![ORCID Icon]({"type":"image","src":"/templates/jsp/images/orcid.png"} "ORCID Icon")
https://orcid.org/0000-0002-4616-6394
ABSTRACT
ABSTRACT
Using a vector-error-correction model (VECM) with total factor productivity (TFP), domestic and foreign research and development investment (R&D) as well as GDP, we find that for the Netherlands for the period 1968–2014, extra investment in public and private R&D has a clear positive effect on TFP growth and GDP. Taking into account the costs of these extra investments, we find that the rate of return to such a policy is positive and high. We also find dynamic complementarity of public and private stocks of R&D for a long period after the initial shock. However, our results also show that the productivity effects on the Dutch economy are weaker when they are part of an internationally concerted policy effort, i.e. when other OECD countries implement policies with the same effects on R&D stocks in their countries. While complements in the long run equations of the model, in the adjustment process Dutch domestic private R&D appears to consider foreign public R&D as a substitute, i.e. when foreign public R&D rises, Dutch private R&D tends to shrink.
1. Introduction
A number of studies have attempted to estimate the economic impact of public investment in research for different OECD countries. Until recently, a broad consensus existed concluding that the rate of return of such public investment would be high. The most well-known study (Guellec and van Pottelsberghe de la Potterie 2004) found a long-term elasticity of Multifactor Productivity (MFP) with respect to business R&D in OECD countries over the period 1980–1998 of 13%, increasing over time. For public research, the study found that the long-term elasticity of government and university-performed research on productivity was even higher around 17%.1 At the same time the impact of ‘cross-border’ spillovers appeared higher for smaller countries than for larger ones, reflecting the higher shares of international co-publication and co-patenting of smaller nations. To achieve such benefits though, the smaller country would need to become more R&D intensive and more specialized.
However, a recent empirical study on the subject by van Elk et al. (2015), using R&D data going back to 1963 for some countries, arrived at different macroeconomic evidence for the economic returns on public R&D, with results depending strongly on the specific model used in the estimations. They implemented three classes of econometric models: (i) a Cobb–Douglas production function, (ii) a translog production function, and (iii) an ‘augmented’ production function as proposed by Khan and Luintel (2006). Only in the augmented model, which allows for different rates of return of public (and private) R&D between countries, did the impact of public R&D appear robustly positive. These estimates also showed that public R&D has widely different impacts between countries.
In the present paper we provide an alternative methodological approach to estimate the economic impact of public investment in research. Our starting point is that the literature (as reviewed above) may have reached the intrinsic limit of econometric estimations based on cross-country comparisons. Countries are very different from each other, particularly with respect to research investment activities: not just with respect to the size of such investments, but also with respect to the nature of those investments (e.g. military vs civilian), the particular institutional set-up (e.g. whether universities are fully autonomous or government controlled, see Aghion et al. 2010), the presence and funding structure of research technology organizations (RTO’s), the dominance within private firms of large multinational corporations versus small, new firms, the international openness of the country, etc.
Taking all such differences together, we argue that an econometric model that (implicitly) assumes parameter homogeneity between countries may not be very adequate. Cross-country econometric analysis, such as carried out in Guellec and van Pottelsberghe de la Potterie (2004) or more recently van Elk et al. (2015), (usually) assumes slope homogeneity between countries. But when heterogeneity between countries is strong, it is likely to make results insignificant in this type of analysis. As Khan and Luintel (2006) put it most explicitly, ‘the best empirical strategy would be to conduct country-by-country econometric analyses … given cross-country slope heterogeneity’. Therefore, we adopt a time series approach within a single country to estimate the impact of public (and private) research.
We focus on the Netherlands, a relatively small country, but with significant public and private research investment and a strong scientific and technological base as reflected in both scientific output and number of patents filed. The country has had relatively high R&D intensity for decades – in the 1960s it actually had the highest civilian R&D intensity of any OECD countries – with the presence of large public and private research laboratories owned by large, domestically grown multinational companies (Philips, Shell, Unilever, Akzo, DSM), newcomers (ASML, NXP) as well as an increasing number of foreign firms. The country is characterized by an extensive public research sector with a large number of autonomous, high quality universities, an extensive set of RTO’s, a limited emphasis on military research, and an international, open research culture with a growing presence of many foreign researchers. The Netherlands is also characterized by major policy efforts with respect to research support, either through increased public funding of public research organizations such as universities and RTO’s or through subsidies and tax credits to private firms. This all makes the country an interesting case to investigate the impact of public R&D using the approach that we propose below.
In the next section, we briefly review the literature, with special emphasis on studies that argue why a country specific estimation methodology seems particularly appropriate when attempting to measure the economic impact of research. In section 3 we introduce the econometric approach and data that are the basis of our estimations, and present the basic estimation results and the dynamic solution to the model. Section 4 presents simulation effects for various cases of policy shocks: (i) transitory and permanent domestic public R&D shocks, and (ii) an internationally coordinated symmetric shock to public and private R&D. These results represent the core of our findings about the impact of private and public R&D on productivity growth in the Netherlands. Section 4 also presents estimates of rates of return and net gains from policy for future generations. Section 5 summarizes the analysis and outcomes.
2. A brief literature review
Our main interest is on the economic benefits of investing in public R&D. The paper that is methodologically closest to our approach is Luintel and Khan (2004). They relate Total Factor Productivity (TFP) to foreign and domestic R&D, following an earlier paper by Coe and Helpman (1995). As we will do below, they apply a vector-error-correction model (VECM). They do not consider public R&D explicitly, but provide separate estimates for total (domestic) R&D and just private (domestic) R&D. In their results for 10 OECD countries, they find less significant effects for business R&D than for total domestic R&D. We interpret this as suggesting that non-business R&D might play a strong role. This is confirmed in the study by Guellec and van Pottelsberghe de la Potterie (2004), which includes 16 OECD countries and found a slightly higher output elasticity for public R&D than for private R&D, as was already mentioned above. They use, among other models, a single equation error-correction model.
Bottazzi and Peri (2007) also use a VECM approach, but their interest in R&D explaining the dynamics of patents, as an indicator of ideas, rather than productivity (TFP). They also do not deal with the economic impact of public R&D expenditures, which is our main focus. They use R&D employment rather than R&D expenditures.2 The broader measure of R&D costs and TFP as the direct measure of productivity rather than the indirect one of patents seems more adequate to us and is also more in line with the literature in this area (see Hall, Mairesse, and Mohnen 2010).
