Economic Modelling 36 (2014) 108–119
Contents lists available at ScienceDirect
Economic Modelling journal homepage: www.elsevier.com/locate/ecmod
Do business and public sector research and development expenditures contribute to economic growth in Central and Eastern European Countries? A dynamic panel estimation☆ Monica Ioana Pop Silaghi ⁎, Diana Alexa, Cristina Jude, Cristian Litan Babeș-Bolyai University, Faculty of Economics and Business Administration, Cluj-Napoca, Str. Theodor Mihali No. 58–60, 400591, Romania
a r t i c l e
i n f o
Article history: Accepted 29 August 2013 JEL classification: O32 O33 O52 Keywords: Economic growth Research and development Human capital Central and Eastern European Countries Generalized Methods of Moments estimator
a b s t r a c t This paper empirically estimates the role of private and public research and development in explaining growth of Central and Eastern European Countries (CEE) during 1998–2008. We employ a dynamic panel model using the Arellano–Bond's Generalized Methods of Moments (GMM). Our findings suggest that a 1% increase in business R&D intensity boosts economic growth by 0.050 (0.213) % in these countries in the short (long) run. Public R&D is found to be statistically insignificant. When introducing human capital in the regression, the contribution of business R&D to economic growth decreases, although it remains significant. We argue that part of its effect may be accounted for by human capital. While various robustness checks are performed (such as adding different control variables, sub-periods and dummies for the entrance years to the EU), most of the results imply significant business R&D coefficient. Some policy implications are addressed based on our results. © 2013 Elsevier B.V. All rights reserved.
1. Introduction The aim of the present paper is to assess the role of public and private R&D stock (expressed as percentages of GDP) for the economic growth of all 10 new member states of European Union (EU), the CEE countries1 during 1998–2008.2 In the EU enlargement context, our focus on the CEE countries is of real interest, given the specificity of this group of members. After the fall of communism, R&D intensities (i.e. R&D expenditures expressed as percentages of GDP) have been very low in these countries, as they experienced
☆ We thank the participants of the annual conferences EEFS and INFER for their useful suggestions, to Prof. Subrata Ghatak† and to Prof. Vincent Daly, Kingston University, London, UK for their helpful comments and remarks. We also thank the anonymous referees for their constructive suggestions which helped us to improve the paper. The usual disclaimer applies. ⁎ Corresponding author. Tel.: +40 722837987 (mobile). E-mail address:
[email protected] (M.I. Pop Silaghi). 1 Bulgaria, Czech, Estonia, Latvia, Lithuania, Hungary, Poland, Romania, Slovenia, and Slovakia. 2 The choice for the period is motivated by the fact that starting with the year 1998, positive dynamics for R&D could be found for most of the countries in the sample. To avoid the effects of the highly volatile data of the crisis period on our estimations, we also limit our analysis up to the year 2008. 0264-9993/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.econmod.2013.08.035
recession. In the last decade, the dynamics of R&D spending observed in many CEE countries are positive. Fig. 1 (see Appendix 1) presents the average R&D intensity for the period 1998–2008 for all CEE countries in comparison to EU27. However, most of the countries have R&D intensity below 1% of GDP, lower than one might expect given their income level (Kravtsova and Radosevic, 2012). Slovenia and the Czech Republic are the only countries accounting for higher shares of R&D expenditure (1.44% and respectively 1.3% of GDP on average). Within the Lisbon strategy (The Lisbon Review, 2004), the EU set the ambitious goal of becoming “the most competitive and dynamic knowledge-based economy in the world” by allocating 3% of GDP to R&D link spending (with 2/3 realized by the private spending). Nevertheless, by 2010, these goals were far from being achieved. More recently, the Europe 2020 strategy “for smart, sustainable and inclusive growth” established the same target of investing 3% of GDP in R&D (European Commission, 2010). Among the initiatives, the Commission invites the Member States to prioritize “growth-enhancing items” such as education and skills, R&D, innovation, and infrastructure. Given the Lisbon objectives, the R&D intensities in CEE countries are low. The weak R&D intensities can partially be explained by the fact that the research systems of these countries still largely depend on public funding, which is sometimes volatile and under restrictive conditions.
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119
Therefore, the need for European Union is to focus not only on the amount of R&D but also on its composition and one strong recommendation of the EU Commission is to improve the conditions for private R&D sector in the union. The above presented aspects enhance the motivation of our work. It is an ongoing debate on how to allocate between different types of R&D, such as public and private. Fig. 2 (Appendix 1) shows the changes of the two types of R&D between 1998 and 2008. The distribution of R&D expenditures by public and private funds has evolved quite differently. On average, the dynamics of private R&D has surpassed public R&D, in line with the Lisbon objectives. However, the increase in private R&D is very heterogeneous among the CEE countries: significant increase in Estonia and Slovenia, while Romania, Slovakia and Poland show a decrease. The high shares of public R&D in the latter group of countries partly compensate the weak business R&D intensity. Nevertheless, compared with the modest levels of R&D investment, all CEE countries are characterized by relatively high levels of human capital. Despite a certain mismatch between supply and demand of labor in CEE countries, an important level of human capital may favor social returns of R&D. To the best of our knowledge there is no panel study to assess the role of R&D for the economic growth for the whole group of CEE countries.3 Our original approach relies in splitting between private and public R&D, which brings an important contribution to the existing literature. Moreover, since possible complementarity between human capital and R&D is depicted in the growth theory (see Redding, 1996; Romer, 1990), we also include human capital in our empirical setting. To accomplish our goals, we employ a dynamic panel estimation using the Arellano–Bond Generalized Method of Moments (GMM), based on a production function approach. The advantage of the first-differenced GMM estimator is that it is robust in the presence of endogenous covariates, allowing for individual fixed effects, heteroskedasticity and autocorrelation within the cross-section units.4 Our findings depict a significant coefficient for business R&D and confirm the hypothesis that human capital could play a role for the absorptive capacity of new technologies. The strong implication of our research is that governments should stimulate business (private) R&D. Based on our results, reflective points may be raised for the policy makers regarding the differences of the quality of the two types of R&D that should be analyzed and also the channels of cooperation between all the participants in the innovative systems that need to be improved. The remainder of this paper is as follows: Section 2 presents a brief review of the relevant theoretical and empirical literature, Section 3 presents the methodology and the data, Section 4 presents the results, while Section 5 concludes and discusses the policy implications of the current work. 2. Theoretical and empirical literature background It is acknowledged even starting with Solow (1956) that the new capital, based on known technology, which improves in time, has a more valuable role than the old (vintage) capital. Romer (1986) and Lucas (1988) pioneered an endogenous growth by introducing knowledge spillover, usually associated with R&D, respectively with human capital. Later on, models of horizontal product innovations (Romer, 1990) or vertical product innovations (Aghion and Howitt, 1992) are developed. The latter model implies a negative externality known in 3
Scarce literature on individual CEE countries exists (Dragomir et al. (2008), for Romania; Verbic et al. (2011), for Slovenia). 4 The work of Caselli et al. (1996) is significant in recommending the first-differenced GMM estimator for empirical growth models.
