生技新藥產業發展條例對公司創新影響

The Impact of the Act for the Development of Biotech and New Pharmaceuticals Industry on Firm Innovation in Taiwan

生技新藥產業發展條例對公司創新影響

Woan-lih Liang, Department of Information Management and Finance, National Yang Ming Chiao Tung University

梁婉麗 / 國立陽明交通大學資訊管理與財務金融學系

Tai-Cheng Liu, Institute of Finance, National Yang Ming Chiao Tung University

劉泰承 / 國立陽明交通大學財務金融所

Received 2019/8, Final revision received 2021/2 Abstract

This study investigates how the innovation activities of the biotech and new pharmaceuticals industry (biopharmaceutical industry) change after the exogenous shock of the enactment of the Act for the Development of Biotech and New Pharmaceuticals Industry (Biopharmaceutical Act). We adopt Propensity Score Matching (PSM) and Difference-in-difference (DID) approaches to deal with sample selection bias and endogeneity problems. We find that approved biopharmaceutical firms engage in more innovation investments after the passage of the Biopharmaceutical Act than unapproved firms. We also confirm the policy effectiveness of the Biopharmaceutical Act, finding that approved biopharmaceutical firms are more encouraged to engage in innovation activities than high-tech firms after the Biopharmaceutical Act. In addition, the stimulation effect on innovation exists only for pharmaceutical and low R&D intensity firms. In the subsample analysis of the inter-industry effects, the stimulation effect is driven primarily by low R&D intensity firms and small firms. These findings consolidate the effectiveness of the Biopharmaceutical Act for biopharmaceutical firms with more serious R&D underinvestment problems. Our investigation also shows the effectiveness of tax credits granted by the Biopharmaceutical Act for the R&D investment of biopharmaceutical firms in Taiwan.

【 Keywords】innovation, R&D, tax credits, propensity score matching, difference-in- difference

摘 要

本研究探討生技新藥業發展條例（簡稱生技條例）對生技新藥業（簡稱生技業）研發 創新的影響。我們採用「傾向分數配對法」與「差異中之差異法」來處理樣本選擇誤 差與內生性問題。研究發現，在生技條例施行後，經生技條例核可的公司比未經核可 的公司有較多的研發創新。我們也發現在生技條例施行後，經條例核可的公司會比高 科技公司有較多的研發創新。再者，經生技條例核可的公司中，新藥公司與研發強度 低的公司會致力於較多的研發創新。與高科技公司比較時，生技條例核可的公司中，

研發強度較低的公司與小公司會有較多的研發創新。這些發現隱含條例對於研發較嚴 重投資不足的公司的激勵效果較強。我們的研究亦驗證政府對於生技業所提供的課稅 減免優惠對刺激該產業投資研發創新有正向的鼓勵作用。

【關鍵字】研發創新、課稅抵減、傾向分數配對法、差異中之差異法、生技新藥業發 展條例

領域主編：陳忠仁教授

1. Introduction

Government policies are commonly used to encourage firms to increase their investments in Research and Development (R&D). In general, R&D investments and innovation activities are key sources of growth and sustained operations of a firm and/or a nation (Romer, 1990; Aghion and Howitt, 1992). However, the nature of R&D activities, which includes high uncertainty, long-run cumulative effect, imperfect appropriability, and high information asymmetry between investors and managers, causes R&D investment to fall below the socially optimal level (Aghion, Van Reenen, and Zingales, 2013; Arrow, 1962; Brown, Martinsson, and Petersen, 2012; Bushee, 1998; Jones and Williams, 1998;

Porter, 1992). To solve the R&D underinvestment problems, many countries adopt public policies such as tax credits and direct subsidies to stimulate R&D because these policies can ameliorate funding issues and indirectly eliminate the financing difficulty problem of R&D (Aerts and Schmidt, 2008; Czarnitzki and Hussinger, 2004; Cerulli and Potì, 2012;

Czarnitzki, Hanel, and Rosa, 2011; Yang, Huang, and Hou, 2012; Hall and Van Reenen, 2000).

In Taiwan, the problem of R&D underinvestment in the biotech and new pharmaceuticals industry (biopharmaceutical industry) is particularly severe. The biopharmaceutical industry and high-tech industry are respectively the first and second highest R&D intensity industries in Taiwan for the period 2007-2017.

Brown, Martinsson, and Petersen (2017) argue that the industries with more R&D intensity are more likely to experience innovation underinvestment. In 1982, the Science and Technology Development Plan of Taiwan listed these two industries as important development items for the nation. After decades of struggles, Taiwan’s economy is mainly supported by the high-tech industries, including electronic components, information services, semiconductors, and optoelectronics technology.

However, the growth of the biopharmaceutical industry is relatively limited and

1　 The biopharmaceutical industry includes pharmaceutical, biomedical material, and health care firms.

The high-tech industry consists of computers and peripherals, semiconductors, consumer electronics

retailing, optoelectronics, telecommunication and networking, electronic components, information

technology services, and other electronics. Our untabluated results show that R&D intensities, the ratio

of R&D expenditure to total assets, for the Biopharmaceutical and high-tech industries are 4.08% and

3.04%, respectively.

slow due to several reasons.

First, most Taiwan biopharmaceutical firms cannot obtain substantial funding to engage in R&D activities because their firm sizes are relatively small. Second, Taiwan’s venture capitalists have less incentive to invest in domestic biopharmaceutical firms than in foreign ones because of less successful domestic cases in developing both new medicines and high risk medical devices.

Third, owing to the restrictions of new drug prices under the national health insurance policy, the relatively low ratio of health expenditure to GDP, and the domestic market-oriented patent authorization and technology, biopharmaceutical firms in Taiwan tend to have less incentive to engage in high risk innovation activities.

To address the R&D underinvestment problem and to provide a more favorable environment to the development of the industry, the Taiwan government adopts several public policies.

Among these public polices, the Biotech and New Pharmaceuticals Development Act

（生技新藥產業發展條例） , which was promulgated in 2007, is intended to promote innovation investment in the biopharmaceutical industry.

The Biopharmaceutical Act primarily uses tax credits to spur R&D. Theoretically, the effectiveness of tax credits appears to be stronger in encouraging R&D than direct subsidies for several reasons. First,

2　 The reasons are summarized from reports of Biopharmaceutical Industry White Paper of Industrial Development Bureau, Ministry of Economic Affairs.

3　 According to Taiwan Venture Capital Association (2008)（2008 年 臺 灣 創 業 投 資 年 鑑 ）, capital invested in biopharmaceutical industry accounts for less than 10% of annual venture capital from 2001 to 2007. By contrast, capital invested in high-tech industry accounts for more than 25% of annual venture capital in the corresponding time period.

4　 According to the “Development of the biotechnology and pharmaceutical industry in Taiwan (The report on the monthly meeting of the presidential palace)” of Wong (2007)（政策焦點：生技製藥產 業在台灣的發展（2007 年 7 月總統府月會專題報告））the ratio of health expenditure to GDP in Taiwan, which is about 5.3-5.4%, is lower than the average 7.8%, of developed countries. In addition, Chen (2017) finds that the restrictions on new drug prices under the national health insurance policy tend to discourage companies from developing and selling new drugs in Taiwan. Further, Sun (2003) argues that the main sources for companies to obtain patent authorization and technology introduction are domestic, which shows that the development of Taiwan’s biotechnology industry is limited to the domestic market and local technology status.

