Does the national level of banking competition affect innovation levels : a cross-country analysis

(1)

University of Amsterdam BSc Economics & Business Finance & Organization

Does the national level of banking competition affect

innovation levels: a cross-country analysis.

Bachelor thesis

Author: Floris Meijer

Student number: 10283749 Supervisor: Robin Döttling

(2)

Hierbij verklaar ik, Floris Meijer, dat ik deze scriptie zelf geschreven heb en dat ik de volledige verantwoordelijkheid op me neem voor de inhoud ervan.

Ik bevestig dat de tekst en het werk dat in deze scriptie gepresenteerd wordt origineel is en dat ik geen gebruik heb gemaakt van andere bronnen dan die welke in de tekst en in de referenties worden genoemd.

De faculteit Economie en Bedrijfskunde is alleen verantwoordelijk voor de begeleiding tot het inleveren van de scriptie, niet voor de inhoud.

(3)

Abstract

This research aims to provide further evidence for the relation between banking competition and innovation. Several research has been done with opposite outcomes. In contrast to previous work, this research has an international focus. A panel data set, consisting of 59 countries and 3 years, was created and used to test the aforementioned relation. The dependent variable is the innovation index as composed by Insead. The Lerner index in the banking market, as well as bank concentration, are used as measures of banking competition. GDP, Education, domestic credit and time dummies were included as control variables. Finally, no conclusive result for the relation between banking competition and innovation is found.

(4)

1. Introduction

In recent decades, the financial sector has grown fast, with 13 basis points per year between 1980 and 2006, as measured by US Bureau of Economic Analysis, (Greenwood & Scharfstein, 2013). Research, from as early as Schumpeter (1911), shows how the financial sector is an essential part for capital allocation and economic performance. The growth of the financial sector does lead one to question whether or not this is a good development for the economy, even aside from financial crises. Goldsmith (1969) presented that an increase in the size of the financial sector leads to long- run economic growth. King and Levine (1993) argued financial depth could be used as estimation for economic growth.

Solow (1957) presented, in the famous Solow Growth model, that economic growth is foremost driven by technical change (innovation). The aim in this paper is to assess the relation between the growing financial sector and economic growth. Therefore, we look at the effects that the financial sector has on innovation (in the spirit of the main driver of Growth according to the simple Solow model).

More recent, Cornaggia et al. (2015) found that an increase of banking competition has a negative effect on innovation in the United States (on state level). This effect is primarily driven by three channels, which will be explained in the literature review. However, the counter intuitive result that this paper provided was a reason to test the external validity. In this research a panel data set including over 60 countries and multiple years was compiled to test whether this effect also exists on a global level. The research question answered in this paper can be formulated as follows:

Does the national level of banking competition affect innovation levels: a cross-country analysis

In this paper, the Random Effects and the Fixed effects model is used to find the relation between banking competition and innovation. Multiple tests (Hausman test, F- test, fixed time effects test) have been done, to get the appropriate equation and test method, which includes some control variables as well as time dummies, and the most efficient test method. A positive relation between banking competition and innovation is found in 3 out of 5 estimations, as can be seen by the negative sign for the Lerner index. Moreover, for all estimations, the results are not statistically significant (at the

(6)

5% level). Altogether, in this research we were unable to provide further evidence for the (negative) relation between banking competition and innovation.

This paper starts with a review of the existing relevant literature. Subsequently, in the data and methodology section the compiled data will be described in addition to an explanation of the research method. In the section results output of the regression will be presented and discussed. Afterwards, the limitations of this research are outlined. Finally, a summary of the research and most important findings will be presented in the conclusion.

(7)

2. Literature review

During the last decade, the financial sector has been subject of public and political debate. The peak was in 2006, when the financial sector covered 8.3 percent of US GDP from 2.8 percent in 1950. This is measured by BEA, US Bureau of Economic Analysis. Following to the methodology of Philippon (2012), the growth of the financial sector was, in the years between 1950 and 1980, 7 basis points per year. In the years between 1980 and 2006, the growth increases to 13 basis points per year, according to Greenwood and Scharfstein (2013). Schumpeter (1911) argues that the financial sector is an essential part for capital allocation and economic growth. Greenwood and Scharfstein (2013) argue growth of finance comes together with two activities: asset management and the provision of household credit. The household credit grew from 48 percent of GDP in 1980 to 99 percent in 2007.