109
the literature as a “business-stealing” effect of R&D investment5 which may promote over-investment in R&D activities. In Jones (1995) “semi-endogenous” R&D based growth models consistent with time series evidence for the advanced countries, growth is endogenous in the sense that it is driven from the acquirement of new technologies by agents that are rational and seek to maximize profits. However, the growth rate is determined by exogenous parameters that are not affected by policy manipulation. Some recent theoretical works give to R&D and human capital accumulation essential roles in driving economic growth. Sequeira (2008) develops an endogenous growth model with physical capital, human capital and R&D. He concludes that the R&D subsidies have an overall positive effect on growth, wealth and welfare while human capital policy is simultaneously the most income-generating and welfareimproving. In Gomez (2011), innovation is subject to externalities associated to the duplication of research effort, as well as to R&D spillovers which significantly increases the model's fit on the observed data. Mattalia (2012) employs human capital as a production factor in the final and intermediate goods sector, together with the embodied nature of the technological progress and the important role of R&D and concludes that the productivity of schooling affects the long run growth of the economy. Empirically, the impact of R&D intensity on economic growth has been explored mostly for advanced countries. The empirical studies that enable a strong relationship between R&D and economic growth suggest that a 1% increase of the R&D stock will generate an output increase of 0.05–1% (Coe and Helpman, 1995; Grilliches, 1992). Some papers place human capital next to R&D as explanatory variables for productivity (see Coe et al., 1997; Engelbrecht, 1997; Frantzen, 2000, among others). Their findings suggest an overestimated coefficient for R&D in the absence of human capital. Also, on aggregate data a large body of literature seeks to estimate the social and private returns of R&D6, despite the measurement and specification errors encountered. The main conclusion is that the estimated social returns are greater than the private returns and this could explain the under-investment in R&D. Regarding the source of R&D funding, a lower rate of return is found for public R&D than private R&D, both at the private and social level (Grilliches, 1980; Park, 1995). Therefore, an important research question that should be posed is whether the public policy should take into account the types of R&D when promoting them. David et al. (2000) make a review of the econometric evidence on the relationship between public and private R&D expenditures, at various levels of aggregation. They find complementarity more prevalent than substitution relationship in many industry or national-level studies for the US economy. They suggest further work in this area based on international panel data since a lot of variations may affect the expected private rates of return of R&D. Bassanini et al. (2001) estimate the impact of public and private R&D among other determinants of economic growth for OECD countries during 1980–1990 and find significant R&D business coefficient. For public R&D coefficient, the authors depict a negative impact of public R&D on growth. Their results suggest that the research expenditures in the public sector crowd out resources that could have been used by the private sector. Coccia (2012) finds that when R&D spending of the business enterprise sector exceeds R&D spending of the government sector, the labor productivity tends to grow in advanced countries. Moreover, they show a strong positive association between public and private R&D expenditure.
5 While not shown here in detail, readers may refer to Jones (2005) for a detailed presentation of the externalities. These externalities (positive or negative) may promote either under-investment in R&D activities or over-investment in R&D activities. 6 Hall and Mairesse (2009) offer a comprehensive survey of the literature that sought to estimate the rate of return to R&D at country and international panel data set levels.
110
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119
However, for the less developed countries the literature is generally scarce (Coe et al., 1997; Goel and Ram, 1994; Samini and Alerrasoul, 2009) and moreover, it does not distinguish between private and public R&D. Answering a question regarding which type of R&D can boost economic growth in countries such as CEE countries is of crucial importance. Depending on the answer, relevant policy recommendations can be formulated for government interventions, through direct funding of private and/or public R&D and this paper is meant to bridge a gap in this respect. 3. Methodology and data We estimate a standard growth equation corresponding to a production function that adds R&D intensity among other types of capital (see Goel and Ram, 1994), in our case domestic and foreign capital, expressed as percentages of GDP. We split total R&D in its two main components: business R&D and public R&D. In this way we can assess which of the two types of activities has an effect on economic growth. The econometric analysis presented in the remainder of this paper has been conducted also for the total R&D; no significant impact on economic growth has been revealed.7 We extend the model by adding human capital, based on theoretical and empirical reasons already exposed in the literature review section. To check for the robustness of the results we include different sub-sets of additional control variables that are broadly representative for the existing growth empirical literature. Therefore, the starting equation can be described as:
lnGDP i;t ¼ αlnGDP i;t−1 þ β1 lnFDIi;t þ β2 lnKDi;t þ β3 lnRDbusinessi;t þβ4 lnRDpublici;t þ β5 lnHCi;t þ β6 Xi;t þ ηi þ μ i;t
ð1Þ
where: GDPi,t is GDP per working age population; FDIi,t is FDI stock as a percentage of GDP; KDi,t is the domestic investment, computed as a difference between Gross Fixed Capital formation and FDI flows, as a percentage of GDP; RDbusinessi,t is the research and development in business sector, computed as the share of R&D business expenditure stocks in GDP; RDpublici,t is the research and development in public sector computed as the share of public R&D expenditure stocks in GDP; HC is proxied by the share of students enrolled in tertiary education in their corresponding age group.8 Xi,t represents a set of control variables that are used to check for the robustness of the results. All the data are expressed in logarithms. ηi represents the individual fixed effects specific to each country and constant in time and μi,t is an error term, in principle assumed to be homoskedastic and with no serial correlation. Eq. (1) can easily accommodate for certain period dummies, if that is the case. The detailed description of all variables is relegated to Appendix 1, in Table A1.1. Note that our dependent variable is the GDP per working age population (see also Bassanini et al., 2001; Falk, 2007; Nonneman and Vanhoudt, 1996), thus capturing the impact of labor in the dependent variable. Mankiw et al. (1992) offer a tractable theoretical basis for the derivation of a functional form of the type expressed in Eq. (1). Nonneman and Vanhoudt (1996) extend Mankiw et al. (1992) model by adding the ratio of R&D to GDP. However, like in Goel and Ram (1994) we do not 7
Results on the estimates of R&D total can be provided upon request. Due to data availability of annual data for all CEE countries in our sample, we used tertiary education as a proxy for human capital. Moreover, for the purpose of depicting a possible complementarity with R&D, we considered it as a more suitable proxy. In a recent paper, Mattalia (2012) also proposes as proxy for human capital the share of population that has attained qualifications at the tertiary level, which for the developed countries shows a significant increase. In the case of CEE countries, we also noticed a significant increase. 8
explicitly consider restrictions that are to be imposed on the production function. Moreover, we acknowledge that by adding in the econometric specification control variables like the percentage of government balance in GDP, or trade openness, we depart even more from a possible interpretation of our results in the standard neoclassical framework. In its basic form, neoclassical setup is a closed economy model, solved under the assumptions of unsustainable long-run budget and trade deficits, an approach that applies to Mankiw et al. (1992). However, in the current work we have no specific purpose in testing neoclassical interpretations versus endogenous growth ones. Nevertheless, by adding the above mentioned control variables our work differs from the specification approach of Goel and Ram (1994). The first step of the differenced GMM procedure is to remove the individual effects, by differentiating Eq. (1) with respect to time: ΔlnGDP i;t ¼ αΔlnGDPi;t−1 þ β1 ΔlnFDIi;t þ β2 ΔlnKDi;t þβ3 ΔlnRDbusinessi;t þ β4 ΔlnRDpublici;t þ β5 ΔlnHC i;t þβ6 ΔX i;t þ Δμ i;t
ð2Þ
the correlation between the errors Δμi,t and the regressor ΔlnGDPi,t − 1 is corrected by instrumenting ΔlnGDPi,t − 1 with lnGDPi,t − s, i.e. values of y that are lagged 2 periods or more. The differenced GMM estimator exploits the following moment conditions (Bond et al., 2001): E[lnGDPi,t − s, Δμi,t] = 0, E[Δxi,t, Δμi,t] = 0, where x comprises the explanatory independent variables. It is possible that certain explanatory variables suffer endogeneity problems with respect to the economic growth, for example FDI, but also business and public R&D. For such variables the latter moment conditions may not be valid. Therefore, besides instrumenting ΔlnGDPi,t − 1 with lagged values of lnGDP, in the same manner values that lagged more than 2 periods behind can be used to instrument the corresponding explanatory variables susceptible of endogeneity problems. Different lags combinations used to instrument the explanatory variables FDI, R&D business, R&D public gave, mainly, the same qualitative outcomes9 as the ones reported in the main results section (see Table 1). The coefficients of Eq. (2) are short-term estimates, reflecting immediate changes in the dependent variable due to a temporary increase in the explanatory, ceteris paribus. We computed also the long-run coefficients. The long-term coefficients can be easily deduced from the following error correction form: Δ lnGDP i;t ¼ −ϕ lnGDP i;t−1 −θ1 lnFDI i;t −θ2 lnKDi;t −θ3 lnRDbusinesi;t −θ4 lnRDpublici;t −θ5 lnHC i;t −θ6 X i;t Þ þ ηi þ μ i;t ;
ð3Þ
β
j where θ j ¼ ð1−α Þ and ϕ ¼ ð1−α Þ; j ¼ 1:::6:
Arnold et al. (2011) were able to theoretically derive an error correction form to be estimated for both a neoclassical augmented Solow model, respectively for constant returns to a scale endogenous growth model. In order to empirically discriminate between these setups, they exploit different non-linear restrictions implied by the models regarding the relationship between factor shares and speed of convergence. While acknowledging that there are theoretical derivations of error correction forms under both neoclassical and endogenous growth setups, we do not consider restrictions on the production function, since we have no interest in discriminating between the two frameworks. In this context the long-run coefficients estimated here can be viewed simply as the standard concept of long-run propensities, i.e. the long-run change in the dependent variable given a 9
Details can be provided upon request.
