August 2011
University of Groningen
Faculty of Economics and Business
MASTER THESIS
Christine Elisabeth Praamsma; 1690221
Corruption and firm performance in Asian transition
economies
Supervisor: Gjalt de Jong
August 2011
Handelshøyskolen BI in Oslo
Faculty of Economics and Business
MASTER THESIS
Christine Elisabeth Praamsma; 0913355
Corruption and firm performance in Asian transition
economies
Supervisor: Lars Huemer
Title: Corruption and firm performance in Asian transition economies Author: Christine Elisabeth Praamsma
Author’s e-mail address: christine@praamsma.com
Author’s address: Betuwestraat 6, 6811 MA Arnhem, The Netherlands Department: Faculty of economics and business
Supervisors: Gjalt de Jong and Lars Huemer
Abstract: This thesis addresses the effect of different types of corruption on firm
performance. A sample of Asian transition economies is selected representing a group of corrupt, albeit growing countries. Next to overall corruption, a distinction is made between institutional, operational, and performance corruption. Institutional corruption is found to have an inverted U-shape relationship with firm performance for all the countries in the sample. The results for corruption suggested that a negative linear relation exists between corruption and firm performance. The other types of corruption did not provide us with significant results.
1. Introduction --- 1
2. Literature Review --- 3
2.1 Corruption --- 3
2.2 Firm performance --- 6
2.3 Corruption and firm performance --- 7
2.4 Institutional corruption --- 9
2.5 Operational corruption --- 11
2.6 Performance corruption --- 12
2.7 The theoretical model --- 13
3. Methodology --- 13
3.1 Data and sample selection --- 13
3.2 Measurement --- 17
3.4 Econometric Techniques --- 19
4. Empirical results --- 24
4.1 Descriptive Statistics --- 24
4.2 Characteristics of corrupt versus non-corrupt firms --- 27
4.3 Corruption and firm performance --- 28
4.4 Robustness Tests --- 35 5. Conclusions --- 35 5.1 Limitations --- 37 5.2 Future research --- 38 References--- 39 Appendix --- 42
Appendix A: Survey items and scales --- 42
Appendix B: Factor analysis --- 45
1
1. Introduction
The research in the field of corruption has been approached in several ways. Firstly, there are distinguishable units of analysis. Hitherto most research concentrates on the country as the unit of analysis (Mauro 1995, Hakkala, Norbäck and Svaleryd 2008, Swaleheen 2009, Treisman 2000). Recently, research is being published with the firm or the individual as the unit of analysis (Gaviari 2002, Gao 2010, Kaufman and Wei 2004, Javorcik and Wei 2009, Svensson 2003, Wu 2008, de Jong, Anh Tu and van Ees 2010). A second distinction that can be made is between the antecedents of corruption (Gao 2010, Chen, Yasar and Rejesus 2007, Lee, Oh and Eden 2010, Treisman 2000, Wu 2008, Svensson 2003) and the consequences of corruption (Mauro 1995, McArthur and Teal 2002, Kaufman and Wei 2004, Hakkala, Norbäck and Svaleryd 2008, Asiedu and Freeman 2009). Lastly, a number of papers address the demand side of corruption while others focus on the supply side of corruption. This thesis is concerned with the effects of the supply side, firm-level corruption.
A thorough understanding of the effects of firm-level corruption is essential for firms to strengthen their market position and respond to policy reforms that are put in place to build and strengthen market-promoting institutions. Additionally this broadened understanding of the effects of corruption in organizations is useful for governments when designing policy reforms since it clarifies where they should aim at. The persistence of corruption is evident from the subjective measurements undertaken by Transparency International. Since 1999 they publish the Bribe Payers Index (BPI) which attests the supply of corruption in 26 countries. The BPI indicates that the supply side of corruption is just as significant, if not even more, as the demand of corruption. For numerous firms bribery is a routine business practice. This applies in particular to developing countries where bribery is omnipresent. The present study will focus on transition economies where new opportunities for bribery arise due to the transition from a state planned to a market economy.
Research objective
2
At country-level, different types of corruption have already been distinguished; examples are bureaucratic, economic and political corruption (Jain 2001). At an organizational level no clear distinctions are pronounced so far. The aim of this thesis is to fill this gap in the literature. This is done by examining four Asian transition economies, such as, the Central Asian Republics of Kazakhstan, Kyrgyzstan, Uzbekistan and Tajikistan. Although these countries are thus far not listed in the BPI index they are found in the Corruption Perception Index (CPI). Their CPI scores establish that these economies have high and persistent levels of corruption. The main research question that will be answered is:
What is the relationship between different types of corruption and firm performance in Asian transition economies?
In order to answer this research question a number of sub questions are addressed: o What is corruption and how can it be measured? (2.1)
o What is firm performance and how can it be measured? (2.2)
o What is the relationship between corruption and firm performance? (2.3) o What is institutional corruption and how is it related to firm performance? (2.4) o What is operational corruption and how is it related to firm performance? (2.5) o What is performance corruption and how is it related to firm performance? (2.6)
Thesis overview
3
2. Literature Review
This section defines the concept of corruption in the context of the firm and provides an overview of a range of characteristics of corruption that have been discussed in the literature. An overview is presented of the different determinants and effects of corruption on three units of analysis; the country, organizational and individual level. In section 2.2 a brief discussion on firm performance is presented. This is intended to illustrate the different ways to define firm performance.
The remainder of the section is dedicated to the hypothesis development concerning different types of corruption and firm performance. First, the effect of corruption on firm performance is examined. Subsequently, the research is elaborated by recognizing the fact that corruption is a multidimensional concept including institutional, operational and performance corruption. It is anticipated that different types of corruption will, next to a common element, have a distinguishable unique effect on firm performance. Before proceeding to an analysis of the relationship between corruption and firm performance both concepts are now clarified.
2.1 Corruption
4 Table 1: Characteristics of corruption
Types of corruption
Bribery: Offering (non-)pecuniary benefits in order to influence a public official. Nepotism: Preferential treatment for relatives or close friends of public officials. Fraud: Deceiving the government by cheating.
Embezzlement: The theft of money or other government property.
Kind of corruption
Bureaucratic (administrative) petty corruption: Corruption that affects policy
implementations with small bribes that typically involve junior officials. (Jain 2001).
Political (legislative) grand corruption: Corruption that affects laws, regulations, and
policies with substantial pecuniary bribes that typically involve high level officials (Jain 2001).