Goldsmith (1969) has found a positive relation between the size of the financial sector and the long- run economic growth. He argued the increasing size of the financial sector improved the efficiency, what resulted in economic growth. After Goldsmith (1969), more economists started looking for positive relations between the size of the financial sector and economic growth. King and Levine (1993) showed financial depth could be used as an estimation for growth of economy. Levine and Zervos (1998) founded a relation between the liquidity of a countries’ stock market and it’s GDP.

Solow (1957) argues that technical change is one of the main drivers of economic growth. In addition, the increase of the financial sector in the last decennia (Cecchetti & Kharroubbi, 2015) has enhanced competition between banks. However, the (empirical) effects of increased competition in the banking sector on innovation, have been documented pretty well in recent years.

There has been a lot of criticism on finance, which goes as far back as the Old Testament (Zingales, 2015). More recent problems arisen in the finance sector are e.g. the Libor scandal and the financial debt crisis. Zingales (2015) argued that, as a consequence, trust towards bankers decreased tremendously. Besides, finance academics should be aware of the important role that the financial sector has in society nowadays (Zingales, 2015). The author thinks that the positive role of finance depends on the publics’ perception of the sector.

(8)

Demetriades and Hussein (1996) used a sample of 16 countries in time series techniques and they did not find evidence of a causal relation between finance and economic growth. Demetriades and Law (2006) provided evidence that financial depth has no effect on growth in poor countries. Furthermore, Rousseau and Wachtel (2002) concluded financial depth has no impact on the national economy beyond a certain (double- digit) inflation threshold. In addition, Arcand et al. (2012) find that economic growth becomes negative when the percentage credit to private sector reaches 80-100%.

Most related to this research, the paper “Does banking competition affect innovation?” by Cornaggia et al. (2015) intends to study how innovation reacts to an increase in competitiveness in the banking sector. The authors exploit the deregulation of interstate bank branching laws, in the U.S. from 1994 to 1997, to test this effect. The authors find that an increase in banking competition decreases aggregate state-level innovation. In addition, the paper shows this decrease is mainly driven by (larger) public firms. For the effect on private firms, no conclusive evidence is found. Cornaggia et al. (2015) argue that banking competition could explain innovation through 3 different channels. First, firms that have high external finance dependence could benefit from the increase in competition, since these firms are likely to have access to more funds after the deregulation. This should mainly affect (smaller) private companies; the data provides evidence for this theory. Second, companies that had out-of-state banking relations before the event were, assumedly, unable to fulfill their financial needs close to home, previous to the deregulation. The increase in bank finance availability should drive innovation (if the funds are allocated to R&D). Finally, mergers and acquisitions should explain an observed decrease in innovation by large corporations. Since owners of small firms are more likely to get bank loans in the new situation and prefer control (funding with debt instead of shares), there are less willing targets for acquisition available to the large firms. To complete the argument, it is argued that a significant part of innovation of large corporations exists of buying smaller, innovative firms (‘M&A innovation’).

The main contribution of Cornaggia et al. (2015), the result that banking competition decreases innovation in aggregate and for larger firms, in combination with the findings of Arcand et al. (2012), brings up an interesting research question. These results could lead one to think that there is inefficiency in the market. With more banking competition, young innovative firms are less likely to be acquired by

(9)

larger companies (they have access to enough funds to operate independently). In this way, one could hypothesize that valuable experience and expertise is not transferred from larger companies to smaller (start-up) firms. This inefficiency could (at least partially) explain the overall decrease in patents observed. To test whether this hypotheses holds, it might be a good start to figure out whether the effect found by Cornaggia et al. (2015) is also true worldwide (as opposed to only the interstate in the United States).