Khan and Luintel (2006) do not follow an error-correction approach. Instead, they adopt a generalized production function model in which the intercepts and slopes differ between countries. To vary the slopes of the R&D variables per country, they constructed interaction terms of a range of variables (such as Foreign Direct Investment and share of high-tech industries in exports) with the country-specific averages of the knowledge stock variables. The heterogeneity of country-specific regression coefficients that result from this are thus crucially related to the variety of the interaction variables. In this approach it is not possible to analyse the feedbacks from productivity on the left-hand side of the equation to the knowledge stocks and the various dynamic interactions between the knowledge stocks, productivity and output.3 This essentially excludes indirect effects from the analysis, which is potentially an important drawback.
van Elk et al. (2015) compare the approach of Khan and Luintel (2006) to a range of models based on either the Cobb–Douglas or translog production function (see Hall, Mairesse, and Mohnen 2010 for an overview of the use of these models in the R&D literature), using a panel dataset in which most OECD countries were present, sometimes going back to 1963. In the ‘raw’ Cobb–Douglas estimations, none of the R&D measures appeared significant. In the error-correction model (ECM) version of the Cobb–Douglas model, private R&D had a positive impact, but public R&D a negative impact (but often not significant). In the translog model, results varied widely depending on the exact implementation of the profit-maximization conditions. In the ‘augmented’ model based on Khan and Luintel, the impact of public R&D appeared overall positive.
Complementarities between public and private R&D seem to be an important topic in trying to estimate the impact of public (or private) R&D. If public R&D stimulates private R&D, there are direct and indirect effects of public R&D investments. Jaumotte and Pain (2005a, 2005b) analyzed 19 OECD countries over a 20-year period to 2001 and found evidence of significant complementarity between public sector and business sector R&D. They argued that such complementarities more than offset any negative effect from extra public R&D on labour costs in the business R&D sector. They found that ‘an increase of one standard deviation in the share of non-business R&D in GDP (an increase of 0.06 percentage points for the average economy) raises business sector R&D by over 7% and total patenting by close to 4%.’ (Jaumotte and Pain 2005b, 38). By contrast, Khan and Luintel (2006), dealing more specifically with the heterogeneity between countries using the interaction terms of variables with country-specific averages of the knowledge stocks, found a negative interaction effect, meaning that the effect of an increase in public or private R&D is smaller if the other variables are higher.
One dimension of complementarity between public investment in research and private sector R&D arises from the attraction it exerts on internationally mobile R&D, what could be called a R&D ‘crowding-in’ effect. Factors such as the prospect of high quality collaborators, recruitment opportunities, and the presence of a local knowledge cluster, often accompanied by a knowledge technology transfer infrastructure, feature variously in such studies (Cassiman and Veugelers 2002). The overall message is that particularly for small countries, a high quality research base will attract international R&D. The same factors will encourage domestic companies to retain and expand their R&D investments domestically. A positive impact of foreign R&D on productivity has been confirmed through country-panel analyses by Luintel and Khan (2004) and Khan and Luintel (2006). More recent evidence on a large country such as the UK, found that besides public sector financed R&D also foreign R&D had a significant impact on TFP growth (see e.g. Haskel and Wallis 2010).
Thus, when estimating the economic impact of public R&D, it seems to be rather important to account for heterogeneity between countries, and for the possible (indirect) effects of interactions between the main variables involved (public R&D, private R&D, foreign R&D and productivity). The VECM approach, when applied to a single country, incorporates both of these aspects. Because it exploits the time series nature of the data rather than the cross-section, it will yield country-specific effects, and the modelling of interactions between the variables comes naturally with the approach because all variables are considered to be endogenous and impacting on each other. To our knowledge, our study is the first to adopt the VECM approach in trying to estimate the impact of public R&D.
3. Econometric model, data and estimation
3.1. Data
The model includes the following variables, which are all endogenous: total factor productivity (TFP, denoted by A), GDP (denoted by Y), the domestic public R&D capital stock (G), the domestic private R&D capital stock (P), the foreign public R&D capital stock (G*), the foreign private R&D capital stock (P*). All variables are specified as natural logs (log or ln), and TFP (non-log) is normalized to unity for 2011. Data are an updated version of those in van Elk et al. (2015), using more recent sources that extend the time period to 2014. GDP and TFP are from the Penn World Tables (version 9.0; Feenstra, Inklaar, and Timmer 2015). We use the national accounts version (RGDPNA) for GDP; the TFP variable uses data on employment that is not augmented with the human capital index that is available in PWT.
R&D data come from the OECD, and as in the case of van Elk et al. (2015), we use older versions of the OECD database kept at UNU-MERIT to extend the coverage of R&D data back into the 1960s. Gaps in the R&D data are filled by interpolating R&D intensity (R&D as a share of GDP) and using GDP data to recover the implied R&D expenditures. The time series for R&D expenditures are then converted into R&D capital stocks, to represent the idea that it is not only current R&D expenditures that influence productivity, but rather the accumulated knowledge that results from present and past R&D expenditures. It is also assumed that this accumulated knowledge depreciates (we use a rate of 15% as common in the literature, Hall, Mairesse, and Mohnen 2010). We use a perpetual inventory method to construct the stocks: St = (1–0.15)St−1 + Rt, where S is the stock and R is current expenditure.4 We apply this to both public and private R&D, yielding a stock for both types of R&D. Private R&D expenditures are expenditures by business enterprises, public R&D expenditures are total domestic expenditures minus business enterprise expenditures (higher education and public labs are the largest categories of public expenditures defined in this way).5
The foreign R&D capital stocks, private and public, are distance-weighted averages of the stocks of countries in the sample of van Elk et al. (2015), excluding the country under analysis, the Netherlands. Using patent applications as weights instead as Khan and Luintel (2006) do, would have to deal with the strong structural change in the data for the period 2000–2010.6 The broad discussion of this issue in Hall, Mairesse, and Mohnen (2010) does not lead to any better alternative than our choice.
Figure 1 shows the yearly growth rates (ln differences) of all variables in the estimation (i.e. growth rates of the R&D variables are growth rates of the stocks).7 Growth rates are positive but falling with ups and downs, leaving open the question whether they have a lower limit, and at which value. Especially TFP and public R&D may come close to such a lower limit at the end of the period. The growth rates of GDP and TFP have positive and negative outliers in 1964 and 1966 respectively, as well as 2009, biasing the estimate of the relation between TFP and GDP. To avoid this, we used 1968 as the beginning year in the estimations below. We ignore the 2009 outlier, i.e. include it in the estimations, because it may well reflect a long or medium term effect of the financial crisis on growth. Moreover, in 2009, TFP and GDP have a symmetric outlier of −0.04, which does not bias the estimate.
Published online:
22 February 2019![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/gein20/2020/gein20.v029.i01/10438599.2019.1580813/20191219/images/medium/gein_a_1580813_f0001_oc.jpg"})
Figure 1. Growth rates 1964–2014 for all variables in the analysis.