Economic corruption: Use of public office for private gains, where an official entrusted
with carrying out a task by the public engages in a malfeasance for private enrichment which is difficult to monitor for the principal (Bardhan 1997).
Sector of corruption
Public sector: Corruption that involves government institutions.
Private sector: Corruption amongst firms or between firms and private individuals
Effect of corruption
Cost-reducing: The corrupt government official is able to reduce the costs for the producer,
but demands a payment in return (Bliss and Di Tella 1997).
Benefit enhancing / surplus shifting corruption: There is a surplus in the business which can
be taken by the corrupt government official so that it is not lost (Bliss and Di Tella 1997).
Initiator of corruption
Briber-initiated: The bribe is initiated by the demand side of corruption. Bribee-initiated: The bribe is initiated by the supply side of corruption.
Degree of centralization
Centralized: A centralized government control can reduce the number of agencies that an
agent needs to ask for permits to, thus decreasing the chance of having to pay more bribes per permit (Tanzi 1998).
Decentralized: A weak central government control enables different government agencies
to obtain independent bribes from a private agent in need of permits from several agencies. (Shleifer and Vishny 1993, Bardhan 1997)
Degree of cooperation
Coercive: Occurs when a government official asks for a bribe in order to, for example,
arrange a cost-reduction for a firm. Both benefit from this situation (Tanzi 1998).
Collusive: When the government official forces the firm to pay a bribe for example in order
to acquire a permit (Tanzi 1998). The benefits mostly fall to the government official.
Extent to which bribes are expected
Pervasive: Bribes that are expected to be demanded. The firm will take this into account
when for example considering making an investment (Lee et al. 2010).
Arbitrary: Bribes that are not expected, the firm is not able to take them into account when
making a decision and the damage is potentially higher compared to a pervasive bribe (Lee et al. 2010).
Type of transfer
5
The discretionary power of the agent is created by either market imperfections or an institutional position, giving the agent a position of power over the firm. In this study the definition of firm bribery is: the exchange of pecuniary or non-pecuniary benefits between a
firm and an agent with discretionary power in order to gain an entrepreneurial advantage, overcome an institutional obstruction, or continue operations.
As suggested by Wu (2008) it is essential to understand the determinants of corruption in order to understand its consequences. Throughout the literature several factors, such as cultural and historical factors, appear to have an essential role in the determination of corruption (Gaviari 2002). In Table 2 several determinants are projected at a country, firm and entrepreneurial level. The focus of the present study lies on the effects of corruption on the level of the firm.
Table 2: determinants and effects of corruption on country, firm and entrepreneurial level
Determinants Effects
Variable Study Variable Study
Country level
Discretionary power Jain (2001) Slow economic growth
Shleifer and Vishny (1993), Mauro (1995) Low income, transition economies Svensson (2005) Enhance economic growth Leff (1964) Economic rents and unaccountable
public servants
Gaviari (2002) Jain (2001)
Decrease spending on education and environmental quality
Mauro (1998)
Strength of government institutions and political processes
Jain (2001)
Shleifer and Vishny (1993)
Damage development of financial, economic and political institutions
Wu (2008)
Quality of government service and
taxation Wu (2008)
Reduce legitimacy of
democratic government Tanzi (1998) Perceived corruption, salary,
detection probability, discount rate, high bribes
Andvig and Moene (1989)
Jain (2001) Moral and political value of society Jain (2001) British legal origin and years of
schooling
Chen et al. (2008), Treisman (2000) Protestant, level of development and
imports, exposure to democracy Treisman (2000)
Institutional transparency, fairness
6
Organizational level
Control power, bargaining power Chen et al. (2008) Reduce sales or enhance sales
Gaviari (2002), de Jong et al. (2010) Firm size, growth rate, corporate
governance, competitiveness, taxes Wu (2008)
Enhance growth and
efficiency Leff (1964)
Firm age and ownership De Jong et al. (2010)
Lower output per worker and efficiency of the firm
McArthur and Teal (2002) Firm‟s ability to pay and refusal
power Svensson (2003)
Creates need for secrecy Shleifer and Vishny (1993)
Firm competitiveness Wu (2008) Composition and level of FDI, entry strategies.
Gaviari (2002), Mauro (1995) Intertwined gift-giving culture and
bribery, strong link between officials and family businesses
Luo (2002)
Evolutionary hazard, strategic impediment, competitive disadvantage, organizational deficiency
Luo (2002)
Alternative authorities, dependence
on public infrastructure, exports Chen et al. (2008) Increase transaction costs
Rose-Ackerman (1975)
Mimetic isomorphism, market
competition, perceived benefits Gao (2010
Entrepreneur level
Masculinity Chen et al. (2008)
Higher level of social capital, boost trust and build a shared belief or reciprocity
De Jong et al. (2010) Years of education and work
experience
De Jong et al. (2010)
Network investment to overcome liability of newness of smallness.
De Jong et al. (2010)
2.2 Firm performance
7
of the non-financial measures mentioned above is that data is not easily obtainable, while a financial measure such as sales is usually available and therefore often used in the literature on firm performance.
2.3 Corruption and firm performance
At the firm level the literature on the effects of corruption is dispersed. Arguments for the positive effects of corruption often originate from Leff (1964). In their study it is argued that corruption can enhance efficiency and improve growth by removing impediments developed by the government which hinder investment and growth enhancing factors (Bardhan 1997, Leff 1964, Tanzi 1998). This can take place through several mechanisms (Mauro 1995). The first is a commonly used argument in the literature in favour of corruption and states that it can „grease the wheels‟ or „oil the mechanism‟ of business (Jain 2001, de Jong, van Ees and Nhu 2009, Kaufman and Wei 1999, Rose-Ackerman 1975, Tanzi 1998). Corruption then acts as speed money and allows individuals to circumvent bureaucratic delay. Tanzi (1998) adds that the firm whose time is most valuable, due to for instance higher efficiency, will offer bribes to jump ahead in the queue. This increases the overall efficiency of the industry. The second mechanism establishes that a bribe will increase the level of social capital for the firm (de Jong et al. 2010). A higher level of social capital enhances performance through two different channels. Firstly, bribes may increase trust and establish a shared belief of reciprocity between the government official and the firm. This will increase firm performance through an increase likelihood for the firm to obtain favorable government contracts and treatment. The second channel is investment in networks that overcome liabilities of newness and smallness. Firm performance can increase through this channel by favorable relationships with government officials that give firms legitimacy and additionally decrease their risk for closure.
