University of Groningen
Faculty of Economics and Business
Master Thesis
An analysis on new product innovation
and corruption
Groningen
2012
Student: A. Anca (s2016028)
Student e-mail: alexa_anca@hotmail.com
1
Abstract
This thesis analyses the relationship between corruption, measured by the percentage of annual sales paid as bribes, and new product innovation, measured by the introduction of new products. The study uses a sample of 7,500 firms from 29 transition economies. The analysis focuses on an estimation of a standard probit model but endogeneity is accounted for in the robustness check. The results indicate that there is a non-linear relationship between corruption and new product innovation. This relationship was not accounted for in the two studies found on product innovation and corruption.
2
Table of contents
1. Introduction ... 3
2. Literature review... 4
2.1 Corruption ... 4
2.2 Innovation and corruption ... 7
2.2.1 Theoretical studies ... 7
2.2.2 Cross-country empirical studies ... 8
2.3 Product innovation and corruption ... 9
2.4. Hypothesis ... 10
3. Data and methods ... 11
3.1. Data sample ... 11
3.2. Dependent variable ... 12
3.3. Independent variable ... 12
3.4. Control variables ... 13
3.5. The econometric model ... 16
3.5.1. Model specification ... 16 3.5.2 Econometric model ... 16 3.5.3 Issues ... 18 4. Empirical results ... 19 4.1 Descriptive statistics ... 20 4.2 Empirical results ... 21
4.2.1 Entire sample estimation ... 22
4.2.2 CEE and CIS sample regression estimates ... 24
4.3 Robustness check ... 26
5. Conclusions and limitations... 27
References ... 30
Appendix ... Error! Bookmark not defined.4 Appendix 1. Distribution of observations ... 34
Appendix 2. Descriptive statistics ... 35
3
1. Introduction
Innovation is one of the main drivers of progress and consumer well-being. Endogenous growth theory (Aghion and Howitt, 1999) states that innovation is a key driver of long-run growth. The authors acknowledge that innovation is a social process because the incentive of entrepreneurs to engage in innovative activities is conditioned by laws, institutions and regulations which affect the size of profits that can be generated. Innovative firms that activate in an environment with poor governance are often subject to corruption since corruption is present where poor governance is (Kaufmann, 2005).
Corruption can distort resource allocations and market competition (Shleifer and Vishny, 1993). These distortions can represent a major disincentive for firms to engage in innovative activities such as the creation of new products, processes and so on (Oslo manual, 2005). On the other hand, certain levels of corruption can be beneficial to development (Acemoglu and Verdier, 1998) or they can offer advantageous opportunities for entrepreneurs (Tanzi, 1998). Recent firm-level studies have found a positive non-linear relationship between corruption and firm performance (de Jong et al., 2012). Thesamepositive non-linear relationship has been reached in cross-country studies analyzing the link between corruption and growth (Mendez and Sepulvelda, 2006) and the control of corruption and innovation (Anokhin and Schultze, 2009).
4
Commonwealth of Independent States (CIS) members and Turkey. The privatization process experienced by the transition economies has been found to increase the occurrence of corruption (Fan et al., 2009) and this makes the sample suitable for the study of corruption.
Taking into consideration the findings in the literature for corruption and innovation the research question used for this thesis is the following: Is there a non-linear relationship between
corruption and new product innovation?
The analysis is made using a standard probit model and the robustness of the estimates against the endogeneity issue of corruption is assessed by means of a two-stage instrumental variable probit model. The results obtained show a non-linear relationship between corruption and new product innovation. However, the functional form of the relationship differs depending on the estimator.
The remaining of the thesis is organized as follows. Section two addresses the theoretical and empirical studies relevant to this study. A description of the data and the methodological issues are dealt with in section three. Afterwards section four presents the empirical results and in section five conclusions and limitations of the study are included.
2. Literature review
The literature review is divided into three sections. In the first section a review is presented consisting of studies dealing with corruption. The second section incorporates a review of studies on innovation and corruption, while the last section presents a review of the literature which strictly deals with product innovation and corruption.
2.1 Corruption
5
al., 2012) or both. The definition used in this thesis is given by de Jong et al. (2012), and states that corruption is “the payment of cash by an organization with the aim of influencing the actions
of a private official”. In this definition the firm is seen as the initiator of the illicit action;
corruption is narrowed to bribery, and it is implied that through bribery firms can achieve a particular advantage for them. Apart from bribery other forms of corruption are lobbying, nepotism, embezzlement and fraud (Hallack and Poisson, 2001).
Other distinctive types of corruption have been conceptualized. For instance there can be bureaucratic/administrative petty corruption or grand corruption. Petty corruption is the engagement of a public official in a corrupt act when he is dealing with the public or with politicians, where small amounts of bribery are offered to speed up the bureaucratic process (Jain, 2001). Whereas, grand corruption is the use of power by the elite officials to influence economic policies, where typically higher bribes are involved (Jain, 2001). Hellman et al. (2003) suggest that small firms have the highest chances to be involved in petty corruption. Whereas, larger “de novo” firms are involved in grand corruption. They also find that unlike petty corruption grand corruption can increase firm performance. Furthermore, there have been noticed differences in the degree of cooperation between the involved parties, as such corruption is regarded as coercive or collusive (Tanzi, 1998). Another difference noticed in corruption is the effect it produces for the firm - cost-reducing or benefit enhancing (Tanzi, 1998). An example of cost-increasing coercive corruption used by Tanzi (1998) is when public officials harass enterprises, especially small firms, to force them to pay a bribe. Undoubtedly such type of corruption has a negative effect on entrepreneurship, firm performance, the probability to innovate and so on. On the other hand, the payment of bribes can be benefit-enhancing to firms when they can acquire profitable contracts and gain advantage in the face of competitors.
The different conceptualizations for corruption point out that it can provide several advantages and disadvantages for firms. Concerning the literature on growth, development and corruption, the views are as well set in favour of or against corruption.
6
government, and the bureaucratic administrative process in an environment where the government does not stimulate the businesses (Leff, 1964). Moreover, bribery can aid innovators in overcoming the opposition of incumbent producers that have lobbying power (Leff, 1964). In a similar line of reasoning Nye (1967) suggests that corruption can help economic development by stimulating capital formation, offering firms the means to cut red tape, and providing incentives for entrepreneurs. Moreover, certain levels of corruption are conducive to investments because prevention of corruption is highly expensive (Acemoglu and Verdier, 1998). They demonstrate that an optimal organization of society can function with corruption, rents and misallocation of talent.
In the long run, when firms give bribes to cut red tape officials are stimulated to raise regulation (Aidt and Dutta, 2008), thereby increasing inefficient allocation of resources. Misallocation of talent is also a result of corruption; this is produced when entrepreneurs invest more time in corrupt activities than in increasing productivity (Murphy et al. 1991). Moreover, Murphy et al. (1991) provide arguments why corruption is especially detrimental to innovators. The reasons are that innovators often lack financial means and bribery can be viewed as a burden to them. They are also subject to future expropriation risks because they have long term projects.