3.2. Econometric approach and estimation results
We use a so-called vector error correction model (VECM) that explicitly captures all forms of multi-way causality, i.e. all variables can influence each other. Because the VECM is a standard and well-specified model, we only present it formally (below) in its estimated form. The model is estimated in Eviews, using the standard VAR procedure of the software (which also allows estimating VECM). Note that the VECM approach is based on the very principle of all variables being endogenous on each other. The standard methods for estimating VECM therefore do not account for endogeneity in the way as is done in other regression approaches, e.g. by instrumental variables. We did check for correlation between residuals in our model and the explanatory variables (i.e. the lagged variables), and no correlation was larger than 0.12. We therefore adopt the standard VECM estimation approach without accounting for endogeneity.
The model assumes that all variables in the model affect each other mutually, and therefore are endogenous, going possibly beyond the assumptions of Luintel and Khan (2004) where foreign R&D is weakly exogenous and Haskel and Wallis (2010) where it is fully exogenous. This implies that we also assume that variables for the Netherlands may affect the foreign R&D stocks. Thus, we treat the Netherlands as a ‘large’ country, as imperfect competition theory8 and, in regard to government decisions, its G20 status would suggest. However, the estimations may still show that the effect of Dutch variables on the foreign R&D capital stocks is negligible, suggesting rather a ‘small country’ effect.9
The VECM estimates one or several long-run relationships between the variables called the co-integrating equations, CEs. It then assesses how far the economy is from these long-run relationships by calculating the residual in the co-integration relationships and includes these in the VAR model. When the estimated model is stable, the residuals of the co-integration equations, commonly termed the error-correction terms, will tend to zero, representing a long-run ‘equilibrium’. Such a model can be used in simulations of the effects of exogenous shocks in one or more of the variables in the model. The reactions to the shocks take into account the dynamic interaction of all variables and in this way go beyond the merely partial effect of a statistically significant regression coefficient in a long-term relation.
In order to determine the lag order of the model (how many lags of each variable to use in the estimations), we estimate a VAR with endogenous GDP and TFP, domestic and foreign private and public R&D, a constant and a trend. This has as optimal lag length two (for three of the standard criteria) or three (for the AIC and FPE criteria) (we allow for a maximum of three lags in these tests10). However, for lag lengths two or three, the VAR model is not stable, which makes it unsuitable for econometric and simulation analysis. With just one lag the VAR is stable and therefore only this model can be used as a basis for the VECM. In other words, in regard to the choice of the lag length the instability problem forces us to be even more restrictive than the Schwarz Information Criterion would suggest. An implication is that the VECM does not have lagged differenced terms. Business cycle effects are then captured by the residuals of each equation and the dis-equilibrium deviations from long-term relations. For a steadily growing economy like the Dutch one over the last 50 years this offers sufficient flexibility to capture all movements. This can be seen from Figure 2 where all observations remain in the confidence intervals and, more intuitively, in Figure 3 where returns to an approximate steady state takes more than 50 years after a transitory shock.
Published online:
22 February 2019![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/gein20/2020/gein20.v029.i01/10438599.2019.1580813/20191219/images/medium/gein_a_1580813_f0002_oc.jpg"})
Figure 2. Baseline scenarios (dynamic stochastic model solutions) and data.
Published online:
22 February 2019Figure 3. Impulse responses to a 1 standard deviation transitory shock in public and private R&D stocks.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/gein20/2020/gein20.v029.i01/10438599.2019.1580813/20191219/images/medium/gein_a_1580813_f0003_oc.jpg"})
Figure 3. Impulse responses to a 1 standard deviation transitory shock in public and private R&D stocks.
The maximum eigenvalue test and the trace test suggest, at the five percent significance level, two or three co-integrating equations (CEs) under a quadratic trend and three or four under a linear trend.11 We worked through all the four cases, using maximum likelihood as the estimation method. The highest log likelihood is obtained under a linear trend and four co-integrating equations.12 With K = 6 endogenous variables and r = 4 co-integrating equations the number of I(1) variables or unit roots is K-r = 2 (i.e. a VECM can include both I(1) and I(0) variables, Lütkepohl 2005, 250).
For univariate unit-root analysis we use the augmented Dickey-Fuller tests including break point tests, both the additive outlier test and the innovational outlier test. Domestic private R&D has almost certainly no unit root. However, domestic public R&D and foreign private R&D have a unit root according to the augmented Dickey-Fuller tests including break tests, in both the additive outlier test and the innovational outlier test. Other variables have probabilities for a unit root near twenty percent and all have coefficients below unity, indicating that these are probably near-unit roots.13
With six variables and four co-integrating equations, identification requires that each long term relation has at least four constraints on the coefficients. One of these constraints is implemented as a normalization of a coefficient to one (making the corresponding variable look similar to a ‘dependent’ variable in a common regression framework), while the other three constraints are implemented by setting a coefficient to zero, i.e. exclude the corresponding variable from the equation. By implication, each long-term relation can have only two regressors with unconstrained coefficients besides the constant and the trend. The effects of other variables then come via the feedback from the dis-equilibrium in the long-term relations.
We first do the renormalization in a way that all coefficients in the long term relations are statistically significant and in line with economic intuition, and then set adjustment coefficients with low t-values to zero as long as the p-value for the chi-square test for the whole constraint constellation is increasing. When doing the renormalization we follow the suggestion of Boswijk (1996) and Lütkepohl (2005) to normalize the coefficients of those variables to unity which are least likely to have unit roots according to the univariate unit root analysis, because they are less likely to have zero coefficients.
We now briefly discuss the long term relations of the VECM. The estimation period is 1968–2014. The first co-integration equation represents a long-term equilibrium relation between TFP, domestic private and public R&D capital (absolute t-values in square brackets):
Table
Domestic public and private R&D capital both translate positively into TFP: a one percent increase in public R&D capital would lead to a 0.59% increase in TFP. A one-percent increase of private R&D capital increases TFP by 1.78%. Moreover, there is a negative time trend of about six percent. The large negative intercept comes from the fact that TFP is scaled to 1 in 2005 (i.e. logA = 0 in 2005), while the R&D variables are in constant PPP dollars. However, the negative 0.061 should not be attributed to the TFP alone, but is also de-trending the other variables (see Wooldridge 2013).
The functional form of this equation comes close to the single-equation Cobb–Douglas estimations by van Elk et al. (2015). But here the equation is part of a larger model with three other long-run relations and other endogenous variables in an explicit long- and short-run adjustment framework in which all variables are involved. In the long-run equation here, we have a clearly positive impact of both public and private domestic R&D. This also confirms the result in the literature that private R&D has a stronger effect than public R&D (see Hall, Mairesse, and Mohnen 2010). However, this is only a partial effect, as the complete effect should also look at the other co-integration equations and the adjustment equations.
