The value-relevance of R&D and advertising expenditures: Evidence from Korea

Transcription

1 The International Journal of Accounting 39 (2004) The value-relevance of R&D and advertising expenditures: Evidence from Korea Bong H. Han a, David Manry b, * a College of Business Administration, Ajou University, Suwon , South Korea b College of Business, University of New Orleans, 2000 Lakeshore Drive, New Orleans, LA , USA Abstract This study examines the value-relevance of R&D and advertising expenditures of Korean firms, using a regression model based on the Ohlson [Contemp. Account. Res. (1995) 661] equityvaluation framework. Results indicate that R&D expenditures are positively associated with stock price, suggesting that capitalizing R&D expenditures is appropriate. The association is stronger for the portion of R&D expenditures that is capitalized, rather than expensed, suggesting that investors agree with management that the capitalized expenditures represent greater future economic benefits. Investors also appear to interpret fully expensed R&D expenditures as positive net present-value investments, however, suggesting that these expenditure should also be capitalized. Additional results indicate that advertising expenditures are negatively associated with stock price, and the magnitude of this negative association is similar to the association between other expenses and stock price. These findings suggest that investors believe the economic benefits of advertising expenditures expire in the current period, similar to other expenses. D 2004 Published by University of Illinois. All rights reserved. Keywords: R&D expenditures; Advertising expenditures; Value-relevance 1. Introduction In the current environment of rapid scientific advancement and intensified global competition, many firms in Korea are increasing their R&D investments and launching new advertising projects. Whether R&D and advertising expenditures represent expected future economic benefit (are value-relevant ) is an important accounting issue in both * Corresponding author. Tel.: ; fax: addresses: bhhan@ajou.ac.kr (B.H. Han), dmanry@uno.edu (D. Manry) /$30.00 D 2004 Published by University of Illinois. All rights reserved. doi: /j.intacc

2 156 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) academia and practice. If expenditures provide future economic benefits, then capitalizing them is desirable; otherwise, expensing in the current period is preferred. This study investigates the value-relevance of R&D and advertising expenditures of Korean firms, using an empirical model derived from the Ohlson (1995) valuation framework. R&D expenditures may be capitalized in Korea, whereas they are fully expensed in the United States. Studies using U.S. data provide evidence generally consistent with the value-relevance of R&D expenditures (Bublitz & Ettredge, 1989; Hirschey & Weygandt, 1985; Lev & Sougiannis, 1996). Using Korean data, Choi (1994) shows that capitalized R&D expenditures are value-relevant, but expensed R&D expenditures are not. On the other hand, evidence provided by Cho and Chung (2001) supports the value-relevance of both capitalized and expensed R&D expenditures. Advertising expenditures are expensed universally. Findings of previous studies, regarding the value-relevance of advertising expenditures are mixed. Hirschey and Weygandt (1985) provide evidence supporting the value-relevance of advertising expenditures, whereas Bublitz and Ettredge (1989) and Choi (1994) report evidence to the contrary. The research methods of previous studies either associate R&D and/or advertising expenditures with future earnings (Cho & Chung, 2001; Lev & Sougiannis, 1996) or relate them to firm market value (Aboody & Lev, 1998; Bublitz & Ettredge, 1989; Choi, 1994; Hirschey & Weygandt, 1985). This study adopts the latter approach; however, unlike previous research, we base our empirical tests on a rigorously derived theoretical model, the Ohlson (1995) valuation framework. Ohlson expresses firm value as a linear function of accounting earnings, the book value of net assets, and other information. By fitting R&D and advertising expenditures into this framework, we find that R&D expenditures are positively associated with stock price, indicating that market participants view the expenditures as providing future economic benefits. This association is stronger for capitalized R&D than for the expensed portion, however, to the extent that capitalized R&D is judged to be a positive net present-value investment. Advertising expenditures are negatively associated with stock price, suggesting that they are regarded as expenses. To control for the possibility of a self-selection bias, we partition our sample into two groups based on whether firms fully or partially expense R&D expenditures. We find that the expensed R&D of firms that at least partially capitalize has no significant price impact. When firms fully expense their R&D expenditures, however, their R&D expense is significantly associated with price. Because the association between price and fully expensed R&D expenditures is greater than the association between price and net book value, the market appears to regard fully expensed R&D as a positive net present-value investment. The significant difference we find in the value-relevance of expensed R&D between our two subsamples may provide an explanation for the conflicting findings of previous research on Korean R&D expenditures; previous research does not control for a possible self-selection bias. Advertising expenditures are negatively associated with stock price. Because the magnitude of this negative association closely resembles the association between other expenses and stock price, these results are consistent with a market belief that the economic benefits of advertising expenditures expire in the current period, similar to other expenses.