8
„guidelines and shortcuts on a relatively small sample of actors, which may impair the entrepreneur‟s ability to bring a new perspective on business activities.‟ (de Jong et al. 2010). The liability of sameness implies that network embeddedness may initially benefit the firm through favorable relationships and increased legitimacy. However, this same embeddedness can also decrease firm performance by making firms blind to information outside their network.
There are three additional arguments that support the argument that the positive effects of corruption on firm performance are not infinite. Firstly, firms are constrained in their growth by the increased bureaucratic costs that arise in more complex organizations that are also subject to information processing constraints (Cyert & March 1963). Secondly, bribes may crowd out alternative investments and eradicate incentives for instance in innovation (Luo 2004). The optimal level of growth is attained when the economic benefits from previously underutilized resources are outweighed by the bureaucratic costs of managing the additional size of the firm (Peng and Heath 1996). Thirdly, bribes need bribes (de Jong et al. 2010). Government officials recognize the possibility to demand bribes and ask for more. As a result the volume of bribes will increase, while high volumes of bribes are expected to be less effective than small volumes. The increased bribes will absorb the firm‟s resources and limit firm performance.
In summary, the above arguments suggest that for firms that operate in a weak formal institutional environment bribes are potentially advantageous. The arguments supporting a positive effect of corruption on firm performance are; “greasing the wheels” of business, and investing in social capital. Investing in social capital means both investing in networks to overcome liabilities of newness/smallness and increasing trust and establish a shared belief of reciprocity. On the other hand, it is acknowledged that “greasing the wheels” of business may actually increase administrative costs, networks have the risk of the liability of “staleness and sameness”, a larger firm will also have increasing bureaucratic costs, bribes may crowd out alternative investments and finally bribes need bribes. It is assumed that these arguments hold for the different types of corruption as well.
9
constant, the benefits diminish relative to the costs when the corruption level increases. This is mostly applicable for small firms as they have limited production capacity and fixed resources in the short term. The arguments above lend support to the following hypothesis:
Hypothesis 1: There is an inverted U-shape relationship between corruption and firm performance
This thesis recognizes that different types of corruption exist and that they are likely to have independent effects on firm performance. In the last section of this chapter the overall concept of corruption is therefore disentangled into institutional, operational, and performance corruption. A theoretical explanation of these different types of corruption is provided in the next sections and is intended to enhance the understanding of the sub-dimensions. The first sub-dimension to be discussed is institutional corruption. This is intended particularly as grease money to deal with laws and regulations. Next to the general corruption effects on firm performance, this type of corruption has the added benefit of potential spillover effects. The second sub-dimension is operational corruption. This type of corruption is meant to deal with local governments when a firm is faced with obstacles for ongoing operations. Finally, the last sub-dimension is performance corruption. In this case the firm considers corruption simply as a cost of doing business in a certain country. A more thorough theoretical explanation per sub-dimension is provided in the following sections.
2.4 Institutional corruption
10
collusive form of corruption. In that case the firm is the initiator of the bribe and both parties gain.
There are five main causes for institutional corruption. The first is the lack of transparency in rules and laws, which in turn decreases the likelihood of being detected and/or receiving a penalty, making it more likely for a firm to pay bribes (Tanzi 1998). The second reason, the level of public sector wages, is an indirect cause of institutional corruption. A low wage can increase the frequency of corrupt acts as underpaid employees risk less by demanding bribes. A third common facilitator of institutional corruption is the amount of red tape (Mauro 1995 and Tanzi 1998). The amount of red tape indicates the degree to which the regulatory environment faced by firms when seeking approvals and permits presents an obstacle to business. The fourth cause is that a firm that faces a decentralized government is forced to go through a sequential corruption system by having to approach several government officials before it can obtain all the necessary licenses and permits. As Huntington (1968, p.386) stated in Kaufman and Wei (1999) “in terms of economic growth, the only thing worse than a society with a rigid, centralized, dishonest bureaucracy is one with a rigid, over-centralized and honest bureaucracy”. Firms facing such a over-centralized government control are characterized by being under a high level of bureaucratic control (Svensson 2003) which gives discretionary power to the government agencies who use this to extract bribes. In such environments firms are thus more likely to pay bribes in order to circumvent these laws and regulations. A final essential determinant of institutional corruption is the government officials‟ opportunity to demand bribes. This depends on the extent of their control over a firm‟s business decisions and cash flows (Svensson 2003). It is argued that for private firms the control rights originate from the regulatory system in place and to what extent the government official can implement, execute and enforce rules and benefits for firms.
11
possibility for other firms in the industry to make investments which benefits the industry as a whole due to an increased demand for the goods produced in that sector. In addition, with a high level of demand for institutional corruption local firms are attractive partners for foreign firms entering the market because they know their way through the institutional maze. A surge in FDI can increase the amount of capital a firm has available to invest and this can in turn enhance the performance of the firm. Institutional corruption is subject to the same rationale that limits the benefits of general corruption. We argue that institutional bribery is also subject to diminishing returns to firm performance. Therefore, the following hypothesis is formulated:
Hypothesis 2: There is an inverted U-shape relationship between institutional corruption and firm performance
2.5 Operational corruption
Operational corruption is the exchange of pecuniary or non-pecuniary benefits between a firm and an agent in order to overcome obstacles for the firm to continue or establish operations. This type of corruption is expected to be initiated by the government official and is forced upon the firm, making it a coercive form of corruption.
Operational corruption occurs because firms seek to embed themselves in a network of local firms and government officials in order to establish and/or continue their operations. In transition economies relationships with local governments are essential as they control investment allocation, financing and business formation.
The effect of investing in operational corruption is a solid social network that help firms„ ongoing operations in a country. When firms invest in these networks with bribes this can also help them overcome liabilities of newness and/or smallness (de Jong 2010). In addition, it gives them increased legitimacy and reduces its risk for closure. Operational corruption is subject to the same rationale that limits the benefits of general corruption. Hence, we argue that operational corruption is subject to diminishing returns to firm performance. The following hypothesis is formulated:
12
2.6 Performance corruption
Performance corruption is defined as the exchange of a pecuniary or non-pecuniary benefit between a firm and an agent in order for the firm to obtain an entrepreneurial advantage. This can be classified as an economic type of corruption. This type of corruption can be both firm and government official initiated and is likely to be mutually beneficial and thus a collusive form of corruption. An example of performance corruption is a firm paying a bribe to obtain a favorable contract. The benefits obtained with this type of bribery are likely to be scarcer, since the corrupt official can only prefer one high paying briber over other potential beneficiaries for a contract.