5　 These policies include “Action Plan for Strengthening the Biotechnology Industry”（加強生物技術產 業推動方案） in 1995, “Relaxing the Listing for Biopharmaceutical Industry”（放寬生技公司上市

（櫃）標準） in 2001, and “Act for the Development of Biotech and New Pharmaceuticals Industry”

(Biopharmaceutical Act)（生技新藥產業發展條例）in 2007.

6　 We focus on the Biopharmaceutical Act of 2007 rather than other related public policies or regulations

of the Biopharmaceutical industry because data on listed firms from 1995 to 2001 are not available.

previous studies argue that the government may not grant subsidies to proper projects or firms with high social return (e.g., Wallsten, 2000; Winston, 2006) because of government failure resulting from political pressure, corruption, and bureaucratic objectives. By contrast, tax credits decrease the marginal cost of R&D. Thus, the incentive function of tax credits is determined by the benefit and cost analysis of R&D rather than government failure. Second, the recipients of subsidies may not devote efforts to R&D after receiving the funds from government. Tax credits generally do not have this moral hazard problem because the firms must increase R&D investment to obtain the credits. Accordingly, the tax credits used in Biopharmaceutical Act should be more effective for R&D than direct subsidies.

In addition to tax credits, the Biopharmaceutical Act also adopts non-tax credit tools. First, the Biopharmaceutical Act grants managers and employees the firms’ shares and share warrants to increase managers’ motivation to invest in R&D. Second, the Act relaxes certain restrictions on employees to increase collaboration between industries and academic institutions, which allows the biopharmaceutical firms to obtain the knowledge of R&D incoming spillover from other institutions. Therefore, these non-tax credit tools could also alleviate the R&D underinvestment problems.

Although previous studies have investigated the effects of Biopharmaceutical Act on innovation (Chen, 2013; Hsu, 2018; Kao, 2012), they do not have consistent results.

By comparing the data before and after 2007, Chen (2013) finds significant decreases in

R&D expenditure after the Biopharmaceutical Act. Kao (2012) adopts the direct subsidy

of R&D from the government, the amount of tax credit from the Biopharmaceutical Act,

and the amount of government direct investment as policy proxies for the influence of the

Biopharmaceutical Act, and finds no significant impact of these policies on biotechnology

patents. Moreover, Chen (2013) and Kao (2012) both examine the change in innovation

for the biopharmaceutical industry before and after the Biopharmaceutical Act, but they

may suffer from endogeneity issues. Specifically, the difference between the pre- and post-

Biopharmaceutical Act outcome may result from other exogenous influences such as the

global financial crisis of 2007-2008. Hsu (2018) shows that the R&D expenditures of

biopharmaceutical industry significantly increase after the Biopharmaceutical Act. Hsu

(2018) also finds that firms which are approved under Biopharmaceutical Act (approved

biopharmaceutical firms) tend to have higher R&D expenditures than those not approved

(unapproved biopharmaceutical firms). Hsu’s (2018) findings thus support the positive

influence of Biopharmaceutical Act on innovation. However, Hsu (2018) may also suffer

from endogeneity problems because the result may be driven by omitted variables.

Accordingly, the abovementioned studies may not identify the actual effect of the Biopharmaceutical Act because these papers do not address the potential endogeneity.

To prevent the endogenous problem, we adopt Difference-in-difference (DID) approach to investigate. Namely, relative to control firms (i.e. firms could not obtain the benefits of the Biopharmaceutical Act), how treated firms (i.e. approved biopharmaceutical firms) respond to the exogenous shock of the Biopharmaceutical Act.

Before the DID approach, we use the Propensity Score Matching (PSM) method to identify comparable control firms which have characteristics similar to those of approved biopharmaceutical firms. For the control firms, we first consider biopharmaceutical firms which are not approved by the Biopharmaceutical Act. The intra-industry comparison between approved and unapproved firms enables us to identify the change in innovation from obtaining benefit from the Biopharmaceutical Act. Further, industries with higher R&D intensity are more likely to experience the innovation underinvestment problem (Brown et al., 2017).

Yang et al. (2012) find that tax credits are more effective for R&D in industries with higher R&D intensity and suggest that the government should establish tax credits across industries. As a result, we use the high-tech industry as a comparable control group since high-tech firms also have great R&D underinvestment problems; however, they cannot benefit from the Biopharmaceutical Act. This inter-industry comparison helps to evaluate whether the Biopharmaceutical Act is effective only for biopharmaceutical firms.

We collect firms approved under the Biopharmaceutical Act from the “2018 Biotechnology Industry White Paper of Bureau of Industry”, Ministry of Economic Affairs (2018)（2018 生技產業白皮書）. From 2007 to 2018, there are 134 approved biopharmaceutical firms, with 65 of these being listed firms. Accounting information is collected from the Taiwan Economic Journal (TEJ) database. We use R&D intensity (the ratio of R&D expenditure to total assets) as the proxy of R&D investments and adopt patent adjusted citation as the proxy of innovation quality. The patent data are collected from the European Patent Office (EPO) Worldwide Patent Statistical Database (PATSTAT), from which we choose patents applied with the United States Patent and Trademark Office

7　 The control firms are matched firms for the approved biopharmaceutical firms. Because the approved

biopharmaceutical firms receive the Biopharmaceutical Act benefits, the control firms are the ones

which do not obtain the benefits from the Act.

(USPTO), since we only focus on U.S. patents applied by Taiwanese firms.

Our empirical results show that innovation activities are encouraged by the Biopharmaceutical Act. First, unlike unapproved firms, approved biopharmaceutical firms significantly increase their R&D investments after the Biopharmaceutical Act.

However, these firms have not significantly improved innovation quality yet. In addition, the encouragement effect in the intra-industry analysis only occurs for pharmaceutical and low R&D intensity firms. Second, the approved biopharmaceutical firms exhibit significantly more innovation activities and higher innovation quality than high-tech firms after the Biopharmaceutical Act. This result of the inter-industry comparison over innovation investment primarily exists in low R&D intensity firms and small firms while the result of inter-industry comparison over innovation quality only exists among large firms. Therefore, our findings support the policy effect of the Biopharmaceutical Act on innovation.

Our paper contributes to the literature regarding the effect of the Biopharmaceutical Act on innovation in several ways. First, we conduct DID approach to avoid the endogeneity problem, which previous studies did not particularly address (Chen, 2013;

Hsu, 2018; Kao, 2012). This makes our results more reliable and also reduces the influences of other exogenous shocks and other potential factors. Second, we find that the approved biopharmaceutical firms (beneficiaries of the Act) have greater investments in innovation than unapproved biopharmaceutical firms and high-tech firms (non-beneficiaries of the Act) have after the Biopharmaceutical Act. Our results confirm with previous studies that tax credit policy can encourage innovation activity. In addition, our finding from the inter-analysis supports the argument of Yang et al. (2012) that government should establish tax credits across industries. This finding also confirms the policy effectiveness of the Biopharmaceutical Act, which grants tax credits only for biopharmaceutical firms. Third, we examine what types of firms derive greater benefits from the tax credit provisions of the Biopharmaceutical Act. We find that pharmaceutical firms and low R&D intensity biopharmaceutical firms exhibit greater innovation investments than unapproved biopharmaceutical firms after the Biopharmaceutical Act. We also find that small and low R&D intensity approved biopharmaceutical firms have higher innovation investments than

8　 There are fewer citations received by the patents that are applied with and granted by the Taiwan

Intellectual Property Office (TIPO). Thus, using TIPO patents could be more difficult to gauge patent

citations. In addition, USPTO patents are usually regarded as more valuable (Huang and Chang, 2020).

high-tech firms after Biopharmaceutical Act. These findings imply that firms with more serious R&D underinvestment problems are more strongly encouraged to invest in R&D by the Biopharmaceutical Act.