The next long-term equilibrium relation is between domestic private R&D capital P, foreign private R&D capital P* and foreign public R&D capital:
Table
A one percent increase in foreign private R&D capital translates into 0.3% fall of domestic private R&D capital (i.e. substitution of own R&D by spillovers from abroad), which is plausible in the presence of strategic substitutes in oligopoly models, deterrence effects and low cost strategies of absorbing spillovers. This substitutability is in line with the static result of Khan and Luintel (2006). On the other hand, a one percent increase in foreign public R&D leads to a 0.4% increase in domestic private R&D, indicating positive (long-run) spillovers from the foreign public sector.
The third long-term relation is between foreign public R&D capital G*, domestic public R&D capital G and foreign private R&D capital P*:
Table
If domestic public R&D increases by one percent foreign public R&D decreases by 0.11%, indicating slight international substitutability in the field of public R&D. A one percent increase of foreign private R&D translates with 0.3 into foreign public R&D. A positive time trend of 0.02 de-trends all variables.
The fourth and final long-run relationship is for GDP and domestic TFP:
Table
If Dutch TFP increases by one percent, GDP increases by 2.44%, because higher TFP increases capital, employment and wages. Chakraborty and Lahiri (2007) suggest that with public capital included there is an upper bound for this coefficient of 2.5. The time trend increases the GDP beyond that of TFP at an order of magnitude of the labour growth rate.
As stressed already, all these effects are only partial long-run relations among endogenous variables. The total system may respond differently if we take into account the interaction between these four co-integration equations, as well as the disequilibrium adjustment dynamics, to which we now turn.
As already explained above, the disequilibrium dynamics of the model takes into account the ‘error terms’ that can be derived from Equations (1)–(4), by subtracting the right hand side of each equation from the left hand side. We denote these error terms CEi, with i = 1,…, 4, corresponding to the long-run equations. Estimation results for this VAR part of the model are as follows (t-values in brackets; variables normalized to unity in CEi in the headline):14
Table
The adjusted R2 for the six equations are 0.47, 0.63, 0.24, 0.87, 0.83, 0.34, respectively. The error probability of rejecting the constraints on the adjustment coefficients for the long-term relations is p-value = 0.999994. The lowest absolute t-value accepted in this way is 1.65 in the last equation. We will not discuss these results in detail, but instead point to a few salient adjustment coefficients that will have a large impact on some of the outcomes of the simulation experiments below. In the equation for private domestic R&D (P), we see only negative coefficients. For P itself (CE2), this implies a tendency towards equilibrium (whenever this stock is above its long-run equilibrium, it will fall). The same holds for foreign public R&D (G* / CE3), which points to a tendency for foreign public R&D to substitute for domestic private R&D in the adjustment process. Interestingly, domestic public R&D (G) has positive coefficients for all error terms. For example, domestic public R&D will tend to rise when foreign public R&D and domestic private R&D are above equilibrium values. TFP (A) will react negatively to above equilibrium values of itself and domestic R&D.
Whenever the CE terms deviate from zero, they trigger changes on the left-hand side. Unlike Luintel and Khan (2004) where foreign total R&D is weakly exogenous here foreign private and public R&D are both endogenous. As G* and P* are endogenous, our result also differs from Haskel and Wallis (2010) where foreign R&D is exogenous. As the expected values of the CE terms in long-run equilibrium (i.e. Equations 1–4 have zero residuals), the model can be solved for the expected long-run growth rates, which are then equal to the constants in Equations 5–10. These implied long-run growth rates are roughly between one percent for the TFP and 3.5% for foreign private R&D capital. Both foreign growth rate variables are higher than the domestic ones. As growth rates of R&D capital indicate also cumulative costs, the fact that they are higher than those of TFP indicates that the costs of making TFP are increasing more than the TFP itself. This confirms a point made by Bloom et al. (2017). However, as can be imagined by the use of a production function and marginal productivity condition for capital, the returns do not consist of TFP effects only but rather of an additional effect through capital movements on GDP that is increased 2.4 times as much as the TFP in the policy scenarios discussed below. Moreover, semi-endogenous growth would be favoured if we had found quadratic time trends, i.e. small negative trends in the long-run growth rates. But we find that the highest likelihood is with only linear trends in the long-term level equations leading to constant long-term growth rates. It takes a structural break to come to possibly worse results in the future. Fernald and Jones (2014) discuss some possible reasons for future structural breaks.
As the final stage of the econometric analysis, we solve the model dynamically with one thousand stochastic repetitions using normal random numbers. These model solutions will be the baseline in the simulations, and are documented in Figure 2. One salient finding from this dynamic solution is that Dutch GDP for the period 1979–1989 is below the long-run trend, while for the period 1996–2008 it is above the long-run trend. The crash in 2009 brings the economy back to the trend line, immediately for TFP and slowly for GDP.15
3.3. Internal rate of return
In the next section, we will implement a number of policy simulations, by introducing exogenous shocks to the R&D stock variables. As the values of TFP, GDP and the R&D expenditures increase after such a policy shock, we have both benefits (productivity increases leading to extra GDP) and costs (extra R&D expenditures). The question is therefore whether the benefits are larger than the costs. Both depend on the size of the shock and the length of the period under consideration. To answer this question, we will calculate rates of return to R&D investment for each of the scenarios. We adopt the notion of the ‘internal rate of return’, which is defined as the discount rate that yields a zero sum of discounted net flow gains over the time horizon (which we take as the start of the policy shock until 2040, i.e. a long period of over 70 years).
We specify the benefits as the yearly additional GDP due to increases of TFP. This is calculated as the percentage difference between TFP in the scenario and baseline TFP, multiplied by the baseline GDP, in each year. An alternative would be to use the increase of GDP that is generated in the scenario itself (because GDP is one of the six endogenous variables, the scenarios produce a time path for it). The scenario increases in GDP are usually higher than those for TFP, i.e. besides productivity, there is also a ‘factor-intensive’ part of growth. We do not follow this road for the calculation of the internal rate of return, because this would also require us to estimate the costs of the increased factor use (capital and labour) that is associated to this part of growth.
The current costs are the yearly additional gross investments in domestic R&D (private and public). These costs consist of net investment and depreciation, and are calculated as the yearly gross investment in the scenario minus gross investment in the baseline. Foreign R&D investments are not considered in the rate of return calculations, because we also do not include foreign GDP increases as welfare benefits. In other words, we look at national benefits and costs, not the international ones.
With yearly benefits and costs available, we calculate yearly net benefits (benefits minus costs; these may be negative), and then numerically solve for the discount rate that sets the discounted sum of these net benefits to zero. This definition of the rate of return differs from the cross-section or panel literature, which tends to estimate the rate of return as the elasticity of output with regard to R&D capital divided by the capital-output ratio.16 The first (elasticity) is obtained by econometric estimation, while the latter (capital-output ratio) is calculated from the data as an average (see Hall, Mairesse, and Mohnen 2010, for slightly different procedures for direct estimates). Such a procedure is impossible in our VECM approach, because we do not estimate directly comparable elasticities.
