my complexity ranking.

I organize the remainder of this paper as follows. Section II provides background information and develops the hypotheses in the context of prior research. In Section III, I discuss research design issues, including the models used, the complexity-ranking measure, and the sample. Section IV discusses the results of the empirical tests; Section V concludes the paper.

II. BACKGROUND AND HYPOTHESES

The Tax Reform Act of 1986 and Effective Tax Rates

The Tax Reform Act of 1986, formally adopted in October 1986, materially changed a number of specific tax-law provisions that affected corporate ETRs. This exogenous change provides a natural setting for examining the effect of complexity on analysts' forecasts. Some of the tax-law changes took effect immediately, while others became effective January 1, 1987. Accordingly, my analysis uses the forecasts of 1987 ETRs that analysts issued in the fourth quarter of 1986.

A firm's ETR equals its total income tax expense divided by its pretax accounting income. Total income tax expense is the product of taxable income and the tax rate applied to that income, reduced by any tax credits available to the firm. Tax-law changes typically affect ETRs through changes in total income tax expense stemming from changes in (1) the tax rates applied to taxable income; (2) the calculation of taxable income; or (3) the calculation or availability of tax credits. This study focuses on six tax-law changes included in the Tax Reform Act of 1986: two changes in tax rates (a decrease in the statutory tax rate and an increase in the capital gains tax rate); one change in tax rate and the calculation of taxable income (implementation of an alternative minimum tax); and three changes to tax credits (elimination of the investment tax credit; computation of the available research and development [R&D] tax credit; extensive changes to the foreign tax credit).

Tax-law changes do not directly affect pretax accounting income, the denominator of ETR. However, prior research documents firms' strategic responses to anticipated tax-law changes that could affect their ETRs through changes in pretax accounting income that do not proportionally affect total income tax expense (e.g., Gramlich 1991; Boynton et al. 1992; Collins and Shackelford 1992; Dhaliwal and Wang 1992; Manzon 1992; Scholes et al. 1992; Guenther 1994; Maydew 1997). I consider the potential implications of firms' strategic responses in my model (in Section III) and on the results (in Section IV).

Using ETR Forecasts

I use Value Line analysts' explicit forecasts of ETRs to examine how the complexity of the Tax Reform Act of 1986's tax-law changes affected analysts' use of that information. Analysts can more directly estimate the effect of tax-law changes on firms' ETRs than on firms' overall EPS, because each firm itemizes material items that directly affect its ETR in the annual report income tax footnote. I use these footnote data to estimate the effect of each tax-law change on the firm's ETR. Thus, testing the effect of the tax-law change on analysts' forecasts of the specific component of earnings most directly affected by the tax-law change (i.e., ETRs) is sharper than testing the effect of the tax-law change on overall earnings forecasts, which may be both directly and indirectly affected by the tax-law change. Provided that analysts exercise due care in forming ETR forecasts, using ETR forecasts increases the power of my tests and provides more reliable evidence on information used by analysts than I could obtain by analyzing their EPS forecasts.

I took several steps to assess whether Value Line analysts take care in forming their ETR forecasts. First, I carefully read the firm-specific text included with Value Line's numerical forecasts for evidence that analysts use tax-related information. During the third and fourth quarters of 1986, Value Line analysts frequently referred to the expected effect of proposed changes in the tax code on firm ETRs and earnings, including: (1) the direct effect of specific tax-law changes on ETRs; (2) the net effect of tax-law changes on earnings; (3) the offsetting effects of the tax-law changes on ETR and earnings; and (4) the effect of the tax-law changes on consumer and firm behavior. (2) These findings suggest that analysts carefully considered how the Tax Reform Act of 1986 tax-law changes would affect firms' ETRs and earnings. Second, I conducted empirical analyses that further support my assumption that Value Line analysts exercise care in developing their ETR forecasts. Specifically, using data described in Section II, I find that when an analyst forecasted an increase in a firm's ETR, she tended to forecast a decrease in the firm's related earnings; there is a negative correlation between the ETR forecasts revisions and earnings forecast revisions (untabulated correlation = -0.15; at p-value of 0.01). Moreover, when an analyst overestimated a firm's ETR, she tended to underestimate the firm's earnings; there is a negative correlation between the ETR forecast errors and earnings forecast errors (untabulated correlation = -0.20; at p-value of 0.001). These data suggest that analysts do take care in forming their ETR forecasts. Third, in personal interviews, two Value Line analysts assured me that they carefully consider the effects of intermediate forecasts (including ETRs) when forming future earnings forecasts.

Prior Literature and Hypothesis Development

Prior research documents that analysts apparently fail to impound all available information into their forecasts, but provides little insight into why (e.g., Biddle and Ricks 1988; Bernard and Thomas 1989; Klm and Schroeder 1990; Abarbanell 1991; Abarbanell and Bernard 1992; Ali et al. 1992; Abarbanell and Bushee 1997, 1998). More recent studies (Mikhail et al. 1997, 1999; Clement 1999) explore how analyst attributes or incentives contribute to variation in analysts' accuracy (which may stem from variation in their use of information, among other factors).(3) In contrast, I focus on information attributes, not analyst attributes.

Two streams of research suggest that an information attribute, complexity, may affect how efficiently market participants use that information. Theoretical and empirical judgment/decision-making research concludes that increased complexity of a task adversely affects judgment quality (e.g., Payne 1976; Einhorn et al. 1977; Iselin 1988; Paquette and Kida 1988; Payne et al. 1988). (4) This literature suggests that task complexity impairs judgment through decision-makers' strategy selection, where a strategy is the method or set of procedures an individual uses to incorporate information into decision making (e.g., expected utility maximization, satisficing, elimination by aspects). For example, Payne (1976) finds that, at a high level of task complexity, individuals use strategies that are analytically simpler to complete the task. Subsequent studies report similar findings (e.g., Payne 1982; Smith et al. 1982; Earley 1985; Bettman et al. 1990). These studies suggest that higher task complexity leads decision makers to adopt analytically simpler strategies that may result in incomplete use of available information.

A second stream of research suggests that information complexity likely impairs analysts' abilities to assimilate the information. For example, Hirst and Hopkins (1998) document that analysts fail to access comprehensive income information under certain reporting formats and suggest that "clear reporting" of information increases analysts' use of it. However, Hirst et al. (2002) hypothesize and find that analysts experienced with relevant comprehensive income information are unaffected by differences in reporting format. (5) In another experimental study, McEwen and Hunton (1999) document that analysts who forecast more accurately tend to emphasize different information than other analysts who forecast less accurately. McEwen and Hunton (1999, 14) suggest that the accurate analysts' tendencies to ignore certain information "may be a function of its relevance, complexity, or both," but they do not test this conjecture. Finally, Chen and Schoderbek (2000) find that analysts failed to incorporate the deferred tax effect of an increase in the corporate tax rate from 34 percent to 35 percent stemming from the Omnibus Budget Reconciliation Act of 1993 into their forecasts of earnings for the third quarter of 1993. However, analysts appropriately revised their forecasts of 1993 annual earnings. In a follow-up study, Chen et al. (2002) report that some analysts inappropriately revised their earnings forecasts for the fourth quarter of 1993 and the first quarter of 1994 to reflect a continuing effect of this one-time change in tax rates. Chen et al. (2002, Abstract) suggest that analysts' incorrect responses to the increase in tax rates may arise because of "the complexity of this deferred tax adjustment." They do not measure the relative complexity of the adjustments or other transitory items to test this conjecture, however.

In summary, judgment/decision-making research consistently finds that higher task complexity leads decision makers to select analytically simpler strategies to complete a task, and analyst-related research finds that analysts fail to use all available information in forming their forecasts. If less complex information is less costly to use than more complex information (i.e., it takes less time, effort, or training) and there are constraints on an analyst's time, effort, or ability, then, all else equal, the net benefit of incorporating information into a forecast should be a decreasing function of the information's complexity.

