Master Thesis
University of Amsterdam
Department of Economics
Innovation and Competition Agency Performance: A Cross-National
Analysis
Maarten van Oostrom
Student Number: 10467130
Supervisor: Dr. Carmine Guerriero
Abstract
What is the relation between competition policy performance and innovation rates? This thesis inquires into the several ways in which these may influence each other. Two distinct and opposite channels are described from academic literature, which would support either a positive or negative relation. An analysis using World Bank data on investment levels and competition agency data from the World Economic Forum in 2004 investigates the relation empirically. By using both an OLS as well as an Instrumental Variable specification, a significant positive relation emerges for agency performance as a function of past research & development levels. Investment that is not geared towards cost reduction is shown not to be a significant predictor of agency performance. The result supports the academic literature that predicts that countries that are particularly concerned about increasing innovation rates are hesitant to conduct strict competition policy.
Acknowledgments
The acknowledgment of my sincere gratitude is in order for my supervisor, Dr. Carmine Guerriero, for guiding me in the selection of the topic and in the writing of the thesis; to A.E. Rodriquez and Lesley DeNardis, who were so kind as to provide me with their data; to my professors, for giving me the tools to conduct this research; and to the University of Amsterdam, for providing the environment in which it could be accomplished.
Table of Contents
1. Introduction 4
2.Theoretical framework 5
2.1 The case for a negative relation 5
2.2 The case for a positive relation 8
3. Data & Methodology 12
4. Analysis 17
4.1 OLS Estimation 18
4.2 Two-Stage Least Squares Estimation 19
4.3 Robustness tests 22
5. Conclusion 25
6. Appendix 27
1. Introduction
This thesis examines the relation between innovation and competition agency performance. Specifically, it considers whether current levels of competition authority performance can be explained as a response to past research and development levels.
There is a great wealth of academic literature on the topic of competition policy – its proper aims, necessity, optimal methodology, and limitations. All of this has established the field as a subject which is firmly grounded in conventional, accepted economic theory. Despite this fact, there appears to be a great deal of variation in the performance of competition authorities between nations. In a 2007 article
published in the Journal of Business & Economic Studies by A.E. Rodriguez and Lesley Denardis1, on
which this thesis paper relies for part of its data, and whose methodology it extends to incorporate the effects of past innovation levels, the authors identify several causes for this variation.
The nature of these causes can be said to broadly fall within two categories: those country-specific parameters which prevent competition agencies from achieving their ideal performance levels, and those parameters which determine the appropriate level for each nation's economy individually. The former category of parameters are therefore descriptive: they explain why nations are not uniform in applying the findings of academic literature when it comes to competition policy even if it were so desired; the latter category contains those variables which are normative, and therefore explain variation in optimal application. These variables deal with the question why different nations may set different targets and goals in competition policy. The subject of this paper concerns the latter category -why, and if, countries may influence their policies based on past levels of research and development in the economy.
This paper is structured as follows. The next section outlines the theoretical framework and identifies the ways in which research and development expenditure and competition agency performance may be linked, identifying two opposite effects. Section 3 describes the data and methodology of the empirical analysis. Section 4 contains the OLS and 2SLS analysis, robustness tests, and the interpretation of the result. Section 5 concludes the findings. The appendix describes the provenance of the data, and lastly
1 Rodriguez, A.E. and DeNardis, L. (2007). Examining the Performance of Competition Policy Enforcement Agencies: A Cross-Country Comparison. Journal of Business & Economic Studies, Vol. 13, No. 1, Spring 2007.
the bibliography lists the sources used.
2. Theoretical framework
The theoretical route through which R&D levels may affect competition policy can be deduced from a basic understanding of what goes on in an efficient market. In perfectly competitive markets, each producer charges the same price for its goods, which is, by definition, equal to its marginal cost of production. However, most, if not all markets, do not fit this description: instead, they leave each individual firm in the market with some degree of market power, which allows the firm to charge a markup over its marginal cost and thereby increase its profits. This market power can be achieved through several routes of firm behavior, some of which are a natural and possibly inevitable consequence of the market or product in question. In particular, highly heterogeneous goods fit this framework. It is an accepted thesis of economic theory that when such product heterogeneity is the result of research and technological innovation, it is desirable that firms are permitted to make short-run surplus profits to recoup the cost of their investment, thereby incentivizing firms to invest in R&D in the first place.
The field of competition policy concerns itself with the alternate case: when market power is the result of anti-competitive market characteristics, which can ensure market power and long-run surplus profits for a firm even in the absence of innovation. Under the assumption that individual firms behave rationally in the economy, a particularly low rate of innovation can therefore be indicative of anti-competitive elements present in the market – whether the result of high barriers to entry, strongly increasing returns to scale, or, in more incriminating fashion, the result of anti-competitive behavior by the firms themselves.
As will be discussed, the degree of innovation in an economy can affect policy decisions from competition agencies in several ways.
2.1 The case for a negative relation
the connection of both to the level of competition in the industry that allows them to influence each other.
The relation between competition intensity and R&D spending in an industry is examined empirically
by Thakor and Lo (2015)2. They consider competition and R&D in a single industry, the
bio-pharmaceutical industry, over a long time-frame. Bio-pharma may be particularly salient for this analysis due to both its reliance on research for the profitability of individual firms, and the strong societal need for research in this field that extend beyond the private returns of investment. Competition intensity and R&D expenditure over time are illustrated in Figure 1, which shows these variables in the period between 1950 and 2012.
