Firm heterogeneity, R&D, and economic growth

Economic Modelling 36 (2014) 149–156

Contents lists available at ScienceDirect

Economic Modelling journal homepage: www.elsevier.com/locate/ecmod

Firm heterogeneity, R&D, and economic growth☆ Hyunbae Chun a,1, Joonkyung Ha b,⁎, Jung-Wook Kim c,2 a b c

Department of Economics, Sogang University, Seoul 121-742, Republic of Korea Division of Economics, Hanyang University, Ansan 426-791, Republic of Korea Graduate School of Business, Seoul National University, Seoul 151-916, Republic of Korea

a r t i c l e

i n f o

Article history: Accepted 17 September 2013 JEL classification: O31 L16 G11 Keywords: Firm heterogeneity R&D Technological change Economic growth

a b s t r a c t In this paper, we establish a link between firm heterogeneity and long-run economic growth both theoretically and empirically. We show that firms' technological heterogeneity creates the diversification effect for R&D financiers, facilitating R&D investment, and thus leading to long-run economic growth. This result holds even when heterogeneity limits the possibility of a synergy effect between firms with similar technologies. In testing the model's prediction using U.S. firm-level data, we define industries with higher firm-specific or idiosyncratic stock return volatility as those exhibiting higher firm-level technological heterogeneity and find a positive link between this measure and R&D intensity. Our paper implies that an economic growth policy aimed at increasing the diversity of the corporate sector may be more effective in attracting private R&D investments than the one aimed at concentration of resources on homogeneous projects due to the foregone diversification benefit of the latter. © 2013 Elsevier B.V. All rights reserved.

1. Introduction In this paper, we attempt to establish a link between firm heterogeneity, R&D, and economic growth, both theoretically and empirically. Morck et al. (2000) report that countries with higher level of firm performance heterogeneity exhibit higher level of per capita GDP. However, the underlying reason for why firm heterogeneity and economic growth should be positively related is yet to be explained. Existing literature on related topics focuses on the implication of diversity in products (Romer, 1990; Young, 1928) and in sectors (Acemoglu and Zilibotti, 1997) on long-run economic growth. We supplement this literature by providing a link between firm-heterogeneity and R&D financing in a simple economic model. In our model, firm heterogeneity is viewed as diversified technologies, success of which is determined randomly. Firms with the same technology are regarded to be homogenous. Homogeneity among firms

☆ We thank Peter Howitt and the seminar participants at the Bank of Korea. This work was supported by the National Research Foundation of Korea Grant funded by the Korean Government (NRF-2010-330-B00069), the Sogang University Research Grant of 2012 (201210069), the Institute of Finance and Banking and the Institute of Management Research at Seoul National University. We are also most grateful to the editor and an anonymous referee for their particularly helpful comments. ⁎ Corresponding author. Tel.: +82 31 400 5622; fax: +82 31 436 8180. E-mail addresses: [email protected] (H. Chun), [email protected] (J. Ha), [email protected] (J.-W. Kim). 1 Tel.: +82 2 705 8515; fax: +82 2 704 8599. 2 Tel.: +82 2 880 6986; fax: +82 2 878 3154. 0264-9993/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.econmod.2013.09.028

would imply that each firm's growth would exhibit higher comovement and thus, be driven more by systematic component rather than firmspecific component. In contrast, heterogeneous technologies among firms would imply that each firm's growth would be driven more by firm-specific component rather than by systematic component. We assume that the financial sector is well organized so that there is no friction in mobilizing investment funds. Households provide necessary funds for R&D activities and the R&D portfolio is constructed by a fund manager. With these assumptions, we show that the increased firm heterogeneity provides a valuable diversification opportunity for the R&D portfolio fund manager, thus increasing the overall R&D activity of the economy. This firm-level diversification effect is shown to outweigh the potential synergy effect that may come from the benefit of focusing on a small number of technologies. This has an important implication that ‘focus’ and ‘concentration’ may increase systematic risk and thus, exert a negative impact on the long-run economic growth. We test the model's prediction using U.S. firm-level data by examining whether industries with more heterogeneous firms exhibit higher R&D intensity, which would increase the long-run industry growth. We define industries with higher firm-specific or idiosyncratic stock return volatility as those exhibiting higher firm-level technological heterogeneity. We find a positive link between this heterogeneity measure and R&D intensity. We use monthly stock returns covered in the Center for Research in Security Prices (CRSP) data from 1971 to 2006 to calculate the firm-specific heterogeneity for each firm each year and then aggregate this at the 2digit industry-level. R&D and other variables are obtained from

