China’s IPR regime and provincial patenting activity

ABSTRACT

This paper examines whether regions in China with stronger IPR regimes have better patenting performance, and whether they exhibit positive spillovers on neighbouring regions’ patenting. The results affirm positive and significant effects on both regards. This is likely due to the fact that a stronger IPR regime with its accompanying hard and soft knowledge infrastructure facilitates knowledge sharing and exchange, hence promoting innovation efficiency. The negative spillover effect of neighbour’s patenting activity may reflect inter-regional R&D collaboration, where patents are appropriated only by some participants. The results are robust across specifications.

KEYWORDS:

• Intellectual property rights
• innovation
• China
• knowledge production function

JEL CLASSIFICATION:

• H11
• O31
• O33

I. Introduction

With the fast increase in R&D inputs in China in recent years, we see a corresponding rise in innovation output. The high technological content output is mainly consolidated in the form of invention patent. To understand this patenting activity in China, we depart from the New Growth Theory and Romer’s (1990) knowledge production function (KPF), and consider heterogeneity in the strength of intellectual property rights (IPR) protection, while controlling for economic structure in various aspects, such as the share of non-state sector in manufacturing, the level of the rule of law, and the maturity of the factor market. With certain capital and labour inputs, those factors exhibit an influence on the efficiency of knowledge creation by their respective channels. There are two contradicting forces concerning the role played by IPR regime in innovation efficiency; positive effects such as the facilitated disclosure of technologies by hard and soft infrastructure and, negative effects such as increasing the cost for accessing the already published knowledge by enforcing more rigid and complicated legal constraints.

On the positive side Hall and Harhoff (2012), Brusoni et al. (2006), Landes and Posner (2003), Menell (1999), Denicolo and Franzoni (2003), Kitch (1977), and Machlup and Penrose (1950), conclude that stronger IPR protection contributes to more and better disclosure of the already developed technology, hence to its dissemination and improved innovation performance. On the negative side, a higher level of IPR protection further limits the accessibility of the already available knowledge stock, lowering the efficiency of R&D investment, thus lowering growth. Others, such as Magerman, Van Looy, and Debackere (2015), conclude that patenting hampers the dissemination of research outcomes, while Murray and Stern (2007) discovered an ‘anti-commons’ effect showing that citation rate declines on average by 10 to 20% after a patent is granted.

In this paper, we look into the effect of the strength of China’s provincial IPR regimes on patenting activity, particularly with invention patents. Other variables mainly serve as control variables. Possible endogeneity caused by omitted variables and reverse causality are limited by employing fixed effect, IV, and spatial regressions. Different specifications and diverse sets of control variables are used to check the robustness of the results. It is unlikely there will be a significant amount of innovators hiding their cutting edge inventions by not patenting them in today’s competitive environment. As a result, patenting activity translates to innovation performance.

Considering positive and negative effects on knowledge dissemination by IPR regimes, we assume that the effects are not only locally valid but also affecting nearby regions. This assumption is tested in our benchmark model, followed by a spatial model. Our paper introduces geography as an additional component, enabling us to examine how knowledge production in one region is related to IPR protection in surrounding regions. Results confirm a positive relation between the strength of IPR regime and patenting activity regionally, and with positive spillover to neighbour’s patenting. Using different sets of explanatory variables and specifications, the effect of IPR protection remains the same, indicating strong robustness of the result.

The spillover effect by neighbour’s patenting activity on local patenting is negative, this is reasonable because closer regions tend to collaborate in their R&D efforts more than distant ones (Marek et al. 2017), and researchers are competing to patent similar new technologies in both business and academic environments, meaning one’s success entails others’ loss; also, patents are sometimes attributed to the most significant participants in collaboration. For this line, we suggest further research.

Section 2 discusses the model, data and variables. Section 3 presents the results, and section 4 concludes the paper.

