how important are financing constraints? - world...
Post on 10-Apr-2018
221 Views
Preview:
TRANSCRIPT
1
How Important Are Financing Constraints?
The role of finance in the business environment
Meghana Ayyagari Asli Demirgüç-Kunt Vojislav Maksimovic*
August, 2005
Abstract: What role does the business environment play in promoting and restraining firm growth? Recent literature points to a number of factors as obstacles to growth. Inefficient functioning of financial markets, inadequate security and enforcement of property rights, poor provision of infrastructure, inefficient regulation and taxation, and broader governance features such as corruption and macroeconomic stability are all discussed without any comparative evidence on their ordering. In this paper, we use firm level survey data to present evidence on the relative importance of different features of the business environment. We find that although firms report many obstacles to growth, not all the obstacles are equally constraining. Some affect firm growth only indirectly through their influence on other obstacles, or not at all. Using Directed Acyclic Graph (DAG) methodology as well as regressions, we find that only obstacles related to Finance, Crime and Political Instability directly affect the growth rate of firms. Robustness tests further show that the Finance result is the most robust of the three. These results have important policy implications for the priority of reform efforts. Our results show that maintaining political stability, keeping crime under control, and undertaking financial sector reforms to relax financing constraints are likely to be the most effective routes to promote firm growth.
Keywords: Financing Constraints, Firm Growth, Business Environment
JEL Classification: D21, G30, O12
_______________________________
*Ayyagari: School of Business, George Washington University; Demirgüç-Kunt: World Bank; Maksimovic: Robert H. Smith School of Business at the University of Maryland. We would like to thank Daron Acemoglu, Gerard Caprio, Stijn Claessens, Patrick Honohan, Leora Klapper, Aart Kraay, Norman Loayza, David Mckenzie, Lant Pritchett, Dani Rodrik, L. Alan Winters and seminar participants at George Washington University for their suggestions and comments. This paper’s findings, interpretations, and conclusions are entirely those of the authors and do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent.
2
I. Introduction
Understanding firm growth is at the heart of the development process, making it a
much researched area in finance and economics. More recently, the field has seen an
resurgence in interest from policymakers and researchers, with a new focus on the
broader business environment in which firms operate. Researchers have documented
through surveys that firms report many features of their business environment as
obstacles to their growth. Firms report being affected by inadequate security and
enforcement of property rights, inefficient functioning of financial markets, poor
provision of infrastructure services, inefficient regulations and taxation, and broader
governance features such as corruption and macroeconomic instability.1 Many of these
perceived obstacles are correlated with low performance.
These findings can inform government policies that shape the opportunities and
incentives facing firms, by influencing their business environment. However, even if
firm performance is likely to benefit from improvements in all dimensions of the business
environment, addressing all of them at once would be challenging for any government.
Thus, understanding how these different obstacles interact and which ones influence firm
growth directly is important in prioritizing reform efforts. Further, since the relative
importance of obstacles may also vary according to the level of development of the
country and according to firm characteristics such as firm size, it is important to assess
whether the same obstacles affect all sub-populations of firms.
1 For example, Batra, Kaufmann, and Stone (2003), Dollar, Hallward-Driemeier, and Mengistae (2004) and Carlin, Fries, Schaffer, and Seabright (2001).
3
In this paper we examine which features of the business environment directly
affect firm growth. We use evidence from the World Business Environment Survey
(WBES), a major firm level survey conducted in 1999 and 2000 in 80 developed and
developing countries around the world and led by the World Bank.2 We use this data to
assess (i) whether each feature of the business environment that firms report as an
obstacle affects their growth, (ii) the relative economic importance of the obstacles that
do constrain firm growth, (iii) whether an obstacle has a direct effect on firm growth or
whether the obstacle acts indirectly by reinforcing other obstacles which have a direct
effect, and (iv) whether these relationships vary by different levels of economic
development and for different firm characteristics.