Based on this prior research I expect the following. If analysts impound the information in less complex tax-law changes into their forecasts, then analysts' revisions, or changes in their ETR forecasts will incorporate their estimates of the effects of the less complex tax-law changes on firms' ETRs. As a result, the new levels of their ETR forecasts should fully impound the estimated effects of these tax-law changes on the firms' ETRs, and thus the error in the level of analysts' ETR forecasts will be unrelated to the estimated effect of these less complex tax-law changes. Conversely, if analysts fail to incorporate the information in the more complex tax-law changes in revising their ETR forecasts (either because the more complex information affords constrained analysts a smaller net benefit or because analysts are less proficient in using more complex information), then the magnitude of the error in analysts' ETR forecasts will be related to the estimated effects of these complex tax-law changes, but their revisions of ETR forecasts will not. These expectations are formally stated below.

[H.sub.1]: The less complex the economic effects of the tax-law change on the firm, the more likely analysts' ETR forecast revisions reflect the effects of the tax-law change.

[H.sub.2]: The more complex the economic effects of the tax-law change on the firm, the more likely the magnitude of analysts' ETR forecast errors will be associated with the effects of the tax-law change.

My research design isolates when analysts appear to impound information into their revisions of ETR forecasts ([H.sub.1]) and when they do not ([H.sub.2]). I infer that information complexity adversely affects analysts' use of that information if the less complex information explains forecast revisions and the more complex information explains forecast errors.

III. RESEARCH DESIGN AND SAMPLE

Regression Models

Testing [H.sub.1]: Model 1

I test [H.sub.1] by regressing analysts' ETR forecast revisions (which capture changes in analysts' expectations due to the tax-law changes) on proxies for the effects of six tax-law changes enacted by the Tax Reform Act of 1986:

[FR.sub.i] = [alpha] + [[beta].sub.R1][STATUTORY.sub.i] + [[beta].sub.R2][CAPITALGAIN.sub.i] + [[beta].sub.R3][ITCREDIT.sub.i] + [[beta].sub.R4][R&DCREDIT.sub.i] + [[beta].sub.R5][FOREIGNCREDIT.sub.i] + [[beta].sub.R6][ALTERNATIVE.sub.i] + [[delta].sub.1][FRPRETAX.sub.i] + [[delta].sub.2][INTERACTION.sub.i] + [[epsilon].sub.i]

where, for firm i:

[FR.sub.i] = Value Line forecast of firm i's 1987 ETR made in the fourth quarter of 1986, less the Value Line forecast of firm i's 1987 ETR made in the third quarter of 1986;

[STATUTORY.sub.i] = 46 percent (the previous corporate tax rate) less the blended tax rate (based on firm i's fiscal year end--varies from 45 percent to 34 percent);

[CAPITALGAIN.sub.i] = mean percentage effect of the capital gains tax rate on firm i's ETR reported for 1983-1985;

[ITCREDIT.sub.i] = mean percentage effect of the investment tax credit on firm i's ETR reported for 1983-1985;

[R&DCREDIT.sub.i] = mean percentage effect of the R&D credit on firm i's ETR reported for 1983-1985;

[FOREIGNCREDIT.sub.i] = mean percentage effect of the foreign tax credit on firm i's ETR reported for 1983-1985;

[ALTERNATIVE.sub.i] = 1 if firm i reported a net operating loss carryover for the fiscal year ended prior to September 1986 in excess of 90 percent of the firm's average pretax income over the prior three years, 0 otherwise;

[FRPRETAX.sub.i] = Value Line forecast of firm i's 1987 pretax income made in the fourth quarter of 1986, less the Value Line forecast of firm i's 1987 pretax income made in the third quarter of 1986, scaled by absolute value of the Value Line forecast of firm i's 1987 pretax income forecast made in the fourth quarter of 1986; and

[INTERACTION.sub.i] = [STATUTORY.sub.i] x mean percentage effect of the foreign tax rate differential on firm i's ETR reported for 1983-1985.

To form proxies for the effects of the tax-law changes, I use three types of measures. First, to proxy for the change in the statutory tax rate (STATUTORY), I use the explicit difference between the old and new statutory rate, which varies across firms as a function of their fiscal year-ends. Second, to proxy for the expected effects of changes to the capital gains tax rate (CAPITALGAIN), the investment tax credit (ITCREDIT), the research and development credit (R&DCREDIT), and the foreign tax credit (FOREIGNCREDIT), I use the mean of the percentage effect on ETR of the related tax law that each firm reported in its tax footnote over the three-year period prior to the enactment of the Tax Reform Act of 1986 (i.e., the three fiscal years ended prior to September 30, 1986). This period yields estimates of the effects of these tax laws that do not reflect the effect of changes enacted in the Tax Reform Act of 1986. (6)

Third, a firm is generally subject to the alternative minimum tax if it reduces its taxable income to less than 10 percent of its current year income by using net operating loss carryovers. Thus, I estimate the effect of the alternative minimum tax-law change on firm i's ETR using an indicator variable (ALTERNATIVE) that equals 1 if, for the fiscal year ended prior to September 1986, the firm reports a net operating loss carryover in excess of 90 percent of the average pretax book income over the prior three years, and 0 otherwise. (7)

The regression includes two control variables to capture possible sources of cross-sectional differences in ETR forecast revisions that may be correlated with my variables of interest: (1) forecast revisions of pretax income scaled by the absolute value of the fourth-quarter forecast of pretax income (FRPRETAX), (8) and (2) STATUTORY x the mean percentage effect of the foreign tax rate differential on ETR reported for 1983-1985 (INTERACTION). I include FRPRETAX as a control variable for two reasons. First, since ETRs are a function of pretax income (Wilkie and Limberg 1992), FRPRETAX controls for the cross-sectional variation in ETR forecast revisions due to anticipated changes in pretax income that are unrelated to taxes. Second, as prior research documents that firms responded strategically to tax-law changes that affect both pretax income and ETRs (e.g., Gramlich 1991; Boynton et al. 1992; Collins and Shackelford 1992; Dhaliwal and Wang 1992), I include FRPRETAX as a control for the cross-sectional variation in ETR forecast revisions related to these anticipated strategic responses.

I control for STATUTORY x the mean percentage effect of the foreign tax rate differential reported for 1983-1985 (i.e., INTERACTION), because firms with foreign taxable income will not recognize the full effect of the change in the statutory rate, due to reported foreign taxable income.

Prior research documents that analyst-specific attributes, such as experience, employer size, and the number of firms and industries an analyst follows, are associated with analyst accuracy (Mikhail et al. 1997; Clement 1999). Value Line's employment of all the analysts in my sample controls for variation in analyst accuracy associated with employer size (Clement 1999). Although I am unable to control for differences in other analyst attributes, such as firm-specific forecasting experience, these other attributes generally have a modest effect on analyst accuracy (Jacob 1997), so this limitation is unlikely to affect my inferences.

I expect five of the six tax-law changes to have unambiguous directional effects on firm ETRs, allowing for one-tailed tests of significance. Reductions in the statutory rate and the extension of the R&D credit will reduce ETRs; therefore, I expect the coefficients on STATUTORY and R&DCREDIT to be negative. Eliminating the capital gains tax rate and the investment tax credit and limiting the use of the foreign tax credit will increase ETRs; therefore I expect the coefficients on CAPITALGAIN, ITCREDIT, and FOREIGNCREDIT to be positive.

The net effect of changes to the alternative minimum tax across firms is ambiguous. Many of the "tax adjustment" items included in the calculation of the alternative minimum tax are temporary differences, which will generate an alternative minimum tax credit for use in future years. For these items, the alternative minimum tax decreases expected ETR because previously no alternative minimum tax credit was generated.9 Conversely, "preference items" used to calculate the alternative minimum tax that arise due to permanent differences do not generate an alternative minimum tax credit. For these items, the alternative minimum tax increases future ETR. Finally, the limitation applied to the alternative minimum tax net operating loss may effect whether a firm eventually uses a net operating loss, which changes the net operating loss from a temporary difference to a permanent difference. As I am unable to form an unambiguous expectation for the ALTERNATIVE variable, the test of significance of its coefficient in the regression is two-tailed.