3 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) The financial press has been critical of various Korean accounting practices, including the capitalization of amounts that do not represent future benefits (Ehrlich & Mann, 1998). R&D investments and advertising activities are essential for securing a firm s technological superiority and enhancing its brand values. Our findings on the information content of R&D and advertising expenditures may prove useful in future deliberations about, on modifying current accounting standards that govern the extent to which these expenditures should be capitalized. The next section discusses previous research and the Korean accounting standards that govern R&D expenditures. Section 3 covers the model, hypotheses, sample selection, and data. Details of tests and results are included in Section 4, and the final section provides a summary and conclusions. 2. Background 2.1. Accounting standards for R&D Press reports have criticized the amount and transparency of Korean accounting disclosures (Condon, 1998; Webb, 1998). Recently, the Korean government has encouraged Korean firms to adopt Western accounting standards, which it feels will help the country avoid problems like the recent International Monetary Fund (IMF) crisis (The Economist, 1999). The IMF criticized Korean accounting and auditing practices and has required reforms (Ehrlich & Mann, 1998; Macrae, 1998). Difficulties cited by the IMF include disclosure lapses, such as allowing the omission of key liabilities, and permitting the capitalization of amounts that do not represent future economic benefits. Under the previous Korean accounting standard for R&D, promulgated in 1987, R&D expenditures could be capitalized when the expenditures were individually allocable to particular products or technologies, and reasonably expected future economic benefits were sufficient to cover the expenditures. If these conditions were not met, R&D expenditures were to be expensed. Under a new standard, effective in 1999, R&D expenditures are partitioned into an expensed research component and a capitalized development component. Development costs are expenditures made in the development stage that are individually allocable to new products or new technologies (including software), after technological feasibility is proven, and from which future economic benefits are probable. The new standard thus reduces the portion of R&D expenditures that can be capitalized, by requiring all expenditures incurred in the research stage to be expensed. 1 While the new capitalization rule moves Korean accounting closer to Western standards, there is still significant flexibility in the capitalization of development costs. 1 Although considerable subjectivity over capitalization arguably exists under both old and new rules, the capitalized portion of R&D expenditures dramatically decreased when the new standard was implemented. In 1998, the year before the new standard became effective, the capitalized portion was 0.90% of sales on average; this portion decreased to 0.25% of sales in The expensed portion of R&D expenditures increased from 0.53% of sales in 1998 to 1.01% in 1999.

4 158 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) Because the new standard has only recently become effective, R&D data reflecting the new standard are not sufficient to provide robust evidence. Thus, data from under the previous standard ( ) are used in this paper. Regardless of differences between the new and previous standards concerning capitalizing or expensing R&D expenditures, the sum of capitalized and expensed amounts is the same under both rules. For capitalization, the condition that future economic benefits are expected is equally applicable before and after the new standard. Results would thus be similar if data under the new standard were used instead. 2 U.S. GAAP requires the full expensing of R&D expenditures. It allows the capitalization of software development costs, but only after the technological feasibility of new products is proven. Advertising expenditures are expensed under both Korean and U.S. GAAP Prior studies Aboody and Lev (1998), Bublitz and Ettredge (1989), Choi (1994), and Hirschey and Weygandt (1985) infer the value-relevance of R&D and/or advertising expenditures from their associations with market value proxies, such as Tobin s q or stock returns. On the other hand, Cho and Chung (2001) and Lev and Sougiannis (1996) investigate valuerelevance by examining its contribution to the generation of future earnings. Hirschey and Weygandt (1985) provide evidence that R&D and advertising expenditures are positively associated with Tobin s q. This indicates that these expenditures are priced by the market, and thus merit capitalization as intangible assets. Choi (1994) replicates Hirschey and Weygandt (1985) using Korean data, but reports that only the portion of R&D expenditures that is capitalized under Korean GAAP is positively associated with Tobin s q. Bublitz and Ettredge (1989) examine the association of unexpected R&D and advertising expenditures with stock returns, after controlling for unexpected revenues and other ordinary expenses. They find that unexpected R&D expenditures are not negatively associated with stock returns, but unexpected advertising expenditures are more negatively associated with stock returns than are other unexpected ordinary expenses. They interpret these results as suggesting that R&D expenditures are valued as assets, but advertising expenditures are recognized as expenses. Although U.S. GAAP requires the full expensing of R&D expenditures, it allows the capitalization of software development costs when the technological feasibility of new products can be proven. Aboody and Lev (1998) find that changes in capitalized softwaredevelopment costs are positively related to stock returns, unlike changes in expensed software-development costs. This supports the position of U.S. GAAP on the capitalization of software-development costs. Lev and Sougiannis (1996) estimate, by industry, the contribution provided by R&D expenditures to the generation of future earnings. Their results suggest that for each 2 The value-relevance of the capitalized portion of R&D expenditures could be found to be higher under the new standard. This would occur if future economic benefits are found to be more highly assured when the additional condition is imposed: that technological feasibility should be proven in the development stage prior to capitalization.