We define an obstacle to be binding if it has a significant impact on firm growth.
Our regression results indicate that only Finance, Crime and Political Instability emerge
as the binding constraints with a direct impact on firm growth. In order to rule out
reported obstacles that are merely correlated with firm growth but are unlikely to be
causal we also use the Directed Acyclic Graph (DAG) methodology implemented by an
algorithm used in artificial intelligence and computer science (Sprites, Glymour, and
Scheines (2000)).3 This algorithm uses the correlation matrix of a set of variables to
determine whether a variable meets certain criteria, derived from probability and graph
theory, for it to be classified as a direct or indirect cause of another variable.
2 The World Bank created the steering committee of the WBES and several country agencies from developed and developing countries were involved under the supervision of EBRD and Harvard Center for International Development. For a detailed discussion of the survey see Batra, Kaufmann, and Stone (2003). 3 See Knill and Maksimovic (2005) for an application of this methodology to international finance. The methodology has been used recently in economics and finance to analyze price discovery and interconnectivity between separated commodity markets and the transportation markets linking them (Haigh and Bessler, 2004), to model the US economy (Awokuse and Bessler, 2003) and to study the interdependence between major international stock markets in the world (Bessler and Yang, 2003).
4
The DAG algorithm also confirms Finance, Crime and Political Instability to be
the binding constraints, with other obstacles having an indirect effect, if at all, on firm
growth through the binding constraints.
In further robustness tests, we find that the Finance result is the most robust, in
that Financing obstacles are binding regardless of which countries and firms are included
in the sample. Regression analysis also shows that Financing obstacles have the largest
direct effect on firm growth. These results are not due to influential observations, reverse
causality or perception biases likely to be found in survey responses. Political Instability
and Crime, the other two binding constraints in the full sample, are driven by the
inclusion of African and Transition economies where, arguably, they might be the most
problematic.
We also find that the relative importance of different factors varies according to
firm characteristics. Larger firms are affected by Financing obstacles to a significantly
lesser extent but being larger does not relax the obstacles related to Crime or Political
Instability to the same extent.
Examining the Financing obstacle in more detail, we find that although firms
perceive many specific financing obstacles, such as lack of access to long-term capital
and collateral requirements, only the cost of borrowing directly affects firm growth.
However, we find that the cost of borrowing itself is affected by imperfections in the
financial markets. Thus we find that the firms that face high interest rates are the ones
that perceive banks they have access to as being corrupt, under-funded, and requiring
excessive paperwork. We also find that difficulties with posting collateral and limited
5
access to long-term financing are also correlated with high interest rates. It is likely that
these latter obstacles are also aggravated by underdeveloped institutions.4
The extensive literature on institutional obstacles to firm growth is reviewed in
the next section. Several papers have specifically pointed to the importance of financing
obstacles. Using firm level data, Demirguc-Kunt and Maksimovic (1998) and others
provide evidence on the importance of the financial system and legal enforcement in
relaxing firm’s external financing constraints and facilitating their growth. Rajan and
Zingales (1998) show that industries that are dependent on external finance grow faster in
countries with better developed financial systems.5 Although these papers investigate
different obstacles to firm growth and their impact, they generally focus on a small subset
of broadly characterized obstacles faced by firms.
More recently, Allen, Qian, and Qian (2005), argue that China is an important
counterexample to the findings in the law, finance and growth literature. China is one of
the fastest growing economies although neither its legal nor financial system is well
developed by existing standards. Thus, they argue that the role of different factors in
contributing to the growth process is not well understood. We investigate the impact of a
wider set of potential obstacles and evaluate their relative importance as well as
interactions between them in constraining firm growth.