150

H. Chun et al. / Economic Modelling 36 (2014) 149–156

Compustat for the same period. We find that industries with higher firm-specific stock return heterogeneity exhibit higher industrylevel R&D intensity. Our model's predictions are related with recent empirical and theoretical literature which examines the relationship between volatility and economic growth. Using cross-country data, Ramey and Ramey (1995) find a negative relationship between the GDP growth volatility and GDP growth. They explain that low aggregate growth volatility provides a stable environment for investment, thus prompting capital accumulation. However, an opposite relationship between volatility and growth is reported from sectoral and firm level analyses. Imbs (2007) finds that sectors with higher growth rate volatilities grow faster in his cross-country study. Chun et al. (2008) analyze U.S. firm-level data and find that industries with higher firm-specific volatilities in stock returns and sales growth grow faster. Both Imbs (2007) and Chun et al. (2008) argue that the high level of sectoral or firm-level growth volatility could be consistent with the low level of aggregate level volatility if sector-specific or firm-specific component in growth, which could be diversified, is larger than systematic component. Acemoglu and Zilibotti (1997), Koren and Tenreyro (2013), and Michelacci and Schivardi (2013) find that economies that could benefit from reduced risk through diversification would grow faster. Our paper supplements these findings by focusing on the role of firm heterogeneity in facilitating R&D funding. Our empirical evidence obtained from firm-level analysis supplements Imbs (2007) who examines the impact of diversification on economic growth using industry-level data. Our finding is different from those studies which examine the implication of the within-firm diversification and R&D activity (Garcia-Vega, 2006; Link and Long, 1981; Peyrefitte and Brice, 2004; Tanriverdi and Lee, 2008) since our work focuses on the implication of firm heterogeneity in the industry-level R&D funding. This paper is organized as follows; Section 2 presents the model. Section 3 shows the empirical analysis and Section 4 concludes the paper.

2. The model The model formally shows the relationship between firm heterogeneity and R&D investment. The model follows the basic framework of Acemoglu and Zilibotti (1997), but differs in that the model does not assume any fixed cost for setting up a sector. Furthermore, the model considers the possibility of a synergy effect from concentration, and also analyzes the case where there are an integer number of firms instead of a continuum of points. The last feature enables us to analyze shocks to individual firms and provides the connection between theory and empirics. We interpret technological diversification as something happening among firms, not necessarily among sectors. We will first analyze how firm heterogeneity affects R&D activities in the macroeconomy, and then we will turn to the issue of concentration versus diversification. The basic channel through which firm heterogeneity affects R&D is that (1) firm heterogeneity provides a richer opportunity for risk diversification, and (2) this leads to more investments in R&D through financial markets. Here, firm heterogeneity is viewed as diversified technologies: the firms with the same technology are regarded to be homogeneous. We also assume that the financial sector is well organized so that there are no frictions in mobilizing investment funds. The model analyzes overlapping generations; individuals live two periods (young and old). The total population is constant and normalized, so that each generation has a unit mass. For simplicity, individuals born at date t are assumed to work and save in t and consume only in t + 1.

There is an uncertainty over the state of nature in t + 1: there are N different states with equal probability in t + 1. We have: States of nature : s∈f1; 2; …; Ng with equal probability 1=N:

ð1Þ

2.1. Benchmark case: no externalities of clustering We will first analyze the benchmark case where there are no externalities driven by the synergy effect of clustering in the same technology. The expected utility function of households can be written as follows: N
 X
s  1 log ctþ1 Et U ctþ1 ¼ N s¼1

ð2Þ

where cst + 1 is consumption in state s at time t + 1. Households can choose between risky asset and safe asset to maximize the expected utility. Specifically, there are two different assets that can be purchased in period t. First, one can choose the safe asset whose rate of return is r in any state for t + 1, which means that if you invest one dollar in t, you get r dollars in t + 1. Second, there is the risky asset or R&D fund, whose average rate of return is R, which is strictly greater than r (i.e., R N r).3 The variance of R is determined by the firm sector, which will be shown later. Households make decisions on the amount of safe asset, which is B, and risky asset, which is F, given their income W in t. In a sector, there are N firms engaged in innovative activities. Here, the number of firms, N, is the same as the number of states in t + 1 for convenience. We assume that each firm should get financed from the R&D fund, F, to survive. Each firm uses technology i that becomes productive only if s = i in t + 1, which means a firm's return becomes positive only if the state of nature turns out to be favorable to the firm. Therefore, we have: Firm’s return : R; if i ¼ sðif its bet is successfulÞ;

ð3Þ

0; otherwise: As for the R&D fund, we assume that the fund is managed by a fund manager, who chooses the portfolio of the fund. Specifically, the fund manager chooses M different technologies that will be invested by the fund, given the amount of F. This means that only M different technologies are in operation in period t + 1.4 It is also assumed that the manager distributes the R&D fund F across M technologies equally. Therefore, the number of firms in each technology group i is simply N ∕ M. The return of the R&D fund in period t + 1 depends on the realization of s. If s = i, which means that if the state of nature turns out to reward technology i, then each firm in group i generates R, so that (N ∕ M) R is the fund's return in state i. If s is not equal to i, group i's return is 0. Therefore, the expected return of the fund is: EðR F Þ ¼

M X 1N R þ 0 ¼ R: N M s¼1

ð4Þ

3 In fact, R&D is only part of risky assets. However, the major proposition of this paper is that more heterogeneity leads to more R&D and can be rewritten as follows: higher firm heterogeneity leads to more risky investments, and more risky investments lead to more R&D. This means that the results of this paper are not affected by the fact that R&D is only part of risky investments as long as more risky investments lead to more R&D. We believe that this is warranted by reality as well as the convention of related researches such as Acemoglu (2009), Aghion and Howitt (2009), and many others where R&D cannot be undertaken without sufficient supply of funds which is subject to high risk and proper management of risks. The return of R&D is inherently uncertain and usually follows stochastic processes, so private R&D requires well developed financial markets and risky investments. 4 This implies that firms without R&D fund do not survive, which may sound unrealistic. However, this assumption makes the model much simpler without changing the core results. In order to make the model more realistic, one could assume that no R&D firms generate lower returns than R&D firms. Our paper's main results are not affected by the new assumption.