The remainder of this paper is organized as follows. Section 2 reviews prior literature including studies on the R&D underinvestment problem, theoretical papers on public policies that encourage R&D investment, and important related public policies of other countries. Section 3 describes the data, defines the variables, and introduces the PSM and DID methodology. Section 4 shows the results of the intra-industry and inter-industry analyses, and the subsample analysis. Section 5 summarizes our findings and makes the conclusion.

2. Literature Review

The first part of this section discusses the nature of private R&D and the corresponding theoretical concepts that explain the R&D underinvestment problem. To address the R&D underinvestment problem, the government often adopts policies such as tax credits, direct subsidies, construction of national laboratories, and encouraging the cooperation between industry and academia. Among these public polices, the two primary policies used by the government to encourage private R&D are direct subsidies and tax credits. Therefore, we respectively use two subsections to illustrate and compare these two policies. Finally, to understand the effectiveness of the Biopharmaceutical Act in Taiwan, we consider the influence of tax credits and non-tax credits on R&D.

2.1 Theories that Explain the Underinvestment in Private R&D

Many papers argue that the nature of private R&D activities leads to underinvestment in R&D. First, the agency theory posits that the conflict of interest between managers and shareholders gives managers less incentive to engage in R&D. Managers usually have less incentive to innovate because the innovation process is long, unpredictable, heterogeneous, complicated, and has a high probability of failure. Porter (1992) and Bushee (1998) argue that myopic managers, who focus on profits to meet short-term goals, may cut R&D expenditures because R&D activities are usually long-term. In addition, Aghion et al.

(2013) propose that risk-averse managers have less incentive to do R&D because these

managers may be fired merely for having bad luck with risky investments. Further, lazy

managers like to have a quiet and happy life and may not want to engage in R&D because

of the complications and long-term effort required (Aghion et al., 2013). Overall, the R&D underinvestment problem is derived from the agency problem of managers. Previous studies (Aghion et al., 2013; Bushee, 1998; Chang, Liang, and Wang, 2019) propose that institutional investors could alleviate the agency problem and thus ameliorate the problem of R&D underinvestment.

Second, spillover theory proposes that a firm cannot appropriate all the returns from its R&D investments and accordingly has less incentive to invest in R&D at the socially optimal level. Arrow (1962) and Jones and Williams (1998) suggest that firms may underinvest in innovation because it is difficult to conceal all knowledge of innovation from competitors. The free rider problem (or the imperfect appropriability) of R&D allows competitors to take advantage of firms’ knowledge of R&D to reduce production costs or to increase profitability. Chen, Chen, Liang, and Wang (2013) further examine the outgoing spillover effect of R&D and discover that firms which are less able to appropriate their R&D benefits are more likely to underinvest in R&D. The protection of patent and intellectual property may help to eliminate the free rider problem of R&D investments.

However, such protection is incomplete in the real world because of patent and intellectual property litigation (Bessen, Neuhäusler, Turner, and Williams, 2014).

Third, the financial constraint theory argues that relative to other investments,

the R&D investment is more affected by financial constraints because of its greater

uncertainty and its higher information asymmetry. Li (2011) argues that when a firm

cannot raise enough funds to conduct the required tests of R&D, it may suspend the R&D

project. Thus, R&D intensive firms with financial constraints are more likely to cut R&D

investments. Brown et al. (2012, 2017) and Hsu, Tian, and Xu (2014) further investigate

the external financing for R&D investments because R&D intensive firms often easily

exhaust internal financing due to their lack of tangible assets and highly asymmetric

information. They argue that equity markets are more suitable for the innovative firms

to finance their R&D investments because equity markets provide investors with upside

returns without collateral requirements and allow feedback of valuable information about

the prospects of innovative projects. By contrast, Hall (2002) states that debt finance is less

suited for R&D investment because of the limited collateral value of intangible assets and

the high probability of failure in R&D. Accordingly, the improvement of equity financing

for R&D investments (i.e. reducing the financing constraints on R&D investments) helps

to reduce the R&D underinvestment problems of innovative firms.

2.2 The Effect of Government R&D Subsidies on Innovation

In order to promote R&D for economic growth, the government should grant subsidies to projects with high expected social benefit but with low returns for the private sector. There are two competing theories that explain the relationship between private R&D and public expenditures such as R&D subsidies: the substitution and complementary theories.

Theoretically, public subsidies can positively contribute to the private sector because the recipients of subsidies directly receive the profits of funds while non-recipients of subsidies indirectly obtain knowledge from R&D spillovers (David, Hall, and Toole, 2000; Özçelik and Taymaz, 2008; Chen, Chen, Liang, and Wang, 2020). Klette and Møen (2012) use the dynamic and long-run model and argue that government R&D subsidies produce positive learning-by-doing effects on private R&D. Takalo and Tanayama (2010) also use a theoretical model and suggest that government R&D subsidies directly reduce the financing constraints and can decrease the capital costs of innovative firms because the firms receiving the R&D subsidies provide informative signals to the market. Accordingly, these theoretical papers suggest that the complementary theory under which the subsidies produce additional effects for private R&D investments is the correct explanation.

Although above theoretical papers support the complementary theory, empirical studies show conflicting results (David et al., 2000; Wallsten, 2000; Wu, 2005). To assure the effectiveness of government subsidies, the government should choose target projects with high social returns that private firms would not undertake on their own. However, Wallsten (2000) finds that firms which devote more R&D tend to be more easily to receive government subsidies. Further, Wallsten (2000) does not find positive influence of subsidies on innovation because he finds that federal R&D grants decrease firm-financed R&D.

Wu (2005) and Toivanen and Niininen (2000) also find that the government direct subsidies crowd out firm R&D investments, meaning that direct subsidies substitute for private R&D expenditures. Becker (2015) argues that this crowding-out effect may result from the problem of sample selection bias. In fact, the government may favor certain

9　 Substitution theory suggests that these two mechanisms replace each other: private R&D is reduced when the government infuses funds into the private sector. Complementary theory suggests that the public fund infusion of government policy ameliorates underinvestment in private R&D.

10　 Wallsten (2000) posits several reasons that the government funds may provide firms with the ability to

attract private funds and may allow the firms to delay refinance.

projects and firms with more promising outcomes (David et al., 2000; Klette, Møen, and Griliches, 2000; Lach, 2002). Such premeditated selection by the government appears to violate the assumption of the effectiveness of government subsidies. Thus, since these papers do not consider the endogeneity of the sample selection, their results may be biased.

By adopting new econometric techniques to control for selection bias, more recent studies are shifting away from the crowding-out effect of subsidies on private R&D to their stimulating effect on private R&D (Becker, 2015). Using a matching methodology for the samples, Czarnitzki and Hussinger (2004), Duguet (2004), and Carboni (2011) find that government subsidies have a positive effect on private R&D in German, French, and Italian firms, respectively. In addition, Aerts and Schmidt (2008) use a conditional DID estimator and also reject the crowding-out effect on private R&D in Flanders and Germany. Further, adopting a treatment analysis, Özçelik and Taymaz (2008) also support the additionality effect of government subsidies on private R&D in Turkish manufacturing firms. For Italian firms, Cerulli and Potì (2012) adopt matching methods and the DID estimator and also obtain a similar positive effect. Therefore, after controlling for the endogeneity problem, most empirical papers find that government subsidies did stimulate private R&D investments.