4 Fleisig (1996) highlights the problem with posting collateral in developing and transition countries with the example of financing available to Uruguayan farmers raising cattle. While cattle are viewed as one of the best forms of loan collateral by the US, a pledge on cattle is worthless in Uruguay. Uruguayan law requires for specific description of the pledged property, in this case, an identification of the cows pledged. The need to identify collateral so specifically undermines the secured transaction since the bank is not allowed to repossess a different group of cows in the event of nonpayment. 5 There is a parallel literature on financial development and growth at the country level. Specifically, cross-country studies (King and Levine 1993; Beck, Levine, and Loayza 2000; Levine, Loayza, and Beck 2000) show that financial development fosters economic growth. Also see Levine (2005) for a review of the finance and growth literature.
6
Our work is most closely related to Beck, Demirguc-Kunt, and Maksimovic
(2005). They select, on a priori grounds, the financing, legal and corruption obstacles,
and examine, one at a time, the relation between these obstacles and growth rates of firms
of different sizes. By contrast, we begin by examining a large set of business environment
obstacles and focus on empirically identifying the subset of binding ones.
Our paper also contributes to the policy debate generated by Hausman, Rodrik,
and Velasco (2004) who describe a theoretical framework for analyzing reform priorities
that focuses on identifying and targeting the most binding constraints in a particular
economic setting.
The paper is organized as follows. The next section presents the motivation for
the paper and describes the methodology. Section III discusses the data and summary
statistics. Section IV presents our main results. Section V presents the conclusions and
policy implications.
II. Motivation and Methodology
Numerous studies argue that differences in business environment can explain
much of the variation across countries in firms’ financial policies and performance.
While much of the early work relied on country-level indicators and firms’ financial
reports, more recent work has relied on surveys of firms which provide data on a wide
range of potential obstacles to growth.
The obstacles that have been investigated can be broadly divided into Financing
(such as problems with access to and cost of financing), Judicial Efficiency (security and
protection of property rights, effective functioning of the judiciary), Taxes and
7
Regulation (taxes, regulations, anticompetitive practices), Infrastructure (quality and
availability of roads, electricity, water, telephone, postal service etc.), Corruption
(corruption of government officials, crime), and general Macroeconomic Environment
that makes financing costly (political instability, exchange rate instability and inflation).6
A large literature in law and finance has identified the importance of Financing
and Judicial Efficiency for firm growth. Many studies, starting with LaPorta, Lopez-de-
Silanes, Shleifer, and Vishny (1998), argue that differences in legal and financial systems
can explain much of the variation across countries in firms’ financial policies and
performance.7
There are, however, few systematic multi-country studies of how other general
business environment obstacles faced by firms affect their growth. Many of the extant
studies have a regional or single-country focus and concentrate on a single obstacle. For
instance, recent studies have focused on the importance of Infrastructure and
Regulations. Klapper, Laeven, and Rajan (2005) use firm level data from Western and
Eastern Europe and show that anti-competitive regulations such as entry barriers lead to
slower growth in established firms. Dollar, Hallward-Driemeier, and Mengistae (2004)
use firm-level survey data and show that the cost of different bottlenecks such as days to
clear goods through customs, days to get a telephone line, and sales lost due to power
6 Several other papers using the same survey have analyzed specific financing obstacles. Beck, Demirguc- Kunt, and Maksimovic (2005) focus on the role of country-level financial and institutional development in overcoming the constraining effect of financing obstacles and Beck, Demirguc-Kunt, Laeven, and Levine (2005) analyze firm characteristics that explain differences in reported financing obstacles. None of these papers tries to prioritize the importance of specific obstacles for growth. 7 Related to judicial efficiency is the absence of secure property rights. Johnson, McMillan, and Woodruff (2000) analyze employment and sales growth from 1994 to 1996 in five countries and find that insecure property rights are more inhibiting to private sector growth than the lack of bank finance. In a study centered on SMEs in Russia and Bulgaria, Pissarides, Singer, and Svejnar (2003) find the opposite result that while constraints on external financing limit the ability of firms’ to expand production, insecurity of property rights is not a major constraint. Using Chinese data Cull and Xu (2005) also show that protection of property rights as well as access to finance plays an important role in explaining firm reinvestment rates.