Figure 1
Competition and R&D expenditure in the bio-pharma industry between 1950 and 2014, from Thakor and Lo (2015)
Figure 1 shows how over time, as competition increased, R&D expenditure in the industry grew. Of particular interest is how the innovation growth in this industry outstripped the general innovation growth of other industries, suggesting that the increase in research spending was not the result of an economy-wide trend. They test the apparent relation statistically, and one of their findings suggests that competition indeed increases R&D spending. They argue that while the bio-pharmaceutical sector is the only sector examined, this result should hold validity at least for other industries with comparable reliance on research and development.
2 Thakor, R. and Lo, A. (2015). Competition and R&D Financing Decisions: Theory and Evidence from the Biopharmaceutical Industry. Available at SSRN Electronic Journal.
But not only can low R&D rates be indicative of the absence of competition within the industry, thereby meriting increased investigation and more active intervention - policymakers can also view increasing R&D as a goal in itself. An innovative economy has long-term benefits besides strengthening the competitiveness of innovative firms both domestically and abroad, through spillover
effects. Indeed - Hall, Mairesse, and Mohnen (2009)3 argue that while social returns to R&D are
difficult to measure and can vary greatly, they are much higher than the private returns to the firm. Additionally, the social returns to R&D investments are larger than those of investments in ordinary capital, which emphasizes the societal need for an economy which is not only developed, but also innovative.
For the above reasons, it seems likely that competition policy can, and possibly should, be used as a tool to boost innovation in the economy. By limiting other sources of market power and encouraging market entry, firms will turn to other, more economically desirable ways of generating surplus profits. It therefore seems plausible that competition agencies may be encouraged to pursue more active inquiry concerning anti-competitive effects and behaviors in the economy when innovation is markedly low. Active competition policy will boost innovation through its direct effect on firm entry in the economy. This relation between competition policy and equilibrium R&D intensity was modeled, among others,
by Martin (1998).4 The implications of his model suggest that stricter policy indeed results in higher
non-cooperative R&D intensity.
Kee & Hoekman (2007)5 investigate this relation empirically, and note that competition laws facilitate
increased entry of firms, reducing industry markups even in the long term, but mention an important caveat. While it is one way of fostering a competitive environment, it is not necessarily the most cost-effective. Nations have other policy tools at their disposal to foster competition, including trade liberalization and the reduction of bureaucratic restrictions. In particular, import liberalization shows to have a powerful effect on competition in industries which produce tradable goods. As such, they question where the priority should lie for nations looking to elevate domestic competition – in more general liberalization policies, or through the methods traditionally associated with competition law
3 Hall, B., Mairesse, J. and Mohnen, P. (2009). Measuring the Returns to R&D. Handbook of the Economics of
Innovation, published 2010.
4 Martin, S. (1998) Product market competition policy and technological performance. Included in Market Structure and Competition Policy: Game-Theoretic Approaches, ed. Norman & Thisse, published 2004.
5 Kee, H. and Hoekman, B. (2007). Imports, entry and competition law as market disciplines. European Economic
and a formal competition agency.
2.2 The case for a positive relation
The previous section discussed why low R&D rates may prompt nations to pursue more active and comprehensive competition policy. However, the opposite can also be argued; that low R&D rates should merit a more laissez-faire approach.
Von Graevenitz (2005)6 notes how even the best intentions of competition authorities may impact
innovation negatively. In his paper, he constructs a model reflecting firms' decision making regarding cooperative R&D efforts. The benefits and costs of research and development may be distributed by cooperating firms in two ways: ex ante, which is most notably the case in the creation of a joint venture between firms, or ex post, whereby one of the firms performs the research and development, and retroactively recoups its costs, and shares the benefits, by licensing its technology to another firm. Depending on the strength of competition and the level of technological opportunity in the industry, either approach may be favorable from a welfare point of view. These approaches are, however, not viewed equivalently by competition authorities, particularly in Europe, and not without good reason. Ex ante cooperation is preferred, as the uncertainty of success prior to the development of new technologies makes it less likely that it will result in anti-competitive effects, and certainly limits the degree to which firms can deliberately shape these effects. His conclusion is that more expansive competition authorities can therefore limit R&D through this route.
It is not surprising that Von Graevenitz mentions the case of Europe explicitly. After all, EU member states exist within a non-standard framework which handles competition policy on both the national and supra-national level. Within this dual framework, it seems likely that unintended direct effects and efficiency loss through bureaucratic inefficiency are magnified. The possible detrimental effects that strict competition agency scrutiny and intervention can have on innovation have since been acknowledged by the European Commission. In its 2014 Competition Policy Brief, titled Supporting
R&D and innovation in Europe: new State aid rules7, the Commission updates its policies concerning 6 Von Graevenitz, G. (2005). Integrating Competition Policy and Innovation Policy: The Case of R&D Cooperation.
Available at SSRN Electronic Journal.
7 European Commission (2014). Supporting R&D and innovation in Europe: new State aid rules. Competition Policy
the investigation of competition effects in the case of state aid to firms for innovation purposes. In short, it outlines its plan to allow for more autonomy in member states' decision-making in the case of innovation stimulation, and to “cut red tape” to ensure more flexibility.
While this is a fairly unique example, it does illustrate a change in the viewpoint from one of the more expansive competition agencies in existence. Because although the competition effects of state aid to research and development will still be subject to investigation by national competition authorities, the European commission grants that the very act of investigating these effects can result in antithetical consequences. More tellingly still: in a sense, it relinquishes some of its duty to investigate subsequent competition effects between member states, an aim for which it, and it alone, is properly equipped. Simply put, the decision of the European Commission constitutes a loosening of competition policy for the express aim of providing more room for innovation within its economy.