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119 Table 1 Regression results using first difference GMM estimation. Short and long run coefficients.
Short-run coefficients lnGDPi,t−1 lnFDIi,t lnKDi,t lnRDbusinessi,t lnRDpublici,t
1
2
3
4
0.766 *** (0.101) 0.078 *** (0.019) 0.089 * (0.046) 0.050 *** (0.014) −0.109 (0.133)
0.675 *** (0.132) 0.073 *** (0.023) 0.085 (0.058) 0.047 ** (0.017) −0.120 (0.138) 0.111 * (0.057)
0.749 *** (0.108) 0.069 *** (0.022) 0.049 (0.050) 0.048 *** (0.013) −0.038 (0.115)
0.667 *** (0.129) 0.061 ** (0.023) 0.047 (0.062) 0.047 ** (0.016) −0.038 (0.120) 0.104 ** (0.046) 0.019 (0.045) 0.005 (0.003)
lnTertiaryi,t
0.034 (0.053) 0.005 (0.003)
lnTradei,t Gbalancei,t Long-run coefficients −φ lnFDIi,t lnKDi,t lnRDbusinessi,t lnRDpublici,t
−0.234 ** (0.101) 0.334 *** (0.075) 0.380 ** (0.165) 0.213 *** (0.061) −0.465 (0.599)
lnTertiaryi,t
−0.325 ** (0.132) 0.224 *** (0.059) 0.262 (0.184) 0.143 ** (0.048) −0.369 (0.463) 0.342 ** (0.146)
lnTradei,t Gbalancei,t No. of countries No. of observations Hansen Testa AR(2) b No. of instruments c
10 90 0.996 0.189 6.000
10 90 0.411 0.104 7.000
−0.251 ** (0.108) 0.276 *** (0.078) 0.197 (0.180) 0.192 *** (0.054) −0.152 (0.462)
0.137 (0.190) 0.020 (0.016) 10 90 0.912 0.242 8.000
−0.333 ** (0.129) 0.184 ** (0.059) 0.142 (0.184) 0.142 *** (0.041) −0.113 (0.372) 0.312 ** (0.127) 0.058 (0.127) 0.016 (0.013) 10 89 0.243 0.219 9.000
Note: Standard errors in brackets. *, ** and *** denote significance levels of 10%, 5% and 1%. Method used is First difference GMM of Arellano and Bond (1991) with robust standard error, consistent with panel-specific autocorrelation and heteroskedasticity in one-step estimation. Instruments: Arellano-Bond (AB) type: the second and the third lag of the dependent variable. Standard instruments: differences of all the regressors. Different lags combinations were used for the explanatory variables. a p-values are reported. The Hansen test reports the validity of the instrumental variables. The null hypothesis is that the instruments are not correlated with the residuals. b p-values are reported. The Arellano–Bond (AB) test failed to reject the null hypothesis of no autocorrelation in the residuals. c The number of instruments was reduced to the minimum by applying the collapse option when using the xtabond2 command in Stata.
ceteris paribus permanent increase in the explanatory variable of interest. We also report −ϕ, considered by the literature as the convergence coefficient towards the steady state (Neuhaus (2006), Arnold et al. (2011), etc.). Turning to the set of control variables denoted by Xi,t, the first subset reported in the main results section are government balance as a percentage in GDP and trade openness calculated as the ratio of the sum of exports and imports to GDP. By the first sub-set of the variables we intend to see if the relation between economic growth and R&D (business and public) intensities in CEE countries is enhanced in any way by inside country factors (such as government policies) or by interaction with the outside world (through trade). Since raising business R&D would become a government priority for these countries based on our empirical evidence, we include also government balance as a growth determinant. Also, a more open economy could have a positive effect on the linkage between R&D and economic growth, due to trade spillovers and a greater chance for technological diffusion (Coe and
111
Helpman, 1995).10 For CEE countries, a positive impact of trade openness would promote rapid absorption of technological knowledge from the developed world. For robustness checks, we include subsequently the role of freedom (monetary, financial and fiscal) and also the role of credits and market capitalization (see Appendix 1, Table A1.1 for variables definitions). The reforms that CEE countries faced after the fall of communism meant a lot of changes in terms of financial, fiscal and monetary policies. We are interested to see if the inclusion of variables that could stress such aspects affects our results. Since financial reforms were the most important features of the transition process of these countries, we include in the specification some financial development variables (such as credits and market capitalization). The data we used are mainly from the Eurostat database and are expressed in year 2000 constant prices. Data for FDI comes from the UNCTAD FDI database (2011, 2013). The control variables used for robustness purpose are from the World Bank databases. We have chosen for our analysis the period 1998–2008 in order to avoid the transition period during the nineties, when economic restructuring fundamentally affected the macroeconomic framework and might lead to ambiguous results. Moreover, availability and quality of data increased with the last years of the nineties and beginning of the new decade. We chose not to include the years of economic crisis (2009–2010), starting to affect the CEE countries later than Western Europe (the end of 2008, beginning of 2009). Such turmoil periods (for growth but also for R&D) could introduce important distortions in our estimations. To obtain series for country's R&D stock, we used the methodology designed by Coe and Helpman (1995) i.e. the perpetual inventory method to R&D investment data with a depreciation rate of 15%.11 We applied the same method for computing the domestic capital stock. For physical capital stock we considered the 5% depreciation rate also often used in the literature for developing economies. We followed the rule of thumb recommended by Roodman (2009) to keep the number of instruments smaller than the number of individuals by collapsing the instrument matrix. The coefficients of the variables obtained in this way were very similar to those obtained with the un-collapsed form of the matrix, proving that our models are quite robust.12 The p-values obtained for the Hansen test in the collapsed estimation ensured us that the instrument sets are orthogonal to the regressors and therefore valid for estimation. The Arellano– Bond (AB) test failed to reject the null hypothesis of no autocorrelation in the residuals. We also checked if the coefficient of our lagged variable lies between OLS and FE estimators (which are biased in the opposite directions) and results were favorable.13 Finally, Bond et al. (2001) point out that, when time series is persistent, then the first-difference GMM estimator can behave poorly, since lagged levels of the time series provide weak instruments for the first difference variables, possibly generating serious finite sample biases. To fix this problem, they suggest using an asymptotically efficient estimator i.e. the system GMM estimator in the context of empirical growth research. We also estimated the growth equations using system GMM estimator. In the short run, the results respect the FE b GMM b OLS rule and are qualitatively the same as those obtained with difference GMM. Nevertheless, in the long run all the explanatory variables of interest from the main specification prove to be insignificant and the estimation becomes not reportable.14 A possible explanation could be the small value of 10 It would be interesting in a further study to split between domestic and international R&D, under the condition of data availability for large time spans after the entrance of the countries in the EU. For our period under consideration, no significant results were obtained for international R&D. 11 This methodology is widely used in the literature (see among others Coe et al., 1997; Frantzen, 2000). Also, the value for the depreciation rate is widely used in the literature. See Grilliches,1998; (Krammer, 2010). 12 Tables with results in the un-collapsed matrix can be provided upon request. 13 Tables with results for OLS and FE can be provided upon requests. 14 Tables with system GMM results can be provided upon request.