2.3 The Effect of Government Tax Credits on Innovation

The policy on R&D tax credits is a more market-oriented method, since it permits firms to decide the amount and timing of investing in R&D activities. In theory, David et al. (2000) and Czarnitzki et al. (2011) use the relationship between marginal return of R&D and marginal cost of R&D to explain firms’ decisions to engage in private R&D.

Both studies find tax credits decrease the marginal cost of R&D and thus result in higher

private R&D investments. Namely, tax credits do not have crowding-out effects on private

R&D because the tax credits shift the marginal cost curve downwards. Tax credits may

thus result in two possible situations. First, tax credits may cause some firms that would

not invest in R&D to engage in R&D investments instead when the marginal cost of R&D

in these firms is greater than the marginal return of R&D. Second, the recipients of tax

credits may be induced to infuse more investment projects with less profit. Under this

micro-level framework, tax credits decrease the marginal cost of R&D. Thus, these two

papers support the financial constraint hypothesis under which the tax credit ameliorates

the financial constraint problem and increases R&D investments.

The empirical studies find that tax credits have a positive effect on R&D investment, though the results of estimators vary depending on the sample data, model specification, and methodology. Without the tax credits for R&D, the expense labelled or not labelled as the R&D investment is indifferent for a firm. However, when the government provides preferential tax treatment for R&D investments, firms generally prefer to label expenses as R&D investments. Considering the relabeling problem of R&D, which may cause overestimation of the effect of R&D, Hall and Van Reenen (2000), using the data of OECD countries, still find that each dollar of tax credits for R&D increases R&D investments by a dollar.

In addition, assessing the effect of R&D tax credits suffers from the selection bias problem because the recipients of tax credits may have characteristics different from those of non-recipients. Czarnitzki et al. (2011) adopt the non-parametric matching method to remove the selection bias problem and support the effectiveness of tax credits on innovation output for Canadian firms. Yang et al. (2012) investigate the tax credit policy of Taiwan and use the PSM approach to eliminate selection bias problems. By adopting detailed information about the amount of R&D tax deduction as the instrumental variable and a generalized method of moment methodology to control for endogeneity and firm heteroskedasticity, Yang et al. (2012) also find that tax credits stimulate additional R&D investments.

2.4 Direct Subsidies versus Tax Credits

In terms of the effects on the private R&D decisions, direct subsidies are different from tax credits. In practice, target firms or projects may not be randomly granted subsidies. Winston (2006) proposes that political pressure, corruption, and bureaucratic objectives may result in government’s failure to select proper firms or projects. In addition, direct subsidies may cause moral hazard problems because the recipients of subsidies may not devote their efforts to R&D activities after obtaining funds from the government. By contrast, tax credits do not engender moral hazard problems because firms must increase R&D investments to obtain the credits. Further, firms would recalculate benefits and costs to decide their investments in R&D while tax credits reduce the cost of R&D. The credit granted depends on the market-oriented R&D decision of firms rather than the government’s discretionary decision. Accordingly, the problems of government failure and moral hazard from the direct subsidies imply that tax credits are more effective in encouraging R&D.

The time patterns of the stimulus effects of tax credits and direct subsidies are also

different. David et al. (2000) find that the recipients of tax credits tend to use additional funds from tax credits to preferentially invest in projects with the highest private return because tax credits reduce the marginal cost of R&D. Based on this concept, David et al. (2000) suggest that the stimulating effect of tax incentives is strong in the short-run because the recipients of tax credits tend to concentrate in the projects with short-term prospects. By contrast, firms or projects granted subsidies by government are selected because they can benefit the general social welfare but the firms themselves may not receive benefits. In theory, projects with high expected social benefits but with insufficient private expected returns usually have long-term nature. Therefore, the stimulus effect of tax credits on innovation is generally faster than that of direct subsidies.

2.5 The Effect of the Taiwan Biopharmaceutical Act on R&D

In Taiwan, the Biopharmaceutical Act is established to promote R&D investment of biopharmaceutical industry. The Biopharmaceutical Act adopts two main policy tools to stimulate R&D: tax credits and non-tax credits. First, Biopharmaceutical Act provides tax credits on profits if the expenditure of R&D and personnel training of a particular year exceeds the average of the expenditure of the previous two years. This regulation decreases the cost of R&D investment, which is consistent with the theoretical concept of tax credit (David et al., 2000; Czarnitzki et al., 2011). The Act also offers tax credits for holding shares of biopharmaceutical firms and grants tax credits to top executives and technology investors for their new shares in biopharmaceutical firms. Such a regulation can increase equity financing opportunities and is therefore consistent with the argument of previous papers (Brown et al., 2012, 2017; Hsu et al., 2014) that the equity market is more suited to financing R&D investments. Therefore, the decreasing cost of R&D investment and the incentive of equity financing tend to support the financial constraint theory in explaining R&D underinvestment.

Based on the comparison of literature between the two policies, if the government wants to more strongly promote R&D, it should adopt tax credits rather than direct subsidies. In fact, the Biopharmaceutical Act primarily uses tax credits. Thus, we infer that the government’s policy is good. In addition, we predict that the effectiveness of Biopharmaceutical Act is stronger in the short run than in the long run because the stimulating effect of tax credits is more rapid than that of direct subsidies (David et al., 2000).

Second, the biopharmaceutical industry also provides non-tax credit treatment as

follows. The Biopharmaceutical Act grants the managers and employees of the firms a share and share warrant. This regulation reduces the agency problems between managers and shareholders and thus increases the managers’ incentive to invest in R&D. In addition, the Biopharmaceutical Act also relaxes restrictions on employees to increase the collaboration between firms and academic institutions. This can increase the effectiveness of R&D spillover from other institutions to biopharmaceutical firms. Therefore, the improvement effects encouraged by these non-tax credit treatments such as spurring managers’ motivation and increasing cooperation opportunities tends to support the agency theory and spillover theory in explaining R&D underinvestment.

3. Data and Methodology

3.1 Data

The study investigates the influence of the Biopharmaceutical Act on firm’s innovation for biopharmaceutical industry and high-tech industry. The biopharmaceutical industry and the high-tech industry are classified based on the definitions from the Taiwan Stock Exchange.

We collect firms which are approved under the Biopharmaceutical Act from the 2018 Biopharmaceutical Industry White Paper of Bureau of Industry, Ministry of Economic Affairs.

From 2007 to May 2018, 134 biopharmaceutical firms are approved by the Biopharmaceutical Act, 65 of which are listed firms.

There are 85 firm-year observations from the listed approved firms. From 2007 to 2017, 148 biopharmaceutical firms were not approved by the Biopharmaceutical Act. In addition, there are 11309 firm- year observations of the high-tech industry. We obtain the accounting information from

11　 The definitions of the biopharmaceutical industry and high-tech industry are presented in Section 1.

12　 If companies seek to be approved as biopharmaceutical companies, they should meet the requirements of the Biopharmaceutical Act. These regulations include “Regulation of Shareholder Investment Deduction for Biotech and New Pharmaceuticals Industry”（營利事業適用生技新藥公司股東投資 抵減辦法） and “Regulation of Investment Deduction for Research and Development and Personnel Training Expenditures of Biotech and New Pharmaceuticals Industry”（生技新藥公司研究與發展 及人才培訓支出適用投資抵減辦法） . Of these regulations, the most important requirement is that companies whose R&D expenses should have a significant proportion of their total operating income or paid-in capital.

13　 The approved firms include firms listed in the stock exchange market, over-the-counter market, and

emerging stock market, and also include unlisted companies.

the Taiwan Economic Journal (TEJ) database. We collect U.S. patent data of these target firms from the European Patent Office (EPO) Worldwide Patent Statistical Database (PATSTAT) because it is more detailed and comprehensive and because it is widely used in the literature (Bena and Li, 2014; Chang et al., 2019; Chen et al., 2020; Li, Lai, D’amour, Doolin, Sun, Torvik, Yu, and Fleming, 2014).