Testing [H.sub.2]: Model 2

If analysts fully incorporate less complex information into their forecasts of ETRs, but they do not fully incorporate more complex information, then the magnitude of the ETR forecast errors will (1) not be associated with the less complex information, but (2) will be associated with the more complex information. To test this hypothesis, I regress the absolute value of the analysts' ETR forecast errors on the proxies for the effects of the six tax-law changes and two control variables:

|[FE.sub.i]| = [[alpha].sub.i][[beta].sub.R1][STATUTORY.sub.i] + [[beta].sub.R2][CAPITALGAIN.sub.i] + [[beta].sub.R3][ITCREDIT.sub.i] + [[beta].sub.R4][R&DCREDIT.sub.i] + [[beta].sub.R5][FOREIGNCREDIT.sub.i] + [[beta].sub.R6][ALTERNATIVE.sub.i] + [[beta].sub.1]|[FEPRETAX.sub.i]| + [[beta].sub.2][INTERACTION.sub.i] + [epsilon]

where, for firm i:

|[FE.sub.i]| = absolute value of the error in the Value Line forecast of firm i's 1987 ETR--the actual 1987 ETR less the Value Line forecast of firm i's 1987 ETR forecast made in the fourth quarter of 1986; and

|[FEPRETAX.sub.i]| = the absolute value of the ratio of the error in the Value Line forecast of firm i's 1987 pretax income, scaled by the actual 1987 pretax income. The error in the Value Line forecast of firm i's 1987 pretax income is the actual 1987 pretax income less the Value Line forecast of firm i's 1987 pretax income forecast made in the fourth quarter of 1986.

All other variables are as defined above.

The dependent variable is the absolute value of the ETR forecast errors. I expect the effects of tax-law changes that are not fully incorporated into analysts' forecasts of ETRs to be positively associated with the magnitude of the errors in analysts' ETR forecasts, allowing for one-tailed significance tests.

Complexity Rankings of the Six Tax-Law Changes

For each of the six tax-law changes, three certified public accountants practicing during the enactment of the Tax Reform Act of 1986 completed the 15 questions that comprise the AICPA's tax-complexity index. (10) For each question, participants responded on a scale ranging from -3 (an expected decrease in complexity) to +3 (an expected increase in complexity). The sum of the scores from the 15 questions measures the complexity of each tax-law change, for each participant. I calculate the mean of these sums across the three participants and use them to rank the tax-law changes by complexity. (11)

I assess the consistency of the rankings obtained across the participants using Kendall's coefficient of concordance, a measure of consistency of rankings across more than two individuals. Kendall's W is 0.91, indicating high consensus. (12) The only inconsistency in complexity ranking across the individuals is a disagreement, in one case, about whether the change in the capital gains tax rate or the change in investment tax credit was more complex.

This process produces a relative measure of complexity, not an absolute measure. Table 1 shows that three of the tax-law changes score as relatively simple: change in statutory rate = 1.7; change in capital gains rate = 2.3; and change in investment tax credit = 2.7. For the first two items, the Tax Reform Act of 1986 changed the rate applied to taxable income; for the third item, the Act effectively eliminated the tax credit. In contrast, the remaining three tax-law changes score as much more complex: change in R&D credit = 20.3; change in foreign tax credit = 27.0; change in alternative minimum tax -- 31.0. The Act extended the R&D credit, but at a reduced rate of 20 percent with stricter guidelines about what qualifies as research. The Act substantially changed a firm's ability to use foreign tax credits by requiring it to compute the foreign tax credit and apply the credit limitations separately for different categories of income. The Act also replaced the minimum tax with an alternative minimum tax system. Overall, this complexity ranking appears appropriate, given the tax-law changes made by the Tax Reform Act of 1986.

In the context of the tax laws examined in this study, information complexity is a function of both rule complexity (understanding and applying a tax law to a static set of facts) and strategic complexity (understanding firms' tax-planning responses to new tax laws). One limitation of this study is that my empirical measure of complexity ranks tax-law changes in terms of their rule complexity only. The validity of my tests rests on whether this ranking reflects the overall information complexity of these tax laws. If the unmeasured strategic complexity would alter the rankings based on the rule complexity, then the conclusions drawn may be incorrect. I mitigate this measurement error problem by controlling for pretax income to capture firms' tax-planning responses to the Tax Reform Act of 1986 that affected pretax accounting income and ETRs. In addition, in my discussion of the results, I evaluate anecdotal evidence of firm strategic responses documented in prior studies and the likely effects of these strategic responses on the complexity rankings.

Sample

My sample consists of 355 firms that Value Line followed from 1984 through 1988. Value Line provides both forecasts and reported actual amounts of ETRs and earnings. (13) It does not provide ETR forecasts for firms with either negative or zero taxes or earnings; thus my analysis includes only firms Value Line expected to be profitable in the near future.

Value Line forecasts annual ETRs and earnings once every 13 weeks (i.e., once each calendar quarter) for each firm, issuing annual forecasts for the current year, the subsequent year, and three to five years ahead. Since Congress enacted the Tax Reform Act of 1986 in October of that year, I consider Value Line forecasts made in the third quarter of 1986 (July-September) as prior to enactment and forecasts made in the fourth quarter of 1986 (October-December) as after enactment. (14) I use Value Line ETR forecasts for the fiscal years ended May 31, 1987, through April 30, 1988 (the "subsequent year"), to ensure that my tests use forecasts for the year following the tax-law changes. (15)

Value Line issued forecasts of ETRs for 1,225 firms in the third and fourth quarters of 1986. Additional data (Value Line reported "actual" ETRs for 1984, 1985, and 1987 and Value Line forecasted ETRs for 1985 and 1986 issued in the fourth quarter of 1983 and 1984) required to form the forecast-error measures in the primary analysis and the robustness tests limit my sample to 424 firms. I found the firm-specific tax footnote data from annual reports (required to estimate the proxies for the effects of the tax-law changes) for 360 of these firms. (16) I eliminated five firms considered influential in my regressions based on the Belsley et al. (1980, 28) procedure. Thus, the final sample includes 355 firms. (17)

Table 2 provides descriptive statistics for this sample. As shown in Panel A, the mean and median ETR forecast revisions are negative (-1.48 and -2.0, respectively), consistent with analysts, on average, expecting ETRs to decrease around 1.5 percentage points after the implementation of the Tax Reform Act of 1986. (A firm with a forecasted ETR of 42 percent prior to the Act and a forecasted ETR of 40 percent after the Act would have a forecast revision of -2.0.) The mean (median) absolute value of the Value Line ETR forecast error is 3.63 (2.2), which indicates that, on average, analysts missed firms' ETRs by 3.6 percentage points. (A firm with a forecasted ETR of 40 percent after the Act and an actual ETR of either 36.4 percent or 43.6 percent would have an absolute ETR forecast error of 3.6.)

Panel B of Table 2 provides descriptive statistics for the proxies for the effects of the six tax-law changes. The mean values of all the tax-law proxies are greater than zero. In contrast, median values equal zero for all but STATUTORY and ITCREDIT, indicating that less than half of my sample firms are affected by the other four tax-law changes. Specifically, STATUTORY and ITCREDIT are nonzero for 99.7 percent and 85.6 percent of my sample firms, respectively, but CAPITALGAIN, R&DCREDIT, FOREIGNCREDIT, and ALTERNATIVE are nonzero for just 18.6 percent, 14.9 percent, 12.4 percent, and 10.4 percent of my sample firms, respectively. Some firms are affected only by the change in the statutory rate, while others are affected by multiple tax-law changes.

Panel C of Table 2 provides descriptive statistics for the control variables. The mean (median) pretax income forecast revision is -0.08 (-0.03); it is zero for 5.4 percent of the firms. (A value of -0.08 represents an expected 8.0 percent decrease in pretax income.) The mean (median) absolute value of the pretax income forecast error is 0.29 (0.15), indicating an average forecast error for pretax income of 29 percent. The mean (median) interaction term, which captures the mitigated effect of the change in statutory rates for firms with foreign operations, is -0.07 (0.00). The mean value for the interaction term suggests that, for the 65 firms that reported a foreign tax rate differential, that differential was negative.