In view of the advantages and disadvantages of corruption the dominant empirical evidence supports the negative impact of corruption on growth (Mauro, 1995; Mo 2001), on firm productivity (De Rosa et al., 2010), on investment levels (Asiedu et al., 2009), and on product innovation (Mahagaonkar, 2008; Waldemar, 2011).
Although, the empirical evidence shows that corruption is a deterrent to growth and other channels affecting growth; some studies show that there are diminishing returns to bribery. Mendez and Sepulvelda (2006) find evidence, in a cross-country level study for the period of 1960-2000, of an inverted U-shape relationship between corruption1 and growth, measured as annual growth of GDP per capita. De Jong et al. (2012), in a firm-level study on 606 Vietnamese
1
7
firms from 2004, find evidence of an inverted U-shape relationship between corruption, measured by the amount of money paid to public officials, and firm performance, measured by the natural logarithm of total revenue. The positive non-linear relationship found in these studies shows that there are growth and firm performance maximizing levels of corruption.
2.2 Innovation and corruption
The Oslo Manual (2005) provides the most comprehensive and worldwide accepted definition of innovation and a framework for surveys to measure innovation. According to it, innovation comprises several categories: product, process, marketing and organizational innovation. Though, product innovation is a type of innovation this section excludes studies focusing on product innovation and corruption. The findings of the studies on product innovation and corruption are presented in the last section of the literature review.
The body of literature found on innovation and corruption comprises theoretical studies (Murphy et al., 1991, 1993; Shleifer and Vishny, 1993; Blackburn and Puccio, 2009; De Jardin, 2011) and cross-country empirical studies (Anokhin and Schulze, 2009).
2.2.1 Theoretical studies
8
Depending on the structure of corruption – organized or monopolistic and unorganized corruption (Shleifer and Vishny, 1993) – endogenous growth can occur at higher or lower rates.
Blackburn and Puccio (2008) developed a dynamic general equilibrium model where growth occurs endogenously and innovative firms need public goods from corrupt public officials. Their model supports the propositions that “the level of bribes under organized corruption is lower
than the level of bribes under disorganized corruption”; and “growth is higher under organized corruption than under disorganized corruption”. Therefore, innovation activity will be higher in
organized corruption. Under organized corruption the firm is ante-assured that it is not required to pay any other bribes and that it gets full property rights over the goods –e.g. licence, permits – bought (Shleifer and Vishny, 1993). Because the levels of bribes are set independently by various public officers unorganized corruption leads to higher levels of bribe paid and it is more disruptive to entrepreneurship activity since it limits the governmental trust. Unlike the previous case, the levels of bribes are set independently by various public officers. These differences in the organization of corruption explain the features exhibited in the pattern of the growth rate of GDP per capita and corruption in the South and South-East Asia (Blackburn and Puccio, 2008).
2.2.2 Cross-country empirical studies
This segment of study is marked by the research of Anokhin and Schulze (2009) which points out that a decrease in corruption can differ in the effect it has on innovative activities. The authors hypothesize an inverted U-shape relationship between the control of corruption2 and domestic innovation. The hypothesis is based on the argument that high levels of corruption deter innovative activities because they raise the uncertainty levels and costs. As such, modest improvements in the control of corruption affect positively the relationship between corruption and innovation. Their findings are that the control of corruption has a non-linear relationship with innovation. They analyze the implications of corruption on entrepreneurial activity and innovation using longitudinal data from 64 countries. Their findings differ according to the measure of innovation used. When innovation is measured by patents, the relationship between it and the control of corruption displays an inverted U-shape. In regard to another measure of
2
9
innovation - rates of realized innovation - the relationship found between innovation and the control of corruption is positive and linear.
2.3 Product innovation and corruption
Product innovation is defined in the Oslo manual (2005) as “the introduction of a good or service
that is new or significantly improved with respect to its characteristics or intended uses”. The
definition measures broad product innovation. If product innovation is measured by the intensity of the innovation achieved it can be disaggregated into new product innovation and improved product innovation. There are several factors that affect the likelihood of product innovating activities.
Firm-level product innovation is driven because of competitive pressure, the need to increase demand and market shares, to acquire more effective production and delivery methods, and because of the necessity to meet environmental requirements and to reduce environmental impact (Oslo manual, 2005). Several factor are reckoned to restrict product innovations, such as cost factors, knowledge factors, market factors and institutional factors - lack of infrastructure, weak property rights and legislation and regulations (Oslo manual, 2005). The latter two institutional factors mentioned are particularly problematic in corrupt environments.
The empirical research in this category is marked by firm-level studies that find evidence on the negative impact of corruption on product innovation (Mahagaonkar, 2008; Waldemar, 2011). Mahagaonkar’s (2008) uses a sample consisting of 3477 firms from seven African countries for the year 2004. The sample used was obtained from the World Bank Productivity and the Investment Climate Private Enterprise Survey. Waldemar (2011) uses a sample of 1600 Indian firms from 2005 which was obtained from the World Bank Enterprise Survey (WBES). Their estimations are done through the use of a probit model, since the dependent variable is a binary one.
10
corruption using an instrumental variable estimator of the probit model and his instruments are the government efficiency and the faith in the judiciary. It is worthy to point out that the author tests the effect of corruption on the other three types of innovation (process, organizational and marketing). To build his hypothesis Mahagaonkar (2008) takes into consideration both advantages and disadvantages of corruption and recognizes that innovative activities internal to the firm - e.g. process innovation – which do not require interaction with public officials are unaffected by corruption. In line with the advantages presented, the author finds evidence that marketing innovation is positively correlated with the amount of bribes as percentage of annual sales. The results from the estimation for process innovation show that corruption has an insignificant effect, whereas for organizational innovation and product innovation corruption has a negative effect.
The measure used for the dependent variable is the introduction of new products, whereas the measure for corruption used by Waldemar (2011) is the sector-state bribe average of the individual bribes reported as percentage of annual sales .The sector-state bribe average was first used by Svensson et al. (2007) as an instrumental variable and provides a solution to the issue of endogeneity of corruption. Waldemar (2011) uses the sector-state measure of corruption directly in the regression, but for the robustness check he uses it as an instrumental variable for corruption and the results remain unchanged.
2.4. Hypothesis
11
the positive ones. Nevertheless, the functional form of the relationship between corruption and product innovation has not been investigated thoroughly. We need to know the functional form of the relationship to choose suitable policies to combat corruption. Therefore, this thesis tries to fill the existing gap in the literature. In particular, the hypothesis is as follows:
H1: Corruption has a non-linear relationship with new product innovation.
3. Data and methods
In this section I present my firm level data and an overview of the econometric model employed.
