Disclosure versus Recognition of Option Expense: An Empirical Investigation of SFAS No. 148 and Stock Returns

Disclosure versus Recognition of Option Expense: An Empirical Investigation of SFAS No. 148 and Stock Returns Steven Balsam Fox School of Business and Management Temple University Philadelphia, PA 19122 Eli Bartov Leonard N. Stern School of Business New York University New York, NY 10012 Jennifer Yin School of Business Rutgers University Camden, New Jersey 08102 October 2004 Preliminary Comments Welcome

Disclosure versus Recognition of Option Expense: An Empirical Investigation of SFAS No. 148 and Stock Returns ABSTRACT Is quarterly employee stock option expense reported under the recently promulgated SFAS No. 148 value relevant? If so, is there a differential valuation effect related to the placement of the option expense within the financial statements, i.e., on the income statement or only in a footnote? Using a sample of 302 distinct firms (839 firm quarters) that recently began recognizing option expense voluntarily, and thus reported both disclosed and recognized option expenses, we find that whether the quarterly option expense is only disclosed in the footnotes, or whether it is also recognized in the income statement, the market values the cost associated with employee stock options as an expense. More importantly, we also find that the market valuation of that expense does not differ whether the amount is only disclosed in the footnotes, or whether it is also recognized in the income statement. Our findings are especially topical given the current FASB exposure draft, issued on March 31, 2004, Accounting for Share-based Payments. In this draft the FASB, is proposing mandatory income statement recognition for employee stock option expense for all firms. By showing that market participants efficiently and equally value the stock option expense whether it is disclosed or recognized, we show that firms need not worry about the first order effect of mandated recognition on their share prices.

Disclosure versus Recognition of Option Expense: An Empirical Investigation of SFAS No. 148 and Stock Returns I. INTRODUCTION Accounting for employee stock options has been a controversial issue for more two decades now. Under Accounting Principles Board Opinion (APB) No. 25 (APB 1972), which was issued in 1972, the expense for these options was determined using the intrinsic value of the options at the date of grant. Given that the vast majority of options are granted at the then current market price, this meant that for the vast majority of options no expense was recognized. 1 In 1984 the Financial Accounting Standards Board (FASB) concluded that stock options were an expense and in June of 1993 issued an Exposure Draft proposing that: (i) at the grant date the fair value of an option grant be capitalized by creating a deferred-compensation-cost asset with a corresponding credit to shareholders equity, and (ii) the deferred compensation cost be amortized over the options vesting period. The intense opposition to that proposal forced the FASB to back down from mandating expensing. 2 Instead, in SFAS No. 123, issued in 1995, the FASB allowed firms to continue applying the intrinsic value method of APB No. 25 in their financial statements while mandating footnote disclosure of pro-forma income and earnings per share as if the expense had been recognized using the fair value method. While SFAS No. 123 encouraged firms to use the fair value method of accounting for employee stock options on the income statement (see par. 62 of SFAS No. 123), until the summer of 2002 only a few firms elected to recognize option option. Then, in 1 Both Matsunaga (1995, footnote 6) and Murphy (1998) find about 95 percent of corporations granting options with an exercise price equal to grant-date stock price. 2 See for example, par. 376 of Statement of Financial Accounting Standards (SFAS) No. 123, Accounting for Stock-Based Compensation (FASB 1995). 1

response to the accounting scandals of 2001 and 2002 (e.g., Enron, WorldCom, Global Crossing) that, among other things, brought option abuses to the spotlight, a number of large publicly traded companies began voluntarily recognizing option expense using the fair value method. To provide guidance for the transition from disclosure to recognition, in December of 2002 the FASB issued SFAS No. 148, Accounting for Stock-Based Compensation-Transition and Disclosure, which amended SFAS No. 123. SFAS No. 148 also increased the frequency with which firms disclosed their pro-forma numbers. In addition to the annual disclosures, SFAS No. 148 requires new quarterly disclosures for all companies with stock-based compensation starting in the fist quarter of 2003. SFAS No. 148 also improved the prominence and clarity of the pro-forma disclosures required by SFAS No. 123 by prescribing a specific tabular format in which companies would now report the total stock-based compensation cost, as well as the amount recognized on the income statement. However, SFAS No. 148 did not close the books on the options reporting issue. On March 31, 2004 the FASB issued an exposure draft entitled Accounting for Share-Based Payments, which, like its exposure draft of 11 years earlier, would mandate the recognition of the expense associated with employee stock options using the fair value of the options measured at the grant date. On October 13, 2004, in face of opposition from securities regulators and corporate executives, the FASB modified its proposal by moving the proposed implementation date for mandatory expensing of stock options from the first to the third quarter of 2005. 3 Like its earlier proposal, this new exposure draft has been met by vocal opposition, with criticisms taking three broad approaches. The first is the economic 3 The Wall Street Journal (2004) reports that pressure from securities regulators and corporate executives was the reason for the FASB to delay by six months its long-awaited plan to require option expense recognition. 2

consequences argument: Mandated expensing will diminish firms ability to show growing profitability and thus affect negatively their stock price performance. This will reduce their ability to offer options to employees and/or raise capital, and hence adversely affect their ability to compete in the global market. The second line of attack takes issue with the impreciseness of the calculations and notes that the amounts are disclosed anyway. The third argument against option-expense recognition is that employee stock options motivate employees to work harder, thereby leading to higher profitability which (partially) offsets the cost associated with options grants. Prior research has investigated the relation between SFAS No. 123 annual option expense and equity values with conflicting results. Bell et al. (2002) investigated a sample of 85 profitable companies from the software industry (three-digit SIC code 737) in the three-year period, 1996-1998. For this sample, they failed to document the expected negative relation between annual option expense disclosed in the footnotes and the market value of common equity at fiscal year end. Conversely, Aboody et al. (2004a), who used the same three-year sample period but a larger sample of 2,274 firmyears, documented a negative correlation between disclosed annual option expense and both year-end stock prices and annual raw stock returns. Aboody et al. interpret their results as evidence that the Bell et al. findings are sample specific and not generalizable to other samples. The purpose of this paper is two fold. First, by utilizing the quarterly disclosures provided under SFAS No. 148, we assess the value relevance of the newly promulgated disclosure requirement that option expense be disclosed quarterly. Second and more importantly, by utilizing a sample of firms that have voluntarily adopted the recognition 3