5 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) inflation-adjusted dollar of R&D expenditures, earnings of US$ are generated over the following 5 9 years, including the year of the expenditures. They present further evidence that the difference between as-if earnings (obtained by capitalizing R&D expenditures) and reported earnings is positively associated with stock prices. Taken together, these results suggest that the market values R&D expenditures as assets, not as expenses, contrary to the treatment prescribed by U.S. GAAP. Cho and Chung (2001) replicate Lev and Sougiannis (1996) with Korean data. They show that 1 won (Korean currency) of R&D expenditures generates 1.25 won of earnings, on average, over the following 2 4 years, including the year of expenditures. A separate examination of the capitalized and expensed portions indicates that the capitalized portion has an effect on future earnings for a longer period than the expensed portion. 3 Aboody and Lev (1998), Choi (1994), Bublitz and Ettredge (1989), and Hirschey and Weygandt (1985) construct hypothesis tests based predominantly on intuition, rather than on valuation theory. For example, Hirschey and Weygandt and Choi do not include earnings in their models. Their findings may thus be influenced by a correlated omittedvariable problem. 4 In contrast, the empirical tests of this study are based on a model rigorously derived from the Ohlson (1995) valuation framework. Because Korean firms have considerable discretion over whether R&D expenditures may be capitalized or expensed, the possibility also exists that the results of previous studies on Korean data may have been affected by a self-selection bias. We control for this possible influence by partitioning our sample based on whether the firms capitalize R&D. We then examine the value-relevance of expensed R&D expenditures separately for the subsamples. 3. Research design 3.1. Regression model Ohlson (1995) derives an equity-valuation model under the assumptions that (1) the equity value of a firm is the present value of its expected future dividend stream; (2) the clean surplus relation is maintained for the change in book value of a firm s net assets; and (3) abnormal earnings exhibit a first-order autoregressive time-series process. The model is p t ¼ kðux t d t Þþð1 kþy t þ av t ð1þ where p t = market value of a firm s equity at date t, x t = earnings for the period ending at date t, d t = net dividends as of date t, y t = book value of net assets at date t, v t = other 3 Cho and Chung (2001) is based on a comparatively small sample, 212 firm years over 2 years, 1995 and In Choi (1994), the sample is 1453 firm years over 5 years from 1988 to In his regression, R&D and advertising expenditure variables are all scaled by sales, as in our sample. We find that the correlations between earnings (before deducting R&D, advertising expenditures, and extraordinary items) and expensed R&D and advertising expenditures are significantly positive (.23 and.40, respectively), although the correlation between earnings and capitalized R&D is not substantial (.05). Omitting earnings from the RHS of a regression of price on expensed R&D and advertising expenditures, for example, will thus bias the estimated coefficients in a positive direction.

6 160 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) information at date t, u=(1 + r f )/r f, where r f = the risk-free rate (thus u>1), and 0 V k V 1, a>0. 5 In Eq. (1), equity value is the weighted average of the book value of net assets ( y t ) and the earnings multiple (ux t ) less net dividends (d t ), adjusted for the effect of other information (v t ). Rewriting Eq. (1), p t þ d t ¼ð1 kþðy t rdcap t þ d t Þþð1 kþrdcap t þ kuðsale t oexp t rdexp t adexp t Þþav t ¼ð1 kþy* t þð1 kþrdcap t þ kusale t kuoexp t kurdexp t kuadexp t þ av t : ð2þ Above, capitalized R&D expenditures (rdcap t ) are separated from the book value of net assets before dividends ( y t + d t ). Thus, y* t is y t rdcap t +d t, or book value before dividends minus capitalized R&D. The earnings variable (x t ) is decomposed into sales (sale t ) minus R&D expenses (rdexp t ), advertising expenses (adexp t ), and other expenses (oexp t ), and R&D expenditures are in part capitalized (rdcap t ) and in part expensed (rdexp t ). 6 Furthermore, because rdexp t and adexp t are expense items, their impact on market price is denoted as negative in Eq. (2). However, if the market believes these expenditures have future economic benefits beyond what is reflected in financial statements, R&D and advertising expenditures may be positively related to price. Finally, although Eq. (2) depicts capitalized R&D expenditures (rdcap t ) as having a positive effect on price in the same magnitude as other assets ( y* t ), the market may value rdcap t more highly than other assets due to its potential to generate high levels of future economic benefits. Eq. (2) is the basis of the regression model to test the value-relevance of R&D and advertising expenditures. 7 Adding the firm subscript i, and allowing the intercept to vary yearly over the test period ( ) in the equation, the regression model is: P it þ D it ¼ X98 y¼88 d y YR it þ b 1 BV it * þ b 2 RDCAP it þ b 3 SALE it þ b 4 OEXP it þ b 5 RDEXPF it þ b 6 RDEXPC it þ b 7 ADEXP it þ e it ð3þ where P it = market value of common stock three months after the end of year t, 8 D it = cash dividends in year t, YR it = year dummy (one for the test year and zero otherwise), 5 Refer to Ohlson (1995) for definitions of k and a. 6 Tests reveal that the amount of R&D expenditures capitalized and the amount subsequently amortized are highly correlated (Pearson correlation coefficient=.907). Because our research focus is on the amounts expensed or capitalized, and to avoid problems with multicollinearity, we do not include a variable for the amortization of capitalized R&D expenditures in our regressions. 7 Myers (1999) provides empirical evidence that Ohlson s (1995) linear models of information dynamics tend to understate firm value, and the correlation between market price and the model s implied value is chiefly attributable to the correlation between price and net book value. 8 The 3-month buffer is to allow sufficient time for the filing of year t financial statements; see Section 3.3.