To assess whether analysts exercised due care in making their ETR forecasts, I examine (but do not report) the mean and median values of pretax income and ETR forecast errors. Consistent with prior research on forecasts of earnings forecast errors (e.g., O'Brien 1988; Philbrick and Ricks 1991), I find that mean pretax income forecasts are significantly optimistic, but median pretax income forecasts are, on average, unbiased. The mean and median ETR forecast errors (0.13 and 0.20, respectively) are not significantly different from zero, documenting that analysts made unbiased ETR forecasts for 1987. ETR forecast errors range from 2.4 at the 75th percentile to -2.2 at the 25th percentile.

IV. EMPIRICAL RESULTS

Test of [H.sub.1]: Value Line Analysts' Revisions of Their ETR Forecasts

I test [H.sub.1] by estimating Model 1, which provides evidence on the extent to which analysts' ETR forecast revisions are associated with the effects of the tax-law changes. Table 3, Panel A lists the coefficients for each of the six tax laws in order of their relative complexity (R1 [least complex] to R6 [most complex]).

The overall regression is highly significant with an adjusted R2 of 24.6 percent, indicating that the proxies capture cross-sectional variation in the forecast revisions. The results suggest that Value Line analysts revised their forecasts of firms' ETRs in response to the three tax-law changes ranked as least complex. As expected, analysts reduced their forecasts of ETRs in response to the expected reduction in the firm's statutory tax rate (STATUTORY), and they increased their forecasts of ETRs in response to the expected increase in the firm's capital gains tax rate (CAPITALGAIN) and the effect on the firm of the virtual elimination of the investment tax credit (ITCREDIT) (all at p-values of 0.001). In contrast, the revisions of ETR forecasts are unrelated to the three tax-law changes ranked as most complex; R&DCREDIT, FOREIGNCREDIT, and ALTERNATIVE provide no explanatory value in the forecast revision regression. With respect to the control variables, the positive coefficient on FRPRETAX (at a p-value of 0.001), suggests that upward revisions in forecasts of pretax income are associated with upward revisions in ETR forecasts. The coefficient on INTERACTION is not significant. (18)

The evidence presented in Panel A of Table 3 supports my hypothesis that analysts use less complex information to a greater extent than they use more complex information. Two alternative interpretations of these results, however, are that analysts' forecast revisions did not impound the more complex information because it does not help them accurately forecast firms' ETRs, or because the proxy for the effect of the tax-law change is not well specified. Testing [H.sub.2] allows me to distinguish between my hypothesis and these two alternative explanations.

Test of [H.sub.2]: Absolute Value of Value Line Analysts' ETR Forecast Errors

Panel B of Table 3 reports the results of my test of [H.sub.2], where I regress the absolute value of the Value Line analysts' ETR forecast errors on proxies for the effects of the six tax-law changes and the control variables. The overall regression is significant with an adjusted [R.sup.2] of 15.0 percent, evidence that the proxies capture cross-sectional variation in the absolute forecast errors. Consistent with [H.sub.2], analysts do not appear to impound the effects of the most complex tax-law changes on firms' ETRs. Specifically, the two most complex tax-law changes (FOREIGNCREDIT and ALTERNATIVE) are significantly positively associated with the magnitude of analysts' ETR forecast errors (at p-values of 0.001 and 0.01, respectively). In contrast, the three least complex tax-law changes, that are significantly associated with analysts' forecast revisions, are unrelated to the magnitude of the error in analysts' forecasts of ETRs, as if the analysts have fully assimilated the implications of these three tax-law changes for firms' ETRs. The proxy for the effect of the changes to the R&D tax credit (R&DCREDIT) is not significantly associated with either the ETR forecast revisions or the absolute ETR forecast errors, suggesting that either the proxy may not be well specified or is not relevant. Finally, the significantly positive coefficient on the control variable, |FEPRETAX|, indicates that larger errors in analysts' pretax income forecasts are associated with larger errors in their ETR forecasts.

The determinants of forecast revisions and forecast errors together provide strong evidence that complexity adversely affects Value Line analysts' use of tax information. The effects of the three least complex tax-law changes explain much of the variation in analysts' revisions of their ETR forecasts, but are unrelated to the magnitude of the errors in analysts' ETR forecasts, which suggests the analysts correctly used the information to forecast ETRs. In contrast, the two tax-law changes ranked as most complex are associated with variation in the magnitude of analysts' ETR forecast errors, but are not associated with analysts' revisions of their ETR forecasts. These results suggest that higher information complexity reduces analysts' use of that information.

Rule Complexity vs. Strategic Complexity

As discussed earlier, in the context of the tax-law changes, overall information complexity is a function of both rule complexity and strategic complexity, but my empirical measure ranks tax-law changes only in terms of rule complexity. The validity of my inferences rests on the assumption that the ranking based on the measured rule complexity reflects the true (unobservable) ranking based on overall complexity. Evidence from prior research supports this assumption.

Prior research documents that firms responded strategically to three of the tax-law changes included in my study: changes in the statutory rate, changes in the foreign tax credit, and changes in the alternative minimum tax. Several studies (e.g., Manzon 1992; Scholes et al. 1992; Guenther 1994; Maydew 1997) document that firms shifted taxable income or deductions across time in response to decreases in the statutory tax rate (a relatively low-complexity tax-law change). Because this strategic response affected taxable and financial income equally, ETRs are unaffected. The strategic complexity of this firm response seems to be low and is uncorrelated with ETRs.

Collins and Shackelford (1992) document that firms issued preferred stock in lieu of debt in response to the Tax Reform Act of 1986's changes to the foreign tax credit, a more complex tax-law change. By doing so, firms increased financial income but decreased tax expense by increased foreign tax-credit utilization, a strategic response that would reduce firms' ETRs and appears relatively complex. Several studies (e.g., Gramlich 1991; Boynton et al. 1992; Dhaliwal and Wang 1992) examine other seemingly complex strategic responses to changes made to the alternative minimum tax, the tax-law change my study ranked as most complex. Firms' strategic responses typically affected the calculation of the alternative minimum taxable income, which is taxed at a different rate than regular taxable income (20 percent vs. 34 percent), resulting in a net effect on firms' ETRs. The strategic complexity of the firms' responses related to both foreign tax credits and the alternative minimum tax, which may involve issuing preferred stock or changing investment decisions or accounting policies, seems to be high, which would increase the overall complexity of these tax-law changes

In summary, the tax-law change ranked as lowest in complexity elicited a Straightforward strategic response that would not affect ETRs, while two of the tax-law changes ranked as highest in complexity elicited more complex strategic responses that would affect ETRs. These results suggest that unmeasured strategic complexity would accentuate, rather than alter, the complexity rankings of the six-tax law changes.

Robustness Test

I use the magnitude of ETR forecast errors as the dependent variable in Model 2. However, a firm's 1987 ETR forecast error might be related to the underlying complexity of applying the old tax laws, which varies across firms. To control for this potential confounding factor, I reestimate Model 2 using the change in the absolute value of the ETR forecast error as the dependent measure, calculated as follows. I first calculate the absolute forecast errors for 1987 as in Table 3, and I compute analogous absolute forecast errors for 1984 and 1985 (as the absolute value of the year t actual ETR less the Value Line year t ETR forecast made in the fourth quarter of year t-1). I then calculate the difference between (1) the absolute 1987 ETR forecast error and (2) the average of the absolute 1984 and 1985 ETR forecast errors. This difference is the change in the magnitude of the ETR forecast errors--the magnitude of the ETR forecast error less the pre-Tax Reform Act of 1986 average absolute forecast error. (19)

Table 4 reports the results from the respecified regression. The estimated effects of R&DCREDIT, FOREIGNCREDIT, and ALTERNATIVE are associated with increases in the magnitude of analysts' errors in forecasts of ETRs (at p-values of 0.01, 0.01, and 0.05, respectively). Both control variables, change in the absolute pretax income forecast error and INTERACTION, are associated with increases in the magnitude of analysts' errors in forecasting ETRs after the Tax Reform Act of 1986 (at p-values of 0.001 and 0.05, respectively). Although these results are not identical to those documented in Table 3, the relation between complex information and the magnitude of analysts' ETR forecast errors is the same; more complex information items explain increases in absolute forecast errors of ETRs, whereas less complex information items do not.