provisions of SFAS No. 123 recently, we examine the valuation effects of disclosed versus recognized option expense. 4 Our sample consists of 302 distinct firms (839 firm-quarters) spanning the first three quarters of 2003, the first year SFAS No. 148 was in effect. Along the line of prior research (Bell et al. 2002 and Aboody et al. 2004a), we isolate the effect of option expense on stock returns by using a specification that considers seasonal changes in earnings, unexpected earnings growth, and firm size as control variables, in addition to our two test variables, recognized and unrecognized option expense. Using abnormal returns accumulated over a 90-day interval around the quarterly disclosures (CAR), we find that quarterly disclosed option expense is value relevant. More importantly, we find that disclosed and recognized option expenses are reflected in stock returns in a similar fashion. That is, regressing CAR on recognized option expense and unrecognized option expense, after controlling for variables shown by prior research to explain stock returns, indicates that the coefficients on the two option expense variables are similar. Our findings contribute to extant literature on the value relevance of option expense in three ways. First and foremost, we contribute to the ongoing debate on option disclosure versus recognition by showing that the placement of the option expense within the financial statements, whether on the income statement or only in the footnotes, makes little difference in terms of investor perceived costs associated with employee stock option grants. Specifically, corporate executives have been lobbing regulators for more 4 Most of the firms adopting the recognition provision of SFAS No. 123 elected to do so prospectively. Consequently, while they recognized the cost associated with the options granted after the date of adoption, the cost associated with options granted prior to adoption was only recognized in pro-forma income. Thus the firms have both recognized and disclosed amounts. 4

than a decade against option expense recognition primarily due to its potential adverse economic consequence. Our results indicate that the economic consequences of option expense recognition are unlikely to be significant, thereby alleviating a major concern leveled against option expense recognition by opponents of the rule change. Second, we are the first to document that SFAS No. 148 disclosures are valuation relevant. These findings are important because they alleviate reliability concerns leveled against expensing costs associated with stock option grants by opponents of the fair value method. Third, our findings provide independent evidence that once potential benefits associated with options incentive effects are controlled, option expense is negatively associated with changes in firm value. This independent-of-prior-work evidence, generated by using quarterly rather than annual data and a more recent sample period (2003 versus 1996-1998), is important in light of the conflicting results reported in prior research. The remainder of the paper is organized in three sections. Section II outlines the empirical model and the tests, and defines the variables. Section III presents the sample selection procedure and describes the data. Section IV reports the results, and Section V summarizes the findings and states our conclusions. II. RESEARCH DESIGN The primary purpose of this study is to investigate the valuation effects of disclosed versus recognized option expense. The recent spate of voluntary adoptions of the option expense recognition alternative allowed under SFAS No. 123, and the recent promulgation of SFAS No. 148 requiring firms to disclose recognized and unrecognized 5

option expense quarterly have created a unique opportunity to test the valuation effects of disclosed versus recognized option expense. To test this, as well as the valuation relevance of the newly promulgated SFAS No. 148 disclosures, we employ the following model: CAR it = α 0 + α 1 EPSBCX it + α 2 UNRECCX it + α 3 RECCX it (1) + α 4 LOGASST it + α 5 GROWTH it + ε it Where: CAR is size-adjusted buy-and-hold stock returns for the 90-day period, staring 45 days after the end of the prior fiscal quarter; EPSBCX is the seasonal change in pershare net income before extraordinary items, discontinued operations, and any recognized stock-based compensation expense, deflated by the stock price at the beginning of the quarter; UNRECCX is the seasonal change in per-share disclosed, i.e., unrecognized stock-based compensation expense, net of tax, deflated by the stock price at the beginning of quarter; RECCX is the seasonal change in the per-share stock-based employee compensation expense included in reported net earnings, net of tax, deflated by the stock price at the beginning of quarter; LOGASST is natural logarithm of total assets at the beginning of the quarter; GROWTH is the seasonal change in quarter-end I/B/E/S mean analyst earnings growth forecast; and i and t denote firm and quarter. There are two important research design choices underlying the empirical model. First, we cumulate the abnormal returns over a 90-day window as opposed to a two-day interval around the earnings announcement. In general, the relation between an accounting variable of interest (e.g., unexpected earnings or cash flows) and returns can be explored by 6

using either an event study or an association study methodology. 5 In estimating Equation (1), however, the latter must be used, because due to the special attention stockbased compensation drew during our sample period, it is particularly difficult to determine when in the quarter the option-cost information was assimilated into stock prices. This approach is also in line with prior research in this area (e.g., Aboody et al. 2004a). It is standard in the accounting literature to assume that quarterly financial statements are available to investors within 45 days of the end of the fiscal quarter, 6 and that the market is efficient in the semi-strong form (i.e., stock prices immediately and unbiasedly incorporate public information). Given this, our 90-day return interval--which covers the period from the filing date of the previous quarter s 10-Q to the filing date of the current quarter s 10-Q--can be used to test the value relevance of SFAS No, 148 and the valuation effects of the recognized and unrecognized option expense. Our choice of quarterly size-adjusted returns as opposed to level regressions, where the dependent variable is the stock price (Bell et al. 2002, and Aboody et al. 2004a), or annual raw returns (Aboody et al. 2004a), differentiates our methodology from those of prior studies. Our use of abnormal stock returns is preferable to raw returns or stock prices, as this specification is less vulnerable to the correlated-omitted-variable problem, which tends to bias regression coefficients. Another advantage of this choice is that it resolves the serious problem underlying prior research of a mechanical positive 5 In an event study, returns are measured over a short interval (i.e., a few days) around the announcement date, whereas in an association study longer windows are used (e.g., fiscal quarters or even years). The choice of the length of the event window involves a tradeoff. While windows that are too wide increase the noise-to-signal ratio and thereby decrease the power of the experiment, windows that are too narrow might exclude the event of interest and thereby bias the parameter estimate on the accounting variable that captures that event toward zero. 6 Firms are required to file their quarterly reports (10-Q) with the Security and Exchange Commission (SEC) within 45 days of the end of their fiscal quarter. 7