112
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119
R&D (%
GDP (th. euros)
2.0
1.6
1.2
0.8
Business R&D intensity Public R&D intensity Average GDP/ working age population
35 30 25 20 15 10
0.4 5 0.0
0
Fig. 1. Average R&D intensities in CEE countries (1998–2008). Source: Eurostat database (2011). The countries are in descending order by total R&D (as the sum of public and business R&D) intensity. Average GDP per working age population is expressed in thousands of Euros.
1 − α, which increases the standard errors of the long run coefficients obtained by the δ-method. 4. Results 4.1. Main results Table 1 presented the main results of the paper. Model 1 splits total R&D into business R&D and public R&D, while Model 2 introduces human capital proxied by tertiary education. Models 3 and 4 are the estimates of all of the equations mentioned above in the presence of the first sub-set of two macroeconomic control variables i.e., government balance and trade openness. Our results suggest a relationship between business R&D intensity and GDP per working age population, being statistically significant in all specifications. The elasticities imply that a 10% increase in business R&D intensity should generate an increase of 0.5% in the short run growth. A 10% permanent increase in the R&D intensity should generate in the long run a 2.13% increase in the growth of the GDP per working age population, ceteris paribus (Model 1). Public R&D is not significant in any of the models. Since a part of government research is not accounted for by existing measures of GDP, while the R&D performed by universities is not a direct measure of output (Guellec and van Pottelsberghe (2003)), results are not that surprising. A possible explanation could be that research in the public sector is often made in areas not directly related to growth, such as defense, medical research, higher education, and this could lead to a diffused impact on output growth. We do not depict crowding effect as in Bassanini et al. (2001) which find a statistically significant negative coefficient for public R&D in the case of OECD countries. More on analyzing the difference in results for business and public R&D is provided in the discussion section. From a methodological point of view, the above considerations on the impact of the public R&D may imply using longer lags in estimation for this explanatory variable, which is difficult due to the constraints of short time series for R&D expenditures in these countries. Even if the data availability would allow, there is no consensus while answering the questions raised by studies such as Griliches (1979) and Hall et al. (2005), concerning the lag structure to introduce when estimating the effect of R&D on output. As Goel and Ram (1994) suggest there is only the minor risk of
underestimating the lagged effect if having a current period specification for the independent variables. Besides our variables of interest, as expected, foreign capital has a positive impact on growth in these countries, while domestic capital have a small and insignificant contribution in all of our specifications. Our human capital proxy introduced in Model 2 seems to be significant for growth in both, short and long run estimations: a 1% increase in the tertiary enrollment rate will cause a 0.11% (and 0.342%) in GDP per working age population. In the presence of human capital, R&D business coefficient drops but still remains significant at a 5% level. When we introduce government balance and trade, the results remain the same: public R&D proves to have no short-term effect on growth, while business R&D and human capital are robust to the introduction of control variables. The human capital variable gains significance from 5% to 1%, proving that the government policies may have a positive effect on education, enhancing the impact of education on growth. Our result is consistent with the view that a part of the R&D effect is actually accounted for by human capital and that the omission of human capital variable leads to an overestimated R&D coefficient.15 Addition of control variables has no effect on public R&D, but it decreases the long-run coefficient of business R&D from a 5% to a 10% level of significance. Bassanini et al. (2001) depict a drop in R&D coefficient in the presence of trade variable and they explain this drop by possible interactions between R&D in the business sector and international trade. They argue that the domestic R&D in the business sector may have a smaller impact in the countries exposed to foreign R&D, which could be the case of CEE countries. Also, it is possible that in the long run, the government balance has a negative impact on the R&D activity employed by the business sector. The coefficient of convergence towards the steady state − ϕ (named Error Correction term in Table 1) has a negative sign and it is statistically significant in all the models. The sign is a frequent outcome in the empirical works (see Neuhaus (2006)), 15 We have examined a number of variants of our model specification. For instance, we have included in our estimations some interaction terms i.e. human capital with R&D, FDI with human capital and FDI with R&D. We have found out that total R&D*human capital, public R&D*human capital prove to be negative and significant. Also, the interaction terms between FDI and R&D variables are negative and significant. FDI*Human Capital turns out to be positive and significant. Tables with results can be provided upon request.
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119
(Arnold et al. (2011)) when different panel estimators are employed (such as dynamic fixed effects (DFE), mean group (MG) or pooled mean group (PMG) estimators). When employing DFE estimator for CEE countries, Neuhaus (2006) finds a convergence parameter of − 0.2 significant at 1%, for the period 1991–2002.
113
Arnold et al. (2011), allowing the short run coefficients and the speed of adjustment to differ across OECD countries and thus employing PMG estimation, report averages of the convergence parameters ranging from − 0.42 to − 0.04, significant at 1%. Our GMM estimates of the convergence parameters are quite stable
Appendix 1
Romania Latvia Bulgaria Slovakia Poland Lithuania Estonia Hungary Czech Republic Slovenia EU27 -0.4
-0.2
0 Business R&D
0.2
0.4
0.6
0.8
Public R&D
Fig. 2. The changes of public and business R&D intensity (2008 versus 1998). Source: Eurostat database (2011), Science and Technology Indicators. Note: Fig. 2 presents the change in R&D intensity (2008 versus 1998). R&D intensity is computed as the share public and private R&D expenditure in GDP.