7 BV* it =BV it (book value of net assets at the end of year t) RDCAP it + D it, RDCAP it = R&D expenditures capitalized in year t, SALE it = net sales in year t, OEXP it = other expenses in year t, RDEXPF it = R&D expenditures expensed for full expensers in year t, RDEXPC it = R&D expenditures expensed for capitalizers in year t, and ADEXP it = advertising expenditures in year t. Net dividends (d t in Eq. (2)) are replaced with cash dividends, as in Collins, Pincus, and Xie (1999). The precise nature of Ohlson s v t ( other information in Eqs. (1) and (2)) is unknown, and may be an omitted variable. However, if v t is both omitted and time dependent, then the yearly intercepts may proxy for the information. To estimate the coefficients on earnings component variables in Eq. (3), we begin by decomposing Net Income before Extraordinary Items into SALE it OEXP it RDEX- P it ADEXP it, as in Eq. (2). Following Aboody and Lev (1998), RDEXP it is then broken down into two variables. For firm years with R&D expenditures fully expensed, RDEXP- F it = RDEXP it, but RDEXPC it = 0. For firm years with R&D expenditures at least partially capitalized, RDEXPC it = RDEXP it, but RDEXPF it = 0. In this way, the coefficient on RDEXPF it will capture the market reaction to R&D expensing for full expensers, and the coefficient on RDEXPC it will represent the market reaction to R&D expensing for (at least partial) capitalizers. 9 All variables, including the yearly intercepts, are deflated by SALE it. This causes the variable SALE it on the RHS in Eq. (3) to become a unit vector, and the coefficient on the fourth term then becomes the common intercept Hypotheses B.H. Han, D. Manry / International Journal of Accounting 39 (2004) The coefficients on the Eq. (3) variables BV* it, SALE it, and OEXP it appear in Eq. (2) as b 1 ¼ 1 k; b 3 ¼ ku; and b 4 ¼ ðkuþ: Because 0 V k V 1 and u>1, it is anticipated that 0Vb 1 V1; b 3 z0; and b 4 V0: Similarly, the coefficient on RDCAP it (b 2 )is1 k, and those on RDEXP it, RDEXPC it, and ADEXP it (b 5, b 6, and b 7 ) are all (ku). Thus, 0Vb 2 ¼ b 1 V1; and b 5 ; b 6 ; b 7 V0: We begin by testing whether the coefficient on RDCAP it is consistent with the above reasoning. That is, if the expenditures represented by RDCAP it are expected to provide 9 The number of full expensers, full capitalizers, and partial capitalizers of R&D are 1511 (47%), 305 (10%), and 1375 (43%), respectively. We combine the subsamples of full and partial capitalizers prior to conducting statistical tests because full capitalizers are comparatively few. 10 Barth and Kallapur (1996) report that including a scale proxy as an independent variable and reporting White s (1980) t statistics is more effective than deflation for addressing problems related to scale. In contrast, Easton (1998) finds that regressions of market value on firm characteristics may lead to coefficient estimates that capture only scale effects. Our results are qualitatively similar when tests are repeated using SALES it as an independent variable and deflating by the number of shares outstanding. These results are available from the authors.

8 162 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) future economic benefit, its coefficient, b 2, should be greater than 0. Our first (alternative) hypothesis is thus H 1 : b 2 >0. The other R&D expenditure variables, RDEXPF it, RDEXPC it, and ADEXP it, are expense items. Based on Eq. (2), their coefficients should approximate the coefficient on OEXP it,orb 5, b 6, b 7 V 0. Market participants may, however, believe that these R&D and advertising expenditures represent future economic benefits beyond what is implied by their financial statement presentation. To test for this, alternative hypotheses concerning the coefficients on RDEXPF it, RDEXPC it, and ADEXP it (b 5, b 6, and b 7 )in Eq. (3) are: H 2 : b 5 >b 4, H 3 : b 6 >b 4, and H 4 : b 7 >b 4. The above hypotheses suggest that if the economic benefits from RDEXPF it, RDEXP- C it, and ADEXP it do not expire entirely in the current period, their coefficients will be higher than the coefficient on OEXP it, b 4. Stronger forms of the above four hypotheses are: H 1s : b 2 >b 1, H 2s : b 5 >b 1, H 3s : b 6 >b 1, and H 4s : b 7 >b 1. In H 1s, we are testing whether RDCAP it represents future economic benefits beyond what is implied by its financial-statement presentation. In Eq. (2), the theoretical coefficients on y* t (net book value) and rdcap t (capitalized R&D) are each 1 k. The coefficient on BV* it, b 1, thus provides an estimate for 1 k. We compare the coefficient on RDCAP it, b 2,tob 1 to test whether RDCAP it is viewed by the market as providing future economic benefits beyond this theoretical value. Alternatively, H 1s tests whether the market expects greater future benefit from a dollar of capitalized R&D than from a dollar of ordinary net assets. Moreover, if one considers the cross-sectional estimate of the coefficient on BV* it as the market s estimate of a normal return, the four stronger hypotheses above test whether the market regards each expenditure (RDCAP it, RDEXPF it, RDEXPC it, and ADEXP it ) as a positive net present-value investment, regardless of its classification into asset or expense under GAAP. Under Korean GAAP, the condition that future economic benefits are reasonably expected should be met for the capitalized portion of R&D expenditures; otherwise, these