The resource based view can be used when considering antecedents of performance corruption (Li and Ouyang 2007). Corruption, in a resource based view perspective, changes the relative costs of inputs and outputs as mentioned before in the theoretical explanations of corruption in general. Li and Ouyang (2007) argue that firms can calculate the minimal bribe which provides them with the largest benefits. The firms that are the first to adopt the equilibrium price of bribing are likely to become the market leaders. Others will, according to the behavioral theory, imitate this behavior. Eventually this will lead to an optimal level of bribery and the firms are in a stable, Nash equilibrium. In this equilibrium the firms decide on their individual optimal bribing level, while taking into account other firms‟ bribing behavior. The effect of paying a bribe to enhance performance is rather straightforward. Firms will decide whether or not it is in their best interest to pay the bribe by calculating the current and expected profits. If the resources they can obtain by bribing provide them with a sustained competitive advantage the inclination to continue bribing is higher. However, the firm needs to take into account that it will not be able to absorb all the additional business that comes with new contracts. In addition, at some point the high cost of the bribe will outweigh the added benefit in sales for the firm. The fourth hypothesis is:
13
2.7 The theoretical model
Combining the four hypothesis leads to the following theoretical model for this thesis that is tested in section three:
is the constant term. This is interpreted as the value of firm performance when all the corruption items are equal to zero. The constant term does often not provide a meaningful interpretation. Nonetheless, it is required for the model in order to describe the fit and make predictions within the range of the observed data.
The regression coefficients of the first order variables of corruption are expected to be positive for all types of corruption. This indicates that an increase in corruption will increase firm performance. Therefore it is predicted that , , , > 0. Considering the squared terms it is anticipated that the slope of the linear line will not be constant, but will in fact decrease at higher levels of corruption. Therefore it is predicted that , , , < 0. The control variables include a measure of innovation, firm age, firm ownership and sector the firm operates in. The expected regression coefficients of these variables will be discussed in chapter three. The error term, , in the regression equation represents the effect of all other factors that influence the dependent variable, other than the corruption variables.
3. Methodology
3.1 Data and sample selection
Data
14
reforms in ATEs have gone hand in hand with an increase in corruption and as a result these countries have become prone to inefficiencies and free riding (Gao 2010). This is also caused by the fact that the government still has a significant role to play in the interaction with firms, which makes it more likely for firms to engage in corrupt behavior. Firms are government institutions were tied together through state ownership and reciprocal benefits. Tan and Peng (2003) note that pushed by the government, which is increasingly interested in tightening state-owned enterprises (SOE‟s) budgets, many SOEs are forced to reduce inefficiency. In addition, private startups have grown to compete for market share. In this increasingly competitive market and with weak (financial) markets it is challenging for SOEs to transform. In fact, Bardhan (1997) pointed out that despite the institutional reforms the output of firms in ATE countries remained partially controlled by government officials. Firms can sell the remainder of their output at market prices, which creates new opportunities for bribery. Treisman (2000) established that because ATEs have not been exposed to democracy the likelihood of being corrupt increases.
The countries used in this sample dataset (Kazakhstan, Tajikistan, Uzbekistan and Kyrgyz Republic) are currently experiencing increasing growth rates (Graph 1) and corruption levels (Table 3). Wu (2008) noted that these corruption levels have been quite persistent over time. This leads to serious threats to these countries‟ long run growth potential, since the occurrence of corruption reduces other investment opportunities. Kraar (1995) shows that bribery adds up to 5 % of the cost of doing business in Asia.
Source: data.worldbank.org
The persistence of bribery in ATE countries can be partially explained by their collectivist cultures (Hofstede 1993). These cultures reflect the importance of individuals contributing to
-30 -25 -20 -15 -10 -5 0 5 10 15 G DP grow th (a nn ua l % ) Years Graph 1: Economic growth sample countries
15
the collective. Therefore mimetic isomorphism is more likely to occur in such countries. Mimetic isomorphism is defined as reaching compliance through imitation. Firms tend to copy the bribery behavior of other successful firms when faced with ambiguity and uncertainty (Gao 2010).
In countries with high corruption levels it is not uncommon for firms to pay a facilitation fee for government services they are rightfully entitled to. Firms often initiate the corrupt exchange themselves in order to avoid or reduce tax, pass laws and regulations, secure public contracts or block the entry of potential competitors into the market (Wu 2008). According to the 2010 Corruption Perception Index (CPI) made by Transparency International the countries in our sample rank high in the level of corruption. The 2010 CPI consists of 178 countries. According to the Transparency International results, nearly three quarters of these countries scores below five, on a scale from 10 (highly clean) to 0 (highly corrupt). The four countries selected for our sample dataset all score below 3, indicating they are experiencing substantial level of corruption.
Table 3: CPI score of sample selection over time
Year 2010 2005 2000 Country Country rank CPI score Country rank CPI score Country rank CPI score Kazakhstan 105 2,9 107 2,6 65 3,0
Tajikistan 154 2,1 144 2,1 N/A N/A
Kyrgyzstan 164 2,0 130 2,3 N/A N/A
Uzbekistan 172 1,6 137 2,2 79 2,4
16
Table 4: Nature of bribery in selection of Asian transition economies
Never (%) Seldom (%) Sometimes (%) Regularly (%) Do firms pay some irregular “additional
payment/gift” to get things done? 34.32 16.23 23.17 26.28 Do firms know in advance about how
much this “additional payment/gift” is? 40.86 14.95 14.82 29.37 Can the firm go to another official to get
the correct treatment without recourse to unofficial payments if a government agent acts against the rules?
21.97 22.60 26.84 28.59
Source: World Business Enterprise Surveys. The categories “frequently”, “usually” and “always” are grouped into the category “regularly” (Wu, 2008).
Sample selection
Measures of corruption can be classified in external, hybrid and internal measures (Asiedu and Freeman 2009). The external measures are assessments of corruption made by risk analysts outside of the country. These are semi micro-level studies as they use firm and country-level data on corruption. For this measure the International Country Risk Guide (ICRG) is commonly used. This is a publication from the Political Risk Services (PRS) Group. Hybrid measures combine different sources of corruption data into a composite index. A common source for this data is the Corruption Perception Index (CPI). Finally, internal measures are based on perceptions of firms that operate within a country. These are micro-studies, based on firm level data on investments and corruption. This is the measure that is used in this paper from the World Business Enterprise Surveys (WBES). These measures are subjective to the respondent‟s perception. Still they are relevant as Mauro (1995) argued that also perceptions of corruption will influence investment decisions, growth and political behavior of citizens.