relation between their dependent variable, stock price, and the option expense variable, which biases the coefficient on the option expense (see, e.g., Bell et al. p. 991). Prior research addresses this problem by using elaborate estimation procedures that reduce power and perhaps even make the results hard to interpret. Our use of abnormal returns as the dependent variable should resolve this problem as there is little reason to believe that abnormal stock returns will be correlated with stock prices. Indeed, for our sample the Pearson (Spearman) correlation coefficient between quarter-end stock price and CAR is -0.022 (0.015) with observed significance level of 0.50 (0.65). Second, along the line of prior research we control for: (1) a variable known to have explanatory power for stock returns, EPSBCX, (2) a variable capturing the potential benefits associated with options incentive effects, GROWTH, and (3) a catch all variable, LOGASST, for possible unknown omitted variables. While the need to control for EPSBCX and LOGASST is straightforward, the use of GROWTH to control for the potential benefits associated with options incentive effects deserves further explanation. Recall that the 1993 Exposure Draft proposed that at the grant date the fair value of an option grant be capitalized by creating a deferred-compensation-cost asset. The reason for this according to FASB is that employee stock options represent probable future benefits to the firm (see Para. 63 of the Exposure Draft). If so, the option expense reflects not only costs to the firm but also benefits, which, if not controlled for, will weaken the expected negative relation between the option expense and CAR. To address this potential problem, we use GROWTH as a control variable because future expected earnings growth is likely to capture future expected benefits associated with stock-based compensation. 8

In terms of Equation (1), our first hypothesis that SFAS No. 148 quarterly disclosure of unrecognized option expense is value relevant suggests, H 1 : α 2 < 0. 7 Our second hypothesis of no difference in the valuation effect between recognized and unrecognized option expense implies, H 2 : α 2 = α 3. This hypothesis follows because rational investors should price stocks based on the nature of information provided in the financial statements, not based on the placement of this information (in the income statement or only in a footnote). III. DATA Sample Selection Table 1 summarizes the effects of our sample selection procedure on the sample size. The basis for our sample is the population of firms voluntarily announcing their intention to expense stock options. As noted earlier, firms began announcing their intention to expense stock options in the summer of 2002. A list of these firms was compiled by Bear Stearns & Co. Inc. We use this Bear Stearns list (cutoff date February 12, 2004), which contained 483 distinct firms corresponding to 1,449 firm-quarter observations, as the starting point of our sample. We delete 12 firms (36 firm-quarters) for which the expensing announcement date was missing on the Bear Stearns list, and additional 48 firms (151 firm-quarters) for which we could not find the Form 10-Q on the Securities and Exchange Commission s EDGAR (Electronic Data Gathering, Analysis, 7 Note that α 3 is expected to be negative even in the absence of SFAS No. 148 disclosures, as RECCX it would be recognized in the income statement even in the absence of SFAS No. 148. 9

and Retrieval) database. 8 This reduces our sample size to 423 firms (1,262 firmquarters). Next, we lose 41 firms (140 firm-quarters) that had announced their intention but have not yet begun to expense as of the third quarter of 2003, the end of our sample period. 9 Finally, we delete 80 firms (283 firm-quarters) because of incomplete data on either CRSP and/or Compustat. Our final sample thus consists of 302 firms corresponding to 839 firm-quarter observations. 10 For entities that adopt the fair value method of accounting for stock-based employee compensation, SFAS No. 148 amends the transition method specified in SFAS No. 123 by permitting two additional transition methods. Specifically, SFAS No. 123 required the use of the prospective method, which involves recognition of option expense for all employee awards granted, modified, or settled after the beginning of the fiscal year of adoption. SFAS No. 148 also permits the modified prospective method and the retroactive restatement method. Under the modified prospective method stock-based employee compensation cost is recognized from the beginning of the fiscal year of adoption as if the fair value method had been used to account for all employee awards granted, modified, or settled in fiscal years beginning after December 15, 1994. The retroactive restatement method calls for restatements of all periods presented to reflect stock-based employee compensation cost under the fair value method for all employee 8 We manually check all the data on the amount recognized and disclosed using 10-Qs, as in the course of our examination we noted severe problems with the corresponding Compustat data fields. In particular, because Compustat looks for the pretax amount of recognized expense and most companies only report the after tax amount, this variable is missing most of the time. 9 The reason that the number of firms-years, 140, exceeds the number of distinct firms, 41, is that 17 firmquarters began expensing in the second or third quarter of 2003. 10 We examine the first three quarters of 2003 so we have a maximum of three observations per firm. However for some firms we have less than three observations because the firm may not have adopted the recognition provisions in the first quarter. In addition, 14 sample firms (26 firm-quarters) had both zero recognized and zero unrecognized stock-based compensation expense. Removing these observations from our sample had no effect on the results. 10