4. The economic effects of public and private R&D
We now proceed to analyse the economic effects of public and private R&D in the Netherlands for the period 1968–2014. In our time series-based VECM approach, this is done by simulations that analyse the effects of an exogenous shock to the R&D capital stock variables. When these variables are shocked, either once or permanently, they will invoke deviations from long-run equilibrium (Equations 1–4 above). These deviations will lead to adjustment dynamics in the short run (non-zero CE terms and their repercussions in Equations 5–10). Given that the estimated model is stable, the economy will, over time, return to a new long-run equilibrium, in which the shock may have caused some changes relative to the original equilibrium state. These changes are considered as the (causal) effects of the original shocks, i.e. as the economic effects of public or private R&D (depending on which variables were shocked in the first place).
4.1. Transitory shocks to domestic public R&D stocks
We start by looking at the effects of a transitory shock to the domestic public R&D stock. In Figure 3, all values are measured as deviations from the baseline, which is the long-run solution of the model in Figure 2. The domestic public R&D stock starts with a transitory (i.e. once-off) shock of 0.0059 (one standard deviation). This shock has effects on the whole difference equation system, which first reacts one period later as indicated by the fact that all other variables start at zero and deviate from this in the next period.
In the long run this leads to an enhancement of the public R&D stock by 0.0156 (1.56%), i.e. the transitory shock becomes a larger permanent shock. Domestic private R&D has about thirty years of positive effects, with a peak of 0.0065 after ten years, but shows a negative effect in the long run. This indicates that in the long run, an increase in public R&D crowds out a small part (0.0023/0.0156 = 0.147 in Figure 3) of private R&D in the Netherlands as suggested by Khan and Luintel (2006), but this happens only after 30 years. Before that, looking at the short and medium run dynamics of Figure 3, private R&D is a complement to public R&D in terms of deviation from baseline with some delay under the transitory shock. All of these effects are decreasing though as the changes go the opposite way after 12 years.
The long-run net effect of these two strong forces is to increase TFP by 0.005 (half a percent), which is about one third of the long-run change in the public value and therefore about half of what may be expected from the coefficient of 0.59 in the first long-term relation of Equation (1). The partial crowding-out of private R&D after all feedback effects explains a great deal of this difference. Similarly, a transitory shock to domestic private R&D shows the same long-run relation between public R&D and TFP; but the private shock starts phasing out after 10 years, and is zero after 35 years.
4.2. Permanent shocks to R&D
We now proceed to analyse the effect of permanent shocks to the R&D variables. We will implement this by adding a value to the intercepts of Equations (6)–(9), which are the short-run equations for the R&D stocks. By adding a value (either 0.005 or 0.0025) to this intercept, we effectively increase the growth rate of this stock by a fixed percentage (either 0.5% or 0.25%, respectively). We document the results for 4 distinct experiments, which are defined in the first column of Table 1.
Table 1. Scenario description and rates of return.
In the first experiment, only the domestic public R&D stock is shocked, by 0.5%. This represents a policy in which only public R&D expenditures in the Netherlands are targeted, and the trends in the other variables are reactions to this policy. In the second experiment, domestic public and private R&D are both shocked, by 0.25%, representing a policy in which both public and private R&D are targeted, in more or less equal magnitudes. In the third experiment, only domestic private R&D is shocked, by 0.5%, i.e. a policy in which only private R&D is shocked. In the fourth and final experiment, policy is internationally coordinated, and all four R&D stocks are shocked by 0.25%. Note that when there is a policy shock to private R&D, our model and rate of return calculations assume that there are no direct costs to the public sector for this policy.
Figure 4 shows the results of the four policy simulations for TFP and the two domestic R&D stocks. Each subgraph shows the deviations for the variable from the baseline scenario in Figure 2. These deviations are always statistically significant, as their estimated standard deviations are very small. The policy is always implemented for the period 1968–2040.
Published online:
22 February 2019Figure 4. A permanent shock to the intercept of the short-run R&D equations, deviations to baseline.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/gein20/2020/gein20.v029.i01/10438599.2019.1580813/20191219/images/medium/gein_a_1580813_f0004_oc.jpg"})
Figure 4. A permanent shock to the intercept of the short-run R&D equations, deviations to baseline.
In terms of the overall nature of the movement of the three core variables in Figure 4, i.e. the two domestic GDP stocks and TFP, the four scenarios are very similar. The domestic public R&D stock shows a clear upward movement, i.e. the deviation from the baseline is positive and increasing over time. The same holds for TFP. Thus, there is a clearly positive productivity effect for the Dutch economy from all four policy simulations. For domestic business (private) R&D, we see an inverted U-shaped pattern, with the deviation rising at first, then levelling off and eventually declining. For two scenarios, domestic private R&D ends with a negative deviation from the baseline towards the very end of the period (2035–2040).
Looking in detail at the differences between the four scenarios, we can point to four salient facts. First, in terms of productivity, three of the four scenarios produce almost identical results. The scenario with international coordination of R&D policy is the one with a lower trend for Dutch TFP. In the three scenarios with only Dutch policy, it makes almost no difference for productivity how Dutch R&D is distributed between the public and private sector. The similarity for productivity trends occurs despite the fact that these three scenarios yield total domestic R&D expenditures that differ substantially. Extra R&D expenditures are between 200 and 245 billion euro, over the entire period up to 2040. In the scenario with international coordination, extra domestic R&D is about 170 billion.
Second, although a shock to domestic public R&D can sustain an increase in domestic private R&D for a medium-range period (i.e. for about 15–20 years), a longer period of increases for domestic private R&D only results when there is a shock to that R&D stock itself. In this case, the increasing deviation of domestic R&D from the baseline continues for about 30–35 years. However, even with the shock to domestic R&D remaining positive until 2040, deviations to the baseline start declining after this period. Thus, looking at the (very) long run, domestic public R&D has a much stronger tendency to sustain its own growth than domestic private R&D and this keeps TFP growth increasing.
Third, and contrary to the previous point, domestic private R&D stimulates domestic public R&D for the entire period. Even without a shock to domestic public R&D itself (in the third scenario), the domestic public R&D stock shows increasing deviation to the baseline until the very end of the period. While this deviation amounts to about 60% in 2040, shocks to domestic public R&D itself only add about 10%–20% to this (scenarios 1 and 2).
Fourth, the Dutch economy does not seem to benefit from shocks to the international R&D stocks. The productivity effect in the fourth scenario is much lower than in the other three scenarios, and so are the deviations for the two domestic R&D stocks. The explanation for this result lies in the estimated coefficients in the long-run and short-run equations of the econometric model, in particular Equations (2), (5) and (7).