II. Model specification, data and variables

The benchmark model:

(1) $Ln(P{I}_{i,t})={\beta }_0+{\beta }_{1}IP{R}_{i,t}+{\beta }_{2}Ln(S{O}_{i,t})+{\beta }_{3}Ln(Funds\mathrm{\_}ad{j}_{i,t})+{\beta }_{4}Ln(RD{P}_{i,t})+{\beta }_{5}Ln(NS{O}_{i,t})+{\beta }_{6}M{F}_{i,t}+{\beta }_{7}M{L}_{i,t}+T_{i,t}+{\epsilon }_{i,t}$(1)

where PI is the number of invention patent applications.¹ IPR is an index for the strength of IPRs protection. To capture the effects of knowledge exchange we use social openness (SO) to account for the level of interaction and exchange of a region, Funds_adj is real R&D funding, RDP is number of research personnel full time equivalent, and NSO is the market share of non-state firms in manufacturing industry. MF represents factor market maturity, and ML the market rule of law index. The index of factor market maturity shows the level of marketization and supply of financial services, human resource, and technology. The index of the market rule of law shows how good legal protection and law enforcement are regarding the maintenance of market order. These two variables relate to PI by influencing transaction cost, the ease of utilizing resource, and the stability of the business environment. We do not take log for IPR due to the scale of the variable compared to other variables, same with MF and ML. Subscripts i and t index province and year. The one and two period lags of IPR are also added in a separate regression to check model specification and robustness. $T$ is time dummy. $\epsilon$ may refer to a white noise or individual fixed effect plus white noise depending on the model.

Spillover of the strength of neighbouring IPR regime is taken into account by the estimation of the benchmark spatial regression, also considering the spillover effect of PI:

(2) $Ln(P{I}_{i,t})={\beta }_0+{\beta }_{1}IP{R}_{i,t}+{\beta }_{2}Ln(S{O}_{i,t})+{\beta }_{3}Ln(Funds\mathrm{\_}ad{j}_{i,t})+{\beta }_{4}Ln(RD{P}_{i,t})+{\beta }_{5}Ln(NS{O}_{i,t})+{\beta }_{6}M{F}_{i,t}+{\beta }_{7}M{L}_{i,t}+{\rho }_{1}W\left(IPR\right)+{\rho }_{2}W\left(PI\right)+T_{i,t}+{\epsilon }_{i,t}$(2)

where ρ is the spatial lag coefficient, and $W$ is a block diagonal NT$\times$NT row-standardized inverse distance matrix. We use the geographic distance of two provincial capitals as the distance between provinces. For the effect of a change in IPR protection may last for more than one period, we also take IPR one period lag, and IPR two period lag as independent variables to check model robustness.

Maturity of the factor market (MF), and market rule of law (ML) control for two important aspects, which may affect innovation performance due to their implication in resource allocation and companies’ business decision. MF mainly concerns transaction efficiency, and ML market stability. They are control variables used in some specifications for our purpose. Since when incorporating the two lags of IPR, they are all insignificant at 10% level, we use IPR two period lag to instrument IPR in an IV estimation. Time dummies are included in all specifications.

Statistical models used for estimation are fixed effect (FE) regression, spatial Durbin model with FE (SDM FE), SDM with random effect (SDM RE), IV regression, and spatial auto-regressive model with auto-regressive disturbance (SARAR) with different sets of control variables.

Data

The estimations are based on a balanced panel data set covering 31 of China’s provincial level jurisdictions for the period of 2008–2016, which yields a total of 279 observations. The data are drawn from National Bureau of Statistics of China, Wind, China Regional Openness Index Report 2018, China Marketization Index Report, and China Intellectual Property Index Report. Time dummies (T_year) capture the expanding stock of knowledge and the intercept effect of the knowledge stock. Descriptive statistics are reported in Table 1.

Table 1. Descriptive statistics

Variable	Mean		Standard deviation		Range
	Std. Spec.	Log	Std. Spec.	Log	Std. Spec.	Log
PI	17 786.789	8.78	28 831.601	1.68	184,593	8.46
IPR	0.17	−2.13	0.16	0.96	0.82	6.70
RDP	67 065.43	10.21	92 221.13	1.71	451,866,00	10.08
Funds_adj	1 582 117.00	13.31	2 207 320.00	1.73	11 963 690.00	9.53
SO	19.81	2.86	10.91	0.49	49.01	2.58
NSO	6.46	1.77	2.33	0.47	9.89	2.44
MF	4.91	1.50	2.38	0.56	13.44	4.71
ML	4.83	1.46	2.57	0.65	14.35	4.10