awards granted, modified, or settled in fiscal years beginning after December 15, 1994. Our sample firms predominately selected the prospective method. Specifically, 731 firms-quarters (87.1 percent of our sample) selected this method, while 76 firmquarters (9.1 percent) selected the modified prospective method, and 20 firm-quarters (2.4 percent) selected the retroactive restatement method. 12 firm-quarters (1.4 percent) did not disclose sufficient information to allow us to identify the transition method selected. 11 Descriptive Statistics Table 2 contains some descriptive statistics comparing our sample to the population of Compustat firms not in our sample, i.e., the non adopters (Panel A) and to Execucomp firms not in our sample (Panel B). 12 We use both Compustat and Execucomp as benchmarks because the former covers a relatively large number of firms, whereas the latter provides important compensation data but its coverage is limited to primarily S&P 1500 firms. One consequence of this rather limited coverage is that 137 of our sample firms (45 percent of the sample) are not on Execucomp. 13 Using the Compustat universe as the benchmark, we examine on a quarterly basis, return on equity (ROE), earnings (E), and earnings per share (EPS), both in the level and 11 Given the relatively small number of firms selecting methods other than the prospective, we are unable to perform the tests below by transition method. Still, we replicated the tests below after removing 41 companies (108 firm-quarters) that selected a transition method other than the prospective method, and the results were unchanged. 12 The possibility exists that the remaining Compustat or Execucomp firms includes some adopters that we missed, i.e., that did not make it on to the Bear Stearns list. If these numbers are large, which we doubt, it would bias the descriptive statistics for the Compustat and Execucomp populations towards those of our sample. 13 Similarly, Aboody et al. (2004b, footnote 13) report that 40 of their sample firms (26 percent of the sample) are not on Execucomp. 11

seasonal change form; as well as assets (ASSETS), market value of common equity (MV), the debt-equity (D/E), market to book (MTOB), and price-earnings (P/E) ratios. The results presented in Panel A of Table 2 generally indicate that our sample firms are larger than the average Compustat firm, as both the means and medians of ASSETS and MV are greater for our sample. We also find most of our profitability measures, i.e., ROE, E, E, EPS and EPS, are greater for our sample, although some of these differences, i.e., E, are driven by the larger firm size. Looking at market valuation/growth, i.e., MTOB, P/E, and ROE we find mixed evidence, i.e., while the means are significantly lower for our sample firms, the differences in medians are statistically insignificant. Finally we find our firms are more highly levered as evidenced by significantly higher median D/E for our sample than the average Compustat firm, albeit the difference in means is insignificant. Using the Execucomp population as a benchmark allows us to examine not only profitability and solvency ratios, but also compensation related ratios such as the ratio of the CEO cash bonus to total cash compensation (BONUS), the percentage of the firm s shares owned by the CEO (CEO_OWN), the percentage of shares held by institutions (INSTIT), the percentage of options granted to the top five executives (OPT_TOP5), and the percentage of shares and options granted to outside directors relative to total number of shares outstanding (OSDIR_STK). Reading across Panel B of Table 2 we note that our sample firms are similar to a typical Execucomp firm with respect to CEO_OWN, and INSTIT, but different with respect to other compensation related ratios. More specifically, we find that BONUS is higher for our sample firms, whereas OSDIR_STK and OPT_TOP5 are both lower for our sample firms. To some extent these differences 12

may be driven by firm size, which is larger for our sample firms than for the typical Execucomp firm. In addition, although Execucomp coverage is relatively limited, comparing our sample firms to a typical Execucomp firm in terms of profitability (PROFIT), and leverage (DEBT_EQ) generates the same insights as the ones from the comparison with Compustat. As before, our sample firms are on the average more highly levered and more profitable than the typical benchmark firm (Execucomp firm). Industry Distribution Table 3 contrasts the distributions of our sample and Execucomp universe by twodigit standard industrial classification (SIC) code. 14 As the table shows, there is a difference in industry distribution between our sample and the Execucomp population. Of note, our largest concentration of adopters is in 2-digit SIC 67 Holding and Other Investment offices, which contains slightly more than 20 percent of our sample but less than 2 percent of the Execucomp population. Overall, relative to Execucomp financial institutions, insurance carriers, and utilities are overrepresented in our sample. Similar findings are reported in Aboody et al. (2004b, Table 1, Panel A) for their sample of 155 firms that, beginning in summer of 2002, voluntarily announced their intention to recognize option expense. Also worth noting, SIC code 73, Business Services, which incorporates SIC code 737 the knowledge firms used in Bell et al. (2002), contains only 2.5 percent of our sample, but 10.14 percent of the Execucomp population. This is consistent with 14 Comparing the industry distribution of our sample to that of Compustat firms generates similar observations. 13

knowledge based firms, for which the impact of expensing would be greatest, being less willing to voluntarily adopt option expense recognition. Certain other industries, however, are fairly represented in our sample. For example, SIC 13 Oil and Gas Extraction, where executive stock options play an important role in aligning the interests of shareholders and management (see, Rajgopal and Shevlin 2002), contains 3.58 present of our sample and 3.15 percent of Execucomp population. IV. EMPIRICAL RESULTS We begin our empirical investigation by estimating a benchmark model using a level regression, as follows: PRICE it = α 0 + α 1 BV it + α 2 EPS it + α 3 UNRECCX it + α 4 GROWTH it + ε it (2) Where, PRICE is share price at the end of the fiscal quarter; BV is book value per share of equity; EPS is earnings per share before extraordinary items and discontinued operations; UNRECCX is per-share unrecognized stock-based compensation expense, net of tax; GROWTH is quarter-end I/B/E/S mean analyst earnings growth forecast; and i and t denote firm and quarter. Our study differs from prior published work in this area (Bell et al. 2002, and Aboody et al. 2004a) along two important dimensions. First, our sample period spans the first three quarters of 2003 as opposed to a much earlier time period, 1996-1998, and second, we use quarterly data rather than annual data. The benchmark model--which is similar in spirit to the models used by those studies (Aboody et al. 2004a, Table 2, and Bell et al. 2002, Panel B of Table 6)--assesses the effects of these differences on our findings. 14