H. Chun et al. / Economic Modelling 36 (2014) 149–156

Note that the number of possible successful states is M, since there are only M technologies working in period t + 1. On the other hand, the variance of the fund's return is as follows: Var ðR F Þ ¼

 M X 1 N s¼1

N M

2 R−R

þ

  N X 1 M 2 2 N ð0−RÞ ¼ R − : N M N s¼Mþ1

ð5Þ

Eq. (5) shows that if the number of technology (M) is the same as the number of firms, the fund is fully diversified and thus, has a variance of 0. This is because each firm adopts different technologies and the success or failure of the technology is independent of each other. Fully diversifiable risk means that the fund has no systematic risk and we are focusing purely on firm specific components, which is perfectly consistent with the empirical part of this paper where systematic risk is fully removed from the stock return data to identify firm heterogeneity.5 Now, a household's utility maximization problem can be written as follows: N M N   X X
 X 1 1 1 S max Et U ctþ1 ¼ log ct þ1 ¼ logðRF t þ rBt Þ þ logðrBt Þ Bt ; F t N N N ð6Þ s¼1 s¼1 s¼Mþ1

2.2. Considering externalities of clustering in same technology We can consider externalities that can arise when many firms are in the same technology group. In this case, return to R&D, R, becomes: e ¼ Rm−z R

( 

−z

g ¼m



f ¼

ð7Þ

W tþ1 −W t Rð f t W t − f t−1 W t−1 Þ rfð1−f t ÞW t −ð1− f t−1 ÞW t−1 g ¼ þ Wt Wt Wt

ð8Þ where f ≡ F ∕ W is the R&D intensity. The steady state growth rate can be calculated if we assume that ft = f*, a constant in the steady state: 








g ¼ Rf þr 1−f −1 M ¼ max r−1; R−1 : N

r Rm−z o1 −z

mðRm −r Þ ð1−mÞr

This means that as M increases, F⁎ increases and B⁎ decreases.6 Specifically, more heterogeneity (M) leads to more R&D (F).7 However, if M ∕ N is less than or equal to r ∕ R, the R&D fund cannot be formed. This implies that there can be “no R&D trap” if technologies are too homogenous. The steady state growth rate of the economy is equal to the growth rate of W. Since the expected value of W in t + 1 is W tþ1 ¼ RF t þ rBt , the growth rate of the economy is: g tþ1 ¼

Rf þ r ð1−f Þ−1; where f ¼ max 0;

ð9Þ

Eq. (9) shows that more heterogeneity – a higher value of M – leads to a higher f, and therefore, higher growth. 5 In fact, it may be unrealistic to argue that the fund has no systematic risks at all. If one assumes that the maximum number of working technologies, M, is strictly smaller than the number of possible states, N, then it is easy to derive the systematic risk from Eq. (5). Our paper's main results are not affected by the new assumption. 6 The same result can be derived when the household's utility function is a CRRA. 7 Regarding this, it is interesting to note that Mark Heesen, president of the National Venture Capital Association, argues that venture investment will increase in more sectors “as the start-up ecosystem is better served with more diversity, not less.” (http://www. nvca.org/index.php? option=com_docman&task=doc_download&gid=897&Itemid=317).

) 1−z m R−r : m−z R−r

ð11Þ

If we use a CRRA utility function in Eq. (6) and solve for the solution, the optimal value of f is obtained as follows.9

Subject to F t þ Bt ≤ W t

 ðM=NÞt R−r  Wt F t ¼ max 0; R−r  R−ðM=NÞt R  ; 1 Wt: Bt ¼ min R−r

ð10Þ

where m ≡ M ∕ N is the degree of heterogeneity, and z represents the externalities from the synergy effect. Considering that m is between 0 and 1, if z is positive (negative), there are positive (negative) externalities in clustering, or benefits of concentration. In this case, the steady state growth rate can be calculated as follows8:

n

where B is the amount of safe asset, F is the amount of R&D fund, and W is the income of the household in period t. The optimal levels of F and B can be obtained as follows:

151

θ

ð12Þ þr −1

where θ is the coefficient of constant relative risk aversion. In Eq. (12), it is not clear whether R&D intensity is increasing in m or not. In order to see the relationship between m and f, we conduct simulations using various parameter values. For example, Fig. 1 shows that when R = 1.1 (10% return), r = 1.02 (2% return), and θ = 1 (log utility), f⁎ increases in m for all possible values of z as long as there is R&D. This result does not change for other values of R, r, and θ. Fig. 2 shows that we have similar relations between m and R&D intensity with θ = 10. 2.3. Measuring firm heterogeneity using stock return data We will now establish the link between theory and empirics. We want to show that as firm heterogeneity or firms' technological diversity m – or M given N – increases, our measure of firm heterogeneity that will be derived from the financial market also increases. Our measure of firm heterogeneity is the variance of stock returns of individual firms.10 Therefore, we need to show that as M increases, the variance of individual firm's stock return increases. Recall that individual firm's return is Rm−z if its bet is successful (i = s), and 0, otherwise. In this case, the mean return for each firm – not the fund's return which could incorporate multiple firms' return – is: −z