3.1. Data sample
The World Bank Enterprise Surveys (WBES) and Business Environment and Enterprise Performance Survey (BEEPS) are the main sources of aggregate indicators available and used for firm level analysis of corruption. According to Urra (2007) an advantage of the BEEPS indicator is the margin of error with which it estimates. The disadvantage coming from perception problems due to the nature of the survey and the fact that corruption is an illicit activity is decreased as much as possible by the formulation of indirect questions aimed at capturing corruption.
12
terms of frequency, amounts and to some extent certain specific purposes. The respondents are business owners or top managers and this eliminates the probability of not being suitable representants of the company. The total number of sectors in which the companies activate is 18, under the two digit ISIC3 revision 3.1.
The final sample used in this thesis consists of firms from 29 countries (CEE, CIS and Turkey). An overall distribution of observations across the countries can be found in Appendix 1. The total number of observations used, after removing missing observations, is 7,442.
3.2. Dependent variable
New product innovation (ninnov): measures novelty in product innovation. The variable is
obtained from the question “In the last three years, has this establishment introduced new products or services?” The initial variable in the sample contains the following values: 1 - for an affirmative answer “Yes”, 2 – for a non-affirmative answer “No”, and -9 for an answer saying that the interviewee is unaware if the establishment introduced new products “Don’t know”. The affirmative answers remain coded with 1, the non-affirmative answers are re-coded with 0, and the answers coded with -9 are dropped from the sample.
3.3. Independent variable
Bribes (bribe): measures if the firm is engaged in corruption and to what degree its resources are
reallocated. The measure is obtained from the following question: “It is said that establishments
are sometimes required to make gifts or informal payments to public officials to “get things done” with regard to customs, taxes, licences, regulations, services etc. On average what percent of total annual sales, or estimated total annual value, do establishments like this one pay in informal payments or gifts to public officials for this purpose?” The answers are either in percent
of total annual sales or total annual value. The most suitable measure is the percentage because the value is given in local currency units and it would require currency conversion. As most of the firms answered in annual value, the annual value was converted into percentage of total
3
13
annual sales. Therefore, the independent variable bribe measures the percentage of annual sales given as bribes by the firm.
3.4. Control variables
In order to obtain reliable estimates it is necessary to control for firm characteristics, firm level innovation characteristics and market characteristics that were established in other empirical studies.
14
There are also other firm level innovation characteristics that have an impact on product innovation, such as the skills of employees or technological factors, but due to the unavailability of data, only the following control variables are included.
Firm characteristics
Size (size): measures the number of permanent employed personnel in the firm.
Age (age): is calculated by subtracting the year when the firm was registered from the year when
the survey was collected.
Private foreign ownership (pfo): measures if the firm is owned by foreigners, or not. This is a
dummy variable. The value of 1 is given for firms with foreign ownership higher than 50 percent. Otherwise it takes the value of 0.
Managerial experience (manexp): measures the number of years the manager has been working
in the industry.
Financial loan (finance): measures if the firm has taken a loan from a private bank, or not. This
is a dummy variable. The value of 1 is given for firms that have a credit line from a private bank. Otherwise, the value is 0.
Firm level innovation characteristics
Research & Development (rd): is a dummy variable which has the value of 1 when the firm has
invested in research and development in the past three years, respectively 0 when it has not invested.
Exporters (export): is an indicator of the potential of a firm to be engaged in information
spillovers. It is a dummy variable which takes the value of 1 when the firm has a share of its annual sales from exports. Otherwise it has the value of 0.
Market characteristics
Competition (comp): measures the perceived product market competition. The variable is
15
Industry and Country dummies
Industry dummies are used for the whole set of industries minus the construction sector, which
will be the omitted category.
Country dummies are used for the whole set of countries minus the one which will be the
omitted category.
Using all dummies would produce collinearity. Therefore, one of the dummies needs to be removed from the industry and country dummies.
Table 1 summarizes the expected signs of the coefficients of the independent and control variables used and the expected relationship they have with product innovation.
Table 1. Expectancy of coefficients
Variable Notation Expected sign Relationship expectancy
Percentage of bribes/annual sales
bribe
+/-
Non-linear relationship The squared percentage of
bribes
bribe2
+/-
Managerial experience manexp
+
Inverted U-shape relationship Squared managerial experience manexp2
-
Competition comp
+
Inverted U-shape relationship
Squared competition comp2
-
Size of company size
+
Age of firm age
-
Private foreign ownership pfo
+
Financial loan finance
+
Research & Development rd
+
16
3.5. The econometric model
This section covers a description of the econometric model employed, several issues regarding the variables and the econometric equation used
3.5.1. Model specification
The dependent variable used in this thesis is new product innovation and it is a binary variable representing the choice of introducing new products or not. When the dependent variable is binary or has a limited restrictive range the common models employed are the non-linear models. These models are called logit or probit. The advantage of using these models over linear models is that they overcome the drawbacks imposed by linear models (Wooldridge, 2002, chapter 13). The logit and probit models are very similar; the main difference consists in the distribution function they use. The model used in this thesis is the probit model and was used as well in other empirical studies addressing this topic (Mahagaonkar, 2008; Waldemar, 2011). The model uses the conditional maximum likelihood estimation. It assumes there is a latent variable (y*) where:
y
i*=x
i’
θ+e
i.Therefore, the dependent variable (yi,i=1,...n) indicates the sign of the latent variable (yi*):
y= .
From the above two equations we have the probability of y conditional on x.
P(y
i=1|x
i)=P(
>0|x
i)=P(x
i’
θ
+ e
i>0|x
i)=P(e
i>-x
i’
θ|x
i)= 1 -
Φ
(-x
i’
θ
) =
Φ
(x
i’
θ)
4.
The model uses the cumulative standard normal distribution (
Φ
) as it is presented in the last equation. An assumption of the model is that the disturbance process has a known variance (Baum, 2006).3.5.2 Econometric model
To test the hypothesis the following econometric model is used:
4
17
P(y
i=1|x
i)=
Φ
(x
i’β)
• yi` is new product innovation;
• P(yi=1|xi) is the probability of yi being 1, given xi;
•
Φ
represents the cumulative standard normal distribution function; • xi is a vector comprised of a set of variables;•
β
iis a vector of unknown parameters;• i is a transcript to denote firms. •
x
i’β = x
i1’β
1+ x
i2’β
2where:
x
i1’β
1= β
11bribe
i+β
12• bribe is the percentage of bribes given from annual sales; • bribe2 is the squared functional form of bribe;
x
i2’β
2= β
21ln(size
i) + β
22age
i+ β
23pfo
i+ β
24manexp
i+ β
25+
β
26comp
i+β
27+ β
28export
i+β
29rd
*i
+β
210finance
i+β
21jindustrydummies
j+β
21kcountrydummies
k• size is the natural logarithm of the size of the company;
• age is the age of the company;
• pfo is the dummy variable for private foreign ownership; • manexp is the managerial experience;
• manexp2 is the squared functional form of managerial experience; • comp is the measure for competition;
• comp2 is the squared functional form of competition;
• export is a dummy variable measuring if the enterprise is an exporter; • rd is a dummy variable for investments in R&D;
• finance is a dummy variable measuring if the firm has a credit line from a private bank; • industrydummies is a vector comprised of j (j= 1,17) dummy variables for each sector;
*
18
• countrydummies is a vector comprised of k (k= 18,45) dummy variables for each country.