8
outages affect firm performance in Bangladesh, China, India and Pakistan. Using similar
data for African countries, Eifert, Gelb, and Ramachandran (2005) show that business
environment variables also have an impact on firm productivity. Sleuwaegen and
Goedhuys (2002) use firm-level data from the Ivory Coast and find that inadequate
physical and financial infrastructure impairs the growth of small firms.
Several other papers focus on Corruption and compare it to Taxes. One of the
earliest papers in this area by Shleifer and Vishny (1993) argues that corruption may be
more damaging than taxation because of the uncertainty and secrecy that accompanies
bribery payments. Using a unique dataset of Ugandan firms, Fisman and Svensson (2004)
find that corruption, specifically bribe payments, retards firm growth more than taxation.
Gaviria (2002) also finds that corruption and crime substantially reduce firm
competitiveness amongst Latin American firms.8 While these studies are important
contributions in understanding the effects of business environment in different countries,
they each examine a narrow aspect of the business environment and hence have limited
policy prescriptions.
Firms in the WBES survey also report on the quality of macroeconomic
governance, where we define macroeconomic governance to be the extent to which
Political Instability, Exchange Rate instability and Inflation impede business. While the
effects of inflation on investment and firm growth have been extensively studied in the
finance literature and now controlled for in most firm growth regressions, there is little
micro evidence on the impacts of political and exchange rate instability on firm growth. It
is conceivable that political instability and exchange rate volatility have a more indirect
8 There are several papers in the macro-literature that study the impact of the various business environment obstacles at the country level. For instance Mauro (1995), Wei (1997) and Friedman et al. (2000) look at the effect of corruption, crime and taxation on GDP growth, size of the unofficial economy and investment.
9
impact on sales growth by affecting the type of financing available to firms. For instance,
Desai, Foley and Forbes (2004) argue that exchange rate depreciations increase the
leverage of firms that have borrowed foreign currency denominated debt, constraining
their ability to obtain new equity or to adjust their capital structure. 9
Identifying Binding Constraints
Given the large number of potential obstacles to growth that have been identified
in surveys, we face a number of difficulties in identifying the obstacles that are truly
constraining. First, a potential problem with using survey data is that enterprise
managers may identify several operational issues while not all of them may be
constraining. Therefore, as in Beck, Demirguc-Kunt and Maksimovic (2005), we
examine the extent to which reported obstacles affect growth rates of firms. An obstacle
is only considered to be a “constraint” or a “binding constraint” if it has a significant
impact on firm growth. Significant impact requires that the coefficient of the obstacle in
the firm growth regression is significant and the value of the obstacle is greater than one,
indicating that the enterprise managers identified the factor as an obstacle.10
Second, to the extent that the characteristics of a firm’s business environment are
correlated, it is likely that many perceived business environment characteristics will be
correlated with realized firm growth. It is important to sort these into obstacles that
9 Alesina et. al. (1996) and Alesina and Perotti (1996) find that political instability has a strong negative association with growth and income distribution. However these papers use cross-country analyses and have little information on the effect of political instability on individual firms. 10 In a cross-country setting or even at the individual country level, the significance of the coefficient is actually sufficient to determine whether an obstacle is binding or not since the value of all obstacles exceed one. However, in determining the relative impact, it is important to take into account the level of the obstacles.
10
directly affect growth and obstacles that may be correlated with firm growth but affect it
only indirectly.
Since there is no theoretical basis for classifying the obstacles, we must proceed
empirically. However, if some of the obstacles share a common unmeasured cause with
firm growth, then the estimates of these obstacles as well as other obstacles will be biased
and inconsistent. This could cause obstacles having no influence on growth whatsoever,
not even a common cause with growth, to have significant regression coefficients leading
to an incorrect estimation of the binding constraints.