Rm ; if s∈½0; M  M Ri ¼ 0; if s∉½0; M : Ri ¼

ð13Þ

Then, the variance, or σ2 of each firm's stock return is as follows. 2 σi

(     ) M 1 Rm−z 2 M−1 Rm−z 2 N−M −z Rm − 0− 0 ¼ þ þ N M M N M M    2z 1 1 2 N − R ¼ : N MN M ð14Þ

8 We assume that the fund manager is not a social planner. Therefore, the manager does not internalize the synergy effect when choosing the portfolio. One can justify this assumption of decentralized decision making by viewing the manager as one of the many identical fund managers. 1−θ 9 −1 The CRRA utility function is assumed to be: c 1−θ . 10 Precisely speaking, it should be the firm-specific component of stock return variance as discussed at the end of the section.

152

H. Chun et al. / Economic Modelling 36 (2014) 149–156

z = 0, M (or m) and σ2i have a positive relationship. However, if the synergy effect (z) is positive and very big, then the value in Eq. (15) can be negative. This analysis along with the previous analyses can be summarized in the following propositions. Propositions. A. If z is not too big (z b 1/2(M − 1)), a higher degree of firm heterogeneity m leads to higher level of firms' return heterogeneity σ2i (Eq. (15)). B. A higher m leads to a higher level of R&D intensity f in all plausible cases (Simulations as shown in Figs. 1 and 2). C. Therefore, from A and B, we have the following: If z is not too big (z b 1/2(M − 1)), firms' return heterogeneity σ2i is positively associated with the level of R&D intensity f as long as R&D intensity has a positive value. If z is too big (z N 1/2(M − 1)), firms' return heterogeneity σ2i is negatively associated with the level of R&D intensity f.

Fig. 1. R&D Intensity for Various values of m and z: θ = 1 (log utility). Note: R = 1.1 and r = 1.02.

Here, M affects firm's stock return heterogeneity in two different ways: first, it increases firm's heterogeneity through more technical diversity; second, it lowers firm's heterogeneity by diluting the benefits of concentration. Therefore, the final effect depends on the magnitude of z, or the degree of the synergy effect. More formally, we have: dσ 2i R2 ¼ f1−2zðM−1Þg 2ð1−zÞ 1−2z : dM M N

From Proposition C, one can test whether firm heterogeneity leads to more R&D, or whether the synergy effect of clustering does not outweigh the benefits of technological diversification. One final note on the choice of a proxy for technological heterogeneity is warranted. In Eq. (15), we show that the increase in M increases a firm's stock return volatility. Precisely speaking, it should be the firm-specific component of the stock return variance. Increase in M implies that a technology a firm may adopt becomes diverse, and the firm's performance measured by its stock return may exhibit more firm-specific variations. For example, if the number of technology is the same as the number of firm in an industry, and if each firm adopts different technologies, the stock return variance of a firm would be driven fully by firm-specific component, depending on the success or failure of each firm's new technology which is independent across firms. As a result, the risk can be fully diversified as shown in Eq. (5) when M = N. Thus, in the following section, we capture the diversity of technology in an industry as the average firm-specific stock return variance of firms in the industry and examine the relationship between this measure and R&D intensity.

ð15Þ

This means that if z is not too big (z b 1/2(M − 1)), then firms' return heterogeneity properly represents firm heterogeneity or technological diversification (M). For example, in the benchmark case where

3. Empirical analysis In this section, we construct firm heterogeneity measures using firm-level stock returns obtained from the CRSP. We then analyze the relationship between R&D investment and firm heterogeneity in the U.S. 2-digit-level manufacturing industries. 3.1. Data To measure the firm-specific component of stock return variance, we remove the common effects shared by all firms in an economy. To do this, as in Roll (1988), we decompose each firm's stock return variance into two components: a systematic (related to economy-wide factors) component and a firm-specific (residual or idiosyncratic) component.

Table. 1 Summary statistics.

ln(RD/Total asset) ln(HET) ln(Firm age) ln(Firm size) Herfindahl ln(SYS) Leverage Liquidity Book-to-market Fig. 2. R&D Intensity for Various Values of m and z: θ = 10. Note: R = 1.1 and r = 1.02.

Mean

Standard deviation

Min.

Max.

−4.761 −4.176 2.753 6.747 0.087 −5.651 0.274 1.835 0.644

1.304 0.541 0.363 1.227 0.081 0.725 0.072 0.467 0.368

−11.563 −6.215 1.609 4.594 0.021 −8.955 0.100 0.585 0.107

−2.606 −2.242 4.007 10.425 0.586 −3.353 0.562 3.407 3.082

Notes: Sample consists of 20 U.S. manufacturing industries from 1971 to 2006 that are constructed from Compustat firm-level data. The sample size is 716.