Still, the mere fact that the largest competition authority in Europe has taken this stance on the relation between its policy and innovation should not be taken as conclusive evidence on the general nature of this relation, nor should it spur automatic and immediate emulation elsewhere. The European Commission's task pertains to not only a very large, hard to regulate integrated economy, but also one which is generally well developed. As discussed, for a nation which aims to create a competitive economy, and, by extension, one that is innovative, formal competition law and enforcement of those laws are not the only possible approach.
Furthermore, extensive work on endogenous institutional design in relation to technological efficiency has been done by Guerriero. In the interests of this paper, his work would largely support a positive relation between past research and development levels and competition agency performance. In a 2011
paper8 Guerriero constructs a model of endogenous institutional reform, linking its design – which
comes in the form of either appointed or elected regulators - to technical efficiency and society's investment concerns. In the static case, in which time inconsistencies like investment decisions are inapplicable, the preferred regulatory appointment rule is by election (by peers). Elected supervisors exert more effort, and their policy is less likely to be influenced by the regulated industry. In the presence of investment concerns, however, appointed regulators become the preferred design, and this
8 Guerriero, C. (2011). Accountability in Government and Regulatory Policies: Theory and Evidence. Journal of
effect is magnified when the technical efficiency of the market is limited. The model therefore predicts that less innovative markets are unlikely to lead to extensive competition policy, and, in essence, will coincide with less strict agencies.
A 2015 paper9 by Guerriero is partly in line with arguments discussed earlier. The model put forth in
the paper shows a tradeoff between allocative efficiency and investment inducement. Competitive markets are more efficient but leave firms with lower expected profits, weakening the desire of firms to innovate for the purpose of cost reduction. As a result, the decision of regulators – whether to let the firms compete or to provide regulation that incentivizes cost-reducing investments – is indeed a matter of the degree of societal desire for innovation, and particularly salient in developing economies with a low technological standard.
It is the above argument that offers an interesting and stark contrast to those that suggested a negative relation, which may be best typified as follows. Earlier, the line of reasoning was that markets that lack competition create little incentive for firms to invest in cost-saving technologies. After all, a monopolist is already able to charge a high markup on its cost of production without the threat of being under-cut in price by a competitor. While research and development efforts may still increase their markup, they also represent a significant upfront cost, and the potential to recoup this cost is uncertain ex ante. However, as will be discussed now, it seems that the strictest way of regulating uncompetitive markets do little to alleviate these concerns of underinvestment. The type of regulation – either rate-of-return or
incentive regulation – is central to the discussion of how competition policy and innovation are linked.
Cambini and Rondi (2009)10 investigate this effect of regulatory regime on investment. Rate-of-return
regulation is characterized by the regulator determining a “fair” price point for the regulated firm at which to sell its product in the market, taking into account the firms costs. Since the regulator will select a price point which allows the firm to stay profitable, this price point will be high if production costs are high, and be lowered when the firm manages to produce more efficiently. As such, the firm has little incentive to innovate from a profit-seeking point of view, and instead a large incentive to
9 Guerriero, C. (2015). The Political Economy of (De)Regulation: Theory and Evidence from the U.S. Electricity Market.
Journal of Comparative Economics, 41: 91-107.
10 Cambini, C. and Rondi, L. (2009). Incentive Regulation and Investment: Evidence from European Energy Utilities. J
over-invest in the expansion of its asset base and to over-state its true cost of production.11 Incentive
regulation, contrarily, is constructed in such a way as to reward firms for achieving desired targets with greater profitability. Among the most straightforward of such targets would be cost-reduction, for which either price caps or revenue caps could suffice as a way of regulating the market without taking away the incentive to innovate. In this way, the regulator also leaves more freedom to the firm to determine for itself how to go about achieving those targets. They find in their analysis of a sample of utility companies in the European Union in the ten year period between 1997 and 2007 that investment rates are indeed higher under incentive regulation than rate-of-return regulation; furthermore, that particularly investment focused on cost-reduction, notably research and development, benefits from this regulatory regime, which leaves more freedom and rents for the regulated firm.
This result is explained further by Guerriero in a 2013 paper12, which focuses on incentive regulation.
In order for firms to make the decision to invest in cost-saving technology, they need have some degree of certainty that after the investment pays off, they can benefit from their innovation to the fullest extent. Strong focus on competition policy can, in this way, create an uncertain landscape for firms. As opposed to arguments made in the previous section, where competition in an industry can theoretically spur firms' innovation efforts, this analysis puts the effort of regulators at the center, and describes how there can be a tradeoff between boosting competition and innovation. Regulators who are particularly worried about investment concerns may want to leave more rents at the hands of firms and be reluctant to pursue competition stimulation in a very strict manner.
Since we have identified two distinct and opposing channels through which strict policy and innovation can be related – one positive, one negative - it makes sense that the desirability of such a policy will depend on the particular circumstances of each nation individually. In investigating the overall relation, one must therefore be cautious to use relevant control parameters. These are discussed in the next chapter, which outlines the sources of the data used and the research methodology.
11 Averch, H., & Johnson, L. L. (1962). Behavior of the firm under regulatory constraint. American Economic Review, 52, 1059–1069.
12 Guerriero, C. (2013). The Political Economy of Incentive Regulation: Theory and Evidence from US States. Journal of
3. Data & Methodology
The data used in this analysis is, for the most part, the same set that was used in Rodriguez & DeNardis' 2007 article. In their investigation, they estimate the effect of several variables on the performance of competition agencies, specified as follows:
PERFORMANCEi =
a + ß1 COMMON LAWi + ß2 LN INCOME PER CAPITAi
+ ß3 INTENSITY OF COMPETITIONi + ß4 CORRUPTIONi
+ ß5 LN SIZEi + ß6 EXPERIENCEi + ei
whereby “i” specifies each country. The variables describe the following: agency performance; national legal tradition (a dummy variable which equals 1 when the legal tradition of the nation is based on a common law system); the natural logarithm of gross domestic product per capita; a measure which grades the intensity of local competition; a measure of perceived corruption in the country; the natural logarithm of a measure of the size of the country; and a measure describing the degree of experience that the country has with modern competition law.