9 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) Table 1 Derivation of research sample of 3191 firm-year observations, Listed on Korean Stock Exchange ( ) Necessary accounting and stock price data available for December fiscal year-end firms Book value of net assets positive for years t and t 1 Stock return data available for years t 1, t 2, and t 3 R&D expenditures greater than 0 a 625 firms 11 years = 6875 firm years firm years a 4609 firm years 4118 firm years 3790 firm years 3191 firm years (sample) expenditures should be expensed. We thus expect that the value-relevance of RDCAP it is higher than that of RDEXPF it and RDEXPC it,or H 5 : b 2 >b 5, and H 6 : b 2 >b 6. Together, H 1s H 4s, H 5, and H 6 enable comparisons among the market valuations of ordinary assets (BV* it ) and R&D expenditures Sample selection and variable measurements The sample consists of firms listed on the Korean Stock Exchange from 1988 to 1998, a period during which firms were allowed to capitalize R&D expenditures when future economic benefits were reasonably expected from the expenditures. Accounting data are retrieved from the Korea Investors Service database, and stock price and returns data are from the Korea Securities Research Institute database. To be included in the sample, necessary accounting and market data must be available. In addition, the sample is confined to firms with December fiscal year-ends, with positive net asset book values for the sample and previous years, and with R&D expenditures in the sample year. The resulting sample has 3191 firm years from 1988 to 1998, inclusive. The sample selection procedure is summarized in Table 1. In Eq. (3), the market value of common stock ( P it ) is as of the end of March in year t + 1. This allows a 3-month filing period for year t financial statements, to ensure that market value is measured after the release of the information. Other expenses (OEXP it ) are all expenses typically found in earnings before extraordinary items, including amortization of capitalized R&D, but excluding R&D expense and advertising expense. 4. Results 4.1. Descriptive statistics Table 2 reports the means and medians of the main independent variables for the sample. The number of observations is the smallest (151) in 1988 and the greatest (352) in

10 164 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) Table 2 Descriptive statistics of variables: yearly means/medians, 3191 firm years, (A) Mean/median of sample firm attributes a Year n Total assets (Wg billion b ) Book value (Wg billion) Market-to-book ratio DE ratio c Capitalization intensity /103 70/ / / / /102 82/ / / / /105 99/ / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / /0.000 (B) R&D and advertising expenses as percentages of sales Year n SALE it (Wg billion) d % OEXP it % RDCAP it % RDEXP it % ADEXP it / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / /0.15 Variable definitions: DE ratio = the debt-to-equity ratio; SALE it = net sales for firm i in year t; OEXP it = expenses other than R&D, and advertising, used to determine earnings before extraordinary items from net sales; RDCAP it = capitalized R&D; RDEXP it = R&D expenses; ADEXP it = advertising expenses; and capitalization intensity = R&D capitalized/r&d expenditures = R&D capitalized/(r&d capitalized + R&D expensed). a Total assets, book value of net assets, and market-to-book ratio are as of the end of each year. b US$1 is approximately 1200 won (Wg 1200). c There were four observations with DE>500; two in 1992, one in 1995 and one in Without those extreme DE observations, the mean DEs are in 1992, in 1995, and in d % indicates the percentage value from dividing the yearly mean/median of the variable by the corresponding yearly mean/median of SALE. Thus, each mean/median %variable, if multiplied by the corresponding mean/ median of SALE, will become the yearly mean/median of the variable Total assets, book value, and SALE it exhibit distinctly increasing trends. The means of these variables are much larger than the medians, because there are a few large firms in the right tails of the variable distributions. Similarly, the yearly DE ratio means are consistently larger than the medians. The market-to-book ratio means (and most medians) are consistently greater than 1 until 1997 and 1998, during the Korean IMF financial crisis. Capitalization intensity ratios are calculated as the proportion of total R&D

11 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) Table 3 Pearson product correlations (N = 3099) BV it * RDCAP it OEXP it RDEXPF it RDEXPC it ADEXP it P it + D it 0.609*** 0.196*** 0.259*** 0.092*** 0.057** BV it * *** 0.117*** 0.058** RDCAP it 0.048** 0.176*** 0.121*** OEXP it 0.150*** 0.145*** 0.402*** RDEXPF it 0.156*** 0.093*** RDEXPC it 0.132*** All variables are deflated by SALE it. Variable definitions: P it = market value of firm i 3 months after the end of year t; D it = cash dividend in year t; BV it *=BV it (book value of net assets at the end of year t) RDCAP it + D it ; RDCAP it = capitalized R&D; OEXP it = all expenses (including cost of goods sold and income tax expenses) other than R&D expense, advertising expense, and extraordinary gains or losses in year t; RDEXPF it = R&D expenditures expensed by firms that fully expense in year t; RDEXPC it = R&D expenditures expensed by firms that capitalize in year t; and ADEXP it = advertising expenses. Correlations are calculated with the same data used for regression estimates of Eq. (3). Extreme values are eliminated from the Eq. (3) data if the absolute value of RSTUDENT is greater than three or the absolute value of DFFITS is greater than 3( p/n) 1/2, where p and n are the number of parameters and observations, respectively (Belsley et al., 1980). * Significant at.05 (two-tailed). ** Significant at.01 (two-tailed). *** Significant at.001 (two-tailed). expenditures capitalized. These ratios also exhibit means consistently greater than their medians, signifying a skewed distribution. OEXP it, RDCAP it, RDEXP it, and ADEXP it are each expressed in the table as a percentage relative to SALE it. The mean of OEXP it ranges from approximately 95% to 98% of SALE it over the period, with a median of similar magnitude. R&D expenditures (RDCAP it and RDEXP it ) increase for the most part until 1996, but then decrease slightly in 1997 and 1998 during the financial crisis. The mean of RDCAP it is 0.30% in 1988, but increases to 1.10% of SALE it in The mean of RDEXP it increases from 0.18% in 1988 to 0.69% of SALE it in ADEXP it also shows an increasing trend until 1996, the year before the IMF financial crisis, but the rate of increase is not as distinct as for RDEXP it and RDCAP it. The means of RDCAP it, RDEXP it, and ADEXP it are all considerably larger than the medians, indicating that their distributions are highly skewed. In particular, the medians of RDCAP it are 0 in most of the years. 11 Pearson product correlations among our research variables are reported in Table 3. Outliers are identified from the regression of Eq. (3), using the RSTUDENT and DFFITS diagnostics discussed in Belsley, Kuh, and Welsch (1980), and are eliminated for robustness. 12 This process reduces the number of observations to a sample size of To be included in the sample, R&D expenditures should be positive. Because R&D expenditures are the sum of RDCAP and RDEXP, RDCAP is included if RDEXP is positive. 12 Observations are identified as outliers in regressions if the absolute value of RSTUDENT is greater than 3 or the absolute value of DFFITS is greater than 3( p/n) 1/2, where p and n are the number of parameters and observations, respectively.