3.2 Variables

3.2.1 Innovation Measures

We use R&D intensity as the proxy for R&D investments (innovation investments) and patent adjusted citation as the proxy for innovation quality. R&D expenditure is the innovation input while patents are the innovation output. Since the Biopharmaceutical Act of 2007 specifies the investment tax credit for R&D expenditures, this Act appears to encourage the approved biopharmaceutical firms to engage in innovation input. Several studies examine the sensitivity of R&D expenditure to R&D tax credit (Eisner, Albert, and Sullivan, 1984; Mansfield, 1986; Tillinger, 1991; Hall, 1993). Innovation activities require capital infusions and R&D expenditures. R&D expenditures are related to firm size: large firms can spend more on innovation. Firm size usually does not change substantially over time. Therefore, we first adopt R&D intensity, namely the ratio of R&D expenditure to total assets, as the quantitative measure of R&D investment (i.e. innovation input).

Second, we adopt the patent adjusted citation as the innovation quality measure.

The patent citation is the total number of citations received from all successful patents that are filed by a firm. Patent citations have been widely used to measure the innovation output (Griliches, 1990; Hsu, 2009; Hsu et al., 2014; Trajtenberg, 1990).

However, patent citations suffer from the inevitable truncation problem, under which later patents receive

14　 We adopt R&D/total assets as the measure of R&D intensity because it is the measure mostly used in the finance literature (Brown et al., 2017; Chen et al., 2013; Franzen, Rodgers, and Simin, 2007). We do not use R&D/sales because the amount of sales tends to fluctuate more than total assets over time, leading to unstable results.

15　 Cohen, Nelson, and Walsh (2000) propose that patents are more valuable for appropriating R&D returns in high R&D intensity firms such as pharmaceutical and medical instrument firms than in low R&D intensity firms. Thus, we adopt patents as the innovation measure for biopharmaceutical firms.

16　 In the early studies, patent count, which is the number of patents applied by a firm, is often used to

measure the quantity of innovation (Griliches, 1981). However, recently, most studies use patent

citations to measure the quality of innovation.

fewer citations because of their shorter time in existence. To prevent this problem, we follow Hall, Jaffe, and Trajtenberg (2001, 2005) to measure the patent adjusted citation by correcting the number of citations received by each patent by the application year and by the technology classification.

3.2.2 Other Control Variables

We follow previous studies in adopting several control variables for innovation in the regression analysis. These variables include firm size, the lagged effect of R&D expenditures, firm leverage, firm performance, and firm value. Bhattacharya and Bloch (2004) find that an increment in innovative activities is accompanied by an increment in firm size. By contrast, Shefer and Frenkel (2005) argue that there is a negative relation between firm size and R&D expenditure. Thus, we measure total assets and net sales as proxies for the firm’s size.

Hall (2002) finds that R&D-intensive firms use less debt financing because R&D investment has greater uncertainty and less collateral. Studies also find that the availability of financing influences R&D expenditures (Brown, Fazzari, and Petersen, 2009; Brown et al., 2012, and Hsu et al., 2014). Thus, we adopt the debt ratio, which is the total liability divided by total assets, to measure the firm’s leverage.

Hitt, Hoskisson, Ireland, and Harrison (1991) suggest that when a firm’s profitability increases, managers will become more risk-adverse and reduce intangible investment.

More recently, Greve (2003) and Chang et al. (2019) also find the same results. Hence, we adopt Return On Assets (ROA), which is the Earnings Before Interest, Taxes, Aepreciation, and Amortization (EBITDA) divided by the average of total assets, to measure a firm’s performance.

Tobin’s Q is the market value of equity plus the book value of long-term and short- term debts, divided by book assets. Tobin’s Q is usually used as the financial market- based measure of a firm’s performance (Wernerfelt and Montgomery, 1988). Connolly and Hirschey (2005) adopt Tobin’s Q to evaluate the firm’s value based on R&D expenditures and show that larger firms have a greater valuation effect from R&D. Aghion et al. (2013) also use Tobin’s Q to control for the influence of a firm’s market value on innovation.

17　 The technology classification of the PATSTAT database is based on the International Patent

Classification (IPC) system. Hall et al. (2001) classify the 3-digit IPC code into 6 main industrial

categories and use the simulated cumulated lag distribution of each category to calculate the

truncation adjusted citations.

R&D expenditure usually exhibits a cumulative effect, and previous studies suggest using lagged R&D expenditure or lagged R&D intensity as the variable to explain the innovation output such as patent count, patent citations, or patent adjusted citations (Artz, Norman, Hatfield, and Cardinal, 2010; Beck-Blease, 2011; Kong, 2020). Griliches (1990) finds that there is a lagged relation between patent and R&D expenditure. Artz et al. (2010) examine the effect of R&D, patent, and product innovation on the firm performance and consider the time lag effect for these variables in their regression model. In particular, Artz et al. (2010) set R&D as invested at year t-3 whereas the patent is granted at year t-2 and thus they define the time lag between R&D and patent as about one year. In addition, Beck-Blease (2011) and Kong (2020) use lag R&D one year in explaining the patent output. Thus, following prior literature, we use all control variables and R&D expenditure in year t-1 to explain the patent adjusted citations in year t.

Appendix Table A1 shows the definitions of all variables.

3.3 Methodology

3.3.1 Intra-industry Analysis and Inter-industry Analysis

We could directly and simply examine the innovation of the approved biopharmaceutical firms before and after the Biopharmaceutical Act to show the impact of the Biopharmaceutical Act. However, this methodology may neglect endogeneity problems, which means that changes in innovation may result from omitted variables bias, such as changes in the macroeconomic environment and other unobserved factors.

To eliminate the endogeneity concern, we identify control firms which have characteristics similar to those of approved biopharmaceutical firms (treated firms), and then compare the difference in innovation between these two groups.

Specifically, we perform an intra-

18　 We follow previous studies (Aghion et al., 2013; Becker-Blease, 2011) and use the patent application date to identify the year of the patent. This can reduce the time gap problem between innovation input and innovation output because there are usually 2 or 3 years between the patent application date and the publication (or granted) date. Therefore, several empirical studies (Aghion et al., 2013; Chang, Fu, Low, and Zhang, 2015; Chang et al., 2019) use the contemporaneous R&D to explain the patent number or patent citations without considering the time lag between R&D and patent.

19　 For example, unapproved biopharmaceutical firms may have the same innovation effect after Biopharmaceutical Act as the approved biopharmaceutical firms. Thus, we cannot infer that the Biopharmaceutical Act improves the innovation of approved biopharmaceutical firms.

20　 This is the concept of Difference-in-difference (DID) method, which is widely used to deal with the

endogeneity problem.

industry analysis, which examines the effect of Biopharmaceutical Act by comparing approved firms with unapproved firms (control firms) in the biopharmaceutical industry since only approved firms receive the benefits of the Act.

In addition, we conduct an inter-industry analysis by comparing high-tech firms and approved biopharmaceutical firms to confirm the effectiveness of policy. In Taiwan, the Biopharmaceutical Act grants tax credits only to biopharmaceutical firms, whereas the Statute for Upgrading Industries (SUI)（促進產業升級條例）, established in 1991, provides tax credits for all firms.