H. Chun et al. / Economic Modelling 36 (2014) 149–156

153

Electronic and other electric equipment (36) Chemicals and allied products (28) Instruments and related products (38) Industrial machinery and equipment (35) Transportation equipment (37) Miscellaneous manufacturing industries (39) Rubber and miscellaneous plastics products (30) Furniture and fixtures (25) Fabricated metal products (34) Paper and allied products (26) Food and kindred products (20) Stone, clay, and glass products (32) Primary metal industries (33) Petroleum and coal products (29) Textile mill products (22) Tobacco products (21) Lumber and wood products (24) Apparel and other textile products (23) Leather and leather products (31) Printing and publishing (27)

0

0.01

0.02

0.03

0.04

0.05

0.06

Fig. 3. R&D intensities in U.S. manufacturing industries, 1971–2006. Notes: Each bar indicates R&D intensity by industry averaged over the 1971–2006 period. R&D intensity is the ratio of R&D spending to total asset. Numbers in parentheses are 2-digit SIC codes.

Electronic and other electric equipment (36) Industrial machinery and equipment (35) Instruments and related products (38) Miscellaneous manufacturing industries (39) Chemicals and allied products (28) Apparel and other textile products (23) Rubber and miscellaneous plastics products (30) Lumber and wood products (24) Fabricated metal products (34) Textile mill products (22) Furniture and fixtures (25) Primary metal industries (33) Transportation equipment (37) Leather and leather products (31) Food and kindred products (20) Printing and publishing (27) Stone, clay, and glass products (32) Petroleum and coal products (29) Paper and allied products (26) Tobacco products (21) 0

0.005

0.01

0.015

0.02

0.025

0.03

Fig. 4. Firm heterogeneity measures in U.S. manufacturing industries, 1971–2006. Notes: Each bar indicates firm heterogeneity for each industry averaged over the 1971–2006 period. Numbers in parentheses are 2-digit SIC codes.

For each firm, we run the following OLS regression each year using 12 monthly returns, R j;i;τ ¼ b0; j;t þ b1; j;t Rτ þ ε j;i;τ

We measure firm heterogeneity for industry i in year t by summing up residuals as

where Rj,i,τ is the stock return for firm j in industry i for month τ, and Rτ is the value-weighted market returns for the month.11 11 In addition to the economy-wide factor, Chun et al. (2008) and Chun et al. (2011) include an industry factor. However, our results in this paper remain unchanged when the industry factor is included.

  SSR R j;i;t X T j∈i j;i;t

X

ð16Þ 2 σ ε;i;t

≡

j∈i

ð17Þ

where SSR(Rj,i,t) is the sum of squared variation in the residuals for firm j in industry i and ∑ j ∈ iTj,i,t is the number of annual observations available in industry i given an estimation window. The systematic component

154

H. Chun et al. / Economic Modelling 36 (2014) 149–156

Table 2 Panel regressions of R&D intensity on firm heterogeneity. Panel A.

ln(HET)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

0.349*** (0.004)

0.247*** (0.001) −0.662** (0.047) −0.134 (0.380)

0.342*** (0.003)

0.344*** (0.008)

0.281*** (0.080)

0.287*** (0.089)

−0.008 (0.055) −2.253** (0.905) −0.293 (0.273) −0.317 (0.265) 0.892 716

0.225*** (0.075) −0.515 (0.309) −0.126 (0.141) 0.772 (1.294) −0.009 (0.049) −1.719** (0.778) −0.147 (0.283) −0.230 (0.235) 0.898 716

ln(Firm age) ln(Firm size) Herfindahl

1.179 (0.593)

ln(SYS)

0.007 (0.898)

0.883 716

0.893 716

0.884 716

0.883 716

−2.252** (0.906) −0.294 (0.271) −0.315 (0.262) 0.892 716

(1)

(2)

(3)

(4)

(5)

(6)

(7)

0.318*** (0.103)

0.174* (0.098) −0.806** (0.376) −0.159 (0.200)

0.309*** (0.095)

0.348*** (0.119)

0.256** (0.092)

0.320*** (0.109)

−1.521 (1.382) −0.262 (0.361) −0.301 (0.206) 0.840 238

−0.105 (0.091) −1.531 (1.345) −0.246 (0.352) −0.359 (0.227) 0.840 238

0.215* (0.115) −0.720** (0.337) −0.126 (0.197) −0.413 (2.487) −0.107 (0.086) −1.104 (1.229) −0.165 (0.305) −0.281 (0.183) 0.847 238

Leverage Liquidity Book-to-market Adjusted R2 Sample size Panel B.

ln(HET) ln(Firm age) ln(Firm size) Herfindahl

1.187 (3.573) −0.050 (0.084)

ln(SYS) Leverage Liquidity Book-to-market Adjusted R2 Sample size

0.836 238

0.847 238

0.836 238

0.836 238

Notes: Dependent variable is the log ratio of R&D spending to total asset. The sample in Panel A consists of annual observations (1971, 1972, …, 2005, 2006) and that in Panel B consists of observations with 3-year interval (1971, 1974, …, 2001, 2004). Explanatory variables in Panels A and B are lagged by one-year and by 3 years, respectively. ln(HET) is the firm performance heterogeneity measure. ln(Firm age) is the log of industry-average years listed in CRSP and ln(Firm size) is the log of industry-average sales. Herfindahl is sales-based Herfindahl index. SYS is a systematic component. Leverage is the sum of short and long-term debts over total assets. Liquidity is current assets over current liabilities. Book-to-market is total book value of equity over its corresponding market value. All panel regressions include both industry and year fixed effects. The sample consists of 20 U.S. manufacturing industries that are constructed from Compustat firm-level data. Numbers in parentheses are p-values that are based on industry-clustered standard errors adjusting for both heteroskedasticity and autocorrelation within an industry. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