12 166 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) firm years. All of the independent variables (all scaled by SALE it ) are significantly correlated individually with P it + D it (also scaled by SALE it ), except for ADEXP it. RDEXPF it, RDEXPC it, and ADEXP it are negatively correlated with OEXP it (all P values <.001). RDCAP it is positively correlated with OEXP it ( P <.01). Although not tabulated here, the correlation between earnings (before deducting R&D, advertising expenditures, and extraordinary items) and RDCAP it is negative ( 0.048, P <.01), but the correlations between earnings and RDEXPF it and RDEXPC it are substantially positive (.150 and.145, respectively, both P values <.001). Together, these correlations suggest that firms tend to capitalize R&D when earnings are low (and other expenses are high), but expense R&D when earnings are high. While many of the independent variables are significantly correlated, diagnostics suggest that multicollinearity is not a problem in our subsequent regressions Main results Table 4 presents pooled regression results. The significance of each regression coefficient is tested with a White s (1980) t statistic, against the null that the coefficient value is 0. Panel A of Table 4 reports coefficient estimates and test results. As stated in Section 3.2, the theoretical values of the coefficients on BV* it, and OEXP it are b 1 =1 k, and b 4 = (ku). Thus, it is predicted that 0 V b 1 V 1, and b 4 V 0. The alternative hypotheses about the coefficient on RDCAP it (b 2 ) and the coefficients on RDEXP it, RDEXPC it, and ADEXP it (b 5, b 6, and b 7 )areb 2 >0 and b 5, b 6,andb 7 >b 4. The stronger form of the above alternative hypotheses are b 2 >b 1 and b 5, b 6, and b 7 >b 1. Although our alternative hypotheses imply one-tailed P values, we report two-tailed P values to be more informative. If one assumes that the estimated coefficient on BV* it (0.663, P <.001) approximates its theoretical value of 1 k, then k=.337. If the coefficient on OEXP it ( 1.026, P <.001) may also be regarded as reasonably close to its theoretical value of (ku), then we can infer a value for the earnings multiple (u) of about 3. The coefficients on RDCAP it and RDEXPF it (7.264 and 4.240) are significantly greater than 0 ( P <.001), suggesting that the market views capitalized and fully expensed R&D expenditures as providing future economic benefits, like assets. Furthermore, Panel B of Table 4 shows that the coefficient on RDCAP it is significantly greater than that on BV* it ( P <.001), indicating that the market values capitalized R&D expenditures more highly than other assets, that is, as a positive net present value investment. These results provide support for H 1 and H 1s. Because the coefficient on RDEXPF it is significantly positive, and the coefficient on OEXP it is significantly negative, H 2 is also supported; Panel B of Table 4 shows that the difference between these two coefficients is significant ( P <.001). Moreover, H 2s is also supported; the coefficient on RDEXPF it is significantly greater than the coefficient on net book value, BV* it. As discussed in Section 3.2, this result is consistent with fully expensed R&D representing a positive net present-value investment, and argues in favor of (at least partial) capitalization. 13 Variance Inflation Factors (VIF; Belsley et al. 1980) for all independent variables are less than 1.4. Kennedy (1992) suggests a VIF greater than 10 indicates harmful multicollinearity.

13 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) Table 4 Coefficient estimates from regressions of with-dividend price on R&D-related variables, advertising expenses, and other control variables; all variables deflated by sales Model: P it þ D it ¼ X98 d y YR it þ b 1 BV* it þ b 2 RDCAP it þ b 3 SALE it þ b 4 OEXP it þ b 5 RDEXPF it þ y¼88 b 6 RDEXPC it þ b 7 ADEXP it þ e it (A) Coefficient estimates and test statistics a Variable Coefficient White s t statistic BV* it *** RDCAP it *** OEXP it *** RDEXPF it *** RDEXPC it ADEXP it *** Adjusted R N 3099 (B) Coefficient difference estimates and test statistics b Coefficient difference estimate White s t statistic RDCAP it BV* it *** RDEXPF it OEXP it *** RDEXPC it OEXP it * ADEXP it OEXP it * RDCAP it RDEXPF it ** RDCAP it RDEXPC it *** RDEXPF it BV* it *** RDEXPC it BV* it ADEXP it BV* it *** All variables, including the yearly intercepts, are deflated by SALE it. Thus, the SALE it variable in Eq. (3) becomes a unit vector, and its coefficient is a common intercept. Intercepts are not tabulated here for brevity. Variable definitions: P it = market value of firm i 3 months after the end of year t; D it = cash dividend in year t; YR it = a year dummy variable with 1 for the test year and 0 otherwise; BV* it =BV it (book value of net assets at the end of year t) RDCAP it + D it ; RDCAP it = capitalized R&D; SALE it = net sales for firm i in year t; OEXP it = all expenses (including cost of goods sold and income tax expense) other than R&D expense, advertising expense, and extraordinary gains or losses in year t; RDEXPF it = R&D expenditures expensed by full expensers in year t; RDEXPC it = R&D expenditures expensed by capitalizers in year t; and ADEXP it = advertising expenses. a Extreme values are eliminated if the absolute value of RSTUDENT is greater than 3 or the absolute value of DFFITS is greater than 3( p/n) 1/2, where p and n are the number of parameters and observations, respectively (Belsley et al., 1980). Regression results are based on 3099 firm years, after eliminating extreme values. White s (1980) t statistics test the null hypothesis that the coefficient value equals 0. b White s (1980) t statistics test the null hypothesis that the coefficient values are equal. * Significant at.05 (two-tailed). ** Significant at.01 (two-tailed). *** Significant at.001 (two-tailed). Panel B of Table 4 shows that the coefficient on RDEXPC it is significantly greater than the coefficient on OEXP it ( P <.05). This result supports H 3, suggesting that the economic benefits of expensed R&D do not appear to expire entirely in the period in which the