Table 4 displays the results from estimating the benchmark model (Equation 2). Panel A provides descriptive statistics for all variables included in the regression. Outlying observations, which give inordinate influence to a small number of observations in the regression, are typical of accounting data pooled over time and across firms. To address this problem, we winsorize each variable at two standard deviations from its mean. Reviewing the Minimum value and Maximum value of each variable, the winsorizing procedure seems successful as outlying observations are no longer present in the sample. Further, as might be expected the variable distributions in our sample seem economically similar to those of prior research. For example, the median stock price, book value per share, and long term growth rate in our sample are comparable to those reported in Aboody et al. 2004a, Panel B of Table 1. Panel B presents the estimation results from Equation (2). Similar to prior research, we find that the coefficients on the three control variables, BV, EPS, and GROWTH are all positive and significant. The coefficient on the unrecognized option expense, UNRECCX, is also significantly positive, which is consistent with the mechanical positive relation between the option expense and the stock price mentioned above. While Bell et al. (2002, Table 6, Panel B) also document such positive relation, Aboody et al. (2004a, Table 2) fail to document a significant relation between the disclosed option expense and the stock price. They argue that this failure follows because (p. 258), the coefficient estimates from this regression likely are biased and inconsistent. To gain additional insights into the level specification, we augment Equation (2) by separating the recognized option expense from earnings. More formally, we estimate 15

the following model: PRICE it = α 0 + α 1 BV it + α 2 EPSBCX it + α 3 UNRECCX it + (3) α 4 RECCX it + α 5 GROWTH it + ε it Where, EPSBCX is earnings per share before extraordinary items and discontinued operations and any recognized stock-based compensation expense; RECCX is per-share recognized stock-based employee compensation expense, net of tax; and all the other variables are as defined in Equation (2). The results show that the estimates on BV, UNRECCX, and GROWTH--the three variables common to Equations (2) and (3)--are positive and significant as before, as is the newly defined earnings variable, EPSBCX. However, the recognized option expense, RECCX, is insignificant, which is consistent with a model misspecification problem associated with the level specification, suggested by Aboody et al. (2004a). Next, we turn to the primary tests of our study, those generated by estimating Equation (1). Table 5 reports the results from this analysis. As before, Panel A reports descriptive statistics for all variables used in the regression. The descriptive statistics in Panel A provide little evidence of outliers presenting a serious problem; our winsorizing procedure seems successful in adjusting extreme observations. In addition, for all variables there is little difference between the mean and the median, implying symmetric distributions. Panel B reports the results from estimating four models nested in Equation (1). Examining first the results for the control variables, we note that EPSBCX is positive and significant in all four models. This positive relation between unexpected earnings and unexpected returns, which is expected, increases our confidence in the reliability of our findings. We also note that LOGASST is positive and significant in one of the 16

models (model 2) but not in the other (model 4). These mixed results may follow because our dependent variable is size adjusted, and thus already controls for size. Finally, GROWTH is insignificant in both models in which it is included. Recall that our first hypothesis predicts that SFAS No. 148 disclosures are value relevant. In terms of Equation (1), this prediction implies that the coefficient on the unrecognized option expense, UNRECCX, be negative. Consistent with this prediction, in all four variations of Equation (1), α 2 is negative, ranging between -0.020 and -0.021, and significant with t-statistics ranging from -2.42 to -3.08. Our second hypothesis predicts that stock prices will reflect recognized and unrecognized option expense in a similar fashion, as rational investors should price stocks based on the nature of information provided in the financial statements, not based on the placement of this information within the statements. In terms of Equation (1), this prediction implies that the coefficients on the recognized option expense ( RECCX) and unrecognized option expense ( UNRECCX) be similar. Before discussing the test results for our second hypothesis, we note that as expected the coefficient on recognized option expense is significantly negative for all four permutations. For example, when CAR is regressed on EPSBCX and on the two option expense variables, UNRECCX and RECCX, α 3, the coefficient on the recognized expense variable is negative, -0.014 and significant (t-statistic = -1.92); when the full model is estimated, α 3, is -0.023 with a t-statistic of -2.26. Turning to the test results for our second hypothesis, we fail to reject the null that α 2 = α 3 in all four permutations of Equations (1). Specifically, (untabulated) results from F-tests of equality between α 2 and α 3 for models (1), (2), (3), and (4) yield, respectively, 17

the following F-values (p-values): 0.30 (0.58), 0.49 (0.49), 0.03 (0.87), and 0.02 (0.89). These results suggest that stock prices reflect costs associated with stock option grants in a similar fashion regardless of the placement of the option expense within the financial statements. V. CONCLUSION This paper examines the association between stock returns and quarterly employee stock option expense disclosures under the recently promulgated SFAS No. 148. Our sample consists of 302 distinct firms (839 firm quarters) that recently began recognizing option expense voluntarily and thus reported both disclosed and recognized option expenses. We find that whether the quarterly option expense is only disclosed in the footnotes, or whether it is also recognized on the income statement, the market values the cost associated with employee stock options as an expense. More importantly, we also find that the market valuation of that expense does not differ whether the amount is only disclosed in the footnotes, or whether it is also recognized on the income statement. Our findings are especially topical given the current FASB exposure draft, issued on March 31, 2004, Accounting for Share-based Payments. In this draft FASB, like its international counterpart, the International Accounting Standards Board has already done, is proposing mandatory income statement recognition for employee stock option expense for all firms. By showing that market participants efficiently and equally value the stock option expense whether it is disclosed or recognized, we show that firms need not worry 18

about the first order effect of mandated recognition on their share prices. 15 Consequently the cost of recognition may not be as great as opponents claim. However, while the costs may not be as great as the opponents claim, the benefits may not be as great as the proponents claim either. That is, at least in terms of market value, investors seem to be efficiently processing the information already in the footnotes, and would not benefit from mandatory recognition of the expense. 16 Finally, we note that although our sample size is reasonably large, it consists only of firms that recognize option expense voluntarily. This limitation may lead to a self selection bias of unknown magnitude. Consequently, whether our results extend to firms that continue to only disclose option expense is an open question that may be fully addressed if the FASB makes it mandatory for all firms to recognize option expense on the income statement. 15 Firms may still be adversely affected if, for example debt covenants are written in terms of reported earnings. Managers may also be adversely affected by mandated expensing via the effect of reduced income on their compensation, e.g., lower accounting based bonus payments. 16 Still, it appears that it would be easier for investors to process earnings information if the option expense is recognized, as searching through footnotes and adjusting reported income to footnote information is costly. 19