As a robustness test of our multiple regression analysis, we use the Directed
Acyclic Graph (DAG) methodology. The DAG algorithm begins with a set of potentially
related variables and uses the conditional correlations between them to rule out possible
causal relations among these variables. The final output of the algorithm is a listing of
potential causal relations between the variables that have not been ruled out and shows
(a) variables that have direct effects on the dependent variable or other variables, (b)
variables that only have indirect effects on the dependent variable through other
variables, and (c) variables that lack a consistent statistical relation with the other
variables.
The DAG algorithm imposes a stricter criterion than regression analysis to
identify the variables with direct effects. In OLS regression the variables that are
identified as significantly predicting dependent variable Y are the ones that have
significant partial correlations conditional on the full set of regressor matrix (X’X). By
contrast, in the algorithm used to construct the pattern of directed acyclic graphs, a
variable is identified as having a direct effect on dependent variable Y only if it has a
11
significant partial correlation conditional on the full set of regressors and all subsets of
the regressor matrix (X’X). Thus, if DAG identifies a particular obstacle as having a
direct effect on firm growth, that obstacle would also have a significant coefficient in all
OLS regressions regardless of which subset of other obstacles are entered as control
variables in the regression equation.
The criteria for causality in DAG are derived from the application of Bayes rule
and reasonable assumptions on the probability distributions of the variables, most
importantly, the Causal Markov Condition. The Causal Markov Condition amounts to
assuming that every variable X is independent of all variables that are not its direct
causes.11 It also implies that if two variables, X and Y, are related only as effects of a
common cause Z, then X and Y are probabilistically independent conditional on Z.
Appendix A2 includes a detailed description of the DAG methodology and how it
compares to regression analysis. Ayyagari, Demirguc-Kunt and Maksimovic (2005b)
further illustrate the use of this methodology.
DAG is useful as a data simplification device and identifying indirect effects
among different obstacles which the regression analysis does not show. But we use
regression analysis to do further robustness tests, such as testing for possible endogeneity
bias via instrumental variable methods. We also perform other robustness tests,
controlling for additional variables at the firm and country level, growth opportunities,
influential observations and potential perception biases in survey responses using
regression analysis.
11A common example used to illustrate this is that applying a flame to a piece of cotton will cause it to burn, irrespective of whether the flame (direct cause of fire) came from a lighter or a match or some other spark (indirect causes).
12
The obstacles a firm faces depend on the institutions in each country, but are not
likely to be the same for each firm in each country.12 Thus, our unit of analysis is the
firm. As described below, the regressions have country-level random effects.
III. Data and Summary Statistics
The main purpose of the WBES survey is to identify obstacles to firm
performance and growth around the world. Thus, the survey contains a large number of
questions on the nature and severity of different obstacles. Specifically, firms are asked
to rate the extent to which Finance, Corruption, Infrastructure, Taxes and Regulations,
Judicial Efficiency, Crime, Anti-Competitive Practices, Political Instability, and macro
issues such as Inflation and Exchange Rate constitute obstacles to their growth.13
In addition to the detail on the obstacles, one of the greatest values of this survey
is its wide coverage of smaller firms. The survey is size-stratified, with 40 percent of
observations on small firms, another 40 percent on medium-sized firms, and the
remainder from large firms. Firm size is defined based on employment. Small firms are
those that employ five to 50 employees. Medium firms are those that employ 51 to 500
employees. And large firms are those that employ more than 500 employees. General
information on firms is limited, but in addition to employment, the survey also gives
information on sales growth, industry and ownership.
12 For example, Johnson and Mitton (2003) find that the effect of the 1997 Asian financial crisis on firms in Malaysia depended on whether or not the firms were connected to specific politicians who gained power over the period of the crisis. Similarly, the effect of crime might depend on factors such as the firm’s location, the ethnic group to which the business owner belongs or his political affiliations. Beck, Demirguc-Kunt, and Maksimovic (2005) find that the correlation between firm growth and financing, legal and corruption obstacles varies by firm size. 13 The survey provides two obstacles on crime, one capturing street crime and the other organized crime. Since the correlation between the two obstacles is higher than 70 percent, we use only street crime in our analysis, which is also more strongly correlated with firm growth among the two.