Measuring the performance of competition agencies is no straightforward endeavor. Possibly the simplest metric would be to look at the volume of output - in essence the amount of investigations completed, interventions performed, merger and integration decisions made, fines imposed, et cetera. However, as discussed in the previous chapter, while this may be appropriate for some inquiries, others should take into account the efficiency with which these functions are carried out. After all, nations have several ways of achieving their goals but only limited resources to do so. It can therefore be difficult to accurately distinguish all the relevant factors in the equation, and metrics based on objectively measurable indicators run the risk of either oversimplification, or becoming all-encompassing to the degree of becoming unworkable.
A second approach aspires to quantify agency performance by looking not at its output, but its effects. One can posit the theoretical and stated aims of an agency and examine to what extent these aims have been achieved over time. The difficulty of this approach lies in the isolation of the effect of the agency from general trends in the economy, and it therefore requires an analysis that is not only
econometrically complex in order to avoid simultaneity and reverse causation, but one that is also specified with a great deal of accuracy in order not to suffer from omitted variable bias.
Lastly, a third approach instead measures agency performance more subjectively. An OECD peer-review program that relies on the judgments of national agencies, as reported by other nations, is an example of this method. Ranking and rating agencies in such a way does of course come at a cost to replicability and verifiability, but seems a more pragmatic way of measurement in studies which focus on the economy-wide effects of competition policy.
The measure that is used by Rodriguez & DeNardis follows the last-described formula. It is constructed based on survey data for 2004, supplied by the Global Competitiveness Report from the World Economic Forum of that same year. Agency performance is rated on a scale from one to seven, seven being the best possible rating, on the basis of the results of the Executive Opinion Survey. This survey uses a dual stratification method to randomly select respondents from different industries and from differing firm sizes. As such, the data can be seen as particularly grounded in the workings of the relevant markets, as business leaders make up all of its provenance, while maintaining a reflective sample of the makeup of the economy.
For the purpose of this paper's investigation, the measure described above also seems the most appropriate. The panel data include a wide range of different nations in different economic conditions, which pleads for a more case-specific assessment: judging all these nations by the same paradigmatic standard would disregard the simple fact that these agencies vary fundamentally in the task which they perform and their relation to other economic policies. Two things, however, need to be discussed when applying this measure for the purposes of this paper.
Firstly, whether or not the measure of agency performance is clouded by concerns of the respondents beyond pure competition effects. As discussed in the section that lays out the theoretical background, competition policy may affect economic trends beyond offering ease of entry and a level playing field; in fact, that idea is the very premise of this paper. Are the survey respondents likely to assess the competition agency in their country more favorably as a result of, for example, its influence on investment as a whole, or research and development expenditure?
This does not seem likely. Firstly, a direct or indirect effect of competition policy on innovation or investment as a whole would be difficult to identify by respondents, making it unlikely that they will base their response on such considerations. Secondly, the Executive Opinion Survey includes a whole range of topics, some of which deal specifically with concepts that may be affected by competition agencies indirectly. As a result, it is plausible that respondents considered purely competition effects in the questions that dealt with competition agencies. Since the existence of a confounding variable linking the rating of agency performance and R&D expenditure seems unlikely, and since the model specification controls for several related and highly significant variables, the analysis in this paper considers R&D expenditure as an exogenous factor.
The second complication regards whether survey respondents took into consideration wider beliefs held about the way institutions work in their country. In this way, even uninformed respondents may be prompted to assign a strongly critical rating for its competition agency if they are dissatisfied with the workings of the government as a whole. It is this complication that will be addressed using an instrumental variable specification of our model later.
The independent variables in the regression equation above are less intricate than our measure for competition agency performance, and their specification and provenance are discussed in the appendix. The Ordinary Least Squares regression that Rodriguez and DeNardis ran, detailed at the beginning of this chapter, yielded the estimation results in Table 1 below.
Table 1
They note that two variables stand out in their significance: intensity of competition in the economy, and the level of corruption present. Furthermore, the experience a nation has with modern competition policy did not explain variation in competition agency performance. The income per capita and size of the nation were significant controls. To their surprise, the legal tradition proved a significant predictor only at the 90 percent level, despite more conclusive results in similar previous research. The explanation they give is that competition policy has chiefly common law characteristics independent of the wider legal framework of the nation in question.
Careful examination of the data actually revealed that the analysis above was skewed rather heavily due to an unfortunate occurrence in the dataset. The variable measuring corruption for each nation, for which the data was drawn from the Transparency International Corruption Perception Index, was entered for each country as a negative figure, ranging from -1 (very high perceived levels of corruption) to -9.9 (very low levels). However, for one entry (Taiwan) this figure lacked its sign, giving it a value of (positive) 5.7. This made the data point in question a significant outlier with dramatic effect. A correction of this data point and subsequent re-running of the regression yields the results in Table 2.
