The findings reported in Panel B of Table 7 indicate that ESOASSET is value-relevant.
If its value-relevance is attributable to capturing the growth potential in future cash flows arising from contributed intellectual capital, then controlling for an alternative measure of growth in the valuation equation should reduce the value-relevance of the ESO intangible asset (see Barth et al. 2001, 96-97). If ESOASSET retains explanatory power after con-trolling for the alternative growth measure, then it reflects value-possibly arising from contributed intellectual capital-beyond that reflected in the additional growth measure.
We re-estimate the valuation Equation (4b) after including an indicator variable equal to 1 (0) if sales growth in the most recent year is above (below) the median for our sample firms in that year. Panel B in Table 8 indicates that investors significantly value the growth proxy. In addition, its inclusion reduces the incremental coefficient on ESOASSET to in-significant levels in 1997 and 1998, which suggests that ESOASSET reflects future sales growth potential. However, the total coefficient on ESOASSET (a3 + a4), remains positive and significant except in 1997, indicating ESOASSET does reflect dimensions of value beyond sales growth, perhaps growth in expected future cash flows arising from employees' contributed intellectual capital.
Elimination of Price Endogeneity Bias
Aboody (1996) notes that because ESO values increase with prices of underlying stocks, regressing stock prices on ESO values creates an endogeneity problem. We use two different approaches to ensure that our finding that investors positively value ESOASSET is not merely an artifact of this endogeneity bias. First, we replace ESOASSET with an instrument that is uncontaminated by the firm's stock price-the number of unvested em-ployee stock options (UNVESTED_ESO). We determine UNVESTED_ESO's value-relevance using its incremental coefficient, because the number of unvested employee stock options cannot be a component of equity book value. Second, we adopt a two-stage instru-mental variable approach following Aboody et al. (2001) and Chamberlain and Hseih (1999). The two-stage procedure purges ESO expense of endogeneity bias by regressing ESO expense on various disclosed option-pricing model inputs, the vesting period, and the number of options granted, and then replacing ESO expense with the fitted value from the first-stage regression in their second-stage valuation regression. We follow this same pro-cedure by replacing ESOEXP with its fitted (or predicted) value ESOEXP_PRD and ESOASSET with ESOASSET_PRD, which we estimate in a similar way to ESOEXP_
PRD.p6 We include AEARN instead of AEARNESO and BVE instead of BVE_ADJ be-cause we do not adjust abnormal earnings and equity book value for the predicted ESO expense (which is, of course, not a disclosed amount). Thus, we determine value-relevance for ESOEXP_PRD and ESOASSET_PRD using their incremental coefficients, a2 and a4. Table 9 reports the results of estimating the valuation Equation (4b) regressions incorporating the two approaches to eliminate endogeneity bias. These results reveal that we cannot attribute the positive valuation of ESOASSET to endogeneity bias. In par-ticular, the incremental coefficients on the number of unvested employee stock options
16 The first-stage regression includes an estimate of the firm's stock price volatility, expected option life, the risk-free rate, dividend yield, and the number of unvested options outstanding. Aboody et al. (2001) use the number of options granted instead of the number of unvested options outstanding. Unreported findings reveal that using
the number of options granted does not affect our inferences.
t~~~~~~~~~~~~~~~~~~~~0 o~ 2 M P 4S 9 O gt
Z2
2 WS0 0
CU it ' 2|z 2 'i
.5 § 21<N i ZSCfi t1<N-N Nm215'tt Wi
L nlgl3i
:1fiE|°°
8992
(UNVESTED_ESO) and the predicted ESO asset (ESOASSET_PRD) are significantly pos-itive.'" Notably, in contrast to Aboody et al. (2001) and Chamberlain and Hseih (1999), who find a negative relation between equity market value and predicted ESO expense, Panel B reveals that the valuation coefficient on the ESO expense variable (ESOEXP_PRD), a2,
remains significantly positive when the ESO asset variable (ESOASSET_PRD) is omitted from the estimating equation. However, when ESOASSET_PRD is included in the esti-mating equation, the ESOEXP_PRD coefficient (a2) is insignificantly different from zero.
Differences between our findings and those of Aboody et al. (2001) and Chamberlain and Hseih (1999) are most likely attributable to sample differences. In particular, sample firms in the other two studies represent a broader set of industries and are not necessarily profitable. Employee stock options may very well create a valuable intangible asset for profitable firms in knowledge-intensive industries, but not for firms in other industries. For example, Keating et al. (2002) estimate valuation models for Intemet firms in the period surrounding the Intemet downtum in the first quarter of 2000. They find that their measure of the fair value of employee stock options granted in 1999 is significantly positively associated with equity value, consistent with our finding that the ESO asset is positively asso-ciated with share prices. However, their estimate of employee stock options granted in 1999 is neither a book value variable nor an eamings variable but rather a value-relevant variable outside the accounting system.
ESO Valuation vs. SG&A Valuation
The evidence in Panel B of Table 7 that investors value ESO expenses (ESOEXP) positively even when the valuation model includes a measure of the intangible asset (ESOASSET) raises the question of whether other compensation creates intangible value for the firm. Ideally, we would test this proposition by estimating the valuation of compen-sation expense. However, because we do not have access to compencompen-sation expense data, we instead use selling, general, and administrative expense (SG&A), which is a noisy proxy for compensation expense. First, it excludes salaries embedded in the cost of goods sold.
Second, SG&A includes costs that are not entirely compensation, such as advertising ex-pense and research and development exex-pense, each of which reflects priced but unrecorded intangible assets related to innovative activities (Landsman and Shapiro 1995; Aboody and Lev 1998; Hall 1999; H-lall et al. 1999). The effect of these noncompensation costs should bias toward finding a positive valuation effect for SG&A.
To test the proposition that investors value SG&A differently from other components of AEARNESO, we estimate Equation (4b) including SG&A as a separate regressor, which permits SG&A to have a pricing effect different from AEARNESO. Untabulated findings indicate that the incremental SG&A coefficient is not significantly different from zero, indicating that it is not valued differently from other expense components of income. In addition, including SG&A does not affect inferences relating to ESOEXP and ESOASSET.
17 Finding a positive pricing coefficient on the number of unvested options does not imply that firms can increase their stock prices simply by issuing stock options. Aboody (1996) and our own unreported analyses from a regression in which we substitute the number of vested options (VESTED.ESO) in place of ESOASSET in Equation (4b) yield a negative coefficient on VESTED_ESO. This is consistent with the value to the firm of the intangible capital provided by the vested options being smaller than their diluting effect on ownership value for present common shareholders.
Most significantly, in contrast to SG&A, ESOEXP still has a positive incremental associ-ation with equity value.'8 Thus, our results suggest that for profitable software finns, in-vestors value ESO expense differently from other components of income, and differently from other forms of compensation as reflected in SG&A.
Alternative Definition of Computer Software Firms
A final robustness test estimates Equations (4a) and (4b) using a narrower definition of profitable computer software firms. In their paper on the value-relevance of software cap-italization, Aboody and Lev (1998) include firms from SIC codes 7370 to 7372. Limiting our sample to these three codes could produce a more powerful test of whether investors value ESOASSET as an intangible asset-if intellectual capital is a larger component of these "purer" software firms-although this restriction reduces our annual sample to 66 finns. Untabulated results are similar to those reported for the full sample of 85 firms.