Yang et al. (2012) investigate the SUI and find that the tax credits have more effect on R&D for industries with greater R&D intensity. Based on this finding, Yang et al. (2012) suggest that the government should establish various tax credits across industries. In Taiwan, the Biopharmaceutical and high-tech industries are respective the first and second highest R&D intensity industries during the period 2007 to 2017.

Brown et al. (2017) argue that industries with greater R&D intensity are more likely to exhibit innovation underinvestment. The Biopharmaceutical Act is established to grant tax credits to the biopharmaceutical industry rather than other industries. Therefore, we choose the high-tech industry for comparison. Since the SUI grants all industries the same preferential tax treatment, comparing the Biopharmaceutical and high-tech industries helps to confirm the effectiveness of the Biopharmaceutical Act, which grants tax credits only to biopharmaceutical firms.

3.3.2 Propensity Score Matching (PSM)

To conduct the intra-industry and the inter-industry analyses, we adopt PSM to identify suitable control firms and prevent sample selection bias for treated firms.

Introduced by Rosenbaum and Rubin (1983), PSM is a widely used technique for finding control firms which have characteristics similar to those of treated firms. This method ensures that estimation bias is greatly reduced by comparing the outcome of innovation across treated and control firms.

Previous studies generally use logistic regression to estimate the propensity score

21　 The SUI is implemented from 1991 to 2019 whereas the Biopharmaceutical Act is implemented from 2007 to 2021.

22　 R&D intensity, the ratio of R&D expenditures to total assets, for the Biopharmaceutical and high-tech

industries is 4.08% and 3.04%, respectively.

for binary treatment (Rosenbaum and Rubin, 1983; Caliendo and Kopeinig, 2008). The propensity score is the predicted probability of a firm, P = Pr(D|X), given a vector of observed predictors X, where D equals one if the firm is an approved biopharmaceutical firm and zero otherwise. After the estimation of the propensity score of each firm, we adopt the nearest neighbor matching method to find the control firms for the treated firms.

Dehejia and Wahba (2002) allow control firms to be matched more than one because of the substantial difference in sample size between treated firms and untreated firms.

We select firms which have the nearest propensity score to the treated firm in each year to be control firms. In addition, to consider the sample size effect, we also choose firms which have the second, third, and fourth-nearest propensity scores to the treated firms to be control firms.

For the intra-industry analysis, we use the data of the year before the approval year to find control firms because the approval year (i.e. the event year) of each approved biopharmaceutical firm is different. We use the logistic regression to estimate the propensity score and then use the pre-event total assets, ROA, Tobin’s Q, and R&D intensity as the explanatory variables in the logistic regression. In addition, for the inter- industry comparison, we use a similar concept to identify control group firms from the high-tech industry.

3.3.3 Difference-in-differences (DID) Estimator

We apply the DID approach to examine the effect of Biopharmaceutical Act on innovation because previous studies argue that DID is a useful instrument for evaluating the impact of certain policies which may only influence one part of the population (Blundell and Costa-Dias, 2009; Buckley and Shang, 2002; Lechner, 2011; Heckman, Ichimura, and Todd, 1997). The DID approach helps to eliminate the endogeneity problem because it assumes that unmeasured factors, such as the changes in economic conditions or other unobservable effects, affect both treated and control groups in similar ways. The DID approach can thus reduce the influence of other factors that may contaminate our treatment of the effect of the Biopharmaceutical Act on innovation.

Following past literature on the basic DID approach, we first calculate the DID estimator. In our paper, the DID estimator calculates the effect of the Biopharmaceutical Act by estimating the difference in average innovation measures before and after the

23　 Lane, Looney, and Wansley (1986) use three matched firms for one treated firm to avoid possible

estimation bias.

approval of the Biopharmaceutical Act for both treated firms and control firms, and then comparing the difference between these two groups. Control firms are found using the PSM procedure. In this study, we have two control groups, one for the intra-industry and the other for inter-industry analyses. Finally, we use the t statistic to examine the significance of the DID estimator. The significance of the DID estimator can be used to explain that the innovation in approved biopharmaceutical firms is significantly different from the innovation in control firms (i.e. unapproved biopharmaceutical firms or high-tech firms) after the Biopharmaceutical Act.

3.3.4 Difference-in-differences (DID) Regression

The DID estimator may not be sufficient to explain the influence of the Biopharmaceutical Act because it does not consider the heterogeneous dynamics from other variables (Buckley and Shang, 2002). In addition, most previous studies conduct only DID regressions and do not use the DID estimator. Thus, we follow previous studies (Blundell and Costa-Dias, 2009; Buckley and Shang, 2002; Lechner, 2011) to simply incorporate possible factors into the linear regression to estimate the influence of the Biopharmaceutical Act.

The DID regression is:

Y

= α

+ β∙After

+ δ∙Treatment

+ γ∙After

× Treatment

+ π∙Control variables

+ Year fixed effect + ε

, (1)

where Y

denotes the measure of innovation of firm i in year t; After

= 1 if the firm is in or after the approval year and 0 otherwise; Treatment

= 1 if the firm is approved according to the Biopharmaceutical Act and 0 otherwise. The time period of this regression is from 2002 to 2017.

We respectively use R&D intensity and patent adjusted citations to measure the innovation activities in the regressions. We use firm size (natural logarithm of total assets;

Huang, 2019), lagged R&D expenditure (pre-year R&D expenditure), ROA, and debt ratio to explain the R&D investment (i.e. R&D intensity). When the patent adjusted citations are the innovation measure, we follow Lerner (1994) and Becker-Blease (2011) and use

24　 All results for DID regressions control for the year fixed effect. To save space, we do not show the

results for the year fixed effect in the tables.

the natural logarithm of 1+adjusted patent citations, LN (1+adjusted patent citation), as the dependent variable because a high proportion of the sample has a value of zero for the adjusted patent citations. To explain the adjusted patent citation, we use the natural logarithm of lagged net sales, lagged R&D expenditure, and lagged Tobin’s Q as control variables.

The coefficient γ of interaction term, After

× Treatment

is used to test whether the Biopharmaceutical Act changes innovation for approved biopharmaceutical firms. If the Biopharmaceutical Act improves innovation for approved biopharmaceutical firms, γ will be significantly positive.

4. Empirical Results

4.1 Summary Statistics

Table 1 presents the descriptive statistics, including the mean and median of all variables, for the four groups. The approved biopharmaceutical firms include only the data of biopharmaceutical firms in the year of approval. The other three groups: unapproved biopharmaceutical firms, biopharmaceutical industry, and high-tech industry, comprise firm-year observations from 2007 to 2017.

Table 1　Descriptive Statistics

Variables Approved

Biopharmaceutical Firms

Unapproved Biopharmaceutical

Firms

Biopharmaceutical

Industry High-tech Industry

Mean Median Mean Median Mean Median Mean Median

Patent 0.09 0.00 0.21 0.00 0.19 0.00 3.25 0.00

Adjusted Citation 0.04 0.00 0.09 0.00 0.08 0.00 1.07 0.00

R&D Expenditure 121.63 76.48 66.09 26.30 80.40 31.89 455.74 60.76

R&D intensity (%) 19.10 9.86 4.51 2.79 7.60 3.66 4.67 2.52

25　 These control variables are incorporated into regressions following previous studies (Klassen,

Pittman, Reed, and Fortin, 2004; Becker-Blease, 2011; Aghion et al., 2013; Chang et al., 2019). In

addition, we consider the lagged effect for the patent adjusted citations because the innovation process

takes time from the R&D input to patent application. Thus, following prior literature, we relate all

control variables in year t-1 to patent adjusted citations in year t.