(10)

3. Data & Methodology

In this section we present and explain used data and methodology. To start with, the dependent and independent variables will be explained, together with the used data. Afterwards, a Hausman test will be done to find out which model should be used. The equation for panel data will be presented. To end with, a test for time- fixed effects will be done.

3.1 Data

As the dependent variable, ‘Innindex’, the Global Innovation Index (GII) from Insead is used. The GII covers 141 different economies globally. The GII consists two elements: ‘the Innovation Input Sub-index’ and ‘the innovation Output Sub-index’. The pillars for the input index are: institutions, human capital and research, infrastructure, market sophistication and business sophistication. Knowledge and technology output and creative outputs are used as pillars for the Output Index. In the panel data section the GII of 2011, 2012 and 2013 are used. The measurement of the index changed in 2011 so it was not possible to include prior years.

The Lerner index, ‘Lerner’, measures market-power in the banking market. The Lerner index compares output pricing and marginal costs (relative to prices). Output pricing is the total bank revenue over assets. Marginal costs are calculated by the derivate from a log cost function. The formula used to compute the Lerner index is (P-MC)/P. An increase in the Lerner index indicates a decrease in the competitiveness within the financial intermediaries, so high values of the Lerner index indicate less competition between banks. The index ranges values are between 0 (very competitive) to 1 (not competitive). Lerner index figures follow from the methodology described in Demirgüç-Kunt and Martínez Pería (2010). An advantage of the Lerner index against the H–statistic, is that Lerner index can be calculated each point of time, as it is not long run equilibrium. The H-statistic measures market concentration. The dataset of the Lerner index can be found in the global financial development database of the World Bank.

For the independent variable ‘Educindex’, the education index composed by the United Nations is used. It uses 79 indicators across a range of themes. Much of this weight is given to mean years of schooling and expected years of schooling.

(11)

‘DomCredit’; domestic credit to private sector, refers to all finance resources provided to the private sector by financial corporations. Financial corporations include monetary authorities, deposit bank and other institutions that provide lease and finance activities. Domestic credit is measured as a percentage of the countries’ GDP. It is the country median. The domestic credit data can be found in the global financial development database of the World Bank.

‘GDP per Capita’ is the countries’ gross domestic product divided by midyear population of the country. GDP is the sum of gross value and all resident producers plus any product taxed minus any subsidies, which are not included in the value of the products. Data is in 2005 U.S. dollar. GDP per capita is transformed to log GDP per capita to make the effect of GDP per capita on the innovation index more clearly visible. It is easier to capture the effect of the coefficient since it would be really small if no log transformation were used. If GDP per capita increases with 1 dollar (unit), innovation index will increase with 𝛽 %. GDP per capita is included in the financial development database of the World Bank.

As a robustness check, bank concentration is added. It is the percentage of assets owned by the three largest banks in a country divided by total assets of all the banks. The World Bank includes this data in global financial development database. However, Bankscope originally measured it.

Moreover, an interaction variable is introduced for GDP* Lerner. This measures whether the relation between banking competition and innovation is different for countries with high GDP, compared to poorer countries (or vice versa), as a control instrument for developed or less developed countries. GDP used in this interaction variable is GDP per capita in log.

(12)

Summary Obs. Mean Std. Dev. Min. Max. Innovation index 258 40.12 11.52 18.6 68.2 Lerner 318 0.29 0.13 -0.03 0.94 Education 477 0.62 0.17 0.19 0.93 GDP*Lerner 303 3627 6206 -137 36421 Domestic credit 492 59.21 51.04 4.02 305.09 GDP per capita 546 10372 15283 152 81852 Bank concentration 372 68.38 20.07 7.25 100 Figure 3.1

Correlations Inn. Lerner Educ. Domcr. Bankc. GDP/c GDP*Lerner Innovation index 1.00 Lerner index -0.05 1.00 Education 0.75 -0.11 1.00 Domestic credit 0.79 -0.08 0.55 1.00 Bank concentration -0.11 0.18 -0.13 -0.01 1.00 GDP per capita 0.87 -0.02 0.84 0.74 -0.05 1.00 GDP* Lerner 0.24 0.94 0.17 0.17 0.15 0.30 1.00 Figure 3.2

Singapore is the maximum value in figure 3.1 for the Lerner index. The Lerner index for Singapore is 0.94 in 2013, which results from the concentration of banks, which is 84.29% in Singapore in 2013 (asset of the three largest banks as a share of total commercial bank assets).