Equation (2), the long-run equation for domestic private R&D has a negative effect for foreign private R&D (hence this is a long-run substitute for domestic private R&D), and a positive sign for foreign public R&D (which is therefore complementary to domestic private R&D in the long-run). However, the short-run equation for domestic private R&D, Equation (6), shows a negative sign for foreign public R&D. Because adjustment to long-run equilibrium takes a substantial amount of time, the negative short-run effect of foreign public R&D on domestic private R&D comes to dominate the scenario with a foreign R&D shock. It makes deviations from the baseline for domestic private R&D much weaker than in the scenarios without a foreign shock. Consequently, the deviation from baseline for domestic public R&D also becomes weaker, because there is a positive feedback from domestic private R&D to domestic public R&D (Equation 7).
Looking at the rates of return in Table 1, we notice that these are fairly high, i.e. between 145 and 165%. Wat this really means is that the benefits of the policy shocks (i.e. increased GDP due to increased R&D) quickly outweigh the costs (extra domestic R&D expenditures). Net yearly benefits turn positive very quickly after the policy shock occurs, which requires high discount rates to make the discounted flow of net benefits zero (remember our definition of internal rates of return). The quick turn to positive net benefits after the initial policy shock may be something that is specific to the macro level at which we estimate our model (e.g. because R&D expenditures immediately count as contributions to GDP). Therefore the high (internal) rates of return in our policy scenarios do not necessarily apply at a micro level.
Coe and Helpman (1995, 874) report only slightly lower rates of return (to total R&D): ‘in 1990 the average own rate of return from investment in R&D was 123% in the G7 countries and 85% in the remaining 15 countries’ (the 15 countries are all developed countries and include the Netherlands). NESTI (2017) quotes studies on the rate of return to basic R&D that are in the same order of magnitude as our estimates in Table 1.
The net gains expressed as a percentage of GDP are also sizeable: between 6 and 13% annually. The main driving factor behind these relatively large numbers is the fact that the productivity effects are cumulative. Each yearly extra growth effect for TFP is added to the cumulative effect of previous years, cumulating to more than 25% higher productivity in 2040.
For comparison to other studies, in particular Guellec and van Pottelsberghe de la Potterie (2004), we can also calculate what can be called a pseudo elasticity of TFP with regard to the R&D stocks. This is defined as the average annual (compound) growth rate over 1968–2040 of TFP, divided by the same growth rate of the R&D stock that relates to the policy experiment. For scenario 1 (varying domestic public R&D investment), this elasticity is 0.32 for TFP (Guellec and van Pottelsberghe de la Potterie estimate 0.17). In scenario 3, varying domestic private R&D, the elasticity is 0.26 (Guellec and van Pottelsberghe de la Potterie estimate 0.13). Thus, our estimates of both (pseudo) elasticities and rates of return are on the high side, both for private and public R&D. This can be explained by the various indirect effects that are incorporated in our model.
Finally, Table 1 also shows how total (gross) costs of the domestic R&D shocks are distributed in the scenarios. This closely follows intuition from the definition of the scenarios. When the shock is applied to domestic public R&D only, i.e. scenario 1, the burden is entirely on public R&D (private R&D expenditures are even slightly lower than baseline). This is reversed in scenario 3, where the shock is exclusively on domestic private R&D, and costs are 98% for the private sector. The other two scenarios have shocks to both public and private R&D, and the burden is shared about equally in those cases.
5. Summary and conclusions
Compared to our time series method that considers a wide range of causal effects, single equation panel analysis with slope homogeneity have two major disadvantages. First, if heterogeneity between countries is strong, as is typically the case here, it is likely to make results insignificant. Second, feedback effects are missing unless multi-equation modelling is used, as we have done here, allowing for a better analysis of the multi-way causalities operating in this area. Therefore, we have used here a vector-error-correction model for one country only, the Netherlands, thus avoiding problems of slope homogeneity in panels and dealing with feedback mechanisms explicitly. That such heterogeneity is probably rather important was already shown by Luintel and Khan (2004) and Khan and Luintel (2006) in general and the augmented production function estimates in van Elk et al. (2015) for public R&D; the latter showing strongly different rates of return to public R&D investment between OECD countries.
The simulation results for permanent shocks to the R&D variables suggest that extra public and/or private R&D will increase the growth rate of TFP and GDP in the Netherlands. The rates of return in such simulation experiments are positive, i.e. the (discounted) total costs (extra R&D) are lower than the gains (extra GDP). Thus, we conclude that the empirical evidence for 1968–2014 suggests that there are clear economic benefits to investing in both public and private R&D in the Netherlands.
R&D spending in the Netherlands is still below the 3% of GDP that the European 2020 strategy asks. The Dutch government itself set a 2.5% target, but actual R&D intensity in 2016 (most recent data) also falls short of this target. Our results suggest that raising R&D investment up to the 2.5% target would increase productivity growth and GDP growth in the Netherlands. The targets specify a 1/3 share for public R&D and 2/3 share for private R&D, which in the 2.5% target asks for public R&D expenditures being equal to about 0.85% of GDP. Current public R&D investment in the Netherlands falls about 0.2% points short of this. Raising public R&D investment to the targeted level requires an extra investment of about 1.4 billion euro (in 2016). Our results suggest that this investment will be well spent.
With respect to the question of public and private R&D being substitutes or complements, the simple but contradictory results in the literature based on interaction effects get more complicated in our dynamic model taking into account feedback effects between all variables and showing also where the R&D variables move. For transitory domestic shocks to public R&D, they are first complements and later substitutes. For permanent domestic shocks they are complements but first to an increasing extent and then to a decreasing extent. For simultaneous permanent shocks to domestic and foreign public and private R&D, they are complements, first to an increasing and then to a decreasing extent. Thus, it matters whether shocks are transitory or permanent and also whether foreign countries act simultaneously or only later. These effects are not constant but changing over time.