3.5.3 Issues
In this section several econometrical issues covered are presented.
Multicollinearity: A prerequisite to ensure that the coefficients are consistent and unbiased
estimated is to check for outliers and collinearity in and between the variables from the sample used. The presence of outliers can influence the coefficients estimates. A graph matrix of the variables was used to check for outliers. The few influential outliers were deleted from the sample. To test the collinearity between the independent/control variables, a bivariate correlation test was used. Values higher than the cut-off point of 0.8 indicate collinearity issues (Adkins and Hill, 2008). No collinearity is found between the variables. To analyse the strength of the multicollinearity the variance inflation factor (VIF) analysis is used. The cut-off point recommended is 10 (Adkins and Hill, 2008). The only variables that exceed the recommended cut-off point are the squared variables. The trade off from using the squared functional form to account for non-linearity is multicollinearity. Therefore, these variables will remain in the model.
Normality: To investigate the normality assumption of distribution the Shapiro-Wilk (SW) test is
used on the residuals of the regression. The null hypothesis of normal distribution is rejected since the p-value of the SW test statistic is 0.000, which is lower than the significance level of 0.05. For the variables that present a higher skewness and kurtosis than the accepted levels of 0.3 for skewness and 3 for kurtosis, the functional form is modified. Due to the nature of the data the only adjustment is made on the variable size. The variable was transformed to its natural logarithmic form. Despite these modifications the null hypothesis of normal distribution is rejected. However, this is not a prerequisite of the probit model.
Heteroskedasticity: Another concern is the presence of heteroskedasticity which means that the
19
heteroskedasticity. However, for the probit model this test is not available. Therefore, to detect heteroskedasticity a scatter plot was used. The residuals were plotted against the independent variable bribe. A wider variation of the residuals is witnessed at higher levels of bribes given. This indicates the presence of groupwise heteroskedasticity (Adkins and Hill, 2008). To correct heteroskedasticity the option vce (robust) is used to obtain the robust estimate of standard errors.
Endogeneity: An issue coming from the study of corruption is the concern of endogeneity in
corruption. The presumption is that corruption is correlated with other unobserved characteristics of the firms or the environment affecting innovation and measured through the error term. This generates inconsistent least squares estimations (Adkins et al., 2008). If endogeneity is present then this should be corrected. A proper method used in econometrics to eradicate this issue is the use of instrumental variables. This particular method has been used in several studies dealing with corruption (De Rosa, 2010; Svensson, 2007; Mahagaonkar, 2008; Waldemar, 2011). To assess if corruption is endogenous a probit regression with instrumental variables is performed in Stata. This requires the specification of the endogenous variable and the instrumental variables used. The regression provides the Wald test of exogeneity. The null hypothesis is that the variable is exogenous. The rejection occurs when the p-value of the Wald test is lower than the significance level set to 0.05. To test the strength of the instrumental variables, an estimation of the reduced form equation is performed. As a rule of thumb the F-value of the instrumental variables should be higher than 10 to be considered strong instruments (Adkins and Hill, 2008).
Goodness-of-fit: To test the goodness-of-fit of the model the Hosmer-Lemeshow test is used.
The observations are grouped into 10 groups based on deciles of predicted probabilities (Hosmer and Lemeshow, 2000). In each of the estimation done for the three samples the null hypothesis that the model fits the data well is not rejected.
4. Empirical results
20
firms according to the region they belong is made. One sample is made for the CEE countries and the other is for the CIS region5. Turkey is used only in the full sample since it is neither a member of the CIS, nor does it belong to the CEE countries.
4.1 Descriptive statistics
In this sub-section the descriptive statistics for the variables used in the model are commented. The descriptive statistics for the entire, CEE and CIS samples are found in Appendix 1. The comments are made on the entire sample along with remarks on the differences encountered in the variables from the other two samples.
Dependent variable (ninnov): The mean of new product innovation indicates that 55 percent of
the firms introduced new product. The deviation from this mean is of 5 percent. The highest mean is in the CEE sample (60 percent), while in the CIS sample is the lowest (53). The deviation is consistent throughout the samples.
Independent variable (bribe): The average percentage of bribes given by firms from annual
sales is of 0.93 percent. The deviation from the mean is of 3.81 percent. The maximum value of bribes paid by one/some of the firm/s in the sample is of 60 percent. Moreover, there are firms that do not give bribes. A distinction between the sub-samples is that for the CEE countries the average percentage of bribes (0.42 percent) is considerably lower than in the CIS countries (1.53 percent). The deviation from this mean is of 2.22 percent in the CEE sub-sample and 4.8 percent in the CIS one. The maximum value of bribes reported is of 50 percent from the annual sales in the CEE, while for the CIS is consistent with the entire sample (60).
Control variables: The average managerial experience is of 17 years, but the deviation from the
mean is of 10 years. The average value of competition shows that firms consider competition to have an important influence on their choice to produce new products. The average size of the companies is 111 employees, but there is a very high deviation from this mean. The maximum
5
21
number (18208) of employees is found in a Hungarian private owned firm operating in the retail industry. The deviation from the mean is of 405 employees. No extreme differences are reported in the average firm size throughout the sub-samples. The deviation is lower in the CIS sample (351 employees). The mean of the age of the companies in the sample is of 16 years. The standard deviation from this mean is of 16 years. Ten percent of the firms in the sample have private foreign ownership and the deviation from the mean is of 3 percent. The highest number of firms with private foreign ownership is in the CEE sample (13 percent), while for the CIS sample is the lowest (8 percent). On average, 23 percent of the firms are exporters, 26 percent invest in R&D and 39 percent have a credit line from a private bank. Between the sub-samples there are some differences. In regard to the average number of exporting firms: 30 percent of firms are exporters in the CEE region, while 11 percent are exporters in the CIS countries. The CIS sample has the lowest percent of firms investing in R&D (20 percent), while the CEE has the highest (30 percent). Finally, in regard to financial loans taken from a private bank, the CIS countries remain to have the lowest proportion of firms having a line of credit (27 percent), while the CEE has a proportion of 39 percent of firms with a credit line.