Montenegro has a negative value for the Lerner index in 2012. The banks in Montenegro were hit hard by the European credit crisis and had to deregulate. Therefore the return on assets for Montenegro’ banks were negative in 2012, S. Vuković (2014). For this reason the Lerner index is negative for Montenegro in 2012. This also explains the negative minimum for GDP* Lerner, which is shown in figure

(13)

3.1. This negative outlier didn’t result in any problems in the tests. Moreover, Figure 3.1 shows that some countries have banking concentration of 100%, these are undeveloped countries where the banking sector is highly regulated by their government, for example: Niger, Nicaragua and Mongolia.

Furthermore, the number of observations deviates per series, because some series are not measurable in countries. This reduces the number of countries, which have all the data available. This reduces the test capacity. Figure 3.2 shows the correlation between the different variables. In line with our expectations, GDP per capita has the highest correlation with the innovation index. In section 4, results this relation will be further explained.

3.2 (Durban- Wu-) Hausman test for panel data: fixed or random effects

Since the data for most of the variables is available for several years, it is an option to enter data points for different years for each variable. In the previous section, we started with cross sectional data. If we add time series data for each individual cross section, we create data with multiple entries (since there is more than 1 year) for each variable. The name of this type of data is panel data. On the one hand, an advantage of panel data is the possibilities of describing changes in estimates over time, on the other hand, effects can (often) be estimated in a better way since more data points are available. Especially the latter is relevant in this paper. Initially, a panel data set consisting of 7 years was created, however, in the end only 3 years were usable, more information can be found in section 3.3. Altogether, the characteristics are small t, (t=3) and many units (N=59).

To decide whether it is more appropriate to work with fixed effect or random effects, a Hausman test can be done. The null hypothesis is that the preferred model is random effects and the alternative hypothesis is that the most suitable model is fixed effects. More specifically, this test measures if the errors are correlated with the regressors (H1), or not (H0). The probability Chi2 is 0.3304, which is bigger than 0.05 (5% significance). As a result, H0 cannot be rejected and random effects model will be used. The reason for using random effects instead of fixed effects is that the variation across entities should be random and not correlated with the regressors or

(14)

𝑌𝑖𝑡 = 𝛽1𝑋𝑖𝑡+ 𝛼 + ɳ𝑖𝑡+ 𝜀𝑖𝑡

Equation 3.1

3.3 Test time- fixed effect

To measure whether the introduction of time-fixed effects would be a valuable addition to the model, an F test is used. This test can be used to see if all dummies for the years are jointly equal to 0 (H0), if so, then no dummies are necessary. The alternative hypothesis is that the dummies are (jointly) not equal to 0 (H1).

The result of this test is a p-value of 0.0000. The null hypothesis is rejected and the dummies are jointly not equal to 0, with a significance level of 1%. Therefore, time fixed effects will be used. We control for time effects because special events could possibly affect the outcome variable, as there has been, for example, a financial crisis during the timespan of our compiled dataset.