Source: The authors

III. Results

Table 2 reports the result. IPR has a positive sign in all nine specifications. It is significant at 1% level with (2), (3), (7), (9), significant at 5% with (6), and 10% with (4). Considering that the coefficient of IPR is all with the same sign and significant at least at 10% level in six out of nine specifications, we can confidently say that IPR protection regime exhibits a positive effect on innovation efficiency locally. In all four SDM, the spillover effects of IPR are positive and significant at 1% level. The positive effect of spatial spillover of IPR is confirmed. In all SDM, the spillover effect of local patenting on neighbour’s is significantly negative at least at 10% level, with one at 10%, two at 5%, and one at 1%. This, we believe, shows again the empirical evidence that closer regions tend to collaborate in innovation projects and that researchers are competing to patent similar new technologies in both business and academic environments, meaning one’s success entails others’ loss; also, patents are sometimes attributed to the most significant participants in collaboration.

Table 2. Different specifications of modelling IPR

	(1) FE	(2) SDM FE	(3) SDM RE	(4) SDM FE	(5) SDM RE	(6) IV ipr_2lag	(7) IV ipr_2lag	(8) SARAR	(9) SARAR
lnRDP	–0.0888	–0.122	0.0131	–0.336***	–0.110	0.134	0.0589	0.296**	0.246**
	(–0.91)	(–1.42)	(0.14)	(–3.35)	(–0.95)	(0.79)	(0.41)	(2.33)	(2.15)
lnFund_adj	0.147	0.166*	0.346***	0.197**	0.435***	0.548***	0.508***	0.454***	0.366***
	(1.41)	(1.82)	(3.54)	(1.96)	(3.96)	(3.67)	(3.86)	(3.83)	(3.36)
lnSO	0.799***	0.950***	1.045***	1.150***	1.112***	0.636***	–0.134	0.733***	0.244**
	(4.34)	(5.75)	(6.72)	(5.50)	(5.25)	(4.25)	(–0.89)	(5.95)	(2.11)
lnNSOE	–0.00497	0.104	0.166	0.0150	0.175	–0.262**	–0.283***	–0.185**	–0.209***
	(–0.03)	(0.78)	(1.22)	(0.09)	(1.01)	(–2.49)	(–3.08)	(–2.35)	(–2.75)
ipr	0.371	0.594***	0.651***	0.469*	0.495	0.890**	2.078***	0.200	1.023***
	(1.46)	(2.60)	(2.68)	(1.76)	(1.63)	(2.42)	(5.73)	(0.76)	(3.94)
ipr1lag				0.301	0.353
				(1.03)	(1.04)
ipr2lag				0.310	0.350
				(1.15)	(1.13)
MF	0.0316	0.0267	0.0390*	0.0506**	0.0648***		0.130***		0.104***
	(1.40)	(1.35)	(1.91)	(2.49)	(2.99)		(4.46)		(4.37)
ML	0.0165	0.0161	0.0344**	0.0183	0.0393**		0.117***		0.103***
	(1.04)	(1.15)	(2.30)	(1.20)	(2.31)		(5.18)		(5.67)
2008.year	-							–	–
2009.year	0.0974	0.238***	0.122					–0.0732	0.132
	(1.25)	(2.79)	(1.42)					(–0.64)	(1.17)
2010.year	0.277***	0.568***	0.426***			-	-	0.129	0.508***
	(3.18)	(4.22)	(3.34)					(0.97)	(3.93)
2011.year	0.477***	0.990***	0.665***	0.563***	0.307***	0.114	0.119	0.198	0.565***
	(4.49)	(4.46)	(3.35)	(4.54)	(2.68)	(0.81)	(1.06)	(1.47)	(4.27)
2012.year	0.627***	1.338***	0.860***	1.001***	0.533***	0.217	0.0853	0.275**	0.524***
	(4.85)	(4.52)	(3.30)	(4.78)	(2.86)	(1.55)	(0.73)	(2.07)	(3.85)
2013.year	0.787***	1.572***	0.978***	1.344***	0.715***	0.358**	0.195	0.402***	0.612***
	(5.13)	(4.22)	(3.05)	(4.48)	(2.79)	(2.44)	(1.63)	(2.77)	(4.41)
2014.year	0.919***	1.749***	1.079***	1.590***	0.845***	0.471***	0.145	0.514***	0.594***
	(5.60)	(4.19)	(3.03)	(4.49)	(2.82)	(3.04)	(1.08)	(3.39)	(4.13)
2015.year	1.071***	1.926***	1.222***	1.814***	1.017***	0.624***	0.318**	0.686***	0.783***
	(6.24)	(4.12)	(3.09)	(4.39)	(2.93)	(3.99)	(2.22)	(4.55)	(5.26)
2016.year	1.214***	2.235***	1.457***	2.182***	1.280***	0.739***	0.520***	0.780***	0.941***
	(6.55)	(4.30)	(3.29)	(4.60)	(3.19)	(4.73)	(3.53)	(5.14)	(6.06)
_cons	4.538***		2.540		4.632**	–2.142***	0.613	–2.430***	–0.636
	(4.14)		(1.39)		(2.05)	(–5.39)	(1.28)	(–6.65)	(–1.53)
Spatial Rho one		5.546*** (3.89)	5.201*** (3.46)	4.843*** (3.40)	4.466*** (2.84)
Spatial Rho two		–0.612**	–0.443*	–0.949***	–0.667**
		(–2.39)	(–1.94)	(–3.09)	(–2.45)
Variance sigma2_e		0.0567*** (11.69)	0.0694*** (10.47)	0.0420*** (10.18)	0.0572***
					(8.74)
lgt_theta			–1.805***		–1.820***
			(–8.00)		(–7.59)
lambda_cons						0.250*** (2.86)	0.0420 (0.51)	0.00554 (0.11)	–0.00401 (-0.09)
N	279	279	279	217	217	217	217	279	279