14 168 Table 5 Coefficient estimates (and White s t statistics) from yearly regressions of with-dividend price on R&D-related variables, advertising expenses, and other control variables; all variables deflated by sales (A) Coefficient estimates and test statistics Year BV* it RDCAP it OEXP it RDEXPF it RDEXPC it ADEXP it n Adjusted R *** 4.01*** * *** 8.69*** ** *** 3.85*** *** ** ** *** 4.78*** *** 5.01*** *** 4.08*** 5.09*** *** *** 3.68*** 4.50*** *** *** 4.32*** 4.02*** *** 3.66*** 4.85*** *** 3.51*** 3.84*** 2.13* Mean z *** 17.27*** 11.82*** 4.826*** *** z *** 8.506*** 3.693*** 5.768*** *** Total N = 3099 Mean adjusted R 2 =.582 B.H. Han, D. Manry / International Journal of Accounting 39 (2004)

15 (B) Mean coefficient difference estimates and test statistics RDCAP it BV* it RDEXPF it OEXP it RDEXPC it OEXP it Mean z *** 5.91*** 2.73** z2 8.04*** 5.70*** 4.33*** ADEXP it OEXP it RDCAP it RDEXPF it RDCAP it RDEXPC it Mean z1 3.24** 2.76** 5.86** z2 2.46* 2.51* 5.52*** RDEXPF it BV* it RDEXPC it BV* it ADEXP it BV* it Mean z1 4.04*** *** z2 4.88*** *** All variables, including the yearly intercepts, are deflated by SALE it. Thus, the SALE it variable in Eq. (3) becomes a unit vector, and its coefficient is a common intercept. Intercepts are not tabulated here for brevity. Variable definitions: P it = market value of firm i 3 months after the end of year t, with the effect of stock dividends and splits adjusted for; D it = cash dividend in year t; YR it = a year dummy variable with 1 for the test year and 0 otherwise; BV* it =BV it (book value of net assets at the end of year t) RDCAP it + D it ; RDCAP it = capitalized R&D; SALE it = net sales for firm i in year t; OEXP it = all expenses (including cost of goods sold and income tax expenses) other than R&D expense, advertising expense, and extraordinary gains or losses in year t; RDEXPF it = R&D expenditures expensed for full expensers in year t; RDEXPC it = R&D expenditures expensed for capitalizers in year t; and ADEXP it = advertising expenses. Extreme values are eliminated if the absolute value of RSTUDENT is greater than three or the absolute value of DFFITS is greater than 3( p/n) 1/2, where p and n are the number of parameters and observations, respectively (Belsley et al., 1980). Regression results are based on 3099 firm years, after eliminating extreme values. z1 and z2 statistics test the significance each coefficient across years against the null, that the coefficient equals 0 (Aboody and Lev, 1998). * Significant at.05 (two-tailed). ** Significant at.01 (two-tailed). *** Significant at.001 (two-tailed). B.H. Han, D. Manry / International Journal of Accounting 39 (2004)