Table 2
OLS. Dependent variable: Agency performance
Table 2 shows a marked change in the regression estimates. Since the typo occurred in the corruption data, it is no surprise that it is this variable that is most affected by its correction. Both competition
intensity and corruption are now highly significant and comfortably within the one percent level. More interesting is the implications for the variables for income per capita and legal tradition. While the former was significant in Rodriguez and DeNardis' OLS estimation, it is now far from it, and the existence of common law is no longer a reliable predictor of agency performance at even the most modest of cutoff values. Overall, the explanatory value of the model has increased: where the tested model specification explained approximately 79 percent of variation in agency performance before, its
R² is now a little over 82 percent.
In light of the above, in the investigation of the effect of R&D spending levels on agency performance the two variables for legal tradition and experience will be dropped from the model. Data on research and development spending, as a percentage of a nation's gross domestic product, is obtained from the World Bank, which has made these data freely available. While the correction in the data rendered the natural logarithm of per capita income an insignificant predictor, it makes sense to include it along with R&D data. After all, it is extremely likely that the two variables are related, and income may account for effects that otherwise would cause a bias in R&D's coefficient estimate through an omitted variable. Along the same line of reasoning, an interesting variable to include in the initial OLS estimation is gross capital formation, again as a percentage of GDP. Under certain market conditions, investing in physical capital or research and development can be interchangeable means to the same end in a firm's decision-making. Returns to scale can bring down marginal costs, resulting in surplus profits, or, alternatively, the accumulation of excess capital may hint at anti-competitive elements. Whether, and in what way, capital formation affects competition agency performance is therefore both an interesting question in its own right as well as a method of keeping the estimate of R&D effects unbiased. These data are also provided by the World Bank.
The first specification of our analysis is therefore the following:
PERFORMANCEi =
a + ß1 PAST R&Di + ß2 LN INCOME PER CAPITAi
+ ß3 INTENSITY OF COMPETITIONi + ß4 CORRUPTIONi
And the related hypotheses are naturally the following:
H0 : Past research and development levels are not a significant predictor of competition agency
performance. (Null Hypothesis)
H1 : Past research and development levels are a significant predictor of competition agency
performance, the relation being either positive or negative. (Two-Tailed Research Hypothesis)
4. Analysis
Let us first examine the added variables. Table 3 and 4 show summary statistics for R&D and capital formation for each year in the period 1996-2000. All values concern expenditure as a percentage of GDP.
Table 3
R&D. in Year Obs Mean Std. Dev. Min Max 1996 48 1.038469 .8438176 .01214 2.76501 1997 58 1.014331 .8685127 .00811 3.4675 1998 55 .9963098 .8416642 .06956 3.08146 1999 56 1.114607 .9292954 .09591 3.58067 2000 60 1.055237 .938168 .04453 4.16784 Table 4
Cap. Form. in Year Obs Mean Std. Dev. Min Max 1996 97 22.36253 6.946934 .2986439 41.81622 1997 97 22.77973 6.350054 5.175169 42.97308 1998 96 22.8813 5.601806 4.884402 37.93667 1999 96 21.71972 5.53089 4.755047 36.74463 2000 96 22.13975 5.440944 4.562498 35.11864
The above tables show that countries varied more in terms of R&D spending in this period than they did in capital formation. Furthermore, R&D spending as a percentage of GDP showed more variation year-on-year.
maximum value at the beginning and end of the period. While the means stay relatively constant, the maximum value in the dataset for R&D in this period increases markedly, while that of capital formation goes in the opposite direction. This may hint that there is a degree of substitution between the two variables present.
4.1 OLS Estimation
The regression equation specified in the previous chapter is run, with the “Past R&D” and “Past Capital Formation” variables being the average values of these entities as a percentage of GDP in the period 1996-2000 for each country. This range captures the period nine to four years before the agency performance measure. The result is shown in Table 5.
Table 5
OLS. Dependent variable: Agency performance
Explanatory variable Coefficient estimate t-stat
R&D rate 0.260*** 2.822
Capital form. rate 0.00614 0.491
Comp. intensity 0.638*** 5.196
Corruption -0.133*** -3.335
Per capita income 0.178* 1.994
Country size 0.0944*** 3.170
Constant -2.808*** -3.695
Observations 70
R-squared 0.863
Robust standard errors used *** p<0.01, ** p<0.05, * p<0.1
Unsurprisingly, the intensity of competition, the measure of corruption, and the size of the nation are still highly significant predictors of agency performance. The R² has increased compared to the OLS model described earlier, to slightly over 86 percent, signifying that the model accounts for a great deal of variation in performance between nations. Capital formation does not appear to be a significant factor in competition agency performance.
But the estimation results are also surprising in several ways. Firstly, GDP per capita has become much more significant, now at the 5 percent level. It does not seem likely that this is due to the variables that were dropped: certainly a relation between legal origin or historical experience with competition policy and per capita income seems far-fetched. Instead, it may be the case that GDP per capita had ambiguous effects on policy performance; two opposite effects, one of which is now captured by the R&D variable, which is also highly significant.
Our variable of interest, past R&D expenditure, shows a strong positive effect on agency performance. Concretely, each percentage point of average research and development spending as a percentage of gross domestic product coincides with an increase in the rating of agency performance of more than a quarter of a point. This effect can certainly be considered large; after all, the measurement of performance ranges from one to seven, and variation between nations in R&D spending is considerable. In terms of magnitude, only competition intensity has a larger effect in the estimation model.
While a priori it was argued that the relation could be negative, signifying an attempt from nations to increase competition policy efforts to boost competition and thereby R&D, a significant positive relation emerges. In this way, the result is consistent with the literature that predicts that markets with lower technical efficiency lead to a less strict regulatory environment.