4.2 Empirical results
In this section the estimated results are presented. Table 2 summarizes the estimations for the entire sample. Table 3 incorporates the results from the estimations for the CEE and the CIS countries. Both tables contain the coefficients of the variables used, their significance and the results for the marginal effects after the probit regression. The estimates from the probit model show the impact of the independent variables on the latent dependent variable. Therefore the marginal effects are added in the table. A proper technique interpret the impact of the independent variables on the dependent variable is to estimate the effect of a unit change in the mean values of the independent variables on the probability P(y=1) (Nagler, 1994). The comments are based on the marginal effects.
22
entire sample, Montenegro – in the CEE sample, and Kyrgyzstan (Kyrgyz Republic) – in the CIS sample.
4.2.1 Entire sample estimation
The estimated results are presented in Table 2. The results suggest the following about the independent variables (i.e. bribe and bribe squared) and the control variables.
In the entire sample of firms the value of the coefficient of determination – pseudo R2 – shows that the variation in the probability to introduce new products is explained by 14.40% percent from the variation in the independent and control variable.
Independent variables: The coefficients of the percentage of bribes and the squared percentage
of bribes are statistically significant and have different signs. The positive value of the percentage of bribes and the negative value of its squared form indicate that the relationship between corruption and new product innovation is non-linear. The null hypothesis of a non-linear relationship cannot be rejected at the significance level 0.10. A 1 percent increase in the mean value of bribe (which is 0.93 percent in the sample) corresponds to an increase of 0.013 points in the probability of new product innovation. A 1 percent increase in the mean value of bribe2 (which is 15.36 percent in the sample) decreases the likelihood of new product innovation with 0.0002 points.
The probability that a firm introduces new product developments on the market, ceteris paribus, is 0.5641 points.
Control variables: With the exception of the age of the firm, all the other control variables are
23
competition on new product innovation leads to an increase of 0.11, 0.10 respectively, in the probability to launch new product innovations.
Table 2 Regression output for the entire sample (CEE, CIS and Turkey)
VARIABLES
New product innovation (y)
Marginal effects after probit
(Notation) (ninnov) dy/dx
Percentage of bribes 0.0341*** 0.0134***
(bribe) (0.0087) (0.0034) Squared percentage of bribes -0.0005* -0.0002*
(bribe2) (0.0002) (0.0001) Managerial experience 0.0172*** 0.0068*** (manexp) (0.0051) (0.0020) Sq. Managerial experience -0.0004*** -0.0001*** (manexp2) (0.0001) (0.00004) Competition 0.276*** 0.1087*** (comp) (0.0832) (0.0328) Sq. competition -0.0343** -0.0135** (comp2) (0.0160) (0.0063) Log. of size 0.0486*** 0.0191*** (lnsize) (0.0129) (0.0051) Age of firm -0.0010 -0.0004 (age) (0.0011) (0.0004) Private foreign ownership 0.210*** 0.0812***
(pfo) (0.0574) (0.0216)
Exporters 0.2750*** 0.1062***
(export) (0.0445) (0.0168) Research and Development 0.8350*** 0.3041***
(rd) (0.0410) (0.0130)
Financial loan 0.1720*** 0.0672***
(finance) (0.0346) (0.0135)
Constant -1.760***
(0.1420)
Industry dummies Yes
Country dummies Yes
Observations 7.442
Wald chi2 (56) 1209.11***
Pseudo R2 0.1440
24
4.2.2 CEE and CIS sample regression estimates
The estimated results are presented in Table 3. The results suggest the following about the independent (i.e. bribe and bribe2) and the control variables for two different data samples.
Table 3 Regression output for the CEE and CIS samples
Sample CEE CIS
VARIABLES New product
innovation Marginal effects after probit New product innovation Marginal effects after probit
(Notation) (ninnov) dy/dx (ninnov) dy/dx
Percentage of bribes 0.0111 0.0042 0.0354*** 0.0140***
(bribe) (0.0198) (0.0075) (0.0108) 0.0043) Squared percentage of bribes 0.0005 0.0002 -0.0006* -0.0002*
(bribe2) (0.0006) (0.0002) (0.0003) (0.0001) Managerial experience 0.0136 0.0052 0.0274*** 0.0109*** (manexp) (0.0087) (0.0033) (0.0080) (0.0032) Sq. Managerial experience -0.0004** -0.0014** -0.0006*** -0.0002*** (manexp2) (0.0002) (0.0001) (0.0002) (0.0001) Competition 0.5210*** 0.199*** 0.198 0.0787 (comp) 0.1320) 0.0504) (0.1230) 0.0487) Sq. competition -0.0779*** -0.0297*** -0.0224 -0.0089 (comp2) (0.0247) (0.0094) (0.0239) (0.0091) Log. of size 0.0325* 0.0124* 0.0914*** 0.0362*** (lnsize) (0.0191) (0.0073) (0.0210) 0.0083) Age of firm -0.0024 -0.0009 -0.0017 -0.0007 (age) (0.0015) (0.0006) (0.0016) (0.0006) Private foreign ownership 0.2410*** 0.0890*** 0.1900** 0.0742**
(pfo) (0.0777) (0.0276) (0.0932) (0.0358)
Exporters 0.2800*** 0.1049*** 0.3030*** 0.1174***
(export) (0.0615) (0.0224) 0.0926) (0.0347) Research and Development 0.8780*** 0.3062*** 0.850*** 0.3093***
(rd) (0.0587) (0.0177) (0.0720) (0.0223)
Financial loan 0.1010* 0.0384* 0.2610*** 0.1025***
(finance) (0.0498) 0.0190) (0.0578) 0.0228)
Constant -1.151*** -1.557***
(0.2650) (0.2140)
Industry dummies Yes Yes
Country dummies Yes Yes
Observations 3,360 3,222
Wald chi2 (43/37) 578.28*** 523.53***
Pseudo R2 0.1524 0.1509
25
In the CEE and CIS samples of firms the value of the pseudo R2 shows that the variation in the probability to introduce new products is explained by 15.24 percent, 15.09 percent, respectively, from the variation in the independent and control variable. The independent and control variables have the highest explanatory power for the firms in the CEE sample.
Independent variables: The percentage of bribes and the squared percentage of bribes paid are
both positive and statistically insignificant. For the CEE sample, the null hypothesis of a non-linear relationship with new product innovation is rejected. In the CIS sample the null hypothesis of a non-linear relationship is not rejected. Compared to the previous results on the CEE sample the average bribe paid (1.54 percent) is higher with 1.12 percent. Also, the average of the squared bribe is 25.37. An increase of 1 percent in the mean value of the bribes paid increases the probability of new product innovation with 0.1 points. A 1 percent increase in the mean value of the squared bribes produces a decrease of 0.0002 points in the probability of new product innovation.
The probability of a producer to introduce new product innovations is of 0.6162 points in the CEE sample and 0.5429 points in the CIS sample, ceteris paribus.