References Aboody, David, Mary E. Barth, and Ron Kasznik. 2004a. SFAS No. 123 stock-based compensation expense and equity market values, The Accounting Review 79(2): 251-275. Aboody, David, Mary E. Barth, and Ron Kasznik. 2004b. Firms voluntary recognition of stock-based compensation expense, Journal of Accounting Research, forthcoming. Accounting Principles Board. 1972. Opinion No. 25: Accounting for stock issued to employees. Bell, Timothy, Wayne Landsman, Bruce Miller, and Shu Yeh. 2002. The valuation implication of employee stock option accounting for profitable software firms, The Accounting Review 77(4): 971-996. Financial Accounting Standards Board. 1995. Statement of Financial Accounting Standards No. 123: Accounting for Stock-Based Compensation. Financial Accounting Standards Board. 2002. Statement of Financial Accounting Standards No. 148: Accounting for Stock-Based Compensation Transition and Disclosure an amendment of FASB Statement No. 123. Financial Accounting Standards Board. 2004. Proposed statement of financial accounting standards: Accounting for share-based payments. Financial Accounting Standards Board. 1993. Proposed statement of financial accounting standards: Accounting for stock-based compensation. Matsunaga, Steven R. 1995. The effects of financial reporting costs on the use of employee stock options, The Accounting Review 70(1): 1-26. Murphy, Kevin J. 1998. Executive compensation, Working Paper, University of Southern California. Rajgopal, S., and T. Shevlin. 2002. Empirical evidence on the relation between stock option compensation and risk taking, Journal of Accounting and Economics 33, 145-171. The Wall Street Journal. 2004. Accounting Cops Delay Change. October 14, C1. 20

TABLE 1 Sample Selection Number of Firms Number of Firm-quarters Firms having adopted or announced intention to adopt fair 483 1,449 value expensing stock options as of 2/12/04 a Expensing announcement dates not available (12) (36) Firms on the Bear Stearns list with announcement dates 471 1,413 10-Q reports not available on EDGAR (48) (151) Firms on the Bear Stearns list with announcement dates and 10-Q reports available 423 1,262 Have not adopted fair value expensing by the 3 rd Quarter of 2003 b (41) (140) Firms on the Bear Stearns list with announcement dates and 10-Q reports available that have adopted fair value expensing before the end of the 3 rd Quarter of 2003 382 1,122 Incomplete CRSP and/or COMPUSTAT data (80) (283) Final sample 302 839 a Source: Bear, Stearns & Co. Inc. as of 2/12/04. b Our sample period ends with the third quarter of 2003. 21

Table 2 Descriptive Statistics Panel A: Sample Firms vs. Compustat Firms Sample Observations Compustat Observations Student Wilcoxon Variable N Mean Median Std. Dev. n Mean Median Std. Dev. t-value b z-value b ROE 815 0.021 0.029 0.113 20,064-0.075 0.016 2.164 6.09 ROE 815 0.008 0.004 0.159 19,859 0.043 0.004 2.436-1.92 E 839 129.450 20.210 298.424 24,267 12.984 0.356 59.826 11.30 E 839 35.393 1.758 144.073 23,818 5.231 0.165 61.360 6.04 7.32 EPS 839 0.433 0.340 0.709 22,949 0.235 0.030 58.814 0.51 21.54 EPS 839 0.102 0.050 0.657 22,461 9.505 0.020 791.704-1.78 ASSET 839 28751.622 4017.178 68806.587 24,019 2568.123 184.761 9150.791 11.02 MV 839 7717.657 1597.938 15750.203 23,211 947.283 91.351 2764.678 12.44 D/E 815 1.595 0.853 4.267 20,096 1.398 0.262 10.094 1.19 16.73 MTOB 815 2.233 1.638 3.139 18,328 4.747 1.660 24.178-11.99 P/E 691 22.197 16.662 21.240 12,395 25.907 15.862 35.157-4.28 Panel B: Sample Firms vs. Execucomp Firms Sample Firms Execucomp Firms Student Wilcoxon Variable N Mean Median Std. Dev. n Mean Median Std. Dev. t-value b z-value b ISSUE 165 0.758 1 0.430 2433 0.628 1 0.483 3.70 3.34 DEBT_EQ 165 1.080 0.815 2.115 1632 0.633 0.353 1.761 2.62 7.09 INT_COV 138 0.277 0.266 0.578 1438 0.196 0.121 0.858 1.50 5.29 ΒΟΝUS 150 0.469 0.505 0.268 1377 0.354 0.389 0.249 5.30 5.00 CEO_OWN(%) 53 3.613 1.080 5.465 804 3.579 1.090 5.397 0.04 0.96 OSDIR_STK(%) 164 0.003 0.001 0.005 1543 0.012 0.005 0.015-15.55 INSTIT (%) 150 63.949 64.355 16.617 1551 65.207 69.390 22.202-0.86-1.77 SIZE 165 8.409 8.381 1.521 1662 6.684 6.759 1.922 13.53 PROFIT 165 0.020 0.062 0.204 1595-0.631 0.045 7.014 3.69 OPT_TOP5(%) 147 25.368 19.970 19.295 1356 27.429 24.570 17.247-1.24-2.14 Data for Compustat are obtained for the first three quarters in 2003. Compustat firms exclude sample firms. ROE is earnings before extraordinary 22 * * -7.02 11.03 5.15 11.50-0.56 24.17 3.36 31.54 31.64-0.26 1.25 * **