4.2 Two-Stage Least Squares Estimation
We must however consider the possibility that the OLS estimation renders biased estimates. In particular, the measure for corruption appears susceptible to this. Since both our dependent valuable, which measures agency performance, and the corruption variable are specified according to their
perceived values, it is likely that the two can be conflated. Survey respondents in a country
characterized by high levels of corruption may simply assume that competition agency performance is one of the domains affected negatively. Conversely, respondents from less corrupt nations may take its absence as a proxy for the quality of institutions. This possibility of the corruption and agency performance variable being jointly determined causes simultaneity and calls into question the causal validity of the analysis.
To curb this potential bias, it is appropriate to use ethnic and cultural fractionalization data as an instrumental variable, capturing the effect of corruption. The effect of the population's diversity on
corruption is well documented, by, among others, Alesina et al. (2003)13; Fearon (2003)14; Paldam
(2001)15; and Schultz & Strauss (2007)16. One of the routes through which they are related is the effect
of a fractionalized society on trust; distinct group-forming in society incentivizes officials to allocate resources to their own group and themselves. In addition to these diversity variables, population size is used to capture the effect of market size and possible increasing returns to scale which emanate therefrom.
To test whether corruption perceptions are indeed related to ethnic and cultural fractionalizition, as well as population, a principal component analysis can be used. Since there is strong multicollinearity between the variables on ethnic and cultural fractionalization, a simple OLS estimation would instead give biased estimates. The result is shown in Table 6.
Table 6
Principal component analysis. Dependent variable: Corruption Coefficient
Principal component factors 0.829***
(0.231)
Constant -4.568***
(0.230)
Observations 98
R-squared 0.118
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
The three instrumental variables are indeed strong joint predictors of our corruption perception
variable. They are therefore used in the following Two-Stage Least Squares regression to alleviate the endogeneity and simultaneity issues that may have been present in the Ordinary Least Squares
regression earlier. The results are shown in Table 7 (next page).
13 Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S. and Wacziarg, R. (2003). Journal of Economic Growth, 8(2), pp.155-194.
14 Fearon, J. D. (2003). Ethnic and cultural diversity by country. Journal of Economic Growth, Vol 8.
15 Paldam, M. (2001). Corruption and Religion Adding to the Economic Model. Kyklos, 54(2&3), pp.383-413. 16 Schultz, T. and Strauss, J. (2007). Handbook of development economics. Vol 4, Ch 50.
Table 7
2SLS. Dependent variable: Agency performance
Explanatory Variable Coefficient
Estimate
z-stat p-value
Corruption (instrumented) -0.166* -1.808 0.0706
R&D rate 0.235** 2.368 0.0179
Capital form. rate 0.00715 0.607 0.544
Comp. intensity 0.608*** 4.204 2.62e-05
Per capita income 0.134 0.890 0.374
Country size 0.0989*** 3.022 0.00251
Constant -2.475** -2.056 0.0398
Observations 70
R-squared 0.862
Durbin-Wu-Hausman p-value 0.728
Overidentifying restrictions p-value 0.500
Robust standard errors used *** p<0.01, ** p<0.05, * p<0.1
The instrumental variable specification accomplishes several results. First and foremost, by using ethnic and cultural fractionalization data as a proxy, the effect of corruption on agency performance has decreased considerably in significance; where it was highly significant in previous estimations, it is now a predictor only at the ten percent level. Furthermore, the use of proxies has had a marked effect on the percentage of variance captured by income per capita, which can no longer be said to predict agency performance.
Most importantly, the use of our instruments has not changed the estimate for R&D expenditure a great deal. Both the magnitude and significance of its effect have remained relatively constant, although it is no longer significant at the 1% level. Capital formation, which was included to capture investment unrelated to innovative purposes, still appears to hold no explanatory power when it comes to competition agency performance.
The decreased p-value of R&D expenditure should, however, be taken note of. After all, the hypothesis specification called for a two-tailed test, meaning that its value should be multiplied by a factor of two to give an accurate reading of the significance level of the result. In the section that lays out the
theoretical framework, it was discussed at length that the sign of the effect of past innovation levels could not be known a priori, necessitating stricter boundaries for achieving a positive result. As it happens, the p-value of the variable is comfortably below the 2.5% cutoff value, meaning that the result of the IV estimation above satisfies the condition for a significant result at the 5% level.
Also shown in Table 7, the null hypothesis of a post-estimation Durbin-Wu-Hausman test for endogeneity cannot be rejected. Furthermore, the null hypothesis for overidentifying restrictions is also not rejected. As a result, the Two-Stage Least Squares estimation should be considered an improvement over the OLS estimation, since it alleviates a potential source of endogeneity while evading complications which may arise from instrumental variable analyses.
4.3 Robustness tests
To test the robustness of the instrumental variable specification, different types of tests should be used for comparison. Wildly different estimation results could indicate weak instruments or model misspecification. Therefore, in addition to the Two-Stage Least Squared used above, Table 8 (next page) shows the result of a Generalized Method of Moment and a Limited-Information Maximum Likelihood test.