However, our results also show that the productivity effects of R&D investment on the Dutch economy are weaker when they are part of an internationally concerted policy effort, i.e. when other OECD countries implement policies with the same effects on R&D stocks in their countries. The reason why the effects are weaker in this case is that Dutch domestic private R&D considers foreign public R&D as a substitute during the adjustment process, i.e. when foreign public R&D rises, Dutch private R&D tends to shrink.
| logA = | −22.084 | + 0.586 logG | + 1.780 logP | −0.061 t | (1) |
| [10.59] | [40.69] | [28.25] |
| logP = | 8.222 | −0.296 logP* | + 0.385 logG* | +0.030 t | (2) |
| [12.6] | [7.70] | [19.2] |
| logG* = | 9.355 | −0.110 logG | + 0.305 logP* | +0.020 t | (3) |
| [2.08] | [12.9] | [12.4] |
| logY = | 13.220 | +2.437 logA | + 0.006 t | (4) |
| [29.8] | [7.36] |
| logA | logP | logG* | logY | |||
| dlogA = | −0.381 CE1 | −0.487 CE2 | −0.175 CE3 | +0.010+εA | (5) | |
| [4.125] | [4.09] | [2.52] | [5.76] | |||
| dlogG = | 0.235 CE1 | +0.484 CE2 | +0.296 CE3 | +0.207 CE4 | +0.027 +εG | (6) |
| [3.00] | [4.20] | [4.46] | [4.55] | [31.4] | ||
| dlogP = | −0.123 CE2 | −0.486 CE3 | −0.107 CE4 | +0.027 + εP | (7) | |
| [−.71] | [3.90] | [1.67] | [13.2] | |||
| dlogP* = | 0.060 CE1 | −0.533 CE3 | −0.189 CE4 | +0.035 + εP* | (8) | |
| [6.42] | [15.3] | [11.1] | [59.1] | |||
| dlogG* = | 0.309 CE1 | +0.468 CE2 | +0.179 CE4 | +0.031+ εG* | (9) | |
| [7.84] | [8.91] | [12.2] | [59.0] | |||
| dlogY = | −0.266 CE1 | −0.298 CE2 | −0.237 CE3 | +0.025 + εY | (10) | |
| [1.90] | [1.65] | [2.25] | [9.94] |
| Scenario | Internal rate of return | Average net gain / GDP | Share of private R&D costs in total costs (discounted) |
|---|---|---|---|
| 1. Increasing the intercept of domestic public R&D by 0.5% | 162.5% | 12.8% | −0.8% |
| 2. Increasing the intercept of domestic public R&D and domestic private R&D by 0.25% each | 153.3% | 12.9% | 54.3% |
| 3. Increasing the intercept of domestic private R&D by 0.5% | 145.0% | 13.1% | 98.6% |
| 4. Increasing the intercept of all R&D stocks (foreign and domestic) by 0.25% | 139.2% | 5.7% | 59.5% |
Acknowledgement
The authors are grateful to Jacques Mairesse, Pierre Mohnen, meeting participants at the Royal Dutch Academy of Sciences and ZEW Mannheim and two anonymous referees for useful comments. The views expressed and possible remaining errors remain solely our own.
Disclosure statement
No potential conflict of interest was reported by the authors.
ORCID
Luc L. G. Soete http://orcid.org/0000-0002-4616-6394
Bart Verspagen http://orcid.org/0000-0001-6975-8141
Thomas H. W. Ziesemer http://orcid.org/0000-0002-5571-2238
Notes
1 Within that, the effect of universities was again higher, possibly because in some countries, government laboratories had primarily non-economic objectives such as supporting defense.
2 This is rather crucial as Bottazzi and Peri (2007) support the semi-endogenous growth approach in line with Jones (1995), who emphasized the strong growth in the numbers of researchers to defend his semi-endogenous growth approach, whereas e.g. Ha and Howitt (2007) emphasized the constancy of R&D expenditure (%GDP) and TFP growth in favour of fully-endogenous growth approaches.
3 This is important because it is well-known from basic macroeconomics that impact effects may vanish after a system of equations has run through the various rounds of adjustment. The most well-known example is probably the ineffectiveness of fiscal policy in the Mundell-Fleming model under the special assumption of flexible exchange rates and perfect capital mobility. For our purpose here, this empirically controversial result is not important, but the theoretical possibility highlights the need for more empirical analyses in this area.
4 We also need to assume a value for the growth rate of the stock for the initial period. This is chosen to minimize the difference between the initial growth rate and the next one that results from the formula. In contrast, Khan and Luintel (2006) use the average growth rate over the sample of the flow variables, which is intuitively less likely to represent the initial rate required by the PIM. With little difference in the intensities noticed by the authors, stock differences (emphasized on p.12 of their paper) must be highly sensitive to the method constructing the initial value. As depreciation is also a common rate, the sensitivity comes from the chosen growth rate.
5 Lucas (1988) points out that even in the richest private universities in the USA half of the money comes from governments.
6 For example, UK applications by nonresidents dropped by 20% in 2005/2006. They do so even more strongly for Croatia and several other countries in connection with changes in their relation to the EPO.
7 Note that the fairly large depreciation rate (15%) implies that the gross yearly additions to the R&D stocks are sizable as compared to the size of the stock itself. This also implies that yearly growth rates, due to gross additions, will be fairly large. However, Li and Hall (2018) suggest that the depreciation rates may be even higher.
8 According to basic microeconomics, under fixed costs and therefore imperfect competition, firms are price setters and there are no small countries in the sense of being price takers (Helpman and Krugman 1989). In the literature on SMEs (small and medium enterprises) firms are defined as small in line with convenient statistical indicators (see Loveman and Sengenberger 1991). Firms that are small according to these indicators will normally have some fixed costs and by implication they have to determine their prices; in other words, some firms which are defined as small according to SME literature are not small according to the microeconomic definition. We use the definitions of microeconomics and international trade theory. In terms of examples, when Philips and Siemens both decide to specialize in health technology they will observe each other irrespective of the geographical size of their countries of location.
9 Estimation results suggest that the foreign R&D stocks are endogenous in the model. However, we also estimated a model in which the two foreign R&D stock variables are considered as exogenous. In this model, which we do not document to save space, a permanent shock on public R&D (implemented in the in the same way as our scenario 1 below) has a positive impact on TFP, but private R&D (implemented in the same way as our scenario 3 below) has a slightly negative effect.
10 With six variables and four lags we would have per equation 6 × 4 coefficients for the lags and per equation one for the constant and the trend in a VAR model; 36 coefficients for 47 observation results in a too low degree of freedom.
11 A quadratic trend is a trend in the growth rate equation, which is relevant if growth rates are falling over a long period. A linear trend is a trend in the level equation of the co-integrated variables.
12 A single equation approach would be justified in case of finding only one cointegrating equation (Jusélius, Framroze Møller, and Tarp (2014)).
13 The hypothesis of a coefficient of 0.95 instead of unity would have a higher probability than that of a unit root.
14 As the underlying VAR has only one lag, the VECM has no differenced terms on the right-hand side.
15 If we only use data until 2011, the estimated trend line is above the data of the crisis period. The difference between these trend lines may come close to a long lasting (bad bank) effect on growth.
16 In the medical literature, Sussex et al. (2016) report calculating rates of return for medical research on the basis of a VECM.