V. CONCLUSIONS

This study provides evidence on the relation between information complexity and analysts' use of that information. I examine how six specific tax-law changes enacted by the Tax Reform Act of 1986 affect Value Line analysts' effective tax rate (ETR) forecasts for 355 firms. I show that analysts' revisions of their ETR forecasts appear to impound the effects of the less complex tax-law changes but not the effects of the more complex tax-law changes. As expected, if analysts impound the less complex information but not the more complex information, the effects of the less complex tax-law changes are not associated with the magnitude of analysts' errors in their ETR forecasts, but the effects of the more complex tax-law changes do explain variation in the magnitude of ETR forecast errors. In conjunction, these results are consistent with increased complexity of information reducing the quality of analysts' forecasts based on that information.

Empirical evidence that the complexity of information imposes sufficient costs on even expert users of financial information--analysts--to reduce their use of that information is relevant to the Financial Accounting Standards Board and the SEC as they consider changing what information companies are required to disclose, and when and how they must disclose it. In addition, the results suggest that researchers should consider the effect of the complexity of information in designing and interpreting the results of their research. For example, studies that find no significant relation between specific information and stock prices changes (or other firm-specific changes) without considering the complexity of the information may erroneously conclude that the information is irrelevant, when it is actually relevant but too complex for users to fully incorporate.

My results are subject to two alternative explanations. First, if the proxies for the effects of the tax-law changes are noisier for the more complex tax-law changes than for the less complex tax-law changes, then the results could reflect the differential quality of the proxies rather than the differential complexity of the tax-law changes. For example, the proxy for the least complex tax-law change is a function of a firm's fiscal year-end month--which is precise--while the proxy for the most complex tax-law change is perhaps the most noisy--it is an indicator variable. This is unlikely to provide a complete explanation for my results, however. Four of the proxies used to capture the effects of the tax-law changes are based on detailed data drawn from each firm's tax footnote (CAPITALGAIN, ITCREDIT, R&DCREDIT, and FOREIGNCREDIT). This should reduce the variance in the noise across these proxies and hence the concern that the results reflect the precision of the proxies.

The second alternative explanation for my results is that the more complex tax laws affect fewer firms, as shown in Table 2, so analysts choose not to invest their time analyzing the potential effect of those tax-law changes. However, the potential benefit of properly incorporating the two most complex tax variables appears to be material; the average estimated error in net profit due to the failure to incorporate the effects of these tax laws ranges between $1.93 and $4.61 million. (20)

I capture the key construct in this study, information complexity, by ranking the tax-law changes by complexity; information complexity is not an independent variable in the tests. This is a limitation of my study, as the research design does not provide a direct test of the effect of complexity on analysts' information use. Instead, one must infer the effect by observing the pattern of statistical significance of the coefficients of the proxies for the tax-law changes in two related regressions. Nevertheless, the study's results illustrate the importance of considering information complexity when examining analysts' use of or failure to use information.

This study does not examine why analysts fail to incorporate more complex information into their forecasts. Either analysts' abilities to incorporate information into their forecasts correctly is a decreasing function of the complexity of the information, or analysts choose not to assimilate complex information because the cost of using the information exceeds the benefit of doing so. Distinguishing between these alternative explanations for why analysts under-use complex information may be a fruitful avenue for future research.

TABLE 1

Complexity Index Scores for Six Tax-Law Changes
Enacted by the Tax Reform Act of 1986 (a)

Tax-Law Change               Variable Name    Complexity Score (b)

Statutory tax rate           STATUTORY                1.7
Capital gains tax rate       CAPITALGAIN              2.3
Investment tax credit        ITCREDIT                 2.7
Research and development
   credit                    R&DCREDIT               20.3
Foreign tax credit           FOREIGNCREDIT           27.0
Alternative minimum tax      ALTERNATIVE             31.0

(a) The tax-law changes: decrease in the statutory tax rate
(STATUTORY); increase in the capital gains tax rate (CAPITALGAIN);
elimination of the investment tax credit (ITCREDIT); changes in the
computation of the available R&D tax credit (R&DCREDIT); extensive
changes to the foreign tax credit (FOREIGNCREDIT); and implementation
of an alternative minimum tax (ALTERNATIVE).

(b) The complexity scores are based on responses from three certified
public accountants practicing during the enactment of the Tax Reform
Act of 1986 to 15 questions that comprise a tax-complexity index
designed by AICPA.

TABLE 2

Descriptive Statistics for Dependent, Explanatory,
and Control Variables n = 355 Firms

Panel A: Dependent Variables

                                    Nonzero Observations (b)

Variable (a)      Mean    Median     Percent     Frequency

FR               -1.48     -2.0        87.3         310
|FE|              3.63      2.2        99.4         353

Panel B: Explanatory Variables

                                    Nonzero Observations (b)

Variable         Mean    Median     Percent     Frequency

STATUTORY        5.96     6.00       99.7          354
CAPITALGAIN      0.33     0.00       18.6           66
ITCREDIT         1.92     1.70       85.6          304
R&DCREDIT        0.28     0.00       14.9           53
FOREIGNCREDIT    0.14     0.00       12.4           44
ALTERNATIVE      0.10     0.00       10.4           37

Panel C: Control Variables

                                    Nonzero Observations (b)

Variable         Mean    Median     Percent     Frequency

FRPRETAX         -0.08   -0.03        94.6        336
|FEPRETAX|        0.29    0.15       100.0        355
INTERACTION      -0.07    0.00        18.1         65

(a) Variable definitions:

           [FR.sub.i] = Value Line forecast of firm i's 1987 ETR made
                        in the fourth quarter of 1986, less the Value
                        Line forecast of firm i's 1987 ETR made in the
                        third quarter of 1986;
         |[FE.sub.i]| = absolute value of the error in the Value Line
                        forecast of firm i's 1987 ETR--the actual
                        1987 ETR less the Value Line forecast of firm
                        i's 1987 ETR forecast made in the fourth
                        quarter of 1986;
    [STATUTORY.sub.i] = 46 percent (the previous corporate tax rate)
                        less the blended tax rate (based on firm i's
                        fiscal year end--varies from 45 percent to 34
                        percent);
  [CAPITALGAIN.sub.i] = mean percentage effect of the capital gains tax
                        rate on firm i's ETR reported for 1983-1985;
     [ITCREDIT.sub.i] = mean percentage effect of the investment tax
                        credit on firm i's ETR reported for 1983-1985;
    [R&DCREDIT.sub.i] = mean percentage effect of the R&D credit on
                        firm i's ETR reported for 1983-1985;
[FOREIGNCREDIT.sub.i] = mean percentage effect of the foreign tax
                        credit on firm i's ETR reported for 1983-1985;
  [ALTERNATIVE.sub.i] = 1 if firm i reported a net operating loss
                        carryover for the fiscal year ended prior to
                        September 1986 in excess of 90 percent of the
                        firm's average pretax income over the prior
                        three years, 0 otherwise;
     [FRPRETAX.sub.i] = Value Line forecast of firm i's 1987 pretax
                        income made in fourth quarter of 1986, less the
                        Value Line forecast of firm i's 1987 pretax
                        income made in the third quarter of 1986,
                        scaled by absolute value of the Value Line
                        forecast of firm i's 1987 pretax income
                        forecast made in the fourth quarter of 1986;
   |[FEPRETAX.sub.i]| = the absolute value of the ratio of the 1987
                        error in Value Line's forecast of firm i's
                        pretax income, scaled by the actual 1987 pretax
                        income. The 1987 pretax income forecast error
                        is the actual 1987 pretax income less the 1987
                        pretax income Value Line made in the fourth
                        quarter of 1986; and
  [INTERACTION.sub.i] = STATUTORY x mean percentage effect on firm i's
                        ETR reported for 1983-1985.

(b) Percentage (number) of firms for which the variable is nonzero.
For example, CAPITALGAIN is nonzero for 18.6 percent of the sample
(66 firms) and zero for 81.4 percent of the sample (289 firms).