(SYS) can also be similarly defined. Then, the average total stock return variance of an industry can be represented as the sum of the two components. We measure firm heterogeneity by either the SSR or the relative portion of SSR out of total stock return variance in our empirical investigation. Using the Compustat database, we construct R&D intensity defined as R&D spending (Compustat annual item #46) over total assets (item #6).12 The heterogeneity measure based on stock returns may reflect financial heterogeneity rather than technological heterogeneity. To control possible financial factors that may affect stock return based heterogeneity measure, we include industry level leverage, liquidity, and book-to-market ratio variables (Chun et al., 2008). We also construct variables that could affect industry R&D intensities that include average age (Gort and Klepper, 1982; Klepper, 1996), size (Cohen et al., 1987; Cohen and Klepper, 1996) for industry's firms, and sales-

12 We also define R&D intensity as R&D spending over sales (item #12). Results of the paper are robust to the new definition of R&D intensity.

based Herfindahl index (Aghion et al., 2005; Scherer, 1967).13 Summary statistics for R&D intensity, firm heterogeneity, and other variables are reported in Table 1. Figs. 3 and 4 show the R&D intensity and firm heterogeneity for 20 U.S. manufacturing industries averaged for the 1971–2006 period, respectively. Both R&D intensity and firm heterogeneity are high in chemicals and allied products (including pharmaceuticals), instruments and allied products, electronics and other electrical equipment (including semiconductors), and industrial machinery and equipment (including computers). In contrast, food and kindred products, petroleum and coal products, and tobacco products industries exhibit low levels of both R&D intensity and firm heterogeneity. In the following

13 In fact, these variables could also be associated with stock return based heterogeneity measures. For example, Fama and French (2004), Chun et al. (2008), and Irvine and Pontiff (2009) emphasize that variables related to corporate demography (age and size), competition (Herfindahl index), and intangibles (book-to-market) may affect stock return based heterogeneity measures. Thus, controlling all these variables is required to use stock return based heterogeneity measures to capture technology driven heterogeneity.

H. Chun et al. / Economic Modelling 36 (2014) 149–156

155

Table 3 Panel regressions of R&D intensity on firm heterogeneity: high versus low R&D sectors. Panel A. High R&D sector

ln(HET)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

0.398** (0.131)

0.233** (0.086) −0.972** (0.346) −0.089 (0.169)

0.384*** (0.111)

0.288** (0.101)

0.357** (0.130)

0.279** (0.105)

0.113 (0.082) −1.435 (2.101) −0.023 (0.200) −0.045 (0.361) 0.890 360

0.186* (0.089) −0.846** (0.292) −0.109 (0.148) −2.044 (1.687) 0.086 (0.064) −0.507 (1.778) 0.217 (0.161) −0.040 (0.245) 0.900 360

ln(Firm age) ln(Firm size)

−3.614 (2.094)

Herfindahl ln(SYS)

0.153* (0.069)

0.889 360

0.898 360

0.892 360

0.890 360

−1.603 (2.022) −0.008 (0.201) −0.073 (0.356) 0.890 360

(1)

(2)

(3)

(4)

(5)

(6)

(7)

0.353* (0.181)

0.252*** (0.073) −0.521 (0.435) −0.230 (0.222)

0.308* (0.141)

0.419** (0.184)

0.239* (0.124)

0.306* (0.138)

−2.521** (0.958) −0.467 (0.364) −0.605** (0.209) 0.852 356

−0.109 (0.077) −2.492** (0.950) −0.460 (0.359) −0.624** (0.217) 0.853 356

0.239** (0.101) −0.417 (0.431) −0.157 (0.168) 1.070 (1.568) −0.103 (0.065) −1.983* (0.929) −0.334 (0.423) −0.489 (0.300) 0.863 356

Leverage Liquidity Book-to-market Adjusted R2 Sample size Panel B. Low R&D sector

ln(HET) ln(Firm age) ln(Firm size) Herfindahl

2.613 (2.664) −0.109** (0.048)

ln(SYS) Leverage Liquidity Book-to-market Adjusted R2 Sample size

0.820 356

0.840 356

0.828 356

0.821 356

Notes: Dependent variable is the log ratio of R&D spending to total asset. Panel A reports results for a high-R&D industry subsample while Panel B for a low-R&D subsample. The high-R&D industry subsample contains industries whose ratio of R&D over total asset (R&D intensity) averaged over the sample period of 1971–2006 is higher than the median R&D intensity of all manufacturing industries. The high-R&D sample includes the first 10 industries from the top of Fig. 3. ln(HET) is the firm performance heterogeneity measure. ln(Firm age) is the log of industry-average years listed in CRSP and ln(Firm size) is the log of industry-average sales. Herfindahl is sales-based Herfindahl index. SYS is a systematic component. Leverage is the sum of short and long-term debts over total assets. Liquidity is current assets over current liabilities. Book-to-market is total book value of equity over its corresponding market value. Explanatory variables are lagged by one-year. All panel regressions include both industry and year fixed effects. The sample consists of annual observations. Numbers in parentheses are p-values that are based on industry-clustered standard errors adjusting for both heteroskedasticity and autocorrelation within an industry. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