Control variables (CEE sample): The age of the firm is one of the two statistically insignificant
control variables in the CEE sample. The other insignificant variable is managerial experience. However, its squared term is significant and negative. In line with the literature, all the other control variables have the expected coefficients and are significant. The decision to invest in R&D has the highest impact on probability of introducing new product innovations, increasing it by 0.31 points. The decision to export brings an increase of 0.10 points on the probability to become a new product innovator. The opening of a credit line from a private bank – financial loan (finance), increases the chance to introduce new products with 0.04 points. The impact of
finance on new product innovation has decreased relative to the previous sample. At the same
26
Control variables (CIS sample): The age of the firm has an insignificant effect on the probability
to introduce new products in this sample as well. Unlike in the previous samples the coefficients of competition and the squared form are statistically insignificant. All other controls are significant and have the expected signs. R&D remains to have the highest effect, followed by the decision to export. The decision to invest in R&D increases the probability to innovate new products with 0.31 points. And the decision to export increases the probability to introduce new products with 0.12 points. Compared to the CEE sample, the financial loans have a higher effect on the probability to become a product innovator. The opening of a credit line from a private bank increases the probability of new product innovation with 0.10 points.
4.3 Robustness check
The robustness of the estimates against the endogeneity issue, see Section 3, is assessed using a two-stage instrumental variable probit model. Following Fishman and Svensson (2007), one of the instrumental variables used is the average of the individual bribes by sector-region-country (bribes_src). This variable is presumed to be uncorrelated with firm specific factors that have an impact on both product innovation and bribes. The bribe_src instrumental variable has 319 unique values. The second instrumental variable used is the average of the individual bribes by sector-location size-firm size-country (bribe_slsc) and it has 342 unique values. This instrumental variable is presumed to eliminate the endogeneity of individual bribes (Aterido et al., 2011). The strength of instrumental variables bribe_src and bribe_slsc is tested using an estimation of the reduced form equation. According to the F-value of the instrumental variables, they are qualified as strong instruments. The p-values from the Wald test of exogeneity indicate that for the entire sample and for the CIS sample the variables bribe and bribe2 are endogenous and they should be instrumented. The results for the strength of the instrumental variables and the Wald test of exogeneity are found in Appendix 3, table 7. In Appendix 3, table 8, the results from the estimation are shown as well. Moreover, the F-value for the strength of the instrumental variable is shown in the table with the estimation results.
27
the value of the coefficients can be noticed, consequently the impact of corruption on new product innovation is higher using this estimator. While the results from the probit model indicate an inverted U-shape relationship, the results from the two-step instrumental probit model indicate a U-shape relationship. Nevertheless, the null hypothesis of a non-linear relationship between corruption and new product innovation is not rejected in either the case at the significance level 0.05. It is interesting to point out that the significance and the signs of the estimated coefficients of control variables are almost unchanged. A slight increase in their values is noticeable.
The results from the CIS sample indicate a U-shape relationship between corruption and new product innovation, unlike the inverted U-shape relationship in the probit model estimation. Competition is now significant and positive, while in the probit model the coefficient of competition was insignificant. Nevertheless, the squared form of competition is insignificant. All the other control variables do not differ in their significance and signs. The dissimilarities in reference to the previous model used are the small increases in the values of the coefficients.
5. Conclusions and limitations
28
The objective of this thesis was to see if there is a non-linear relationship between corruption and new product innovation. If not all levels of corruption produce the same effect on new product innovation, it is highly important to prove it. Effective policies aimed at combating corruption or stimulating product innovation need to have a thorough knowledge regarding the relationship between corruption and product innovation, otherwise they are ineffective.
According to the estimations performed the hypothesis of a non-linear relationship between corruption and new product innovation cannot be rejected at conventional significance levels in probit models, regardless of the method of estimations. However, the functional form of the relationship between bribes and new product innovation differs depending on the estimator: a two-stage instrumental variable estimator of the probit model leads to a U-shape relationship between bribes and new product innovation, whereas a standard estimator of the probit model produces an inverted U-shape relationship. This implies that a change in the percentage level of bribery might actually have a different impact on the probability of new product innovation. If it is indeed a U-shape relationship then this means that medium level of bribery severely disrupt the likelihood of product innovation.
When the sample is split between regional and institutional development differences – CEE versus CIS countries – the results show that in the CEE sample, where firms pay on average 0.45 percent of their annual sales in for bribes, the relationship is positive and linear, but not statistically significantly different from zero. For the firms in the CIS countries, where firms pay on average 1.54 percent of their annual sales for bribes, there is evidence of a statistically significant non-linear relationship between corruption and new product innovation.
29
30
References
Acemoglu, D., & Verdier, T. (1996). Property Rights, Corruption and the Allocation of Talent: A General Equilibrium Approach C.E.P.R. Discussion Papers.
Ades, A., & Tella, R. D. (1996). The Causes and Consequences of Corruption: A Review of Recent Empirical Contributions. IDS Bulletin, 27(2), 6-11.
Ades, A., & Tella, R. D. (1999). Rents, Competition, and Corruption. American Economic Review, 89(4), 982-993.
Adkins, L., Hill, C., Lim, G. (2008), Principles of Econometrics – Third Edition, John Wiley and Sons, Inc., Hoboken, NJ, USA
Adkins, L., Hill, C. (2008). “Using Stata for Principles of Econometrics – Third Edition”, John Wiley & Sons, Inc., Hoboken, NJ, USA
Aghion, P., Howitt, P. (1997). Endogenous Growth Theory. The MIT Press.
Aghion, P., Bloom N., Blundell R., Griffith, R., Howitt, P.(2005). Competition and Innovation: An Inverted-U Relationship. The Quarterly Journal of Economics. MIT Press, vol. 120(2),701-728
Toke S. Aidt, T. S., Dutta J. (2008). Policy Compromises: Corruption And Regulation In A Democracy.
Economics and Politics, Wiley Blackwell, vol. 20(3),335-360
Andvig, J. C., Moene, K. O. (1990). How corruption may corrupt. Journal of Economic Behavior &
Organization, 13(1), 63-76.
Anokhin, S., Schulze, W. S. (2009). Entrepreneurship, innovation, and corruption. Journal of Business
Venturing, 24(5), 465-476.
Asiedu, E., Freeman, J. (2009). The Effect of Corruption on Investment Growth: Evidence from Firms in Latin America, Sub-Saharan Africa, and Transition Countries *. Review of Development Economics,
13(2), 200-214.
Aterido, R., Hallward-Driemeier, M., Pages, C. (2009). Big constraints to small firms'growth ? business environment and employment growth across firms. Policy Research Working Paper Series 5032, The World Bank.
Ayyagari, M., Demirguc-Kunt, A., & Maksimovic, V. (2007). Firm innovation in emerging markets : the roles of governance and finance. Policy Research Working Paper Series 4157.The World Bank.