17
to firm information of 2004. As for the surveys themselves, they were conducted by private contractors using face-to-face interviews since the survey questions are sensitive and address the relationship between the business and the government and bribery-related topics. To guarantee integrity and confidence of the quality of the data, confidentiality of the survey respondents is ensured. The respondents of the enterprise surveys are business owners and top managers of formally registered companies with five or more employees. The primary business sectors of interest are the manufacturing and service sectors. The sampling method for the Enterprise Surveys is a stratified random sampling method with replacement. Stratified sampling means that the population is divided into homogeneous, mutually exclusive and collectively exhaustive subgroups. Each member of the population has the same chance of being selected. Using random sampling in a stratum enhances representativeness of the sample as the sampling error is reduced. It may produce a weighted mean that has less variability than an arithmetic mean which is produced by a simple random sample of a population. Most of the firms that are interviewed are small and medium sized enterprises. The enterprise surveys oversample large firms as they tend to facilitate job creation.
The unit of analysis of this thesis is the firm. The data from the Enterprise Survey represents only firms that were willing to participate in the survey. A total of 1287 firms participated in the interview. Occasionally, the sample included missing observations for the dependent and independent variables. For the regression analysis, all observations with missing values on any questionnaire item were deleted. This resulted in a dataset with 834 full observations, giving an effective response rate of 65%. There are 355 firms from Kazakhstan, 182 firms from Uzbekistan, 162 firms from Tajikistan and 135 firms from Kyrgyzstan.
3.2 Measurement Dependent variable
The performance of the firm is measured by the natural logarithm of the firm‟s total sales in 2004 (in thousands of dollars). The natural logarithm is used because the data for this variable exhibits a skewed distribution. The interpretation of the regression coefficients changes as a unit change in the independent variable will now be interpreted as a percentage change in the dependent variable instead of a unit change.
Independent variables
18
corruption are measured on 7-point likert scales. Each type of corruption is measured by adding three questionnaire items. These added scales are used in the regression.
Institutional corruption is measured by the frequency with which firms have to make unofficial payments/gifts in order to (1) deal with courts, (2) deal with customs and imports, (3) influence the content of new legislation
Performance corruption is measured by the frequency with which firms have to make unofficial payments/gifts in order to (1) obtain business licenses and permits, (2) deal with taxes, (3) obtain government contracts.
Operational corruption is measured by the frequency with which firms have to make unofficial payments/gifts in order to (1) deal with occupational health and safety inspections, (2) deal with environmental inspections, (3) deal with fire and building inspections.
Control variables
There are four control variables for firm performance. The first control variable is the level of innovation of the firm. Innovativeness is likely to increase firm performance, as more innovative firms are likely to operate in growing industries. The second control variable is firm age. Older firms are likely to perform less compared to younger firms as their management styles and/or technology becomes outdated and they are too inert to adapt to new trends. The third control variable is the ownership of the firm, defined by their legal status.. Private firms are likely to have better performance compared to state-owned firms as they are usually the ones operating in more profitable industries and have more flexibility to adapt to changing environments. The last control variable is the kind of industry that the firm operates in. Service firms in Asian transition economies are still new and growing quickly, so their performance is likely to be higher than that for industrial firms who operate in older and declining industries.1
1
19
Three of the control variables are measured with dummy variables. A firm is classified as an innovative firm when it has obtained a product license in the last 36 months. An innovative firm is measured by a dummy variable equal to 1 (and 0 otherwise). Private firm ownership is measured by the legal status of the firm at the point of the interview. Private firms get the dummy variable that equals 1 (and 0 otherwise). The industry the firm is operating in is measured the firm‟s main area of activity in terms of annual sales. This is then classified as a dummy variable where firms that have most of their activities in services get a value equal to 1 (and 0 otherwise). The age of the firm is defined by the number of years the company exists from establishment up to the point of the interview. An overview of the variables, their definition and measures can be found in Table 5.
3.4 Econometric Techniques
In this thesis an ordinary least squares (OLS) regression analysis is carried out. The OLS methods minimizes the sum of squared vertical distances between observed responses in the data, and the responses predicted by the linear approximation, . A multivariate regression on a polynomial model is used to investigate the direction and strength of the
Table 5: Variable, definitions and types
Variable Definition Measure
Firm
performance Total sales. Total sales (log) Corruption The exchange of (non-) pecuniary benefits between a firm and an
agent in order to gain an advantage or overcome obstacles Continuous
Institutional corruption
The exchange of (non-) pecuniary benefits between a firm and an agent in order to affect a law, regulation or policy.
Sum of three 7-point likert scales
Performance corruption
The exchange of a (non-) pecuniary benefits between a firm and an agent in order for the firm to obtain an entrepreneurial advantage.
Sum of three 7-point likert scales
Operational corruption
The exchange of a (non-) pecuniary benefits between a firm and an agent in order to overcome any obstruct for the firm to continue or establish operations.
Sum of three 7-point likert scales
Innovation The obtaining of new product license agreements Dummy (Innovative = 1, Not innovative = 0)
Firm age The age of the firm = 2004 – year of establishment Continuous
Private firm
ownership The legal status of the firm
Dummy (Private = 1, State-owned = 0)
Service
sector The firm‟s main area of activity in terms of annual sales.
20
relationship between the dependent variable (firm performance) and the main independent variables (corruption, institutional corruption, operational corruption and performance corruption). Before applying the OLS method, outliers and possible issues with collinearity are examined. We also examine the possibility of country effects. Finally, factor analysis is used to find whether a number of the questionnaire items measure common underlying factors.
OLS assumptions
21
Outliers and collinearity
Outliers and collinearity were also addressed. Outliers are variables that have a large difference between the predicted and the actual observed variable, also referred to as the residual. An outlier can be an odd observation, but could also indicate a mistake in data entry. Outliers did not seem to affect the results substantially, as observed from running a robust regression. In the robustness test the most influential points are excluded, and observations with large absolute residuals are down-weighted. Outliers are excluded based on the Cook‟s distance. The Cook‟s distance is a measure that combines the information of residual and leverage of the observation. Observations with high leverage and/or large residuals may distort the accuracy of the regressions. Leverage measures how far an independent observation deviates from its mean. Any observation with Cook‟s distance exceeding 1 is dropped. After dropping the most influential points the cases with large absolute residuals are down-weighted.