section, we examine the relationship between firm heterogeneity and R&D intensity using panel regressions. 3.2. Empirical results Using data on 20 U.S. manufacturing industries from 1971 to 2006, we run panel regressions with both industry and time fixed effects as14 2

ln ðRDÞi;t ¼ α þ β lnσ ε;i;t−1 þ γX i;t−1 þ μ i þ ηt þ ui;t

ð18Þ

where ln(RD)i,t is the log of R&D spending over total asset, ln(σ2ε,i,t − 1) is

the log of firm heterogeneity (HET), and Xi,t − 1 is a vector of other 2 variables. Both σε,i,t − 1 and Xi,t − 1 are lagged to mitigate endogeneity issues. Standard errors are clustered by industry, adjusting for both

14 As pointed out in Cohen and Klepper (1992), R&D intensity is highly industry-specific; we thus include industry-fixed effects in our regressions.

heteroskedasticity and autocorrelation (Arellano, 1987; Wooldridge, 2002). R&D spending may be persistent over the time period or serially correlated. This could create the endogeneity problem. To address the issue, we construct the sample using non-overlapping 3-year interval (1971, 1974, …, 2001, 2004) and run the following panel regression with both industry and time fixed effects; 2

InðRDÞi;t ¼ α þ βInσ ε;i;t−3 þ γX i;t−3 þ μ i þ ηt þ ui;t :

ð19Þ

In Eq. (19), dependent variables are measured once in three years and both the heterogeneity measure and other explanatory variables are lagged by 3 years. Table 2 reports the results of panel regressions explaining R&D intensity. Panel A uses the sample consisting of annual observations as defined in Eq. (18) and Panel B uses the sample consisting of

156

H. Chun et al. / Economic Modelling 36 (2014) 149–156

observations with 3-year interval as defined in Eq. (19). The coefficient estimates of firm heterogeneity in Column (1) of both Panels A and B are positive and statistically significant at the 1% level. Overall, the positive relationship between firm heterogeneity and R&D intensity holds after including other controls such as firm age, size, Herfindahl index, leverage, liquidity, and book-to-market ratio. In Columns (4), (6), and (7), we include a systematic component (SYS) as a control variable. This is effectively the same as investigating the relative importance of SSR as defined in Eq. (17) out of the total stock return variance (Durnev et al., 2004). This also generates qualitatively similar results. 3.3. Robustness check It is widely known that R&D activities are concentrated in several large industries. Thus, our results may be driven by the subset of industries. To check this, Panels A and B of Table 3 report the results from panel regressions for subsamples of industries according to R&D intensities.15 Panel A of Table 3 presents results for a high R&D industry subsample while Panel B of Table 3 presents results for a low R&D sample. The high-R&D industry subsample contains industries whose ratio of R&D over total asset (R&D intensity) averaged over the sample period of 1971–2006 is higher than the median R&D intensity of all manufacturing industries. The high R&D subsample includes the first 10 industries from the top of Fig. 3. Panels A and B of Table 3 show that the heterogeneity effect on R&D is significant in both high and low R&D industries. Although R&D is highly concentrated in some manufacturing industries, as shown in Fig. 3, our results hold in both high and low R&D industries. We also run regressions with the sample excluding the top 5 industries with high R&D intensity and find the similar results. For further robustness check, we construct an alternative measure of firm heterogeneity defined as changes in ranking of firm sales within an industry over 5-year intervals (Comin and Mulani, 2009; Fogel et al., 2008). Regression results using this turnover measure also show a positive association of heterogeneity with R&D intensity. Overall, empirical findings confirm that the increased firm heterogeneity within an industry attract higher R&D activities of firms. 4. Conclusion In this paper, we develop a model in which firm performance heterogeneity increases the long-run growth rate by providing a diversification opportunity for R&D financiers. The model also shows that the economic significance of the diversification effect outweighs that of the positive spillover effect of clustering among firms in developing the same technology (focus and concentration). We provide the link between theory and empirics using U.S. firm-level data. We define industries with higher firm-specific or idiosyncratic stock return volatility as those exhibiting higher firm-level technological heterogeneity and find a positive link between this heterogeneity measure and R&D intensity, consistent with the model's prediction. The finding of the paper has an important policy implication: strategies aimed at increasing technological diversity are more efficient in attracting private R&D investments, and hence, growth enhancing in the long-run than those aimed at concentrating resources on homogeneous projects. Therefore, when governments make decisions on incentive systems regarding R&D, it may be desirable that they should take into account the relationship between diversification and private R&D financing.