TABLE 3

Regressions of Revisions (Panel A) and Absolute Errors (Panel B) in
Value Line Forecasts of Effective Tax Rates on Proxies for the Effects
of Six Tax-Law Changes of Increasing Complexity Enacted by the Tax
Reform Act of 1986

Panel A: Forecast Revisions (a)

F[R.sub.i] = [[alpha].sub.i] + [[beta].sub.R1][STATUTORY.sub.i] +
             [[beta].sub.R2][CAPITALGAIN.sub.i] +
             [[beta].sub.R3][ITCREDIT.sub.i] +
             [[beta].sub.R4][R&DCREDIT.sub.i] +
             [[beta].sub.R5][FOREIGNCREDIT.sub.i] +
             [[beta].sub.R6][ALTERNATIVE.sub.i]
             + [[delta].sub.1][FRPRETAX.sub.i] +
             [[delta].sub.2][INTERACTION.sub.i] +
             [epsilon]

Complexity Rank (a)

Variable (b)    Intercept    [[beta].sub.R1]   [[beta].sub.R2]

Predictions        ?             -                +
Coefficient     2.39         -0.68             0.38
t-statistic     2.43         -4.28 ***         3.21 ***

Variable (b)    Intercept    [[beta].sub.R3]   [[beta].sub.R4]

Predictions        ?            +                  -
Coefficient     2.39         0.23              -0.15
t-statistic     2.43         5.92 ***          -1.33

Variable (b)    Intercept    [[beta].sub.R5]   [[beta].sub.R6]

Predictions        ?             +                 ?
Coefficient     2.39         -0.10             -0.61
t-statistic     2.43         -0.68             -0.95

Variable (b)    Intercept    [[delta].sub.1]   [[delta].sub.2]

Predictions        ?            ?                 ?
Coefficient     2.39         3.23              0.02
t-statistic     2.43         5.17 (###)        1.55

Variable (b)    Intercept    Adj. [R.sup.2]

Predictions        ?
Coefficient     2.39         24.6%
t-statistic     2.43

Panel B: Absolute Forecast Errors (a)

|F[E.sub.i]| = [[alpha].sub.i] + [[beta].sub.R1][STATUTORY.sub.i] +
               [[beta].sub.R2][CAPITALGAIN.sub.i] +
               [[beta].sub.R3][ITCREDIT.sub.i] +
               [[beta].sub.R4][R&DCREDIT.sub.i] +
               [[beta].sub.R5][FOREIGNCREDIT.sub.i] +
               + [[beta].sub.R6][ALTERNATIVE.sub.i] +
               [[delta].sub.1][FEPRETAX.sub.i] +
               [[delta].sub.2][INTERACTION.sub.i] + [epsilon]

Complexity Rank (a)

Variable (b)    Intercept    [[beta].sub.R1]   [[beta].sub.R2]

Coefficient     4.11         -0.27             -0.15
t-statistic     3.67         -1.39             -1.07

Variable (b)    Intercept    [[beta].sub.R3]   [[beta].sub.R4]

Coefficient     4.11         0.07              0.21
t-statistic     3.67         1.47              1.36

Variable (b)    Intercept    [[beta].sub.R5]   [[beta].sub.R6]

Coefficient     4.11         0.73              1.96
t-statistic     3.67         4.24 ***          2.53 **

Variable (b)    Intercept    [[delta].sub.1]   [[delta].sub.2]

Coefficient     4.11         2.84              0.01
t-statistic     3.67         7.38 ***          0.05

Variable (b)    Intercept    Adj. [R.sup.2]

Coefficient     4.11         15.0%
t-statistic     3.67

*, **, *** p < 0.05, p < 0.01, p < 0.001, respectively, one-tailed; #,
##, ### p < 0.05, p < 0.01, p < 0.001, respectively, two-tailed.

(a) The six tax-law variables are ordered from least complex (R1) to
most complex (R6).

(b) Variable definitions:

           [FR.sub.i] = Value Line forecast of firm i's 1987 ETR made
                        in the fourth quarter of 1986, less the Value
                        Line forecast of firm i's 1987 ETR made in the
                        third quarter of 1986;
         |[Fe.sub.i]| = absolute value of the error in the Value Line
                        forecast of firm i's 1987 ETR--the actual 1987
                        ETR less the Value Line forecast of firm i's
                        1987 ETR forecast made in the fourth quarter
                        of 1986;
    [STATUTORY.sub.i] = 46 percent (the previous corporate tax rate)
                        less the blended tax rate (based on firm i's
                        fiscal year end--varies from 45 percent to 34
                        percent);
  [CAPITALGAIN.sub.i] = mean percentage effect of the capital gains tax
                        rate on firm i's ETR reported for 1983-1985;
     [ITCREDIT.sub.i] = mean percentage effect of the investment tax
                        credit on firm i's ETR reported for 1983-1985;
    [R&DCREDIT.sub.i] = mean percentage effect of the R&D credit on
                        firm i's ETR reported for 1983-1985;
[FOREIGNCREDIT.sub.i] = mean percentage effect of the foreign tax
                        credit on firm i's ETR reported for 1983-1985;
  [ALTERNATIVE.sub.i] = 1 if firm i reported a net operating loss
                        carryover for the fiscal year ended prior to
                        September 1986 in excess of 90 percent of the
                        firm's average pretax income over the prior
                        three years, 0 otherwise;
     [FRPRETAX.sub.i] = Value Line forecast of firm i's 1987 pretax
                        income made in fourth quarter of 1986, less
                        the Value Line forecast of firm i's 1987 pretax
                        income made in the third quarter of 1986,
                        scaled by absolute value of the Value Line
                        forecast of firm i's 1987 pretax income
                        forecast made in the fourth quarter of 1986;
   |[FEPRETAX.sub.i]| = the absolute value of the ratio of the error
                        in the Value Line forecast of firm i's 1987
                        pretax income, scaled by the actual 1987 pretax
                        income. The error in the Value Line forecast of
                        firm i's 1987 pretax income is the actual 1987
                        pretax income less the Value Line forecast of
                        firm i's 1987 pretax income forecast made in
                        the fourth quarter of 1986; and
  [INTERACTION.sub.i] = STATUTORY x mean percentage effect on firm i's
                        ETR reported for 1983-1985.

TABLE 4

Regression of Change in Absolute Errors in Value Line Forecasts of
Effective Tax Rates on Proxies for the Effects of Six Tax-Law Changes
of Increasing Complexity Enacted by the Tax Reform Act of
1986--Alternative Specification (a)

[DELTA]|FE| = [alpha] + [[beta].sub.R1][STATUTORY.sub.i] +
              [[beta].sub.R2][CAPITALGAIN.sub.i] +
              [[beta].sub.R3][ITCREDIT.sub.i] +
              [[beta].sub.R4][R&DCREDIT.sub.i] +
              [[beta].sub.R5][FOREIGNCREDIT.sub.i]
              + [[beta].sub.R6][ALTERNATIVE.sub.i] +
              [[delta].sub.1][DELTA]|[FEPRETAX.sub.i]| +
              [[delta].sub.2][INTERACTION.sub.i] + [epsilon]

Complexicity Rank (a)

Variable (b)     Intercept     [[beta].sub.R1]    [[beta].sub.R2]

Coefficient      5.64          -0.39              0.02
t-statistic      3.53          -1.54              0.52

Variable (b)     Intercept     [[beta].sub.R3]    [[beta].sub.R4]

Coefficient      5.64          0.06               0.43
t-statistic      3.53          1.52               2.37 **

Variable (b)     Intercept     [[beta].sub.R5]    [[beta].sub.R6]

Coefficient      5.64          0.68               1.50
t-statistic      3.53          2.70 **            1.69 *

Variable (b)     Intercept     [[delta1].sub.1]   [[delta].sub.2]

Coefficient      5.64          3.67               0.50
t-statistic      3.53          5.95 (###)         1.97 (#)

Variable (b)     Intercept     Adj. [R.sup.2]

Coefficient      5.64          16.7%
t-statistic      3.53

*, **, *** p < 0.05, p < 0.01, p < 0.001, respectively, one-tailed;
(#), (##), (###) p < 0.05, p < 0.01, p < 0.001, respectively,
two-tailed.

(a) The six tax-law variables in the regression are ranked by
complexity, from least complex (R1) to most complex (R6).