16 170 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) expense occurs, unlike other expenses. H 3s, which hypothesizes that the coefficient on RDEXPC it is greater than the coefficient on BV* it, is a stronger test of the perceived future benefits of expensed R&D. This hypothesis is not supported; however, the t statistic is not significant. The evidence that R&D expense from partial capitalizers may represent future economic benefits as assets is thus somewhat weak. Panel B also shows that although the difference between the coefficients on ADEXP it and OEXP it is significant, the sign of the difference is incorrect. The coefficients on both variables are significantly less than 0 (both P values < 0.001) in Panel A. Thus, these results do not support either H 4 or H 4s, and are consistent with a belief by market participants that advertising expenditures do not represent future economic benefits, and thus expensing is warranted. Our last two hypotheses, H 5 and H 6, hypothesize that the coefficient on RDCAP it exceeds the coefficients on RDEXPF it and RDEXPC it, respectively. The Korean GAAP prohibits capitalization of R&D expenditures unless future economic benefits are reasonably expected; thus, the value-relevance of capitalized R&D is predicted to be greater than that of expensed R&D. t Statistics at the bottom of Panel B of Table 4 are significant for both tests of coefficient differences, providing support for both hypotheses ( P <.01 and P <.001, respectively). 14 Table 5 presents results from the yearly estimation of Eq. (3). White s t statistics are given for each yearly coefficient estimate, as well as z1 and z2 statistics (Aboody and Lev, 1998) to test coefficient significance across years. 15 The means of the yearly coefficients in Panel A of Table 5 are close to the pooled estimates reported in Panel A of Table 4, except for the coefficient on RDEXPC it ; however, the coefficient on this variable is not significant in the pooled regression, or in tests (z1 and z2 statistics) based on the yearly models. The yearly coefficient estimates on BV* it, RDCAP it, and OEXP it are significant in most years. While yearly coefficient estimates on ADEXP it and RDEXPF it are significant in 4 of 11 years and 1 of 11 years, respectively, the z1 and z2 statistics for these variables indicate strong significance across the 11 sample years ( P <.001). The mean yearly coefficient difference estimates in Panel B of Table 5 are essentially the same as the pooled difference estimates reported in Panel B of Table 4. In summary, the results in Tables 4 and 5 suggest the following. First, capitalized R&D expenditures appear to be not only priced positively but also valued more highly than other assets, and thus appear to be regarded by market participants as positive net present-value investments. Second, market participants appear to interpret expensed R&D expenditures as positive net present-value investments also, in cases where firms entirely expense their 14 It could be argued that our sample may be affected by a self-selection bias, because Korean firms have substantial discretion over whether they capitalize R&D. When we added the Capitalization-Intensity variable (CPRATIO, as defined in Table 2) to Eq. (3) as an independent variable to control for the self-selection bias, its coefficient was not significant, and the significance of the other explanatory variables was basically the same. Further tests reveal that CPRATIO is largely explained by RDCAP, RDEXP, SALE, and OEXP, so adding this variable to Eq. (3) is redundant. These data are available from the authors upon request. 15 The z1 statistic is ð1=t 1=2 Þ P T j¼1 t j=ðk j =ðk j 2ÞÞ 1=2, where t j is the t statistic for year j, k j is the degrees of freedom, and T is the number of years. The z2 statistic is [the mean t statistic/(standard deviation of t statistics/ (T 1) 1/2 )], where T is the number of years. z1 statistics assume residual independence, whereas z2 statistics account for cross-sectional and temporal residual correlations (Aboody & Lev, 1998; Barth, Clement, Foster, & Kaznik, 1998; White, 1980).

17 B.H. Han, D. Manry / International Journal of Accounting 39 (2004) R&D expenditures. Third, the portion of R&D expenditures that is expensed by firms that partly capitalize R&D expenditures is priced significantly less negatively than other expenses. While this result suggests that these R&D expenses exhibit future economic benefits, the perceived benefits are not so great as to argue in favor of capitalization; alternatively, this result could also indicate a market preference for the more conservative treatment of expensing. Under Korean GAAP, firms may capitalize R&D expenditures when future economic benefits are reasonably expected. Firms in Korea may thus exercise discretion over the capitalization of R&D; however, the market seems to value capitalized R&D greater than net book value, although fully expensed R&D expenditures are also priced positively. 16 Fourth, as with other expenses, advertising expenditures are not believed by market participants to represent future economic benefits Summary and conclusions This study investigates the value-relevance of R&D disclosures reported by firms listed on the Korean Stock Exchange from 1988 to Korean GAAP allows the capitalization of R&D expenditures when future economic benefits are reasonably expected from the expenditures. Using tests conducted with regression models derived under Ohlson s (1995) equity-valuation framework, we find that R&D expenditures are, in general, positively associated with stock price. Capitalized R&D expenditures appear to be regarded by market participants as a positive net present-value investment. Assuming capital-market efficiency, this suggests that investors generally agree that capitalizing rather than expensing these R&D expenditures is appropriate. Fully expensed R&D expenditures, although priced less than capitalized R&D, are also found to be regarded by market participants as a positive net present-value investment. This may signify that at least a portion of R&D expenditures merits capitalization, even when these expenditures are fully expensed. Although R&D expenses of firms that partially capitalize R&D expenditures were found to have no significant price impact, the regression coefficient on these expenditures is significantly less negative than the coefficient on other expenses. This may indicate a market belief that these expenditures 16 Ayers (1986), Bowen, Noreen, and Lacey (1981), Daley and Vigeland (1983), DeFond and Jiambalvo (1991), Dhaliwal (1980), Dhaliwal, Salamon, and Smith (1982), Duke and Hunt (1990), Lilien and Pastena (1982), Press and Weintrop (1990), Zimmer (1986), and Zmijewski and Hagerman (1981) note that the debt-toequity (d/e) ratio proxies for closeness to binding debt covenants. Results of additional tests (not presented) indicate that the proportion of R&D expenditures capitalized is greater among firms with higher d/e ratios and lower earnings, consistent with earnings management by these firms. These results are available from the authors. 17 We also tested the sensitivity of our results to macroeconomic condition (annual real GDP growth rates obtained from the Bank of Korea economic database), firm size (year-end market value of equity), and firm technology level (based on the standard industry code of Korea). The sample was partitioned into two groups on each of these dimensions, Eq. (3) was estimated separately for the two groups, and results were compared. In each case, our results were not differentiated by the conditioning variable, and the order of magnitude on our coefficient estimates remained the same (i.e., RDCAP>RDEXPF>RDEXPC). Furthermore, the coefficient on ADEXP was more negative than that on OEXP in every comparison. We thus feel that it is unlikely that macroeconomic condition, size, or technology level is behind our results.