13
Insert Table I here
In Table I we summarize relevant facts about the level of economic development,
firm growth and the different obstacles that firms report, averaged over all firms in each
country. We provide details on our sources in Appendix A1. The countries in the sample
show considerable variation in per capita income. They range from Ethiopia, with an
average GDP per capita of $109 to developed countries like U.S. and Germany, with per
capita incomes of around $30,000. Firm Growth is the sales growth rate for individual
firms averaged over all sampled firms in each country. Firm growth also shows a wide
dispersion, from negative rates of 20 percent for Armenia and Azerbaijan to 64 percent
for Malawi and Uzbekistan.14
Table I also reports firm level obstacles. The WBES asked enterprise managers to
rate the extent to which each factor presented an obstacle to the operation and growth of
their business. A rating of one denotes no obstacle; two, a minor obstacle; three, a
moderate obstacle; and four, a major obstacle.
Looking at average obstacles across countries (Table II, panel A) we see that
firms report Taxes and Regulations to be the greatest obstacle. Inflation, Political
Instability and Financing obstacles are also reported to be highly constraining. By
contrast, factors associated with Judicial Efficiency and Infrastructure are ranked as the
lowest obstacles faced by entrepreneurs.
Insert Table II here
Firms in higher income countries tend to face lower obstacles in all areas. Table I
also highlights regional differences: When it comes to Corruption and Infrastructure,
14 Note that some of the countries with very high firm growth rates are also countries with high inflation rates. For instance inflation rate in Malawi was over 80% in 1995 and between 26-30% during the period of the WBES Survey.
14
African firms report the highest obstacles; Latin American Crime and Judicial Efficiency
obstacles are the highest in the world; and Financing obstacles in Asian countries are
lowest in the developing world. Finally, smaller firms face higher obstacles than larger
firms in all areas, except in those related to Judicial Efficiency and Infrastructure, where
the ranking is reversed. 15
Table II, panel B contains the correlation matrix of the variables. As our cursory
examination of Table I suggests, all obstacles are significantly lower in more developed
countries. Larger firms also face significantly lower obstacles in most cases except
Crime, Corruption and Infrastructure.
The correlations among the obstacles reported by firms are significant but fairly
low, with few above 0.5. As expected, the two macro obstacles, Inflation and Exchange
Rate, are highly correlated at 0.58. The correlation of Corruption with Crime and
Judicial Efficiency is also relatively high at 0.55, indicating that in environments where
corruption and crime are wide-spread, judicial efficiency is adversely affected. It is also
interesting that the correlation between the Financing obstacle and all other obstacles is
among the lowest, indicating that the Financing obstacle may capture different effects
than those captured by other reported obstacles. Table II also shows that all obstacles
are negatively and significantly correlated with firm growth. We explore these relations
further in the next section.
IV. Firm Growth and Reported Obstacles
A. Obtaining the Binding Constraints
15 Note that the Judicial Efficiency obstacle captures those obstacles related to working of the court system in resolving business disputes, whereas legal issues related to collateral, credit information, paperwork etc. are captured within the Financing obstacle as further discussed in section IV.H below.
15
In Table III we regress firm growth rates on the different obstacles they report.
All regressions are estimated with firm-level data using country-level random effects. To
maximize the number of observations and country coverage, we limit the control
variables to firm size and GDP per capita.16 Specifically, the regression equations we
estimate take the form:
Firm Growth = α + β1 Obstacle + β2 GDP/capita + β3 Firm Size + ε (1)
To test the hypothesis that a reported obstacle is a binding constraint, that is, it has
a significant impact on firm growth, we test whether β1 is significantly different from
zero. Significant impact also requires that the obstacle has a value higher than one, which
is true for all obstacles. We also report an estimate of the economic impact of the
obstacle at the sample mean by multiplying its coefficient by the sample mean of the
obstacle.17 This allows us to compare the relative impact of different obstacles on
growth.