In the output, which is presented in the section results, the coefficients for the dummies of the years can be seen. The dataset included 7 years (2007- 2013). The estimation of the dummies for the years 2008- 2010, was between 4.37- 4.51 and after 2010 the coefficient were between 9.7- 11.3. This difference in estimation was not easy to explain based on statistics or natural events. Finally, it was found that the measurement method of the innovation index changed, as it was normalized in a scale from 0 to 7 in the year 2007 to 2010. After 2010 it was normalized from 0 to 100. In addition, Insead changed the pillars that measure the output index. After these findings, the sample was restricted, only 2011, 2012 and 2013 were used to get most reliable results. 𝐼𝑛𝑛𝑜𝑣𝑎𝑡𝑖𝑜𝑛𝑖𝑡 = 𝛽0+ 𝛽1∗ 𝐿𝑒𝑟𝑛𝑒𝑟𝑖𝑡+ 𝛽2∗ 𝐸𝑑𝑢𝑐𝑖𝑛𝑑𝑒𝑥𝑖𝑡 + 𝛽3∗ 𝐷𝑜𝑚𝑐𝑟𝑒𝑑𝑖𝑡𝑖𝑡 + 𝛽₄∗ 𝐿𝑂𝐺 𝐺𝐷𝑃 𝐶𝑎𝑝𝑖𝑡𝑎_𝑖𝑡 + 𝛽5∗ 𝐿𝑂𝐺 𝐺𝐷𝑃 𝐶𝑎𝑝𝑖𝑡𝑎_𝑖𝑡∗ 𝐿𝑒𝑟𝑛𝑒𝑟𝑖𝑡+ 𝛽6 ∗ 𝐵𝑎𝑛𝑘𝑐𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛_𝑖𝑡+ 𝛽₇∗ 2012 + 𝛽₈∗ 2013 + 𝜀_𝑖𝑡 Equation 3.2

(15)

4. Results

In section 4 the results will be presented. First, the results of the random effects model will be shown. Results from different models are revealed and the results will be compared. Subsequently, the coefficient for the Lerner index will be discussed, followed by the other control variables. Furthermore, the model will be evaluated. Finally, the outcomes for the model with fixed effects will be discussed briefly.

4.1 Random effects Random Effects Model (1) (2) (3) (4) (5) Lerner 1.68 -0.42 -0.79 -37.22 (2.64) (2.23) (2.22) (21.32)* Education 44.46 45.77 11.35 9.52 10.28 (7.02)*** (6.91)*** (7.63) (6.45) (8.72) Dom. credit 0.08 0.06 0.03 0.42 0.04 (0.02)*** (0.02)*** (0.02)* (0.18) (0.02)*** GDP / capita 5.06 5.12 3.87 (1.08)*** (1.01)*** (1.17)*** Bank concentration 0.01 0.00 (0.02) (0.02) GDP* Lerner 3.72 (2.22)* Year 2012 1.05 1.09 1.00 1.04 (0.31)*** (0.32)*** (0.32)*** (0.31)*** Year 2013 1.87 1.80 1.60 1.75 (0.32)*** (0.32)*** (0.32)*** (0.31)*** Observations 186 186 183 198 177 Adjusted R2 0.73 0.72 0.80 0.82 0.81

Constant included Yes Yes Yes Yes Yes

*Significance at 10% **Significance at 5% ***Significance at 1% Robust standard errors are used. Variable GDP per capita is in log.

(16)

Figure 4.1 shows the results of the random effect model. The coefficient of Lerner index is negative in model 2, 3 and 5. A negative sign implies that an increase in banking competition (a decreasing Lerner index) has a positive effect on innovation. This result is opposite to prior expectations based on the paper of Cornaggia (2015). In model 1 the Lerner coefficient is positive, this indicates a negative relation between banking competition and innovation. But this probably occurred because there is an omitted variable (GDP) in the model, which causes an upward bias for the coefficient of the Lerner index.

In model 3, GDP per capita is added to the original model, which is significant (1%) in all 3 models. The coefficient for the Lerner index is negative. In model 5, the Lerner coefficient has a negative effect on innovation index in this (at a significance level of 10%). This would imply a positive relation between banking competition and innovation.

Education is included in all 5 models as control variable. In model 1 and 2 it has a significant (1%) positive effect on innovation. However, the effect of education is no longer significant starting at model 3. This can probably be explained due an omitted variable bias, GDP per capita, which is added in model 3- 5. It is very likely that the effect of GDP is captured by education in the first two models (since the two variables are correlated).

Domestic credit is highly significant (1%) in the first two models. Domestic credit availability has a positive effect on the innovation index, because credit is needed for innovation.