Table A1.1 Variables description. Variable
Description
Source
lnGDPi,t
Natural logarithm of GDP per Capita of the working age population (15–64 years old), expressed in millions EURO, measured in constant 2000 prices. Natural logarithm of GDP per capita of working age population, as given in the previous period. Natural logarithm of Foreign Direct Investments, as percentage of GDP Natural logarithm of domestic investment, computed as a difference in Gross Fixed Capital formation and FDI inflows, as percentage of GDP. Natural logarithm of R&D expenses in the business sector, as percentage of GDP. Business R&D activities include R&D activities run by firms, organizations and institutions whose primary activity is the market production of goods or services for sale to the general public. Natural logarithm of R&D expenses in the public sector, as percentage of GDP. Public R&D activities refer to both government and educational research activities. According to the methodological references of Eurostat, government R&D activities are run by departments and establishments of the government, as well as other public bodies, institutes and non-profit organizations which are financed by central or local governments. Educational R&D activities are run by universities, colleges of technology and other institutes of post-secondary education, research institutes and experimental stations administered by or associated with higher education establishments. Natural logarithm of number of students belonging to the tertiary education sector, as percentage of the corresponding age group (20–24 years old). Natural logarithm of trade openness (the sum of exports plus imports), as percentage of GDP. Net surplus(+)/Net deficit(−) of the general government, as percentage of GDP. Monetary freedom reflects the price stability and control. Its score lies between 0 and 100, 0 reflects the lowest level of monetary freedom and 100 reflects the highest level. Financial freedom measures the banking system efficiency, the independence from government control and inference in financial sector. Its score lies between 0 and 100, 0 reflects the lowest level of financial freedom and 100 reflects the highest level. Fiscal freedom reflects the tax burden. Its score lies between 0 and 100, 0 reflects the lowest level of fiscal freedom and 100 reflects the highest level. Domestic credit to private sector, as percentage of GDP. Market capitalization of listed companies, as percentage of GDP. Money and quasi money (M2), as percentage of GDP.
EUROSTAT
lnGDPi,t−1 lnFDIi,t lnKDi,t lnRDbusinessi,t
lnRDpublici,t
lnTertiaryi,t lnTradei,t Gbalancei,t lnMonetaryFreedomi,t lnFinancialFreedomi,t
lnFiscalFreedomi,t lnCrediti,t lnMKCi,t lnM2i,t
EUROSTAT UNCTAD EUROSTAT EUROSTAT
EUROSTAT
EUROSTAT EUROSTAT EUROSTAT Heritage Foundation Heritage Foundation
Heritage Foundation World Bank World Bank World Bank
114
Table A2.1 Robustness check using time dummies, different periods and 3 years averages.
Short-run coefficients lnGDPi,t−1 lnFDIi,t lnKDi,t lnRDbusinessi,t lnRDpublici,t
Panel A:
Panel B:
Panel C:
Panel D:
Period:1998–2008 with time dummies for 2004 & 2007
Period: 1998–2003
Period: 2004–2008
3 years averages (1998–2008)
1
2
3
4
1
2
3
4
0.781 *** (0.094) 0.050 *** (0.014) 0.119 (0.067) 0.053 *** (0.016) −0.08 (0.125)
0.696 *** (0.118) 0.041 ** (0.016) 0.115 (0.079) 0.050 ** (−0.02) −0.084 (−0.13) 0.107 * (0.055)
0.766 *** (0.093) 0.039 * (0.018) 0.091 (0.084) 0.048 *** (0.014) −0.04 (0.125)
0.906 *** (0.179) 0.023 (0.037) 0.056 (0.066) 0.046 *** (0.007) −0.012 (0.067)
0.810 *** (0.213) 0.017 (0.043) 0.063 (0.082) 0.043 *** (−0.01) −0.01 (0.084) 0.072 (0.054)
0.979 *** (0.191) 0.005 (0.048) −0.003 (0.082) 0.039 *** (−0.01) 0.019 (0.063)
0.064 (−0.09) 0.002 (0.002)
0.787 *** (0.197) 0.015 (0.049) 0.038 (0.084) 0.043 *** (−0.01) 0.028 (0.059) 0.077 * (0.041) 0.002 (0.085) 0.003 (0.003)
0.710 *** (0.125) 0.099 *** (0.028) −0.014 (0.094) 0.064 (0.077) −0.167 (0.191)
0.075 (0.072) 0.002 (0.003)
0.695 *** (0.105) 0.031 (−0.02) 0.089 (0.095) 0.047 ** (0.017) −0.041 (0.132) 0.091 * (0.044) 0.062 (0.067) 0.003 (0.004)
−0.234 ** (0.093) 0.166 −0.1
−0.305 ** (0.105) 0.103 (0.069)
−0.021 (0.191) 0.252 (0.837)
−0.213 (0.197) 0.072 (0.179)
−0.290 ** (0.125) 0.342 *** (0.105)
lnTertiaryi,t lnTradei,t Gbalancei,t Long-run coefficients −φ lnFDIi,t
−0.219 ** (0.094) 0.230 *** (0.068)
−0.304 ** (0.118) 0.136 ** (0.054)
−0.094 (0.179) 0.239 (0.222)
−0.19 (0.213) 0.088 (0.158)
1
2
3
4
2
0.422 * (0.212) 0.084 *** (0.023) −0.001 (−0.1) 0.136 (0.082) −0.182 (0.141) 0.421 ** (0.139)
0.696 *** (0.137) 0.069 ** (0.026) 0.005 (0.084) 0.051 (−0.07) −0.02 (0.185)
0.582 *** (0.040) 0.083 ** (0.026) 0.066 (0.044) 0.071 ** (0.023) −0.124 (0.069) 0.208 *** (0.062)
0.076 (0.097) 0.009 * (0.005)
0.470 * (0.224) 0.072 ** (0.027) 0.007 (0.099) 0.119 (0.076) −0.083 (0.153) 0.342 ** (0.139) 0.031 (0.054) 0.006 (0.005)
−0.304 * (0.137) 0.227 ** (0.100)
−0.530 ** (0.224) 0.135 ** (0.048)
−0.418 *** (0.040) 0.198 *** (0.060)
−0.578 ** (0.212) 0.146 *** (0.035)
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119
Appendix 2
Panel A:
Panel B:
Panel C:
Panel D:
Period:1998–2008 with time dummies for 2004 & 2007
Period: 1998–2003
Period: 2004–2008
3 years averages (1998– 2008)
lnKDi,t lnRDbusinessi,t lnRDpublici,t
1
2
3
0.543 ** (0.224) 0.242 *** (0.074) −0.364 (0.599)
0.378 (0.255) 0.165 ** (0.057) −0.277 (−0.46) 0.353 ** (0.148)
0.388 (0.303) 0.206 *** (0.061) −0.17 (0.542)
lnTertiaryi,t lnTradei,t Gbalancei,t 10 90 0.87 0.269 8
10 89 0.318 0.256 9
0.292 (0.303) 0.155 ** (0.050) −0.133 (0.050) 0.299 * (0.137) 0.204 (0.191) 0.008 (0.013) 10 89 1 0.335 11
1 0.591 (0.549) 0.493 (0.883) −0.129 (0.706)
10 40 0.456 0.475 6
2
3
0.329 (0.190) 0.227 (0.238) −0.053 (0.424) 0.38 (0.343)
10 39 0.302 0.661 7
4
−0.133 (5.122) 1.891 (7.171) 0.924 (9.044)
3.08 (1.017) 0.103 (0.982) 10 40 0.446 0.931 8
1 0.18 (0.269) 0.202 (0.169) 0.13 (0.275) 0.362 (0.301) 0.008 (0.402) 0.016 (0.020) 10 39 0.372 0.889 9
2
−0.049 (0.316) 0.22 (0.208) −0.576 (−0.69)
10 50 0.011 0.228 6
3
−0.002 (0.172) 0.236 ** (0.096) −0.315 (0.298) 0.729 *** (0.146)
10 50 0.167 0.366 7
0.016 (0.278) 0.169 (0.196) −0.064 (0.612)
0.249 (0.262) 0.031 (0.021) 10 50 0.028 0.223 8
4
2
0.012 (0.186) 0.224 * (0.105) −0.156 (0.306) 0.644 *** (0.137) 0.059 (0.089) 0.011 (0.012) 10 50 0.563 0.14 9
0.157 (0.110) 0.171 ** (0.062) −0.296 (0.178) 0.498 *** (0.144)
10 30 0.085 . 5
Note: Dependent variable: Growth rate of real per worker GDP (lnGDPi,t). Panel A reports the results when using time dummies for 2004 and 2007 over the period 1998–2008. Panel B reports the results when using the same estimates over the period 1998–2003. Panel C reports the results when using the same estimates over the period 2004–2008. Panel D reports the results when splitting the period 1998–2008 in three-year sub-periods, namely 1998–00, 01–03, 04–06, 07–08, and computing three-year averages for each sub-period. We report just the second model, the rest of the models give qualitatively the same results for the variables of main interest. Estimation method: First difference GMM of Arellano and Bond (1991) with robust standard error, consistent with panel-specific autocorrelation and heteroskedasticity in one-step estimation. Instruments: Arellano-Bond type: the dependent variable from lags 2 to 3. Standard instruments: the differences of all other regressors. Standard errors in brackets. *, ** and *** denote significance levels of 10%, 5% and 1%. a p-values are reported. The Hansen test reports the validity of the instrumental variables. The null hypothesis is that the instruments are not correlated with the residuals. b p-values are reported. The null hypothesis of the Arellano–Bond test is that of no serial correlation between residuals. c The number of instruments was reduced to the minimum by applying the collapse option when using the xtabond2 command in Stata.