Table 8
Different IV-estimators. Dependent variable: Agency performance
(1) (2) (3)
Explanatory Variable 2SLS GMM LIML
Corruption (instrumented) -0.166 -0.176* -0.175
(0.0918) (0.0858) (0.113)
R&D rate 0.235** 0.222** 0.228*
(0.0991) (0.0948) (0.110)
Capital form. rate 0.00715 0.00978 0.00743
(0.0118) (0.0119) (0.0119)
Comp. intensity 0.608*** 0.611*** 0.600***
(0.145) (0.146) (0.159)
Per capita income 0.134 0.102 0.122
(0.151) (0.138) (0.176) Country size 0.0989*** 0.0963*** 0.100*** (0.0327) (0.0317) (0.0343) Constant -2.475* -2.262* -2.383 (1.204) (1.152) (1.400) Observations 70 70 70 R-squared 0.862 0.860 0.861
Robust standard errors in parentheses *** p<0.01, ** p<0.025, * p<0.05
Table 8 above shows estimate coefficients and their p-values for the three different tests. On the whole, the estimates do not change a great deal depending on the test used. Certainly none of the coefficient estimates change signs, and the strongest predictors retain their status. However, our variable of interest, R&D expenditure, does have one caveat. As indicated by the asterisks (which, for clarity of this point, represent a non-standard p-value sequence by which two asterisks represent a 2.5% level), the significance level of R&D expenditure rises above 2.5% for the LIML test. As discussed earlier, to reject our null hypothesis, the 2.5% level is required for the adoption of the two-tailed alternative hypothesis. Both other tests, however, do satisfy this condition.
While an interesting discovery, it does not make sense to give more credence to the LIML estimate over the others, for two reasons. Firstly, the most standard IV-estimator is the 2SLS, which is why it was used in the first place, before analyzing the robustness. Secondly, the LIML estimator is developed
and used specifically as an alternative for when weak instruments or overspecification are a concern17,
both of which were tested before and shown not to be an issue.
It makes sense to test the robustness of the coefficient for innovation for different model specifications. After all, not all variables in the original specification proved significant predictors. Table 9 shows what happens to the estimate by adding variables in a step-wise manner. Since the IV-specification was used to alleviate endogeneity in the corruption variable and not in the R&D rate variable, the first and simplest model is an OLS estimate, solely considering the effect of innovation on agency performance. All subsequent models are 2SLS.
Table 9
Stepwise model specification. Dependent variable: Agency performance
OLS (2) (3) (4) (5) (6) Explanatory Variables R&D rate 0.904*** 0.848*** 0.498*** 0.257** 0.234** 0.235** (0.0916) (0.206) (0.135) (0.105) (0.104) (0.0991) Corruption (instrumented) -0.0265 -0.0273 -0.194*** -0.190* -0.166 (0.0810) (0.0679) (0.0587) (0.0914) (0.0918) Comp. intensity 0.885*** 0.635*** 0.610*** 0.608*** (0.147) (0.148) (0.148) (0.145) Country size 0.0906*** 0.0953*** 0.0989*** (0.0314) (0.0313) (0.0327)
Per capita income 0.0561 0.134
(0.146) (0.151)
Capital form. rate 0.00715
(0.0118)
Constant 3.321*** 3.239*** -0.741 -1.252** -1.664 -2.475*
(0.115) (0.268) (0.546) (0.514) (1.199) (1.204)
Observations 71 71 71 71 71 70
R-squared 0.595 0.623 0.811 0.852 0.853 0.862
Robust standard errors in parentheses *** p<0.01, ** p<0.025, * p<0.05
The above table shows that R&D rates consistently explain variation in agency performance at at least the 2.5% significance level, which was required to reject the null hypothesis. Interestingly, the specification which assigns the lowest degree of significance to R&D rates is the one which excludes
17 Anderson, T. and Rubin, H. (1949). Estimation of the Parameters of a Single Equation in a Complete System of Stochastic Equations. Ann. Math. Statist., 20(1), pp.46-63.
capital formation. This result seems easily explainable: capital formation and innovation constitute a related investment decision for firms, and there is likely to be a degree of substitution between the two. To conclude this section, there appears to be no reason to question the validity of the empirical result. The relation suggested by the analysis is consistent with the literature that proposes a causal positive effect between the level of R&D spending in an economy and the vigor with which the nation conducts competition policy.
5. Conclusion
This thesis set out to investigate whether there is a relation between innovation and competition agency performance. In light of the findings described above, the null hypothesis that states that there is no identifiable relation, whether positive or negative, can be rejected. Both Ordinary Least Squares as well as Instrumental Variable specifications support the research hypothesis that there is a relation in the data used. The positive relation between past R&D rates and agency performance discovered in the estimations consistently shows probability values below the 2.5 percent mark, which, due to the two-tailed nature of the methodology design, ensures a Type 1 Error rate Alpha under five percent. This result holds up under robustness tests that incorporate different IV-estimators and different specifications of the regression model. In order to capture the possibly clouding effect of more widely-defined investment levels, capital formation rates were also included in the models. These rates were shown not to be significant predictors, suggesting that it is indeed research and development rates specifically, and not investment in general, that has explanatory power for agency performance.
It goes without saying that the positive regression coefficient does not disprove the existence of channels through which innovation may prompt lower levels of agency performance, whether through direct effects or by influencing the decision of policymakers. Rather, we can say that the findings support the notion that effects in the opposite direction are of greater magnitude, at least within the scope of the data used. It therefore lends additional credence to the work discussed in the first section, of, among others, Cambini, Rondi, Von Graevenitz, Averch, Johnson, and Guerriero.
With this in mind, it is appropriate to call for further empirical research regarding this matter. From the outset, the aim of this thesis was to conduct the investigation using the exact data, where possible, that
was used in Rodriguez & DeNardis' 2007 paper, and to extend their methodology to incorporate innovation. The advantage of this methodology is that the discovered results can easily be cross-examined with already established findings on agency performance, as there can be no question that the result may be due to different samples or measurement strategies.
A conspicuous direction for further research, therefore, is to extend the analysis beyond the scope of these data. Although sufficient to render a positive statistical result, one of the drawbacks of the dataset was that it concerned agency performance data from 2004. While not in and of itself a problem, a limiting consequence of this lies in the availability of R&D rate data in the period before this year. After all, many developing nations do not have extensive records of historical R&D rates before this time. Since then, however, nations and international organizations like the World Bank have kept a closer eye on innovation rates in the economy. Furthermore, the fact that the provenance of agency performance data is also part of an ongoing endeavor, there is surely fertile ground to extend the analysis over a greater empirical picture.