The variable banking concentration is added to the model (4) as an alternative for the Lerner index, as a robustness check. The coefficient for banking concentration is positive in model (4), which means a negative relation between banking concentration and innovation. So, the (positive) sign is conform the theory of Cornaggia et al. (2015). However, the effect is very small and not significant at any level. In model (5) the coefficient for bank concentration is even closer to zero and also not significant. So, this robustness check is in line with findings in earlier models; we do not find any evidence for a conclusive (negative) relation between banking competition and innovation.

The interaction variable GDP*Lerner is added to the model, to test if banking competition has another effect in developed countries then in less developed countries. In the complete model (5), this interaction variable has 10 % significance.

(17)

This implies that there is a difference in the effect of banking competition on innovation between developed and less developed countries. If GDP is higher, the effect of the Lerner index on innovation is higher. A higher coefficient of Lerner index on innovation means a more negative effect of banking competition on innovation. To summarize, this interaction variable shows that in a country with high GDP, the effect of banking competition on innovation is more negative. This is in line with the results of Cornaggia et al. (2015), who finds a negative relation between banking competition and innovation in the US (which has a high GDP).

In section 3.3 (time-fixed effects), it was tested if the model would improve with time fixed effects. Dummies (time – effects) are added for 2011, 2012 and 2013. The dummy for 2011 is not shown in table. This is because setting dummy 2012 and dummy for 2013 to ‘0’, indicates data from 2011 is used (if all three years were included, the model would suffer from multicollinearity). The dummies for 2012 and 2013 both have high significant (1%) positive coefficients. This implies an increase in the innovation index during the years 2011 until 2013. This is an interesting little result. This may be due to political reasons, for example pressure from countries their government to companies or universities, which are composing the index. Or the creators of the report do not want to disappoint governments. Alternatively, this positive effect in innovation comes from the fact that the world is recovering from financial crises. Although, this effect should be (at least partly) offset by domestic credit availability, since domestic credit availability is negatively correlated with financial crises.

R-squared can be used to evaluate how much of the dependent variables are explained by the independent variable. In this particular case, adjusted R-squared is used, since this measure can also decrease upon adding additional variables (e.g. from model (1) to (2)). We cannot conclude to much with use of the regular R-squared because this is misleading, as R2 is always increasing when there are more variables (good or bad) added to the model. Between models 1 and 5, it can be seen that adjusted R-squared is mostly increasing. Eventually, in model 5 adjusted R-squared is 0.81, which implies that approximately 81% of the variation in the dependent variable is explained by the model.

(18)

4.2 Fixed effects Fixed Effects Model (1) (2) (3) (4) (5) Lerner 1.60 -0.82 -1.30 -13.72 (2.31) (2.33) (2.40) (20.43) Education 133.35 18.06 3.80 -18.22 5.41 (82.01) (62.31) (65.57) (70.53) (65.87) Dom. credit -0.02 -0.04 -0.03 -0.02 -0.02 (0.04) (0.03) (0.03) (0.03) (0.03) GDP/ capita -9.71 -11.26 -10.50 (5.77)* (5.54)** (6.85) Bank concentration 0.01 0.01 (0.02) (0.02) GDP* Lerner 1.08 (2.04) Year 2012 1.21 1.39 1.36 1.39 (0.33)*** (0.35)*** (0.33)*** (0.36)*** Year 2013 2.09 2.41 2.35 2.44 (0.32)*** (0.38)*** (0.38)*** (0.44)*** Obs. 186 186 183 198 177 Adjusted R2 0.51 0.00 0.77 0.78 0.76

Constant included? Yes Yes Yes Yes Yes

*Significance at 10% **Significance at 5% ***Significance at 1% Robust standard errors are used. Variable GDP per capita is in log.

Figure 4.2

Figure 4.2 presents results in the fixed effect model. These results are added as a robustness check. In the fixed effect model the results are not robust. The results are not significant, except from the variable GDP per capita in model (3) and (4).