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119
No. of countries No. of observations Hansen testa AR(2) b No. of instruments c
0.319 (0.245) 0.01 (0.017) 10 90 0.926 0.298 10
4
115
116
Panel B: financial depth variables
Panel A: freedom variables
Short-run coefficients lnGDPi,t−1 lnFDIi,t lnKDi,t lnRDbusinessi,t lnRDpublici,t
1
2
3
4
5
6
1
2
3
4
5
6
0.760 ** (0.103) 0.062 *** (0.019) 0.063 (0.051) 0.047 ** (0.015) −0.098 (0.140)
0.682 *** (0.130) 0.060 ** (0.021) 0.060 (0.059) 0.045 ** (0.018) −0.110 (0.144) 0.096 (0.055) 0.055 (0.050) 0.021 (0.023)
0.761 *** (0.107) 0.069 *** (0.019) 0.080 (0.053) 0.046 *** (0.013) −0.101 (0.138)
0.755 *** (0.126) 0.068 *** (0.016) 0.078 (0.050) 0.047 ** (0.017) −0.099 (0.139)
0.059 (0.061)
0.756 *** (0.097) 0.022 (0.022) 0.048 (0.079) 0.048 ** (0.016) −0.059 (0.145) 0.062 (0.041) 0.051 (0.056)
0.734 *** (0.105) 0.064 *** (0.016) 0.080 * (0.042) 0.035 *** (0.008) −0.095 (0.130)
0.046 (0.045)
0.442 *** (0.131) 0.069 ** (0.023) 0.053 (0.051) 0.043 *** (0.011) −0.137 (0.099) 0.120 ** (0.049) 0.032 (0.038)
0.809 *** (0.086) 0.021 (0.021) 0.046 (0.074) 0.050 *** (0.014) −0.049 (0.141)
0.052 (0.056)
0.680 *** (0.135) 0.067 *** (0.021) 0.078 (0.058) 0.043 ** (0.018) −0.114 (0.140) 0.101 (0.062) 0.046 (0.050)
0.568c *** (0.118) 0.073 *** (0.018) 0.059 (0.040) 0.045 *** (0.008) −0.120 (0.104)
0.055 (0.057)
0.679 *** (0.130) 0.066 ** (0.022) 0.078 (0.064) 0.044 ** (0.016) −0.114 (0.142) 0.101 * (0.052) 0.046 (0.052)
0.045 (0.047)
0.649 *** (0.129) 0.061 *** (0.019) 0.078 (0.051) 0.031 ** (0.010) −0.107 (0.133) 0.102 * (0.054) 0.035 (0.044)
0.002 (0.027)
0.002 (0.023) 0.017 (0.062)
−0.003 (0.068) 0.092 ** (0.033)
0.105 *** (0.026) 0.034 ** (0.014)
0.031 ** (0.014) 0.061 (0.034)
0.066 * (0.036)
−0.266 ** (0.105) 0.240 ***
−0.351 ** (0.129) 0.173 ***
lnTertiaryi,t lnTradei,t lnMonetaryFreedomi,t
0.063 (0.053) 0.025 (0.017)
lnFinancialFreedomi,t lnFiscalFreedomi,t lnCrediti,t lnMKCi,t lnM2i,t Long-run coefficients −φ lnFDIi,t
−0.240 ** (0.103) 0.259 **
−0.318 ** (0.130) 0.188 ***
−0.239 ** (0.107) 0.290 ***
−0.321 ** (0.130) 0.207 ***
−0.245 * (0.126) 0.279 **
−0.320 ** (0.135) 0.208 ***
−0.432 *** (0.118) 0.168 ***
−0.558 *** (0.131) 0.124 ***
−0.191 ** (0.086) 0.111
−0.244 ** (0.097) 0.089
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119
Table A2.2 Robustness checks using financial variables.