items and discontinued operations divided by common equity at the end of the quarter; the observation is deleted if common equity is negative or zero; ROE is seasonal change in ROE; E is quarterly net income before extraordinary items and discontinued operations; E is seasonal quarterly change in E; EPS is quarterly earnings per share before extraordinary items and discontinued operations; EPS is seasonal quarterly change in EPS; ASSET is total assets at the end of the quarter; MV is market value of equity, or share price at the end of the quarter * shares outstanding at the end of the quarter; D/E ratio is long-term debt to common equity at the end of the quarter; the observation is deleted if common equity is negative or zero; MTOB is market to book value of common equity the end of the quarter; the observation is deleted if common equity is negative or zero; P/E ratio is price at the end of the quarter divided by 12-month moving earning per share at the end of the quarter. Data from Execucomp are obtained for year 2002, the last year for which complete data are available from Execucomp. Execucomp firms exclude sample firms. ISSUE is an indicator variable equal to one if the firm received funds from issuance of common and preferred stock in the last three years, and zero otherwise. DEBT_EQ is the ratio of long-term debt to shareholders equity. INT_COV is the ratio of interest expense to operating income after depreciation. BONUS is the ratio of CEO cash bonus to total cash compensation. CEO_OWN is the percentage of the company s shares owned by the CEO. OSDIR_STK is the percentage of shares of stock and options granted to each non-employee director to total shares outstanding at the end of the year; INSTIT is the percentage of shares outstanding held by institutional investors. SIZE is the natural logarithm of year-end market value of equity; PROFIT is net income before extraordinary items and discontinued operations deflated by market value of equity; OPT_TOP5 is the percentage of options granted to the top five executives. All variables are winsorized at two standard deviations. Student t-values and Wilcoxon z-values are for two-tailed tests of means and medians between the two 23

Two- Digit SIC TABLE 3 Two-digit SIC Composition of the Sample and Execucomp Firms No. of Sample Firm- Quarters No. of Execucomp Firms a Name Percentage Percentage 7 Agricultural Services 3 0.36% 1 0.04% 10 Metal Mining 3 0.36% 17 0.70% 13 Oil and Gas Extraction 30 3.58% 77 3.15% 14 Nonmetallic Minerals, Except Fuels 3 0.36% 5 0.20% 15 Building Construction General Contractors 7 0.83% 15 0.61% 20 Food and Kindred Products 11 1.31% 57 2.33% 21 Tobacco Manufacturers 3 0.36% 3 0.12% 22 Textile Mill Products 6 0.72% 18 0.74% 23 Apparel and other Textile Products 3 0.36% 18 0.74% 24 Lumber and Wood Products 6 0.72% 15 0.61% 25 Furniture and Fixtures 12 1.43% 12 0.49% 26 Paper and Allied Products 8 0.95% 38 1.55% 27 Printing and Publishing 6 0.72% 39 1.59% 28 Chemicals and Allied Products 37 4.41% 158 6.46% 29 Petroleum and Coal Products 13 1.55% 22 0.90% 30 Rubber and Misc. Plastics Products 6 0.72% 22 0.90% 33 Primary Metal Industries 3 0.36% 51 2.09% 34 Fabricated Metal Products 5 0.60% 33 1.35% 35 Industrial and Related Products 26 3.10% 161 6.58% 36 Electronic & Other Electric Equipment 13 1.55% 184 7.52% 37 Transportation Equipment 14 1.67% 67 2.74% 38 Instruments and Related Products 6 0.72% 123 5.03% 39 Miscellaneous Manufacturing Products 1 0.12% 20 0.82% 40 Railroad Transportation 3 0.36% 10 0.41% 42 Trucking and Warehousing 6 0.72% 18 0.74% 44 Water Transportation 6 0.72% 8 0.33% 45 Transportation by Air 3 0.36% 20 0.82% 47 Transportation Services 3 0.36% 11 0.45% 48 Communications 17 2.03% 74 3.03% 49 Electric, Gas, and Sanitary Services 42 5.01% 165 6.75% 50 Wholesale Trade-Durable Goods 6 0.72% 48 1.96% 51 Wholesale Trade-Nondurable Goods 3 0.36% 29 1.19% 52 Building Materials & Garden Supplies 6 0.72% 10 0.41% 53 General Merchandise Stores 18 2.15% 31 1.27% 54 Food Stores 2 0.24% 17 0.7% 57 Furniture and Home Furnishing Stores 1 0.12% 16 0.65% 58 Eating and Drinking Places 8 0.95% 38 1.55% 60 Depository Institutions 103 12.28% 172 7.03% 61 Nondepository credit Institution 17 2.03% 26 1.06% 62 Security and Commodity Brokers 26 3.10% 40 1.64% 24

63 Insurance Carriers 112 13.35% 103 4.21% 64 Insurance Agents, Brokers & Service 1 0.12% 13 0.53% 65 Real Estate 9 1.07% 1 0.04% 67 Holding and Other Investment Offices 174 20.74% 42 1.72% 70 Hotel and Other Lodging Places 3 0.36% 8 0.33% 72 Personnel Services 3 0.36% 9 0.37% 73 Business Services 21 2.50% 248 10.14% 75 Auto Repair, Services, and Parking 3 0.36% 4 0.16% 78 Motion Pictures 2 0.24% 10 0.41% 79 Amusement & Recreation Services 3 0.36% 25 1.02% 80 Health Services 3 0.36% 51 2.09% 87 Engineering & Management Services 6 0.72% 33 1.35% 99 Nonclassifiable Establishments 4 0.48% 10 0.41% Total 839 100% 2,446 100% a Excluding Sample Firms; data are for 2002, the last year for which Execucomp has complete data. 25