Note: t statistics in parentheses^{. *} p < 0.1, ** p < 0.05, *** p < 0.01. Lambda_cons equals Spatial rho. Equations (1)(1) $Ln(P{I}_{i,t})={\beta }_0+{\beta }_{1}IP{R}_{i,t}+{\beta }_{2}Ln(S{O}_{i,t})+{\beta }_{3}Ln(Funds\mathrm{\_}ad{j}_{i,t})+{\beta }_{4}Ln(RD{P}_{i,t})+{\beta }_{5}Ln(NS{O}_{i,t})+{\beta }_{6}M{F}_{i,t}+{\beta }_{7}M{L}_{i,t}+T_{i,t}+{\epsilon }_{i,t}$(1) – (12) use logPI.

The one period and two period lags of IPR are insignificant at 10% level in both SDM FE and SDM RE. So we interpret the models without the one and two period lags as preferable, and at the same time we may use the lag of IPR to instrument IPR. In the two IV regressions, MF and ML are significant at the 1% level, thus the model with these two variables as control is preferred, and in this case IPR is significant at 1% level instead of 5% with no MF and ML included. Also note in (7), the magnitude and the t-value are both more than two times those in (6).

The sign of SO is always positive except in (7), in which case it is also insignificant. Besides (7) and (9), SO is positive and significant at 1% level, and in (9), it is significant at 5%. This is evidence suggesting that SO and interchange have a critical role in promoting innovation efficiency in China. The signs of MF and ML are all positive in all cases, and significant at least at 10% in nine cases out of fourteen. This suggests the improvement of factor market and the market rule of law has positive effect on innovation efficiency due to lower transaction costs, the stability of the market, and the more efficient utilization of resources. The signs of capital input, Fund_adj, are all significant at least at 10% level except (1). They are significant at 1% level in six cases, indicating robust positive effect on the output of innovation. Finally, it is interesting to see that in all nine cases, the coefficients of the time dummies are in an increasing trend with increasing significance level. This shows the effect of the accumulation of knowledge, and its effect on innovation efficiency.

Regions with a strong IPR regime exhibit a positive spillover towards neighbours. Figures 1 and 2 group Chinese provinces according to the strength of IPR regime and innovation patent activity respectively.

Figure 1. Strength of IPR regime.

Figure 2. Innovation patents (10,000’s).

A stronger IPR regime with its accompanying knowledge infrastructure is on the whole conducive towards innovation.

IV. Concluding remarks

This paper introduces geography to a KPF studying the relationship between IPR regimes and patenting performance in China, thus allowing us to examine how knowledge production in one region is related to IPR protection in surrounding regions. The results suggest that regions with a stronger IPR regime have better innovation performance and also exhibit a positive spillover effect. The negative spillover effect of patenting activity on neighbour’s is a sign of both technology collaboration and competition.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

¹ See Jaffe and Palmer (1997) and Ulku (2007).