Collinearity occurs when predictors have a strong linear relation between them. When examining the bivariate correlations among the independent variables we observed that only the squared variables had values larger than the cut-off point of 0.8. This is a logical result as a variable and its squared term naturally have a high correlation between them. The results show that the correlation is not very strong for the linear model. By obtaining a variance inflation factor (VIF) for the regression we can see how strong the multicollinearity is in the least squares regression. The VIF is an index which measures to what extent the variance of an estimated regression coefficient ( is increased because of collinearity. . In our model the highest VIF is 1.52 (not including the quadratic term with a VIF of 13.16) is 1.52. For the regression the standard error of the independent variables is times as large as it would be if the independent variables were uncorrelated with the other independent variables. These values are well below the cut-off point of 10 recommended by Neter, Wasseman, and Kutner (1985). Therefore collinearity is not a problem in this dataset. Only when we include the squared values of our independent variables does the standard error become much larger, This is logical as the value and its squared value are obviously highly correlated.
Country effects
22
country. Not all dummy variables will be incorporated into the regression equation as does would result in exact collinearity since the sum of the country dummies would be equal to 1. One dummy variable will be omitted and this is defined as the reference group. The test of the equivalence of the different regression is called a Chow test and has a null hypothesis that the coefficients of the country dummies are all equal to zero. The results show that the countries have significant differences and the data should not be pooled without using country dummies. Since there are only four countries the regression in the next section each country will be estimated separately. This will provide additional insights into the regressions per country. The regression for the entire dataset with country dummies can be found in Appendix C.3.
Factor analysis
In order to obtain the independent variables for the different types of corruption an exploratory factor analysis was performed with varimax rotation. Factor analysis is commonly used to find whether a number of questionnaire items measure a common underlying factor. It is based on the correlation matrix of the observed variables where it is assumed that the observed variables are standardized (E(x) = 0 and = 1). There are a number of regression equations that try to explain each observed variable with common factors and unique factors :
It is expected that every variable has a unique factor that account for the variability of that variable which is not due to any of the common factors. The common factors are hypothetical variables explaining why some variables are correlated. The coefficients are
23
minimizes the loadings on the other factors. In Table 6 the rotated factor loadings on questionnaire item 41 are presented.
Table 6: Rotated factor loadings (pattern matrix) and unique variances
Variable Factor 1 Factor 2 Factor 3 Factor 4 Uniqueness
q41a 0.3563 0.1954 0.3727 0.2349 0.6408 q41b 0.3027 0.2833 0.6287 0.1582 0.4078 q41c 0.3088 0.4259 0.363 0.2566 0.5256 q41d 0.6842 0.323 0.1778 0.189 0.3601 q41e 0.6818 0.1799 0.3832 -0.0214 0.3554 q41f 0.7248 0.2879 0.2236 0.0432 0.3399 q41g 0.3679 0.3483 0.5834 -0.0577 0.3996 q41h 0.2826 0.5959 0.3269 -0.0512 0.4556 q41i 0.3112 0.6701 0.2942 0.0865 0.3601 q41j 0.3555 0.5319 0.1252 0.1792 0.5429
The higher the factor loading in the table above, the more relevant it is in defining the factor‟s dimensionality. Here, four factors are retained. The questions representing the items in question 41 can be found in Appendix A.2. It seems that q41d, q41e and q41f define factor 1. Factor 2 is defined by q41h, q41i and q41j. Factor 3 does not have such a clear pattern of factor loadings. Since we have 3 questionnaire items for each factor it is convenient for the regression comparison to use 3 questionnaire items for factor 3 as well. Consequently, next to the highest factor loadings on q41b and q41g another high factor loading should be chosen. After taking a look at the questionnaire it appears that q41c is the most relevant question to add as a third questionnaire item. Factor 4 does not have high factor loadings and is thus not considered as a different type of corruption. Finally, we point out that q41a has a lower factor loading and was not a part of any underlying construct. This is also evident from the high level of uniqueness of this questionnaire item. It has a uniqueness of 0.6408 which indicates that 64.08% of the variance in this item is not accounted for by the other items.
24
4. Empirical results 4.1 Descriptive Statistics
The average annual sales of the firms in the sample were 781 thousand dollars in 2004 (Table 7). There appears to be a substantial amount of deviation from this mean, as reflected by the high standard deviation of 2,533 thousand dollars.
The firms in this full sample claim to spend an average of 1.37 percent of sales on corruption. The mean sales of 2004 this imply that the average firm in this sample pays over ten thousand dollars for corruption (781,000 * 1.37%) in one year. As for the sub-dimensions of corruption, the most common form of corruption is performance corruption. It scores a 6.4 on the likert scale from 0 (never) to 18 (always). This suggests that these firms seldom to sometimes need to make unofficial payments or gifts to, for instance, obtain a government contract. Institutional and operational corruption score approximately the same on the likert scale. The scores of 4.61 and 4.91 suggest that firms seldom need to make an unofficial payment or gift to, for instance, deal with environmental inspections or influence the content of a new legislation.
The control variables indicate that most firms are private firms. 15% of the firms has obtained a product license in the past 36 months and there is an approximately even distribution of industry and service firms in the sample. The average age of the sample firms is 12.6 years in 2004.
24 Table 7: Descriptive statistics complete sample and collinearity (N = 834)
Variable2 Mean SD 1 2 3 4 5 6 7 8 9
Firm performance (in thousand USD) 781 2533 1.00
Private firm ownership .91 .29 -.23** 1.00
Service sector .44 .50 -.28** -.02 1.00
Innovation .15 .35 .10** .02 -.11** 1.00
Firm age 12.6 14.9 .30** -.36** -.06 -.01 1.00
Corruption (in % of sales) 1.37 2.71 -.08* .06 -.04 .12** -.06 1.00
Institutional corruption 4.613 2.86 .12** .06 -.05 .14** .00 .28** 1.00
Performance corruption 6.401 3.75 .02 .09** -.09 .17** -.03 .44** .68** 1.00
Operational corruption 4.911 2.86 .08* .09** -.14** .13** -.05 .28** .59** .67** 1.00
*p < 0.05; **p < 0.01. SD, standard deviation.
2 (1) Firm performance, (2) Private firm ownership, (3) Service sector, (4) Innovation, (5) Firm age, (6) Corruption, (7) Institutional corruption, (8) Performance Corruption,
(9) Operational corruption
3
25 Table 8a: Descriptive statistics per country
Kazakhstan (N = 355) Uzbekistan (N = 182)
Variable Mean Std. Dev.
Min Max Mean Std. Dev.