Table 4 Regression of Share Price on Book Value of Equity, EPS before Stock-based Employee Compensation Expense, Unrecognized SFAS No. 123 Expense, Recognized Stock-based Employee Compensation Expense, and Analyst Earnings Growth Forecast PRICE it = α 0 + α 1 BV it + α 2 EPS it + α 3 UNRECCX it + α 4 GROWTH it + ε it PRICE it = α 0 + α 1 BV it + α 2 EPSBCX it + α 3 UNRECCX it + α 4 RECCX it + α 5 GROWTH it + ε it Panel A: Descriptive Statistics Variables Standard Lower Upper n mean Deviation Min Quartile Median Quartile Max PRICE 635 30.257 18.061 0.320 18.050 28.880 38.540 129.972 ΒV 635 16.753 11.275-16.963 8.878 15.256 21.853 51.945 EPS 635 0.528 0.694-3.209 0.206 0.457 0.815 4.125 EPSBCX 635 0.528 0.694-3.198 0.206 0.457 0.815 4.139 UNRECCX 635 0.015 0.020-0.063 0.002 0.010 0.021 0.095 RECCX 635 0.010 0.015-0.009 0.001 0.005 0.013 0.061 GROWTH(%) 635 11.102 4.501-1.121 8.200 10.930 13.330 23.889 Panel B: Regression Results - Dependent Variable is Stock Price Intercept ΒV EPS UNRECCX GROWTH Adj. R 2 Prediction (+/-) (+) (+) (-) (+) Model (1) 7.955 0.764 8.507 135.497 0.277 56.88% (5.31) (15.50) (10.92) (5.45) (2.60) Intercept ΒV EPSBCX UNRECCX RECCX GROWTH Adj. R 2 Prediction (+/-) (+) (+) (-) (-) (+) Model (2) 7.831 0.756 8.404 133.486 38.201 0.264 56.99% (5.22) (15.25) (10.74) (5.37) (1.18) (2.47) Significant at the 1 percent level, based on two-tailed tests (t-statistics is parentheses). The sample consists of 635 observations. All variables are winsorized at two standard deviations to alleviate the influence of outliers. Variable definitions: PRICE is share price at the end of the fiscal quarter; BV is book value of equity per share; EPS is earnings per share before extraordinary items and discontinued operations; EPSBCX is earnings per-share before extraordinary items and discontinued operations and any recognized stock-based compensation expense; RECCX is per-share recognized stock-based employee compensation expense, net of tax; UNRECCX is per-share unrecognized stock-based compensation expense, net of tax, (equal to the difference between stock-based employee compensation expense determined under SFAS No. 123 fair value method and recognized stock-based compensation expense); GROWTH is quarter-end I/B/E/S mean analyst earnings growth forecast; and i and t denote firm and quarter. 26

TABLE 5 Regression of Size-adjusted Cumulative Abnormal Returns on Change in EPS before Stock-based Employee Compensation Expense, Change in Unrecognized SFAS No. 123 Expense, Change in Recognized Stock-based Employee Compensation Expense, Logged Value of Assets, and Change in Analyst Earnings Growth Forecast CAR it = α 0 + α 1 EPSBCX it + α 2 UNRECCX it + α 3 RECCX it + α 4 LOGASST it + α 5 GROWTH it + ε it Panel A: Descriptive Statistics Variables Standard Lower Upper n mean Deviation Min Quartile median Quartile Max CAR 839-0.015 0.134-0.362-0.095-0.026 0.054 0.343 EPSBCX 839 0.005 0.082-0.496-0.003 0.002 0.009 0.516 UNRECCX 839-0.119 0.672-5.747-0.044-0.010 0 5.504 RECCX 839 0.013 0.625-6.040 0 0.005 0.023 5.793 LOGASST 839 8.255 2.179 3.640 6.935 8.278 9.529 12.831 GROWTH 513-0.010 0.025-0.100-0.018-0.006 0.001 0.083 Panel B: Regression Results - Dependent Variable is Size-adjusted Returns Intercept EPSBCX UNRECCX RECCX LOGASST GROWTH Adj. R 2 Prediction (+/-) (+) (-) (-) (+/-) (+) Model (1) -0.019 0.229-0.020-0.014 3.18% (-4.01) (4.12) (-2.89) (-1.92) * Model (2) -0.056 0.233-0.021-0.014 0.005 3.60% (-3.12) (4.19) (-3.08) (-1.90) * (2.16) ** Model (3) -0.013 0.306-0.021-0.024-0.019 6.59% (-2.23) ** (4.12) (-2.46) (-2.36) ** (-0.09) Model (4) 0.001 0.308-0.021-0.023-0.001-0.011 6.45% (0.03) (4.13) (-2.42) ** (-2.26) ** (-0.52) (-0.05) *, **, Significant at the 10 percent, 5 percent and 1 percent levels, respectively, based on two-tailed tests (t-statistics is parentheses). The sample consists of 839 observations. All variable are winsorized at two standard deviations to alleviate the influence of outliers. Variable definitions: CAR is size-adjusted buy-and-hold stock returns for a 90-day period, staring from 45 days after the prior fiscal quarter-end; EPSBCX is the seasonal change in per-share net income before extraordinary items and discontinued operations and any recognized stock-based compensation expense, deflated by beginning period price; RECCX is the seasonal change in per-share stock-based employee compensation expense included in reported net earnings, net of tax, deflated by beginning period stock price; UNRECCX is the seasonal change in per-share unrecognized stock-based compensation expense, net of tax, deflated by beginning period stock price; LOGASST is natural logarithm of beginning period total assets; and GROWTH is seasonal change in quarter-end I/B/E/S mean analyst earnings growth forecast; and i and t denote firm and quarter. UNRECCX and UNRECCX are multiplied by 100 for ease of presentation. 27