Variables Approved Biopharmaceutical

Firms

Unapproved Biopharmaceutical

Firms

Biopharmaceutical

Industry High-tech Industry

Mean Median Mean Median Mean Median Mean Median

Total Assets 1,176.30 609.74 2,262.25 1,099.15 1,970.91 917.59 14,979.32 2,342.30

Debt Ratio (%) 19.76 9.64 36.38 34.30 31.86 29.16 44.34 39.60

Net Sales 108.21 21.19 1,442.01 692.92 1,074.53 411.20 17,141.92 2,006.76

ROA -18.23 -12.87 -0.21 8.67 -4.71 5.30 8.53 8.95

Tobin’s Q 4.32 2.00 1.69 1.23 2.24 1.31 1.13 0.86

N 85 85 1,311 1,311 1,817 1,817 11,309 11,309

Note: This table shows the descriptive statistics of the variables for the four groups. Except for observations (N), all numbers represent the mean of the variable. The first group consists of approved biopharmaceutical firms, which are the observations of biopharmaceutical firms at the year of approval (only the event year). Unapproved biopharmaceutical firms are the firm- year observations of biopharmaceutical firms which are not approved by the Biopharmaceutical Act, from 2007 to 2017. Biopharmaceutical industry and high-tech industry are the firm-year observations from 2007 to 2017. All variables are defined in Appendix Table A1.

First, in the biopharmaceutical industry, we find that the approved biopharmaceutical firms have smaller total assets, debt ratio, net sales, ROA, but higher Tobin’s Q than unapproved biopharmaceutical firms. In addition, approved biopharmaceutical firms have higher R&D expenditure and R&D intensity, but lower patent numbers and patent adjusted citations than unapproved biopharmaceutical firms. The median of patent and patent adjusted citation is zero, implying that most biopharmaceutical firms do not obtain innovation output such as patents. The median of R&D expenditure and R&D intensity is lower than the mean ones in the biopharmaceutical firms, implying only some or few biopharmaceutical firms may infuse greater R&D expenditures while most biopharmaceutical firms put relatively low amounts of money into R&D activities. Accordingly, compared with unapproved biopharmaceutical firms, approved biopharmaceutical firms tend to be smaller in firm size, leverage, and operating performance, but have higher firm value and higher innovation input but lower innovation output.

Table 1　Descriptive Statistics (cont.)

Second, in the inter-industry comparison, firms in the biopharmaceutical industry have smaller firm size, lower leverage, net sales, operating performance, and higher firm value than firms in the high-tech industry. In addition, compared with those in high- tech industry, firms in the biopharmaceutical industry have lower patent numbers, patent adjusted citations, and R&D expenditure, i.e. lower innovation input and output. However, firm sizes (i.e. total asset and/or net sales) of the biopharmaceutical industry are much smaller than those of firms in the high-tech industry, giving biopharmaceutical firms a higher proportion of R&D expenditure to firm size.

4.2 Effectiveness of PSM

To examine whether PSM helps to prevent the selection bias problem, we compare the pre-event firm characteristics between the treated firms (i.e. approved biopharmaceutical firms) and control firms. Table 2 presents the comparison of the pre- event firm characteristics between these two groups. The data of firm characteristics are at year t-1 where t is the approval year of the treated firms. This table shows one, two, three, and four control firms which have the first, second, third, and fourth-nearest propensity scores to the treated firms.

In Table 2, Panel A shows the comparison between approved biopharmaceutical firms and unapproved biopharmaceutical firms. In the scenario of one control firm, all characteristics for the treated firms and control firms are not significantly different.

Except for the ROA for the second, third, and fourth control firms, other characteristics of these firms are also not significantly different. Accordingly, among the unapproved biopharmaceutical firms, the control firms selected by PSM are quite similar to the approved biopharmaceutical firms. Panel B of Table 2 shows the comparison between approved biopharmaceutical firms and high-tech firms. Among the high-tech firms, the control firms selected by PSM are similar to the treated firms in ROA and R&D intensity.

26

Overall, by assuring that the pre-event firm characteristics of the treated firms and control firms are similar, we confirm that PSM alleviates the sample selection problem of control firms.

26　 Since the firm characteristics in different industries are quite different, we do not obtain similar results

for total assets and Tobin’s Q.

Table 2 　 Effectiveness of Propensity Score Matching Panel A Comparison of Firm Characteristics between Approved Biopharmaceutical Firms and Unapproved Biopharmaceutical Firms One Matched Firm Two Matched Firms Three Matched Firms Four Matched Firms Treated Control Diff. Treated Control Diff. Treated Control Diff. Treated Control Diff. N 68 68 68 136 68 204 68 272 Total Assets 862.21 819.55 42.66 (0.54) 862.21 827.48 34.73 (0.53) 862.21 861.47 0.74 (0.99) 862.21 848.22 13.99 (0.78) ROA -19.78 -15.02 4.76 (0.27) -19.78 -12.54 7.24* (0.05) -19.78 -10.61 9.17*** (0.01) -19.78 -9.96 9.82*** (0.00) Tobin’ s Q 2.75 2.34 0.40 (0.61) 2.75 2.14 0.61 (0.40) 2.75 2.1 1 0.64 (0.35) 2.75 2.15 0.60 (0.37) R&D intensity (%) 14.32 15.17 0.85 (0.64) 14.32 14.37 0.05 (0.97) 14.32 13.96 0.36 (0.76) 14.32 13.12 1.20 (0.31) Panel B Comparison of Firm Characteristics between Approved Biopharmaceutical Firms and High-tech Firms One Matched Firm Two Matched Firms Three Matched Firms Four Matched Firms Treated Control Diff. Treated Control Diff. Treated Control Diff. Treated Control Diff. N 68 68 68 136 68 204 68 272 Total Assets 862.21 722.38 139.83* (0.05) 862.21 681.46 180. 75** (0. 01) 862.21 683.07 179. 14** (0. 01) 862.21 691.62 170. 59** (0. 01) ROA -19.78 -19.13 0.66 (0.81) -19.78 -18.55 1.23 (0.57) -19.78 -18.26 1.52 (0.48) -19.78 -16.97 2.81 (0.19) Tobin’ s Q 2.75 1.84 0.90 (0.10) 2.75 1.78 0.97* (0.08) 2.75 1.73 1.02* (0.07) 2.75 1.74 1.01* (0.08) (R&D/T A)*100 14.32 17.95 3.63 (0.14) 14.32 18.55 4.22** (0.02) 14.32 16.65 2.33 (0.15) 14.32 16.63 2.30 (0.16) No te: This table presents the comparison of pre-event firm characteristics between the approved biopharmaceutical firms and control firms. The data of firm characteristics are at year t-1 where t is the approval year of treated firms. This table shows one, two, three and four control firms which have the first, second, third, and fourth-nearest propensity scores with the treated firms. The variables are defined in Table 1. Panel A shows the comparison between approved biopharmaceutical firms and unapproved biopharm ac eut ical firm s. Panel B shows the compar ison bet ween approved biophar mac eut ical firm s and high-t ech firm s. Diff. is the mean difference of variable (shown in ab so lute va lue ) b etw ee n the tre ate d firms an d co ntro l firms. Nu mb ers in pa re nth ese s are p -va lue s. *** ,** , a nd * de no te sig nifica nce a t th e 1% , 5 % , a nd 10 % le ve ls, respectively .

4.3 Intra-industry Analysis

4.3.1 Difference-in-differences Estimator (DID Estimator)

Table 3 presents the DID estimator of innovation for intra-industry analysis. In order to consider the time span effect of the Biopharmaceutical Act, we consider the time interval of the pre-event and post-event year from (t-1, t+1) to (t-3, t+3) where t is the event year, i.e. the approval year, when the biopharmaceutical firm is approved by the Biopharmaceutical Act. Panel A and B present the DID estimator of the R&D investment and the DID estimator of the patent adjusted citations, respectively.