Min Max
Firm performance
(in thousand USD) 693 1488 5 14400 1213 4307 8 46242 Private firm ownership .93 .26 0 1 .90 .31 0 1
Service sector .28 .45 0 1 .59 .49 0 1
Innovation .13 .34 0 1 .12 .32 0 1
Firm age 10.29 11.39 3 79 14.10 16.41 3 94
Corruption (in % of sales) 1.34 2.85 0 15 .83 2.13 0 10
Institutional corruption 4.45 2.26 3 18 3.68 1.76 3 14
Performance corruption 6.09 3.54 3 18 5.17 2.91 3 17
Operational corruption 4.93 2.81 3 18 4.13 2.16 3 18
Table 8b: Descriptive statistics per country
Tajikistan (N = 162) Kyrgyzstan (N = 135)
Variable Mean Std.
Dev.
Min Max Mean Std.
Dev.
Min Max
Firm performance
(in thousand USD) 535 1513 4 11811 728 2429 8 20155 Private firm ownership .90 .31 0 1 .89 .32 0 1
Service sector .56 .50 0 1 .53 .50 0 1
Innovation .17 .38 0 1 .19 .40 0 1
Firm age 10.59 12.53 3 72 19.23 20.22 3 113
Corruption (in % of sales) .99 1.74 0 10 2.61 3.49 0 10
Institutional corruption 4.69 2.77 3 16 6.19 3.92 3 18
Performance corruption 6.72 3.90 3 18 8.46 4.23 3 18
Operational corruption 5.30 3.08 3 17 5.45 3.31 3 18
Kazakhstan
26
compared to other countries. In 2004 the average sales is 693 thousand dollars, nearly 100 thousand below the average of the sample. In terms of corruption firms in Kazakhstan are almost the same as the mean values of the entire sample. From the control variables it stands out that Kazakhstan has more service firms relative to industry firms compared to the other countries.
Uzbekistan
In terms of firm performance firm in Uzbekistan perform substantially better compared to the other countries. The average firm sales is 1213 thousand dollars, exceeding the average by almost half a million dollars. Investigating the descriptive statistics per country shows that in terms of firm performance Uzbekistan is performing substantially better compared to the other countries. Interestingly the firms in Uzbekistan also claim to be the least corrupt. Only 0.83 percent of sales is paid for corruption. The different types of corruption occur seldom to never in the case of institutional corruption. The control variables are representative for the entire mean sample set.
Tajikistan
Firms in Tajikistan have the lowest average firm performance of the entire sample. The average firm sales is 535 thousand dollars, considerably lower than the mean of 781 thousand dollars. The corruption levels in Tajikistan are close to the mean values of the entire sample. Also the control variables are near to the mean values of the entire sample.
Kyrgyzstan
27
4.2 Characteristics of corrupt versus non-corrupt firms
A dummy variable was constructed where all firms that claim to not pay any percentage of their sales on bribes and in addition never give additional payments/gifts for operational, institutional and performance corruption are assigned a value of 1. There are 236 out of 834 firms that claim never to engage in corrupt transactions. In contrast, 190 firms admit that they engage in all types of corruption. In Table 9 the characteristics of these different groups are compared.
It appears that firms that admit to use all types of corruption have a higher performance than firms that are never corrupt or only sometimes corrupt. Firms that claim to engage in some types of corruption have the lowest performance. These results suggest that a U-shaped relationship exists between corruption and firm performance. The corrupt firms are characterized as private, innovative industry firms with an average age of 11.6 years. While firms that claim never to be corrupt include more state-owned firms, less service and innovative firms and older firms.
Table 9: comparison of characteristics of firms with and without corruption
Never Corrupt (N = 236) Sometimes Corrupt (N = 408) Always Corrupt (N = 190) Variable Mean SD Mean SD Mean SD Firm performance (in
thousand USD) 783 2361 682.16 2600 992 2598 Private firm ownership .86 .35 .92 .27 .94 .23
Service sector .52 .50 .42 .49 .38 .49
Innovation .09 .29 .12 .33 .26 .44
Firm age 14.4 17.2 12.1 14.11 11.6 13.04
Corruption (in % of sales) 0 0 1.36 2.78 3.07 3.22
Institutional corruption 3 0 4.05 2.01 7.81 3.62
Performance corruption 3 0 6.60 2.99 10.17 3.73
Operational corruption 3 0 4.60 2.26 7.96 3.30
28
4.3 Corruption and firm performance
Three regressions are run per country in order to investigate the different relationships between corruption and firm performance in the respective Asian transition economies. The theoretical model is put into 3 steps. In the first regression, Model 1, only the control variables and the error term are included. Model 2 is a regression that comprises both control and linear independent variables. The last regression, Model 3, adds the squared term to the equation.
29 Table 10: regression corruption and firm performance for four ATE countries
Kazakhstan Uzbekistan Tajikistan Kyrgyzstan
30 Kazakhstan
In the sample of firms from Kazakhstan the adjusted R2 in Model 1 is just 0.1218. This means that only 12.18% of the variance in the dependent variable can be explained by the control variables. In Model 2 the explanatory power of the model has improved by approximately 2%. The control variables and the first term corruption variables together explain 13.24 % of the variance in the dependent variable. Finally, when in Model 3 the squared terms are added to the regression improving it with almost 3%. For firms in Kazakhstan the theoretical model explain approximately 15.16% of the variance in firm performance. This indicates that the corruption variables do not have a very strong relationship with firm performance. On the other hand, we can observe that including the linear and squared terms improves the explanatory power of the model by almost 5%.
Corruption
In Model 2 the regression coefficient negative. This indicates that when a firm engages in corruption it will lower firm performance. The regression coefficient is not significant, nor does it become so in Model 3. The sign of the regression coefficient does change in Model 3, suggesting an inverted U-shape relationship between corruption and firm performance. Nevertheless, this relationship is not significant and thus no strong argument can be made for diminishing returns to corruption. In conclusion, hypothesis 1 that there exist diminishing returns to corruption is rejected for firms in Kazakhstan.
Institutional corruption
The regression coefficient of institutional corruption has a positive sign in both Model 2 and Model 3. The squared term in Model 3 is negative, indicating diminishing returns to institutional corruption. Furthermore, all the regression coefficients in Model 3 are significant at the 1% significance level. These results lend strong support to hypothesis 2, and it can be concluded that we fail to reject hypothesis 2 that there are diminishing returns to institutional corruption.
Operational corruption
31
insignificant we can reject hypothesis 3 that there are diminishing returns to operational corruption.