Ayyagari, M., Demirguc-Kunt, A., & Maksimovic, V. (2010). Are innovating firms victims or perpetrators ? tax evasion, bribe payments, and the role of external finance in developing countries. Policy
Research Working Paper Series 5389. The World Bank.
Balsmeier, B., Czarnitzki, D. (2011). How Important Is Industry-Specific Managerial Experience for Corporate Innovative Activity? European Association for Research in Industrial Education, 38.
31
Bardhan, P. (1997). Corruption and Development: A Review of Issues. Journal of Economic Literature,
35(3), 1320-1346.
Blackburn, K., Forgues-Puccio, G. (2009). Why is corruption less harmful in some countries than in others? Journal of Economic Behavior & Organization, 72(3), 797-810.
Cohen, W. M. Chapter 4 - Fifty Years of Empirical Studies of Innovative Activity and Performance.
Handbook of the Economics of Innovation (pp. 129-213) North-Holland.
D. Congleton, A. L. Hillman & K. A. Konrad (Eds.), 40 Years of Research on Rent Seeking: Volume 2 (pp. 213-218) Berlin and Heidelberg: Springer.
Dejardin M.(2011). Entrepreneurship and Rent-Seeking Behavior. Working Papers 1112, University of Namur, Department of Economics.
De Jong, G., Tu, P.A., Van Ees, H. (2012). Which Entrepreneurs Bribe and What do they get from it? Explanatory Evidence from Vietnam, Entrepreneurship Theory and Practice, Vol. 36, Issues 2, 323-345 De Rosa, D., Gooroochurn, N., & Gorg, H. (2010). Corruption and productivity : firm-level evidence from the BEEPS survey. Kiel Working Papers 1632. Kiel Institute for the World Economy.
EBRD-World Bank Business Environment and Enterprise Performance Survey (BEEPS). (2010)
Fan, C. S., Lin, C., Treisman, D. Political decentralization and corruption: Evidence from around the world. Journal of Public Economics, Elsevier, vol. 93(1-2),14-34.
Fisman, R., & Svensson, J. (2007). Are corruption and taxation really harmful to growth? Firm level evidence. Journal of Development Economics, 83(1), 63-75.
Goedhuys, M. (2007). The impact of innovation activities on productivity and firm growth: evidence from Brazil. United Nations University, Maastricht Economic and Social Research and Training Centre on
Innovation and Technology.
Goel, R. K., & Nelson, M. A. (2010). Causes of corruption: History, geography and government. Journal
of Policy Modeling, 32(4), 433-447.
Hallack, Jacques, Poisson, Muriel (2001). Ethics and Corruption in Education, Policy Forum, No. 15, International Institute for Educational Planning/UNESCO, Paris
Hellman, J. S., Jones, G., & Kaufmann, D. (2003). Seize the state, seize the day: state capture and influence in transition economies. Journal of Comparative Economics, 31(4), 751-773.
Hosmer, David W., Lemeshow, Stanley (2000). Applied Logistic Regression, New York : Wiley
Howitt, P., Griffith, R., Aghion, P., Blundell, R., & Bloom, N. (2005). Competition and Innovation: An Inverted-U Relationship. Harvard University Department of Economics.
32
Kaufmann, D. (2005). Myths and Realities of Governance and Corruption. MPRA Paper 8089.University Library of Munich, Germany.
Lambsdorff, J. G. (2002). Corruption and Rent-Seeking. Public Choice 113. Kluwer Academic Publishers. Lederman, D. (2009.The business of product innovation: international empirical evidence, Policy
Research Working Paper Series 4840, The World Bank.
Leff, N. H. (1964). Economic Development Through Bureaucratic Corruption. American Behavioral
Scientist, 8(3), 8.
Mahagaonkar, P. (2008). Corruption and Innovation: A Grease or Sand relationship? Jena Economic
Research Papers 2008-017.Friedrich-Schiller-University Jena, Max-Planck-Institute of Economics.
Mauro, P. (1995). Corruption and Growth.The Quarterly Journal of Economics, MIT Press, vol. 110(3), pp. 681-712.
Mendez, F., Sepulveda, F. (2006). Corruption, growth and political regimes: Cross country evidence.
European Journal of Political Economy, 22(1), 82-98.
Mo, P. H. (2001) Corruption and Economic Growth. Journal of Comparative Economics. Elsevier. vol. 29(1), pages 66-79
Murphy, K. M., Shleifer, A., & Vishny, R. W. (1990). The Allocation of Talent: Implications for Growth
National Bureau of Economic Research, Inc.
Murphy, K. M., Shleifer, A., & Vishny, R. W. (2008). Why Is Rent-Seeking So Costly to Growth?
American Economic Review, American Economic Association, vol. 83(2), pages 409-14
Nagler, J. (1994). Interpreting probit analysis. New York University.
Nye, J. S., (1967). Corruption and Political Development - a Cost-Benefit Analysis. American Political
Science Review, 61, 417-427.
OECD/Statistical Office of the European Communities, Luxembourg (2005),Oslo Manual: Guidelines for Collecting and Interpreting Innovation Data, 3rd Edition. The Measurement of Scientific and Technological Activities, OECD Publishing.
Shleifer, A., & Vishny, R. W. (1993). Corruption. The Quarterly Journal of Economics, 108(3), 599-617. Tanzi, V. (1998). Corruption Around the World: Causes, Consequences, Scope, and Cures. IMF Staff
Papers, Palgrave Macmillan, vol. 45(4), 559-594
Tingvall, G. P., Karpaty, P. (2008). Service-sector competition, innovation and R&D. Working Paper
Series in Economics and Finance 702. Stockholm School of Economics.
33
Urra. F. J. (2007). Assessing Corruption An analytical review of Corruption measurement and its problems: Perception, Error and Utility. Edmund A. Walsh School of Foreign Service Georgetown University
Veracierto, M. (2008). Corruption and innovation. Economic Perspectives, , 29-39.
Waheed, A., (2011).Size, competition, and innovative activities: a developing world
perspective.UNU-MERIT Working Paper Series 052, United Nations University, Maastricht Economic and social Research
and training centre on Innovation and Technology.
Waldemar, S. D. (2011). New products and corruption: evidence from Indian firms. Université Paris1
Panthéon-Sorbonne. HAL.
Wooldridge, J. M.: Econometric Analysis of Cross Section and Panel Data. XXIII, 752 pp. MIT Press, Cambridge, Mass., 2002.
34
Appendix 1. Distribution of observations
Entire Sample
Freq.