Panel B: financial depth variables
Panel A: freedom variables
lnKDi,t lnRDbusinessi,t lnRDpublici,t
2
3
4
5
6
1
2
3
4
5
6
(0.082) 0.264 (0.197) 0.197 *** (0.059) −0.410 (0.613)
(0.056) 0.188 (0.196) 0.142 ** (0.051) −0.346 (0.497) 0.303 * (0.150) 0.172 (0.145) 0.066 (0.077)
(0.091) 0.334 (0.191) 0.193 *** (0.058) −0.424 (0.592)
(0.103) 0.317 (0.198) 0.193 *** (0.056) −0.404 (0.612)
(0.028) 0.096 (0.091) 0.077 *** (0.015) −0.245 (0.185) 0.215 ** (0.073) 0.058 (0.064)
(0.101) 0.242 (0.380) 0.260 *** (0.080) −0.255 (0.771)
0.310 (0.315)
(0.082) 0.195 (0.326) 0.198 *** (0.061) −0.241 (0.622) 0.255 (0.169) 0.211 (0.218)
(0.056) 0.302 * (0.146) 0.131 *** (0.035) −0.356 (0.520)
0.214 (0.235)
(0.058) 0.244 (0.192) 0.136 ** (0.043) −0.358 (0.488) 0.316 (0.218) 0.143 (0.152)
(0.041) 0.137 (0.083) 0.105 *** (0.024) −0.278 (0.229)
0.228 (0.217)
(0.060) 0.243 (0.199) 0.136 ** (0.046) −0.356 (0.476) 0.313 * (0.148) 0.144 (0.150)
0.169 (0.169)
(0.032) 0.223 (0.158) 0.089 ** (0.029) −0.304 (0.424) 0.289 * (0.142) 0.101 (0.121)
0.008 (0.111)
0.006 (0.072) 0.070 (0.221)
−0.008 (0.215)
0.177 ** (0.063)
0.128 ** (0.052) 0.230 * (0.107) 10.000 90.000 0.934 0.160 8.000
0.188 * (0.093) 10.000 89.000 0.293 0.116 9.000
lnTertiaryi,t lnTradei,t lnMonetaryFreedomi,t
0.261 (0.199) 0.105 (0.079)
lnFinancialFreedomi,t lnFiscalFreedomi,t lnCrediti,t
0.106 (0.094)
0.212 *** (0.057)
0.188 *** (0.046)
lnMKCi,t lnM2i,t No. of countries No. of observations Hansen Testa AR(2) b No. of instruments c
10.000 88.000 0.871 0.099 8.000
10.000 87.000 0.429 0.040 9.000
10.000 90.000 0.807 0.161 8.000
10.000 89.000 0.505 0.081 9.000
10.000 90.000 0.863 0.142 8.000
10.000 89.000 0.514 0.086 9.000
10.000 90.000 0.219 0.291 8.000
10.000 89.000 0.442 0.193 9.000
10.000 90.000 0.282 0.204 8.000
10.000 89.000 0.573 0.187 9.000
Note: Dependent variable: Growth rate of real GDP per working age population (lnGDPi,t). Panel A reports the results when using freedom variables from Heritage Foundation database (2013) as controls over the period 1998–2008. Panel B reports the results when using financial depth variables as controls over the period 1998–2008. Estimation method: First difference GMM of Arellano and Bond (1991) with robust standard error, consistent with panel-specific autocorrelation and heteroskedasticity in one-step estimation. Instruments: Arellano-Bond type: the dependent variable from lags 2 to 3. Standard instruments: the differences of all other regressors. Standard errors in brackets. *, ** and *** denote significance levels of 10%, 5% and 1%. a p-values are reported. The Hansen test reports the validity of the instrumental variables. The null hypothesis is that the instruments are not correlated with the residuals. b p-values are reported. The null hypothesis of the Arellano–Bond test is that of no serial correlation between residuals. c The number of instruments was reduced to the minimum by applying the collapse option when using the xtabond2 command in Stata.
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119
1
117
118
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119
in different specifications, usually ranging from − 0.6 to − 0.2 and significant at 1 or 5% (see also the estimations in the section of robustness checks). 4.2. Robustness checks Different methods for checking the robustness of the results were applied. In Appendix 2, Table A2.1 reports results for alternative subperiods (panel B and C) and for a regression with time dummies for the entrance years to the EU (2004 and 2007, panel A). Such specifications are checks regarding how the results can be affected by possible non-liniarities, due to the entrance and preparation for entrance of these countries in the EU and the associated political and economic reforms. Significant results are qualitatively the same with the main reported ones, especially for the period 1998–2003. For the period 2004–2008, in two specifications business R&D coefficient remains significant in the long-run. Introducing, in the same spirit and for the same reasons as with dividing in sub-periods, yearly dummies (2004, 2007) for the entrance years to the EU do not qualitatively change the results in Table 1. In panel D we ran specifications by splitting the period 1998–2008 in a three-year sub-periods, namely 1998–00, 01–03, 04–06, 07–08, and computing three-year averages for each sub-period (as a possibility to mitigate the problem of business cycle effects). Moreover, by such averaging, the number of countries will be relatively larger than the number of years, which could provide more appropriate GMM-based results (even though, averaging on sub-periods significantly reduces the number of observations). In Table A2.1, we just report the second model, while the rest of the models (1, 3 and 4) give qualitatively the same results for the variables of main interest when comparing to those in Table 1. We also ran specifications with 3, 4 and 5 yrs moving averages for the variables. The results, which are available upon request, prove that in the case of 3 and 5 yrs moving averages, at least in two of our specifications (when human capital and control variables are considered) business R&D coefficient remains significant, both in the short and in the long-run. In Table A2.2 we report results in the presence of other variables, such as institutional variables and financial development variables. As proxies for institutional variables, we use monetary freedom, financial freedom and fiscal freedom. We replaced with these variables, subsequently, our initial fiscal measure i.e. government balance. As shown, they do not yield significant results for economic growth of CEE countries, but do not change the significance of our focus business R&D (panel A). While controlling for financial indicators such as credits to the private sectors, market capitalization and monetary aggregate M2 using the World Bank database (2013) (panel B), our focus variable i.e. business R&D remains significant, both, in the short and in the long run, which ensures confidence in our main results. Also, as expected, the financial variables are all significant for the economic growth of CEE countries. 5. Discussion and conclusions Our results show a statistically significant impact of R&D business on economic growth. Public R&D, although not significant, does not crowd out the positive effect of private R&D in the estimations. When human capital is included, it is highly significant at 1% while the business R&D coefficient decreases, confirming its overestimation if human capital factor is not specifically accounted for. In the presence of a variety of control macroeconomic variables, R&D business and human capital remain significant. The significance for business R&D coefficient remains robust to estimating on certain sub-periods, including the dummy years for the entrance to the EU (2004 and 2007) or averaging on subperiods. We acknowledge that there are certain limits of the proposed research. Although instrumenting FDI and R&D for possible endogeneity
with output growth, the causal link between FDI, respectively R&D, with growth is not statistically inferable from our estimations. Moreover, Hauk and Wacziarg (2009) note that with reverse causality in regressors, for small samples, the Arellano–Bond GMM estimator may not eliminate the bias. The Arellano–Bond GMM estimator overstates the speed of convergence under a wide variety of assumptions and bias towards zero the slope estimates on the human and physical capital accumulation variables. However, we applied the small sample correction methods proposed by Windmeijer (2005) to avoid underestimations in small samples generated by standard asymptotic errors estimated through a two-step GMM estimator. Our depicted significant business R&D coefficient may raise important policy implications. R&D expenditure in the private sector is an important indicator of innovation activities. The governments should be actively engaged and promote innovative activity in firms through tax credits and subsidies and through direct spending on education and training, patent protection, regulation and competition policy. Besides these traditional ways, national governments may try to boost innovative activities by offering incentives and prizes to firms that come with exceptional innovations. Our results show that the public R&D has a neutral effect, in the sense that it does not stimulate growth, but it does not crowd out the positive effect of private R&D either. Given our results for both business and public R&D coefficients, some reflective points may be raised for policy makers (and also researchers) in CEE countries. First, the differences that exist in the quality of the two types of R&D should be analyzed. Secondly, one may want to understand if it is possible that higher shares of public R&D to be just a venue for specialists' formation, the best of which being absorbed within the private sector when getting to professional maturity, and thus providing very productive innovations for this sector. If such a context proves real, an important question may be raised: should policy makers stimulate business R&D, stimulate public R&D or stimulate both? Based on this feature of our results as possible hypotheses generators, it would be interesting to test for the future the following hypothesis. That is, would a higher level of R&D spending on academic research, for instance, act as a stimulus for business R&D? All in all, our results suggest that policy in CEE countries should put more focus on the distinction between private and public R&D. More policy implication could be possible if data availability would allow including in the analysis output measures of R&D (such as patents). To see if our results are further validated, it would be interesting to analyse the magnitude of R&D spillovers at the disaggregated level-based on larger and more comprehensive samples. Also, more theory-based hypotheses should be developed to investigate the interactions among human capital, innovation and foreign direct investments in CEE countries. Role of the funding source This work was supported by CNCSIS-UEFISCSU, project number PNII RU TE code 298/2010. The funding source helped the team members to attend conferences, staff seminars, workshops to present the paper, covered infrastructure conditions (statistical software, books) and research salaries. Study design, collection, analysis and interpretation of data, and involvement in writing the report or in the decision to submit the article for publication were not covered explicitly by the funding source. References Aghion, P., Howitt, P., 1992. A model of growth through creative destruction. Econometrica 60, 323–351. Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev. Econ. Stud. 58 (2), 277–297.