15 Results based on the subsamples of 3-year interval also generate qualitatively similar results.

Our paper supplements the work of Acemoglu and Zilibotti (1997), who examine the relationship between sectoral diversification and economic growth, by examining a financing channel for R&D activity made possible due to firm performance heterogeneity. Technological diversity is observed not only in developed countries due to firms' heterogeneous adoption of a new technology (Chun et al., 2011), but also in developing countries due to the economic development process itself (Koren and Tenreyro, 2013). Our paper shows that this could further increase the long-run growth potential of an economy through facilitating R&D financing. References Acemoglu, Daron, 2009. Introduction to Modern Economic Growth. Princeton University Press, Princeton. Acemoglu, Daron, Zilibotti, Fabrizio, 1997. Was Prometheus unbound by chance? Risk diversification and growth. J. Polit. Econ. 105 (4), 709–751. Aghion, Philippe, Howitt, Peter, 2009. The Economics of Growth. The MIT Press, Cambridge, MA. Aghion, Philippe, Bloom, Nick, Blundell, Richard, Griffith, Rachel, Howitt, Peter, 2005. Competition and innovation: an inverted-U relationship. Q. J. Econ. 120 (2), 701–728. Arellano, Manuel, 1987. Computing robust standard errors for within-groups estimators. Oxf. Bull. Econ. Stat. 49 (4), 431–434. Chun, Hyunbae, Kim, Jung-Wook, Morck, Randall, Yeung, Bernard, 2008. Creative destruction and firm-specific performance heterogeneity. J. Financ. Econ. 89 (1), 109–135. Chun, Hyunbae, Kim, Jung-Wook, Morck, Randall, 2011. Varying heterogeneity among U.S. Firms: facts and implications. Rev. Econ. Stat. 93 (3), 1034–1052. Cohen, Wesley M., Klepper, Steven, 1992. The anatomy of industry R&D intensity distributions. Am. Econ. Rev. 82 (4), 773–799. Cohen, Wesley M., Klepper, Steven, 1996. A reprise of size and R&D. Econ. J. 106 (437), 925–951. Cohen, Wesley M., Levin, Richard C., Mowery, David C., 1987. Firm size and R&D intensity: a re-examination. J. Ind. Econ. 45 (4), 543–565. Comin, Diego, Mulani, Sunil, 2009. A theory of growth and volatility at the aggregate and firm level. J. Monet. Econ. 56 (8), 1023–1042. Durnev, Art, Morck, Randall, Yeung, Bernard, 2004. Value enhancing capital budgeting and firm-specific stock returns variation. J. Financ. 59 (1), 65–105. Fama, Eugene F., French, Kenneth R., 2004. New lists: fundamentals and survival rates. J. Financ. Econ. 73 (2), 229–269. Fogel, Kathy, Morck, Randall, Yeung, Bernard, 2008. Big business stability and economic growth: is what's good for General Motors good for America? J. Financ. Econ. 89 (1), 83–108. Garcia-Vega, Maria, 2006. Does technological diversification promote innovation? An empirical analysis for European firms. Res. Policy 35 (2), 230–246. Gort, Michael, Klepper, Steven, 1982. Time path of diffusion of product innovation. Econ. J. 92 (367), 630–653. Imbs, Jean, 2007. Growth and volatility. J. Monet. Econ. 54 (7), 1848–1862. Irvine, Paul J., Pontiff, Jeffrey, 2009. Idiosyncratic return volatility, cash flows, and product market competition. Rev. Financ. Stud. 22 (3), 1149–1177. Klepper, Steven, 1996. Entry, exit, growth, and innovation over the product life cycle. Am. Econ. Rev. 86 (3), 562–583. Koren, Miklos, Tenreyro, Silvana, 2013. Technological diversification. Am. Econ. Rev. 101 (1), 378–414. Link, Albert N., Long, James E., 1981. The simple economics of basic scientific research: a test of Nelson's diversification hypothesis. J. Ind. Econ. 30 (1), 105–109. Michelacci, Claudio, Schivardi, Fabiano, 2013. Does idiosyncratic business risk matter for growth? J. Eur. Econ. Assoc. 11 (2), 343–368. Morck, Randall, Yeung, Bernard, Wayne, Yu., 2000. The information content of stock markets: why do emerging markets have synchronous stock price movements? J. Financ. Econ. 58 (1–2), 215–260. Peyrefitte, Joseph, Brice Jr., Jeff, 2004. Product diversification and R&D investment: an empirical analysis of competing hypotheses. Int. J. Organ. Anal. 12 (4), 379–394. Ramey, Garey, Ramey, Valerie A., 1995. Cross-country evidence on the link between volatility and growth. Am. Econ. Rev. 85 (5), 1138–1151. Roll, Richard, 1988. R2. J. Financ. 43 (2), 541–566. Romer, Paul, 1990. Endogenous technological change. J. Polit. Econ. 98 (5), S71–S102. Scherer, Fredrick, 1967. Market structure and the employment of scientists and engineers. Am. Econ. Rev. 57 (3), 524–531. Tanriverdi, Huseyin, Lee, Chi-Hyon, 2008. Within-industry diversification and firm performance in the presence of network externalities: evidence from the software industry. Acad. Manag. J. 51 (2), 381–397. Wooldridge, Jeffrey M., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press, Cambridge, MA. Young, Allyn A., 1928. Increasing returns and economic progress. Econ. J. 38 (152), 527–542.