Percentage of bribers relative to non-bribers new product innovators
Percentage of bribers relative to non-bribers incumbent producers CEE sample Freq. Country Country Albania 78 55.56 23.08 Albania 78 Armenia 258 25.83 3.03 Bosnia and Herzegovina 239 Azerbaijan 218 79.03 17.07 Bulgaria 147 Belarus 166 12.73 7.69 Croatia 124 Bosnia and
Herzegovina 239 10.45 1.20 Czech Republic 163
Bulgaria 147 12.73 3.75 Estonia 220
Croatia 124 26.56 0.00 Fyr Macedonia 271
Czech Republic 163 12.50 5.36 Hungary 264
Estonia 220 2.99 0.00 Kosovo 182
Fyr Macedonia 271 13.89 4.00 Latvia 186
Georgia 182 3.08 2.68 Lithuania 204
Hungary 264 4.76 3.40 Montenegro 68
Kazakhstan 330 57.89 17.86 Poland 291
Kosovo 182 7.00 4.17 Romania 238
Kyrgyz Republic 143 144.00 30.51 Serbia 290
Latvia 186 11.43 6.25 Slovak Republic 153
Lithuania 204 11.11 0.00 Slovenia 242
Moldova 264 36.79 3.77 Total 3,360
Montenegro 68 18.60 0.00 CIS sample
Poland 291 7.06 2.91 Armenia 258
Romania 238 16.67 8.04 Azerbaijan 218
Russia 717 55.41 19.02 Belarus 166
Serbia 290 25.68 19.48 Georgia 182
Slovak Republic 153 20.00 3.17 Kazakhstan 330
Slovenia 242 2.25 5.26 Kyrgyz Republic 143
35
Appendix 2. Descriptive statistics
Table 4. Entire sample - descriptive statistics
Variable (notation) Observations Mean
Standard
deviation Minimum Maximum New product innovation
(ninnov) 7442 0.55 0.50 0 1
Percentage of
bribes/annual sales (bribe) 7442 0.93 3.81 0 60
Squared percentage of
bribes (bribe2) 7442 15.36 138.52 0 3600
Size of company (size) 7442 111.02 405.43 1 18208
Log of size (lnsize) 7442 3.48 1.45 0 9.81
Age of firm (age) 7442 16.25 16.23 1 183
Private foreign ownership
(pfo) 7442 0.10 0.30 0 1 Managerial experience (manexp) 7442 17.03 10.36 1 61 Sq. Managerial experience (manexp2) 7442 397.45 460.53 1 3721 Competition (comp) 7442 2.74 1.05 1 4 Sq. Competition (comp2) 7442 8.63 5.44 1 16 Exporters (export) 7442 0.23 0.42 0 1
Research & development
(rd) 7442 0.26 0.44 0 1
Financial loan (finance) 7442 0.39 0.49 0 1
Table 5. CEE countries sample - descriptive statistics
Variable (notation) Observations Mean
Standard
deviation Minimum Maximum New product innovation
(ninnov) 3360 0.60 0.49 0 1
Percentage of
bribes/annual sales (bribe) 3360 0.42 2.22 0 50
Squared percentage of
bribes (bribe2) 3360 5.12 60.39 0 2500
Size of company (size) 3360 103.53 429.16 1 18208
Log of size (lnsize) 3360 3.42 1.47 0 9.81
Age of firm (age) 3360 16.99 16.75 1 183
Private foreign ownership
36 Managerial experience (manexp) 3360 17.53 9.67 1 53 Sq. Managerial experience (manexp2) 3360 400.75 423.44 1 2809 Competition (comp) 3360 2.90 1.00 1 4 Sq. Competition (comp2) 3360 9.43 5.35 1 16 Exporters (export) 3360 0.30 0.46 0 1
Research & development
(rd) 3360 0.30 0.46 0 1
Financial loan (finance) 3360 0.47 0.50 0 1
Table 6. CIS countries sample - descriptive statistics
Variable (notation) Observations Mean
Standard
deviation Minimum Maximum New product innovation
(ninnov) 3222 0.53 0.50 0 1
Percentage of
bribes/annual sales (bribe) 3222 1.54 4.80 0 60
Squared percentage of
bribes (bribe2) 3222 25.37 174.38 0 3600
Size of company (size) 3222 105.31 351.05 1 12000
Log of size (lnsize) 3222 3.45 1.40 0 9.39
Age of firm (age) 3222 15.05 16.64 1 166
Private foreign ownership
(pfo) 3222 0.08 0.27 0 1 Managerial experience (manexp) 3222 14.94 9.91 1 59 Sq. Managerial experience (manexp2) 3222 321.30 410.08 1 3481 Competition (comp) 3222 2.59 1.09 1 4 Sq. Competition (comp2) 3222 7.91 5.58 1 16 Exporters (export) 3222 0.11 0.32 0 1
Research & development
(rd) 3222 0.20 0.40 0 1
37
Appendix 3. Robustness check Table 7. Endogeneity tests
Sample Dependent variable
Independent variable
F-value for the strength of the instrumental variables** Wald test of exogeneity p(chi) value Outcome
Entire New prod. innovation
Bribe 1827* 0.0078 Do not reject the
null hypothesis of exogeneity
Bribe2 793*
CIS New prod. Innovation
Bribe 866* 0.0150
Bribe2 418*
CEE New prod. Innovation
Bribe 1167*
0.8222
Reject the null hypothesis
Bribe2 318*
* denotes p<0.01
** the instrumental variables are bribe sector-region-country average (bribe_src)and bribe sector-location size-firm size- country average (bribe_slsc)
Table 8. Two-stage instrumental variable estimation of probit model
Sample Entire CIS
VARIABLES
New product innovation
New product innovation
(Notation) (ninnov) (ninnov)
Percentage of bribesx -0.169** -0.273**
(bribe) (0.0705) (0.1300) Squared percentage of bribesx 0.00706*** 0.0104**
(bribe2) ((0.0026) (0.0046) Managerial experience 0.0193*** 0.0348*** (manexp) (0.0056) (0.0104) Sq. Managerial experience -0.0004*** -0.0008*** (manexp2) (0.0001) (0.0002) Competition 0.320*** 0.312* (comp) (0.0910) (0.1590) Sq. competition -0.0415** -0.0411 (comp2) (0.0174) (0.0307) Log. of size 0.0387*** 0.0788*** (lnsize) (0.0145) (0.0267) Age of firm -0.0008 -0.0008 (age) (0.0012) (0.0020) Private foreign ownership 0.222*** 0.247**
(pfo) (0.0621) (0.1160)
Exporters 0.271*** 0.272**
(export) (0.0479) (0.1100) Research and Development 0.871*** 0.934***
38
Financial loan 0.184*** 0.277***
(finance) (0.0378) (0.0719)
Constant -1.798*** -1.233***
(0.1520) (0.2970)
Industry dummies Yes Yes
Country dummies Yes Yes
Observations 7,442 3,222
Wald chi2 (56) 1115.96*** 403.29***
P-value Wald test of exogeneity 0.0078 0.015 F-test ( 2, 7385)
1827.88***/ 793.23***
866.54***/ 417.69***
Robust standard errors in parentheses
x instrumented using bribe_src and bribe_slsc