Furthermore, the time dummies are the only significant estimations in model (2-5). These results are perfectly in line with the expectations, which are made in section 3.2. In section 3.2 a Hausman test has been done, with the outcome that the best test, regarding our data, is random effect model.

(19)

5. Limitations

5.1 Sample selection bias

Even though the original data set consisted of over 180 countries, the results were only based on observations within 59-66 countries (depending on the amount of variables included). For multiple countries, there were (some) data points missing in the World Bank database. This would not be a problem from the perspective of amounts of data, since there are enough countries and data points left. The problem lies in the fact that countries that are less developed (or poorer) have, on average, weaker institutions. Weaker institutions result in less data availability. Which poses the problem that, altogether, poorer countries have been left out more often than developed countries. Which results in an unfair representation (in the used sample) of the actual population. Even this, stand alone, would not be a problem if effects are likely to be uncorrelated to how developed a country is (for example, the number of days the sun shines in a year). In this specific case, though, innovation and banking competition are both extremely likely to be correlated to welfare and state of development as can be seen in the significant effects of GDP (welfare) and education (development). So there could be a bias in the estimation based on the causal (and non-random) relation between data (un-) availability and the coefficients measured in this paper. This effect is known as a sample selection bias.

The direction of this bias is not so easy to determine. To accurately do so, we would need to know what the difference is between the effect of banking competition on innovation, between developed and developing countries. If we assume that the negative sign we found in estimation 5 of the random effects model is correct, the bias could be described as follows. Rich countries are overrepresented (actually, poor countries are underrepresented but the effect is similar). In addition, Rich countries have a more negative relation between banking competition and innovation than poor countries. Combining these two arguments, the estimations should be too negative in our findings now, so the Lerner index is biased downwards (again, if the interaction effect captured in the model is correct).

(20)

5.2 Causality issues

A second limitation of this paper exists around the causality of the independent variable (innovation) and some of the explanatory variables. In regression analysis, the explanatory variable should explain some of the variation of the dependent variable and not the other way around. The problem is that innovation explains some of the GDP fluctuations; a country with higher innovation levels today is likely to experience higher future GDP. Since innovation levels seem to be pretty consistent over time, it is also correct to argue that innovation levels today are correlated to higher GDP levels today. The final argument can be written as following. A higher innovation level today is related to higher innovation levels yesterday and, consequently, higher GDP today. So, innovation has a second causal relation with GDP (the first causal relation is that higher GDP causes more innovation since among others more funds, education, data and infrastructure are available), this effect is called simultaneous causality and it is most commonly encountered when estimating price and demand (or supply). In our case, the estimation of the GDP coefficient is likely to be biased upwards because of this effect since both effects have a positive correlation. However, this problem is not too much of an issue since GDP is only a control variable.

Furthermore, it is also possible that the estimation of the Lerner index suffers from this bias. Higher innovation levels in the past would relate to more investment demand (in the past, current and future time periods), and investment demand basically is the demand for loans. Continuing, an environment with high demand for loans would make a great place for banks, since there is a lot of potential profit in the loans. And of course, higher concentration of banks means more competition and a lower markup, decreasing the Lerner index. This is another, probably weaker, phenomena of reversed causality (the causal relation that we try to measure is the effect that banks and the availability of funds have on innovation). Even though this relation may seem pretty fuzzy, it could have an impact on the estimations. Moreover, since this reversed causality is likely to affect the effect that we are trying to capture, it poses a threat to the internal validity of this paper.

(21)

6. Conclusion

The aim of this research is to test whether the results of Cornaggia et al. (2015) hold worldwide. Cornaggia et al. (2015) found that (in the US) an increase in banking competition led to a decrease in innovation. This effect was counterintuitive; one would expect a large banking sector in innovative environment. In this paper, the result of Cornaggia et al. (2015) is hypothesized to be explained as follows: in a normal situation (with low banking competition), small companies cannot obtain funds from banks because banks gives their available funds to large, more reliable companies. Therefore, in normal situation owners sell their firms to larger companies (which is the only way to obtain more funds and continue operations). In a situation with increased banking competition, small firms do get funding from banks and are thus not acquired by larger firms (because owners do not like to give away control unnecessary). In this paper it is hypothesized that this negative relation comes from the fact that there is no transfer of knowledge between the small, innovative companies and the larger, experienced companies.