(b) Variable definitions:

      [DELTA]|[FE.sub.i]| = difference between the absolute value of
                            the 1987 ETR forecast error and the
                            absolute value of the average of the firm's
                            1984 and 1985 ETR forecast errors. The 1987
                            ETR forecast error is the actual 1987 ETR
                            less the Value Line forecast of firm i's
                            1987 ETR forecast made in the fourth
                            quarter of 1986. I compute analogous
                            absolute forecast errors for 1984 and 1985;
        [STATUTORY.sub.i] = 46 percent (the previous corporate tax
                            rate) less the blended tax rate (based on
                            firm i's fiscal year end-varies from 45
                            percent to 34 percent);
      [CAPITALGAIN.sub.i] = mean percentage effect of the capital gains
                            tax rate on firm i's ETR reported for
                            1983-1985;
         [ITCREDIT.sub.i] = mean percentage effect of the investment
                            tax credit on firm i's ETR reported for
                            1983-1985;
        [R&DCREDIT.sub.i] = mean percentage effect of the R&D credit on
                            firm i's ETR reported for 1983-1985;
    [FOREIGNCREDIT.sub.i] = mean percentage effect of the foreign tax
                            credit on firm i's ETR reported for
                            1983-1985;
      [ALTERNATIVE.sub.i] = 1 if firm i reported a net operating loss
                            carryover for the fiscal year ended prior
                            to September 1986 in excess of 90 percent
                            of the firm's average pretax income over
                            the prior three years, 0 otherwise;
[DELTA]|[FEPRETAX.sub.i]| = the difference between the absolute value
                            of the ratio of the error in Value Line's
                            forecast of firm i's 1987 pretax income,
                            scaled by the actual 1987 pretax income and
                            the mean of the absolute value of analogous
                            measures for 1984 and 1985. The 1987 pretax
                            income forecast error is the actual 1987
                            pretax income less the Value Line forecast
                            of firm i's 1987 pretax income made in the
                            fourth quarter of 1986; and
      [INTERACTION.sub.i] = [STATUTORY.sub.i] x mean percentage effect
                            on firm i's ETR reported for 1983-1985.

I would like to thank the members of my dissertation committee (Douglas J. Skinner [Chair], Jeffery Abarbanell, Phil Howrey, and Joel Slemrod), Darrell Brown, Martha Eining, Donna Philbrick, David Plumlee, Taylor Randall, Claire Richards, and especially Christine Botosan for their many insightful comments and suggestions. I have also benefited from comments and suggestions from Ken Klassen, Jerry Salamon, three anonymous reviewers, and the workshop participants at the University of Kansas, University of Massachusetts, University of Michigan, University of Utah, Washington University, and the 1998 American Accounting Association Annual Meeting. I am grateful for the financial support of the Andersen Foundation and the Paton Research Fund of the University of Michigan. This paper is based on my dissertation at the University of Michigan.

Submitted February 2001

Accepted July 2002

(1) In my study, complexity relates to the underlying economic changes resulting from a tax-law change, not to the application of an accounting standard. However, in both cases the complexity of the information imposes a cost on the users of that information

(2) For example, the Value Line report for Big B, Inc., on July 25, 1986 (prior to enactment of the Tax Reform Act of 1986), stated: "Big B will benefit if the tax proposals now under consideration are enacted. In fiscal 1987 the company's tax rate would decline considerably below our current estimates" (Value Line 1986a, 788) (bold in the original). The report for Prime Computer issued November 7, 1986 (after the enactment of the Tax Reform Act of 1986) stated: "Our estimates include the effects of recent tax reform, but we do not expect Prime's tax rate to be much affected by the new code. Though Prime will lose investment tax credits, research and development credits will likely produce a negligible net effect in this area" (Value Line 1986c, 1110) (bold in the original).

(3) Mikhail et al. (1997, 1999) conclude that forecast accuracy is important to analysts, and they document that analysts' forecasts become more accurate as analysts gain firm-specific experience. Clement (1999) finds that more experienced analysts and those who work for larger brokers tend to be more accurate, while analysts who follow larger numbers of firms and industries are less accurate.

(4) I consider information complexity as one facet of task complexity.

(5) Maines and McDaniel (2000) find that the reporting format of comprehensive income affects nonprofessional investors' weighting, rather than acquisition, of the information.

(6) As an alternative proxy for the effect of these four tax laws on firms' ETRs, I could have used the difference between the actual amount of the 1987 reconciling item and the average effect of the reconciling item in past years. However, I chose to limit the proxies to information included in firms' footnotes prior to the forecast date to ensure that the information was publicly available to the analysts at the time they made their ETR forecasts.

(7) 0mer and Atwood (1991) report that two-thirds of firms paying the alternative minimum tax did so because the tax law limited their use of net operating loss carryovers when calculating the alternative taxable income. In addition, of the six variables included in their study, Manzon and Plesko (1999) report that NOL is the best indicator of firms that reported paying the alternative minimum tax.

(8) Because Value Line does not explicitly forecast pretax income, I calculate it by dividing forecasted net profit by (1 - forecasted ETR). For consistency with the dependent variable (ETR), I deflate pretax income forecast revisions by the absolute value of the 1987 pretax income forecast made in the fourth quarter of 1986.

(9) During my sample period, corporate alternative minimum tax credits arose only for alternative minimum tax paid due to temporary differences. The Revenue Reconciliation Act of 1989 changed this rule such that, for tax years beginning after 1989, corporations received alternative minimum tax credits for the entire amount of their alternative minimum tax paid.

(l0) The AICPA designed the tax-complexity index for accountants and members of Congress to assess whether a proposed tax-law change would increase or decrease the overall complexity of the tax system. Generally, the questions in the index relate to definitions or concepts that may change due to the tax-law change, changes or additions to taxpayers' calculations, when the tax-law change would be implemented, and taxpayer recordkeeping requirements.

(11) Using median values does not affect the complexity rankings.

(12) If the rankings across all the individuals are exactly the same (no disagreement in relative complexity), Kendall's W will be 1. If there is complete disagreement across all individuals, then Kendall's W will be 0 or very close to 0 (Conover 1980, 305).

(13) For 44 percent of the firms, Value Line's "actual" firms' ETRs differ from ETRs calculated from Compustat. As Value Line ETRs are the appropriate benchmark for assessing Value Line analysts' forecast accuracy, I hand-collect both the actual and forecast data from Value Line.

(14) While the Tax Reform Act of 1986 and many of its related law changes were widely publicized prior to this time, these pre- and post-Tax Reform Act of 1986 periods are consistent with Value Line's statements at the beginning of the fourth-quarter forecasts: "As noted on the 'pink sheet' accompanying this week's Edition, Value Line will henceforth include the effects of tax-law changes in estimates of future earnings" (Value Line 1986b, 99).

(15) For example, for a Value Line forecast issued in October 1986 for a firm with a November fiscal year-end, I collected Value Line's forecast of the annual effective tax rate for the fiscal year ending November 1987.

(16) I was able to locate 364 firms with necessary tax footnote data to estimate the independent variables that proxy for the effects of the four tax-law changes. These proxies are typically positive, although, for four firms the ITCREDIT proxy was negative due to investment tax recapture. I eliminated these four firms because the relation between these negative values and the dependent measures may be different than the relation between the positive values and the dependent measures. Including these firms does not affect the study's inferences.

(17) Retaining these observations does not affect the study's inferences.

(18) I also estimate the regressions including the main effect of the foreign tax rate differential in addition to the INTERACTION term. The main effect variable of the foreign tax rate differential is never significant and the overall results are substantively similar, so I do not include the main effect variable in the reported regressions.

(19) To be consistent, I use the change in the absolute value of pretax income forecast error in this specification of Model 2.

(20) Multiplying the coefficients on FOREIGNCREDIT and ALTERNATIVE of 0.73 and 1.96, respectively (Table 3, Panel B) by their mean values of 0.14 and 0.10, respectively (Table 2, Panel B) yields an average forecast error due to under-using available information in the amount of 0.102 and 0.196, respectively. These tax-law changes affect 12.4 percent and 10.4 percent of the firms in my sample; thus, an estimate of the average error of under-using the information would be 0.82 percent and 1.96 percent, respectively. With a mean forecasted net profit of $235 million for my sample firms, the average estimated error in net profit due to the failure to incorporate the effects of these tax laws ranges between $1.93 and $4.61 million. These values must be interpreted with caution, because the calculations are based on point estimates from a regression and mean firm-specific values. However, they provide a rough estimate of the potential improvement in analysts' forecast accuracy from incorporating information related to these complex tax-law changes.