Performance corruption
Performance corruption is found to have a negative effect on firm performance in both Model 2 and Model 3. In Model 3 this relationship is significant at the 10% level. The sign of the regression coefficient in the squared term is negative and insignificant. We can conclude that hypothesis 4 is rejected. We cannot support the existence of diminishing returns to performance corruption.
Uzbekistan
The adjusted R2 of the regressions for Uzbekistan are the highest in the sample. In Model 1 38.22% of the variance in the dependent variable is explained by the control variables. The R2 increases with 2.68% when the linear terms are included. Finally, the model including the squared terms has an adjusted R2 of 0.3965. Although this is still far from a perfect explanatory power this still means that 39.65% of the variance in firm performance is explained by the corruption and control variables. Including the squared terms improved the model by 1.38%.
Corruption
Model 2 shows a negative, significant regression coefficient. This result suggests a negative linear relationship between corruption and firm performance. In that case corruption would not increase firm performance, but rather decrease it. In Model 3 the first term remains negative, but it is no longer significant. We can conclude that for firms in Uzbekistan hypothesis 1 can be rejected. There is no support for diminishing returns to corruption.
Institutional corruption
In Model 2 the regression coefficient is positive and significant. Once the squared term is incorporated in Model 3 there appears to be an improvement, as the significance increases and the value of the regression coefficient is also higher. The squared term has a negative, significant regression coefficient. Hence, for firms in Uzbekistan we fail to reject hypothesis 2. There is support for the diminishing returns to institutional corruption.
Operational corruption
32
hypothesis. The first operational corruption term is positive and the second is negative. However, we do not obtain any significant results and will therefore have to reject hypothesis 3 of diminishing returns to operational corruption.
Performance corruption
Performance corruption has no significant regression coefficients in either Model 2 or Model 3. The signs of the regression coefficients in Model 3 indicate a U-shaped relationship, instead of an inverted U-shape relationship. Hence, for firms in Uzbekistan we reject hypothesis 4 that there are diminishing returns to performance corruption.
Tajikistan
Model 1 of Tajikistan has an adjusted R2 0.2666. When adding the first corruption terms the R2 is increased by 3.61 percent. In Model 2 27.25% of the variance in the dependent variable is explained by the independent and control variables. Adding the squared terms improves the explanatory of the model by 1.39%. In Model 3 the adjusted R2 is .2679, implying that 26.79% of the variance in firm performance is explained by the corruption and control variables.
Corruption
The regression for firms in Tajikistan do not provide us with any significant results for any type of corruption. Interpreting the signs of the regression coefficients for corruption we find that in Model 3 the first corruption term is negative and the squared term is positive. This indicates an U-shaped relationship. To conclude, we can reject hypothesis 1 that there are diminishing returns to corruption for firms in Tajikistan.
Institutional corruption
Again there are no significant results for the regression coefficients. The signs of the coefficients can tell us that in this case we do observe the inverted U-shape that was hypothesized. Still, we reject hypothesis 2 that there are diminishing returns to institutional corruption for firms in Tajikistan.
Operational corruption
33
Performance corruption
In the case of performance corruption the linear term has a negative sign, while the squared term has a positive sign. This indicates that the returns to performance corruption tend towards a U-shaped relationship. The results are not significant, therefore we reject hypothesis 4 that there are diminishing returns to performance corruption for firms in Tajikistan.
Kyrgyzstan
The adjusted R2 for the regressions in Model 1 is not high, however it does improve a lot compared to the previous regressions. In Model 2 the R2 has improved with 8.15% after adding the linear terms to the regression. After adding the squared term in Model 3 the R2 even improves with 10.51%. In the end, according to the adjusted R2, 28.60% of the variance in firm performance for firms in Kyrgyzstan is explained by the corruption and control variables.
Corruption
Model 2 shows a negative, significant regression coefficient for the linear term. Model 3 does not have any significant regression coefficients. This would lead us to believe that corruption has a negative effect on firm performance. The same significant relationship was observed in the corruption variables in Uzbekistan. An increase in corruption as a percentage of sales might lead to a decrease in firm performance. We can hereby reject hypothesis 1 that there are diminishing returns to corruption for firm in Kyrgyzstan.
Institutional corruption
The regression coefficients in Model 2 show a clear positive, significant relationship between institutional corruption and firm performance. When adding the squared term in Model 3 the inverted U-shape relationship appears. We fail to reject hypothesis 2. There is support for diminishing returns to institutional corruption in firms in Kyrgyzstan.
Operational corruption
34
Performance corruption
Finally, the regression coefficients of performance corruption are not significant in either Model 2 or Model 3. The signs also do not indicate any diminishing returns to performance corruption. Rather they suggest a negative, decreasing relationship between performance corruption and firm performance. We need to reject hypothesis 4 that there are diminishing returns to performance corruption in firms in Kyrgyzstan.
Control variables
Private firm ownership
The ownership variable refers to whether a firm is state-owned or private. We expected that state-owned firms would have a lower level of firm performance compared to private firms. The results contradict this expectation, since the regression coefficient for private firm ownership is negative. This implies that being a private firm will lower firm performance. An explanation for this finding could be that there are too few observations of state-owned firms in the sample, in total only 77 firms out of the 834 observations are public firms. Except for Kazakhstan the ownership control variable had a significant negative effect on firm performance in every country.
Service sector
The findings show that firms that operate in the service sector have a lower firm performance. The result is not in line with our expectations. Since service firms in these countries are still new and growing quickly their performance was expected to be higher than for manufacturing firms who operate in older and declining industries. An explanation of the result could be provided by the size of the firms. The service firms in this dataset are relatively small compared to industry sector firms. Previously it was already mentioned that firm size and firm performance in this sample have a very high correlation. This can also be seen in Appendix C.1. This can partly explain the negative regression coefficients for service firms, because they are small firms and hence will have a lower performance compared to larger manufacturing firms. The countries in the sample all have a negative significant regression coefficient for the service sector dummy.
Firm age
35
older firms have higher firm performance levels compared to young firms. The regression coefficient is significant in Kazakhstan and Uzbekistan. For Tajikistan and Kyrgyzstan the regression coefficients have a positive sign, but in this case they are not significant enough to explain firm performance.
Innovation
Innovation was expected to increase firm performance, as more innovative firms are likely to be operating in growing industries. This expectation is supported by the results of Kazakhstan and Uzbekistan. Innovation will increase firm performance according to the positive and significant regression coefficients in two countries. However, in Tajikistan and Kyrgyzstan the regression coefficients are not significant and negative. This is an indication that the different country environments influence the effectiveness of innovation.
4.4 Robustness Tests