References
- Aghion, P., M. Dewatripont, C. Hoxby, A. Mas-Colell, and A. Sapir. 2010. “The Governance and Performance of Universities: Evidence from Europe and the US.” Economic Policy 25 (61): 7–59. doi: 10.1111/j.1468-0327.2009.00238.x [Crossref], [Google Scholar]
- Bloom, N., Ch. I. Jones, J. Van Reenen, and M. Webb. 2017. “Are Ideas Getting Harder to Find?” mimeo, March 13, — Version 0.7. https://web.stanford.edu/~chadj/IdeaPF.pdf. [Google Scholar]
- Boswijk, H. P. 1996. “Testing Identifiability of Cointegrating Vectors.” Journal of Business & Economic Statistics 14 (2): 153–160. [Taylor & Francis Online], [Web of Science ®], [Google Scholar]
- Bottazzi, L., and G. Peri. 2007. “The International Dynamics of R&D and Innovation in the Long Run and in the Short Run.” The Economic Journal 117 (March): 486–511. doi: 10.1111/j.1468-0297.2007.02027.x [Crossref], [Google Scholar]
- Cassiman, B., and R. Veugelers. 2002. “R&D Cooperation and Spillovers: Some Empirical Evidence from Belgium.” American Economic Review 92: 1169–1184. doi: 10.1257/00028280260344704 [Crossref], [Web of Science ®], [Google Scholar]
- Chakraborty, S., and A. Lahiri. 2007. “Costly Intermediation and the Poverty of Nations.” International Economic Review 48 (1): 155–183. doi: 10.1111/j.1468-2354.2007.00421.x [Crossref], [Web of Science ®], [Google Scholar]
- Coe, D. T., and E. Helpman. 1995. “International R&D Spillovers.” European Economic Review 39 (5): 859–887. doi: 10.1016/0014-2921(94)00100-E [Crossref], [Web of Science ®], [Google Scholar]
- Feenstra, R. C., R. Inklaar, and M. P. Timmer. 2015. “The Next Generation of the Penn World Table.” American Economic Review 105 (10): 3150–3182. www.ggdc.net/pwt. doi: 10.1257/aer.20130954 [Crossref], [Web of Science ®], [Google Scholar]
- Fernald, J. G., and Ch. I. Jones. 2014. “The Future of US Economic Growth.” American Economic Review: Papers & Proceedings 104 (5): 44–49. doi: 10.1257/aer.104.5.44 [Crossref], [Web of Science ®], [Google Scholar]
- Guellec, D., and B. van Pottelsberghe de la Potterie. 2004. “From R&D to Productivity Growth: Do the Institutional Settings and the Source of Funds Matter?” Oxford Bulletin of Economics and Statistics 66: 353–378. doi: 10.1111/j.1468-0084.2004.00083.x [Crossref], [Web of Science ®], [Google Scholar]
- Ha, J., and P. Howitt. 2007. “Accounting for Trends in Productivity and R&D: A Schumpeterian Critique of Semi-Endogenous Growth Theory.” Journal of Money, Credit and Banking 39 (4): 734–774. doi: 10.1111/j.1538-4616.2007.00045.x [Crossref], [Web of Science ®], [Google Scholar]
- Hall, B., J. Mairesse, and P. Mohnen. 2010. “Measuring the Returns to R&D.” In Handbook of the Economics of Innovation, edited by B. H. Hall and N. Rosenberg, 1033–1082.
Amsterdam : North Holland. [Crossref], [Google Scholar] - Haskel, J., and G. Wallis. 2010. “Public Support for Innovation, Intangible Investment and Productivity Growth in the UK Market Sector.” IZA DP 4772. [Google Scholar]
- Helpman, E., and P. R. Krugman. 1989. Trade Policy and Market Structure.
Cambridge ,MA : MIT Press. [Google Scholar] - Jaumotte, F., and N. Pain. 2005a. “From Ideas to Development: the Determinants of R&D and Patenting.” OECD Economics Department Working Paper 457. [Google Scholar]
- Jaumotte, F., and N. Pain. 2005b. “Innovation in the Business Sector.” OECD Economics Department Working Paper 459. [Google Scholar]
- Jones, C. I. 1995. “R & D-Based Models of Economic Growth.” Journal of Political Economy 103 (4): 759–784. doi: 10.1086/262002 [Crossref], [Web of Science ®], [Google Scholar]
- Jusélius, K., N. Framroze Møller, and F. Tarp. 2014. “The Long-Run Impact of Foreign Aid in 36 African Countries: Insights from Multivariate Time Series Analysis.” Oxford Bulletin of Economics and Statistics 76 (2): 153–184. doi: 10.1111/obes.12012 [Crossref], [Web of Science ®], [Google Scholar]
- Khan, M., and K. B. Luintel. 2006. “Sources of Knowledge and Productivity: How Robust is the Relationship?” OECD STI Working Paper 2006/6. [Google Scholar]
- Li, W. C. Y., and B. H. Hall. 2018. “Depreciation of Business R&D Capital.” Review of Income and Wealth: Early View. doi:10.1111/roiw.12380. [Crossref], [PubMed], [Google Scholar]
- Loveman, G., and W. Sengenberger. 1991. “The Re-Emergence of Small Scale Production: An International Comparison.” Small Business Economics 3: 1–37. doi: 10.1007/BF00389842 [Crossref], [Google Scholar]
- Lucas, R. E., Jr. 1988. “On the Mechanics of Economic Development.” Journal of Monetary Economics 22 (1): 3–42. doi: 10.1016/0304-3932(88)90168-7 [Crossref], [Web of Science ®], [Google Scholar]
- Luintel, K. B., and M. Khan. 2004. “Are International R&D Spillovers Costly for the United States?” Review of Economics and Statistics 86 (4): 896–910. doi: 10.1162/0034653043125266 [Crossref], [Web of Science ®], [Google Scholar]
- Lütkepohl, H. 2005. New Introduction to Multiple Time Series Analysis.
Heidelberg ,New York : Springer. [Crossref], [Google Scholar] - NESTI (Working Party of National Experts on Science and Technology Indicators). 2017. “The Impact of R&D Investment on Economic Performance: a Review of the Econometric Evidence.” OECD, Directorate for Science, Technology and Innovation, Committee for Scientific and Technological Policy, document DSTI/STP/NESTI(2017)12. [Google Scholar]
- Sussex, J., Y. Feng, J. Mestre-Ferrandiz, M. Pistollato, M. Hafner, P. Burridge, and J. Grant. 2016. “Quantifying the Economic Impact of Government and Charity Funding of Medical Research on Private Research and Development Funding in the United Kingdom.” BMC Medicine 14 (32): 23. doi:10.1186/s12916-016-0564-z. [PubMed], [Google Scholar]
- van Elk, R., B. Verspagen, B. ter Weel, K. van der Wiel, and B. Wouterse. 2015. “A Macroeconomic Analysis of the Returns to Public R&D Investments.” CPB Discussion Paper 313. [Google Scholar]
- Wooldridge, J. M. 2013. Introductory Econometrics: A Modern Approach. 5th ed.
Mason : Cencage Learning South Western. [Google Scholar]