REFERENCES

Abarbanell, J. 1991. Do analysts' earnings forecasts incorporate information in prior stock price changes? Journal of Accounting and Economics 14 (2): 147-165.

--, and V. Bernard. 1992. Tests of analysts' overreaction/underreaction to earnings information as an explanation for anomalous stock price behavior. The Journal of Finance 47 (3): 1181-1208.

--, and B. Bushee. 1997. Fundamental analysis, future earnings, and stock price. Journal of Accounting Research 35 (1): 1-24.

--, and --. 1998. Abnormal returns to a fundamental analysis strategy. The Accounting Review 73 (1): 19-45.

Ali, A., A. Klein, and J. Rosenfeld. 1992. Analysts' use of information about permanent and transitory earnings components in forecasting annual EPS. The Accounting Review 67 (1): 183-199.

American Institute of Certified Public Accountants (AICPA). 1993. AICPA Tax Complexity Index. New York, NY: AICPA.

Belsley, D. A., E. Kuh, and R. E. Welsch. 1980. Regression Diagnostics: Identifying Influential Data and Sources of Collinearity. New York, NY: Wiley.

Bernard, V., and J. Thomas. 1989. Post-earnings announcement drift: Delayed price response or risk premium? Journal of Accounting Research 27 (Supplement): 1-48.

--, and --. 1990. Evidence that stock prices do not fully reflect the implications of current earnings for future earnings. Journal of Accounting and Economics 13 (4): 305-340.

Bettman, J., E. Johnson, and J. Payne. 1990. A componential analysis of cognitive effort in choice. Organizational Behavior and Human Decision Processes 45 (1): 111-140.

Biddle, G. C., and W. E. Ricks. 1988. Analyst forecast errors and stock price behavior near the earnings announcement dates of LIFO adopters. Journal of Accounting Research 26 (2): 169-194.

Boynton C., P. Dobbins, and G. Plesko. 1992. Earnings management and the corporate alternative minimum tax. Journal of Accounting Research 30 (Supplement): 131-154.

Chen, K., and M. Schoderbek. 2000. The 1993 tax rates increase and deferred tax adjustments: A test of functional fixation. Journal of Accounting Research 38 (1): 23-45.

--, M. Danielson, and M. Schoderbek. 2002. Analysts' interpretation of transitory earnings components: Evidence from forecast revisions after disclosure of the 1993 deferred tax adjustment. Working paper, Rutgers University.

Clement, M. 1999. Analyst forecast accuracy: Do ability, resources, and portfolio complexity matter? Journal of Accounting and Economics 27 (3): 285-303.

Collins, J., and D. Shackelford. 1992. Foreign tax credit limitations and preferred stock issuances. Journal of Accounting Research 30 (Supplement): 103-124.

Conover, W. J. 1980. Practical Nonparametric Statistics. New York, NY: John Wiley & Sons.

Dhaliwal D., and S. Wang. 1992. The effect of book income adjustment in the 1986 alternative minimum tax on corporate financial reporting. Journal of Accounting and Economics 15 (1): 7-26.

Earley, P. C. 1985. Influence of information, choice and task complexity upon goal acceptance, performance, and personal goals. Journal of Applied Psychology 70 (3): 481-491.

Einhorn, H., R. Hogarth, and E. Klempner. 1977. Quality of group judgment. Psychological Bulletin 87 (1): 158-172. Financial Accounting Standards Board (FASB). 2002. Facts about FASB. Stamford, CT: FASB.

Gramlich, J. 1991. The effect of the alternative minimum tax book income adjustment on accrual decisions. The Journal of the American Taxation Association 13 (1): 36-56.

Guenther, D. 1994. Earnings management in response to corporate tax rate changes: Evidence from the 1986 Tax Reform Act. The Accounting Review 69 (1): 230-243.

Hirst, E., and P. Hopkins. 1998. Comprehensive income reporting and analysts' valuation judgments. Journal of Accounting Research 36 (Supplement): 47-75.

--, --, and J. Wahlen. 2002. Fair values, performance reporting, and bank analysts' risk and valuation judgments. Working paper, The University of Texas at Austin and Indiana University.

Iselin, E. 1988. The effects of information load and information diversity on decision quality in a structured decision task. Accounting, Organizations and Society 13 (2): 147-164.

Jacob, J. 1997. Discussion of do security analysts improve their performance with experience? Journal of Accounting Research 35 (Supplement): 159-166.

Kim, K., and D. Schroeder. 1990. Analysts' use of managerial bonus incentives in forecasting earnings. Journal of Accounting and Economics 13 (1): 3-23.

Maines, L., and L. McDaniel. 2000. Effects of comprehensive-income characteristics on nonprofessional investors' judgments: The role of financial statement presentation format. The Accounting Review 75 (2): 179-207.

Manzon, G. 1992. Earnings management of firms subject to the alternative minimum tax. The Journal of the American Taxation Association 14 (2): 88-111.

--, and G. Plesko. 1999. Strategic disclosure for political gain: The case of the corporate alternative minimum tax. Working paper, Massachusetts Institute of Technology.

Maydew, E. 1997. Tax-induced earnings management by firms with net operating losses. Journal of Accounting Research 35 (1): 83-96.

McEwen, R., and J. Hunton. 1999. Is analyst forecast accuracy associated with accounting information use? Accounting Horizons 13 (1): 1-16.

Mikhail, M., B. Walther, and R. Willis. 1997. Do security analysts improve their performance with experience? Journal of Accounting Research 35 (Supplement): 131-157.

--, --, and --. 1999. Does forecast accuracy matter to security analysts? The Accounting Review 74 (2): 185-200.

O'Brien, P. C. 1988. Analysts' forecasts as earnings expectations. Journal of Accounting and Economics 10 (1): 53-83.

Omer, T. and T. Atwood. 1991. Income reporting strategies of alternative minimum tax-based clienteles. Working paper, University of Illinois.

Paquette, L., and T. Kida. 1988. The effect of decision strategy and task complexity on decision performance. Organizational Behavior and Human Decision Processes 41 (1): 128-142.

Payne, J. 1976. Task complexity and contingent processing in decision making: An information search and protocol analysis. Organizational Behavior and Human Performance 16 (2): 366-387.

--. 1982. Contingent decision behavior. Psychological Bulletin 92 (2): 382-402.

--, J. Bettman, and E. Johnson. 1988. Adaptive strategy selection in decision making. Journal of Experimental Psychology: Learning, Memory and Cognition 14 (4): 534-552.

Philbrick, D., and W. Ricks. 1991. Using value line and I/B/E/S analyst forecasts in accounting research. Journal of Accounting Research 29 (2): 397-417.

Scholes, M., G. P. Wilson, and M. Wolfson. 1992. Firms' responses to anticipated reductions in tax rates: The Tax Reform Act of 1986. Journal of Accounting Research 30 (Supplement): 161-185.

Securities and Exchange Commission (SEC). 1997. Plain English Disclosure. 17 CFR Parts 228, 229, 230, 239 and 274. Release Nos. 33-7497, 34-39593, IC-23011, International Series No. 1113; File No. S7-3-97, RIN 3235-AG88. Washington, D.C.: Government Printing Office.

Smith, J., T. Mitchell, and L. Beach. 1982. A cost-benefit mechanism for selecting problem-solving strategies: Some extensions and empirical tests. Organizational Behavior and Human Performance 29 (3): 370-396.

Value Line. 1986a. Value Line Investment Survey. July 25. New York, NY: Value Line Publishing, Inc.

--. 1986b. Value Line Investment Survey. September 26. New York, NY: Value Line Publishing, Inc.

--. 1986c. Value Line Investment Survey. November 7. New York, NY: Value Line Publishing, Inc.

Wilkie, P. J., and S. Limberg. 1992. Measuring explicit tax (dis)advantage for corporate taxpayers: An alternative to average effective tax rates. The Journal of the American Taxation Association 14 (1): 76-91.

Marlene A. Plumlee
University of Utah