Table 3　DID Estimator: Intra-industry Analysis

Panel A: DID Estimator of R&D Investment: Intra-industry Analysis

t-1 t+1 Differences t-2 t+2 Differences t-3 t+3 Differences

Treated 14.3224 14.7030 0.3806 13.5214 14.6223 1.1010 13.0075 13.4002 0.3927

(0.8686) (0.5928) (0.8183)

Control 1 15.1741 10.3605 -4.8135** 13.6871 10.7748 -2.9123 11.8360 10.5049 -1.3311

(0.0222) (0.1002) (0.3658)

Control 2 14.3681 10.5547 -3.8134** 12.4338 10.1346 -2.2992* 10.9985 9.7941 -1.2043

(0.0216) (0.0685) (0.2863)

Control 3 13.9605 11.2267 -2.7338** 11.6300 10.8426 -0.7875 10.3005 10.3891 0.0886

(0.0392) (0.4661) (0.9234)

Control 4 13.1208 10.3647 -2.7561** 11.2082 10.1346 -1.0736 10.2415 9.7688 -0.4726

(0.0231) (0.2412) (0.5553)

Diff.1 -0.8517 4.3425 5.1942** -0.1657 3.8476 4.0133* 1.1715 2.8953 1.7238

(0.0394) (0.0956) (0.3540)

Diff.2 -0.0458 4.1483 4.1941* 1.0876 4.4877 3.4002 2.0091 3.6061 1.5970

(0.0646) (0.1148) (0.3541)

Diff.3 0.3619 3.4763 3.1144 1.8914 3.7798 1.8884 2.7070 3.0111 0.3041

(0.1319) (0.3653) (0.8500)

Diff.4 1.2016 4.3383 3.1367 2.3132 4.4877 2.1745 2.7661 3.6314 0.8653

(0.1240) (0.2695) (0.5662)

Panel B: DID Estimator of Adjusted Patent Citation: Intra-industry Analysis

t-1 t+1 Differences t-2 t+2 Differences t-3 t+3 Differences

Treated 0.0253 0.0648 0.0395 0.0348 0.0477 0.0129 0.0689 0.0796 0.0107

(0.2171) (0.5468) (0.7261)

Panel B: DID Estimator of Adjusted Patent Citation: Intra-industry Analysis

t-1 t+1 Differences t-2 t+2 Differences t-3 t+3 Differences

Control 1 0.1224 0.0219 -0.1004 0.0617 0.0150 -0.0467 0.0099 0.0158 0.0060

(0.1188) (0.2750) (0.7092)

Control 2 0.1651 0.0162 -0.1489*** 0.0685 0.0128 -0.0557* 0.0300 0.0289 -0.0011

(0.0068) (0.0549) (0.9294)

Control 3 0.1270 0.0176 -0.1094*** 0.0585 0.0266 -0.0319 0.0419 0.0483 0.0064

(0.0043) (0.1624) (0.6652)

Control 4 0.1239 0.0207 -0.1033*** 0.0687 0.0356 -0.0331 0.0670 0.0566 -0.0105

(0.0013) (0.1385) (0.6643)

Diff.1 -0.0971 0.0429 0.1400** -0.0269 0.0327 0.0596 0.0590 0.0638 0.0048

(0.0480) (0.2044) (0.8852)

Diff.2 -0.1398 0.0486 0.1884*** -0.0337 0.0349 0.0686** 0.0389 0.0507 0.0118

(0.0024) (0.0499) (0.7213)

Diff.3 -0.1017 0.0472 0.1489*** -0.0237 0.0211 0.0448 0.0270 0.0313 0.0043

(0.0024) (0.1045) (0.8934)

Diff.4 -0.0986 0.0442 0.1428*** -0.0339 0.0121 0.0460 0.0018 0.0230 0.0212

(0.0013) (0.1052) (0.5759)

Note: This table presents the DID estimator of innovation for intra-industry analysis. Panels A and B present the DID estimator of the R&D investment and the DID estimator of adjusted patent citations, respectively. t is the event year, i.e. the year in which the Biopharmaceutical firm is approved by the Biopharmaceutical Act. Treated represents the treated firms, i.e. approved biopharmaceutical firms. Control 1, Control 2, Control 3, and Control 4 respectively represent one, two, three, and four control firms matched to each treated firm. The control firms in the intra-industry analysis are unapproved biopharmaceutical firms. Diff.1, Diff.2, Diff.3, and Diff.4 represent the mean difference of variables between Treated and Control 1, Control 2, Control 3, and Control 4, respectively. Numbers in parentheses are p-values. ***,**, and * denote significance at the 1%, 5%, and 10% levels, respectively.

In Panel A of Table 3, the approved biopharmaceutical firms do not change R&D investment significantly after the approval year. However, when we consider the time interval (t-1, t+1), the unapproved biopharmaceutical firms exhibit significantly lower R&D investment after the approval year. The DID estimators of the R&D investment for one and two control matched firms are significant. This result implies that compared with unapproved biopharmaceutical firms, approved biopharmaceutical firms have a significantly higher proportion of R&D expenditures to total assets after the

Table 3　DID Estimator: Intra-industry Analysis (cont.)

Biopharmaceutical Act.

Panel B of Table 3 shows that approved biopharmaceutical firms do not significantly change their adjusted patent citations, while unapproved biopharmaceutical firms experience significantly reduced adjusted patent citations for the time interval (t-1, t+1).

In this short time interval, the DID estimators of adjusted patent citations are significantly positive, implying that the approved biopharmaceutical firms have significantly higher innovation output than unapproved biopharmaceutical firms after the Biopharmaceutical Act.

Accordingly, the results of the DID estimators show that compared to unapproved biopharmaceutical firms, the Biopharmaceutical Act encourages approved biopharmaceutical firms to increase their input into innovation activities, leading to higher innovation quality. In addition, both panels of Table 3 show that there are no significant DID estimators for the time intervals (t-2, t+2) and (t-3, t+3), implying that the influence of the Biopharmaceutical Act on innovation input and output has only a short-term effect.

This result is consistent with David et al. (2000), who find that the recipients of tax credits tend to concentrate on projects with short-term prospects.

4.3.2 Difference-in-differences Regression (DID Regression)

To obtain more accurate results, we conduct the DID regression by additionally

considering the heterogeneous dynamics of other variables for innovation measures. Table

4 shows the DID regression results for the intra-industry analysis. Panel A of Table 4

shows the DID regression results for R&D investment. For matched firms the significantly

negative coefficients of After show that all biopharmaceutical firms reduce their R&D

investments after the Biopharmaceutical Act. In addition, the significantly negative

coefficients of Treatment indicate that the approved biopharmaceutical firms on average

have lower R&D investments than the unapproved biopharmaceutical firms. Further,

the significantly positive coefficients of the interaction term, After×Treatment, show

that compared with unapproved biopharmaceutical firms, approved biopharmaceutical

firms have significantly higher R&D investments after the approval. By combining the

coefficient results of Treatment and After×Treatment, we find that relative to control

firms, the Biopharmaceutical Act encourages the group of treated firms, which have

lower R&D intensity, to improve their R&D input. Accordingly, unlike the unapproved

biopharmaceutical firms, the approved biopharmaceutical firms, which respond to the

exogenous shock of the Biopharmaceutical Act and receive its benefits, are induced to

improve their input into innovation activities.