M.I. Pop Silaghi et al. / Economic Modelling 36 (2014) 108–119 Arnold, J., Bassanini, A., Scarpetta, S., 2011. Solow or Lucas? Testing speed of convergence on a panel of OECD countries. Res. Econ. 65, 110–123. Bassanini, A., Scarpetta, S., Hemmings, P., 2001. Economic growth: the role of policies and institutions. Panel data evidence from OECD countries. OECD Economics Department Working Papers 282. Bond, S., Hoeffler, A., Temple, J., 2001. GMM estimation of empirical growth models. CEPR Discussion Papers 3048, C.E.P.R. Discussion Papers. Caselli, F., Esquivel, G., Lefort, F., 1996. Reopening the convergence debate: a new look at cross country growth empirics. J. Econ. Growth 1 (3), 363–389. Coccia, M., 2012. Political economy of R&D to support the modern competitiveness of nations and determinants of economic optimization and inertia. Technovation 32, 370–379. Coe, D., Helpman, E., 1995. International R&D spillovers. Eur. Econ. Rev. 39, 859–887. Coe, D., Helpman, E., Hoffmaister, A., 1997. North–south R&D spillovers. Econ. J. 107, 134–149. David, P.A., Hall, B.H., Toole, A.A., 2000. Is public R&D a complement or substitute for private R&D? A review of the econometric evidence. Res. Policy 29 (4–5), 497–529. Dragomir, L., Barbu, M.C., Ciobanu, G., 2008. The correlation between research&development and economic growth in Romania. Ann. Fac. Econ. 1 (1), 294–299 (Oradea). Engelbrecht, H.-J., 1997. International R&D spillovers, human capital and productivity in OECD economies: an empirical investigation. Eur. Econ. Rev. 41 (8), 1479–1488. European Commission, 2010. Europe 2020. A European Strategy for Smart, Sustainable and Inclusive Growth. European Commission, Brussels. Eurostat databases. 2011; 2013. Falk, M., 2007. R&D, spending in the high-tech sector and economic growth. Res. Econ. 61 (3), 140–147. Frantzen, D., 2000. R&D, human capital and international technology spillovers: a crosscountry analysis. Scand. J. Econ. 102 (1), 57–75. Goel, R.K., Ram, R., 1994. Research and development expenditures and economic growth: a cross-country study. Econ. Dev. Cult. Chang. 42 (2), 403–411. Gomez, M.A., 2011. Duplication externalities in an endogenous growth model with physical capital, human capital and R&D. Econ. Model. 28 (1–2), 181–187. Griliches, Z., 1979. Issues in assessing the contribution of research and development to productivity growth. Bell J. Econ. 10 (1), 92–116. Grilliches, Z., 1980. R&D and the productivity slowdown. Am. Econ. Rev. 70 (2), 343–348. Grilliches, Z., 1992. The search for R&D spillovers. Scand. J. Econ. 94, 29–47. Grilliches, Z., 1998. R&D and productivity: the unfinished business. NBER chapters. R&D and Productivity: The Econometric Evidence. National Bureau of Economic Research, pp. 269–283. Guellec, D., Van Pottelsberghe, B., 2003. The impact of public R&D expenditure on business R&D. Econ. Innov. New Tech. 12 (3), 225–243. Hall, B.H., Mairesse, J., 2009. Measuring the returns to R&D. NBER Working Paper No. 15622. National Bureau of Economic Research. Hall, B.H., Jaffe, A., Trajtenberg, M., 2005. Market value and patent citations. RAND J. Econ. 36, 16–38.
119
Hauk, W., Wacziarg, R., 2009. A Monte Carlo study of growth regressions. J. Econ. Growth 14 (2), 103–147. Heritage Foundation database, 2013. Jones, C.I., 1995. R&D based models of economic growth. J. Polit. Econ. 103 (4), 759–784. Jones, C.I., 2005. Growth and ideas. In: Aghion, P., Durlauf, S. (Eds.), Handbook of Economic Growth, vol. 1B. North-Holland, Amsterdam, pp. 1063–1111. Krammer, S., 2010. International R&D spillovers in emerging markets: the impact of trade and foreign direct investment. J. Int. Trade Econ. Dev. 19 (4), 591–623. Kravtsova, V., Radosevic, S., 2012. Are systems of innovation in Eastern Europe efficient? Econ. Syst. 36 (1), 109–126. Lucas, R., 1988. On the mechanics of economic development. J. Monet. Econ. 22 (1), 3–42. Mankiw, G.N., Romer, D., Weil, D.N., 1992. A contribution to the empirics of economic growth. Q. J. Econ. 107 (2), 407–437. Mattalia, C., 2012. Human capital accumulation in R&D based growth models. Econ. Model. 29 (3), 601–609. Neuhaus, M., 2006. The impact of FDI on economic growth: an analysis for the transition countries of Central and Eastern Europe (contribution to economics), 1st edition. Physica. Nonneman, W., Vanhoudt, P., 1996. A further augmentation of the Solow model and the empirics of economic growth for OECD countries. Q. J. Econ. 111 (3), 943–953. Park, W.G., 1995. International R&D spillovers and OECD economic growth. Econ. Inq. 33 (4), 571–591. Redding, S., 1996. Low-skill, low-quality trap: strategic ccomplementarities between human capital and R&D. Econ. J. 106, 458–470. Romer, P.M., 1986. Increasing returns and long run growth. J. Polit. Econ. 94 (5), 1002–1037. Romer, P.M., 1990. Endogenous technological change. J. Polit. Econ. 98 (5), 71–102. Roodman, D., 2009. How to do xtabond2: an introduction to difference and system GMM in STATA. Stata J. 9 (1), 86–136. Samini, A.J., Alerrasoul, S.M., 2009. R&D and economic growth: new evidence from some developing countries. Aust. J. Basic Appl. Sci. 3 (4), 3464–3469. Sequeira, T.N., 2008. On the effects of human capital and R&D policies in an endogenous growth model. Econ. Model. 25 (5), 968–982. Solow, R., 1956. A contribution to the theory of economic growth. Q. J. Econ. 70, 65–94. The Lisbon Review, 2004. An Assessment of Policies and Reforms in Europe. available at https://members.weforum.org/pdf/Gcr/LisbonReview/Lisbon_Review_2004.pdf. Unctad database 2011; 2013. Verbic, M., Majcen, B., Ivanova, O., Cok, M., 2011. R&D and economic growth in Slovenia: a dynamic general equilibrium approach with endogenous growth. Panoeconomicus 58, 67–89. Windmeijer, F., 2005. A finite sample correction for the variance of linear efficient twostep GMM estimators. J. Econ. 126 (1), 25–51. Worldbank database, 2013.