A first step towards testing this hypothesis, would be investigating whether the negative relation found in Cornaggia et al. (2015), also holds worldwide (testing the external validity of the result). To test this, a panel data set consisting data from over 150 countries for 7 years is created. The dependent variable, innovation, is an index figure for each country, as created by Insead. As a proxy for banking competition, the Lerner index (for the banking market) and bank concentration of top 3 banks are used. In addition, control variables include GDP, education and dummies for different years. Unfortunately, there are some missing data points and 4 of the 7 years cannot be used due to a different manner of composing the innovation index before and after 2011. Finally, data from approximately 65 countries over 3 years was used for the estimations.

The results do partly agree with the results found by Cornaggia et al. (2015). Both positive and negative relations between innovation and banking competition are found in equations throughout the paper. Several control variables were included to reduce the chance of an omitted variable bias, this resulted in lower (absolute) coefficients and a non-significant variable of interest. In addition, the sign on the Lerner index coefficient remained negative in 3 of the 4 models. We saw a positive

(22)

was not significant. In conclusion, no significance (at the 5% level) evidence is provided that banking competition (negatively) affects inflation on a global scale.

Finally, two threats to internal validity were discussed. Firstly, a sample selection bias is very likely to influence the estimations since countries with a higher GDP usually have better data infrastructure and, thus, are more likely to show up in our final 65 countries than a poor country. Secondly, reversed causality may have affected the estimations since the dependent variable (innovation) is likely to drive (future) GDP and, to a lesser extent, banking competition (as opposed to the other way around).

(23)

References

Arcand, J. L., Berkes, E., & Panizza, U. (2012). Too much finance?.

Cecchetti, S. G., & Kharroubi, E. (2015). Why does financial sector growth crowd out real economic growth?

Cornaggia, J., Mao, Y., Tian, X., & Wolfe, B. (2015). Does banking competition affect innovation?. Journal of Financial Economics, 115(1), 189-209.

Demetriades, P., & Hook Law, S. (2006). Finance, institutions and economic development. International Journal of Finance & Economics, 11(3), 245-260. Demetriades, P. O., & Hussein, K. A. (1996). Does financial development cause

economic growth? Time-series evidence from 16 countries. Journal of development Economics, 51(2), 387-411.

Demirgüç-Kunt, A., & Martínez Pería, M. S. (2010). A framework for analyzing competition in the banking sector: an application to the case of Jordan. World Bank Policy Research Working Paper Series, Vol.

Goldsmith, R. (1969). Financial structure and economic development. New Haven: Yale University Pres.

Greenwood, R., & Scharfstein, D. (2013). The growth of finance. The Journal of Economic Perspectives, 3-28.

Levine, R., & Zervos, S. (1998). Stock markets, banks, and economic growth. American economic review, 537-558.

Philippon, T. (2012). Has the US finance industry become less efficient? On the theory and measurement of financial intermediation (No. w18077). National Bureau of Economic Research.

Rousseau, P. L., & Wachtel, P. (2002). Inflation thresholds and the finance–growth nexus. Journal of International Money and Finance, 21(6), 777-793.

Schumpeter, Joseph A. 1911. The Theory of Economic Development. Cambridge, MA: Harvard University Press.

Solow, R. M. (1957). Technical change and the aggregate production function. The review of Economics and Statistics, 312-320.

Vuković, S. (2014). Stress Testing of the Montenegrin Banking System with Aggregated and Bank-Specific Data. Journal of Central Banking Theory and Practice, 3(2), 85-119.

(24)

Zingales, L. (2015). Does Finance Benefit Society? (No. w20894). National Bureau of Economic Research.

Does the national level of banking competition affect innovation levels : a cross-country analysis