6. Appendix
Much of the data used in the analysis was taken from the dataset used by Rodriguez & DeNardis in their 2007 paper Examining the Performance of Competition Policy Enforcement Agencies: A
Cross-Country Comparison. Those variables that were sourced from elsewhere came from reliable and
publicly accessible authorities.
Variables description
Agency performance: The specification of this variable is discussed extensively in Section 3. The
source of the data is the Global Competitiveness Report from the World Economic Forum of 2004. Each agency is assigned a rating between one and seven, one being the poorest possible rating, based on survey data.
Common law: A dummy variable which equals one if the legal origin of the nation is common law. The
sources for these data are Lee (2005), Mahoney (2001), and the CIA Factbook.
Income per capita: Specified as GDP per capita. Source is the online database of the World Bank, from
the section “World Development Indicators”
Experience with modern competition policy: Specified as years since the foundation of a formal
competition authority. The data come from several sources: Lee (2004), the Global Competition Forum online database, the Organization of American States (2002), and updated and expanded by Rodriguez & DeNardis (2007).
Intensity of competition: Data on local competitive environment comes from the same source as those
on agency performance, and shares its functional specification. Competition intensity is rated from one to seven, seven being the most competitive. Based on survey data.
Corruption: Discussed in Section 3. Source for these data is Transparency International’s “Corrupt
corruption.
Ethnic and cultural fractionalization: The main source for these data comes from Fearon (2003).
However, for four nations this dataset had no entry; for these nations (Hong Kong, Iceland, Serbia & Montenegro, and Luxembourg) data was used from Alesina et al (2003).
R&D expenditure: Research and development rates in the 5-year period from 1996 and 2000, taken
from the online database of the World Bank.
Capital formation: Capital formation rates in the 5-year period from 1996 and 2000, taken from the
online database of the World Bank.
Country size: A measure that reflects the physical area of the nation in 2004, taken from the World
Bank’s World Development Indicators online database.
Population size: The number of residents of the country in 2004, taken from the World Bank’s World
7. Bibliography
Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S. and Wacziarg, R. (2003). Journal of
Economic Growth, 8(2), pp.155-194.
Anderson, T. and Rubin, H. (1949). Estimation of the Parameters of a Single Equation in a Complete System of Stochastic Equations. Ann. Math. Statist., 20(1), pp.46-63.
Averch, H., & Johnson, L. L. (1962). Behavior of the firm under regulatory constraint. American
Economic Review, 52, 1059–1069.
Cambini, C. and Rondi, L. (2009). Incentive Regulation and Investment: Evidence from European Energy Utilities. J Regul Econ, 38(1), pp. 1-26.
CIA Factbook (2006). www.cia.gov/cia/publications/factbook
European Commission (2014). Supporting R&D and innovation in Europe: new State aid rules.
Competition Policy Brief, paper by the Competition Directorate–General of the European
Commission.
Fearon, J. D. (2003). Ethnic and cultural diversity by country. Journal of Economic Growth, Vol 8. Global Competition Forum. www.globalcompetitionforum.org.
Global Competitiveness Report. World Economic Forum: www.weforum.org.
Guerriero, C. (2011). Accountability in Government and Regulatory Policies: Theory and Evidence.
Journal of Comparative Economics, 39: 453-469.
Guerriero, C. (2013). The Political Economy of Incentive Regulation: Theory and Evidence from US States. Journal of Comparative Economics, 41: 91-107.
Guerriero, C. (2015). The Political Economy of (De)Regulation: Theory and Evidence from the U.S. Electricity Market. Journal of Comparative Economics, 41: 91-107. Manuscript submitted for publication.
Hall, B., Mairesse, J. and Mohnen, P. (2009). Measuring the Returns to R&D. Handbook of the Economics of Innovation, published 2010.
Kee, H. and Hoekman, B. (2007). Imports, entry and competition law as market disciplines. European Economic Review, 51(4), pp.831-858.
Lee, C. (2005). Legal traditions and competition policy. The Quarterly Review of Economics and
Finance, 45(2-3), pp.236-257.
Mahoney, P. G. (2001). The common law and economic growth: Hayek might be right.
The Journal of Legal Studies, 30:503-525.
Martin, S. (1998) Product market competition policy and technological performance. Included in
Market Structure and Competition Policy: Game-Theoretic Approaches, ed. Norman & Thisse,
published 2004.
Organization of American States (OAS). (2002). Inventory of Domestic Laws and Regulations relating to Competition Policy in the Western Hemisphere. Washington DC: Organization of American
States.
Paldam, M. (2001). Corruption and Religion Adding to the Economic Model. Kyklos, 54(2&3), pp.383-413.
Rodriguez, A.E. and DeNardis, L. (2007). Examining the Performance of Competition Policy Enforcement Agencies: A Cross-Country Comparison. Journal of Business & Economic Studies, Vol. 13, No. 1, Spring 2007.
Schultz, T. and Strauss, J. (2007). Handbook of development economics. Vol 4, Ch 50.
Thakor, R. and Lo, A. (2015). Competition and R&D Financing Decisions: Theory and Evidence from the Biopharmaceutical Industry. Available at SSRN Electronic Journal.
Von Graevenitz, G. (2005). Integrating Competition Policy and Innovation Policy: The Case of R&D Cooperation. Available at SSRN Electronic Journal.