Insert Table III here
Table III results show that when we analyze individual obstacles separately, most
are significantly related to firm growth. The only exceptions are Exchange Rate, Anti-
Competitive behavior, and Infrastructure obstacles which are not significantly related to
firm growth. These regressions explain up to 11 percent of the variation in firm growth
16 The random effects control for country-level differences in economic growth and other policy variables. In unreported regressions we also checked the robustness of our results to including additional control variables in the regression. Specifically, controlling for inflation and GDP per capita growth at the country level, and adding variables at the firm level capturing a firm’s industry, number of competitors, organizational structure, and whether it is government or foreign owned, an exporter, or a subsidy receiver reduces country coverage from 80 to 56, but does not affect the results significantly for individual obstacles. Of the three binding constraints identified above, only Political Instability obstacle loses significance. 17 In estimating the economic impact of the obstacles we make the assumption that a one unit change in the obstacle has a similar value regardless of the initial value of the obstacle.
16
across countries. The economic impact of four obstacles- Finance, Crime, Tax and
Regulations, and Political Instability- are the largest, ranging from 7.9 percent for
Political Instability to 9.6 percent for the Finance obstacles. This result implies that a one
unit increase in one of these obstacles leads to 8 to 10 percent reduction in firm growth,
which is substantial given that the mean firm growth in the sample is 15 percent. In
column 11, we include all obstacles in the regression equation. In this specification, only
Financing, Political Instability and Crime obstacles have a significant constraining effect
on growth. Dropping the remaining obstacles from the regression as in specification 12
does not alter the significance levels.18 The economic impact of the Financing obstacle is
the highest followed by that of Crime and then Political Instability. None of these
differences are statistically significant, however.
It is also possible to do such impact evaluation at the regional level, at the country
level or even at the firm level, instead of the sample mean we have used above. Looking
at the mean obstacles for individual countries reported in Table I, it is clear that the
binding obstacles are not equally important in every country. For example, in Singapore,
where the mean value of the binding obstacles are all closer to one, the economic impact
of the obstacles is much smaller compared to their impact in a country such as Nigeria,
where the mean value of all three obstacles are over three, indicating severe constraints.
Thus, it is possible to use these cross-country results to do growth diagnostics at the
country level as discussed in Hausmann et al. (2004). Going further down, there may
18 In unreported regressions we first regressed the three binding constraints individually on all other obstacles. In the second stage we replaced the Financing, Political Instability and Crime obstacles by their residuals from the first stage regressions, that is, the components of the three obstacles that are not explained by any other obstacle. Consistent with specification 11, all three residuals have negative and significant coefficients, indicating that they constrain firm growth independently of the other obstacles. Replicating this process only for the Financing variable- since it is the most likely obstacle to act as a sufficient statistic for the firm’s business environment- also results in a negative and significant coefficient of the finance residual in the second stage.
17
also be some firms in Nigeria for which the constraints are not binding (depending on the
value of the obstacles they report) and others in Singapore for which they are. In fact,
average values of obstacles by firm size suggests that the three obstacles will always be
more binding for smaller firms compared to larger firms.
Overall, these results suggest that the three obstacles- Financing, Crime and
Political Instability – are the only true constraints, in that they are the only obstacles that
affect firm growth directly at the margin. The other obstacles may also affect firm
growth through their impact on each other and on the three binding constraints; however
they have no direct effect on firm growth.
B. Directed Acyclic Graphs Methodology
In this section we use the DAG methodology implemented by the software
program TETRAD III (Scheines, et al 1994) to check the robustness of our regression
findings. As described in Ayyagari, Demirguc-Kunt and Maksimovic (2005b), and in
Appendix A2, DAG is useful in simplifying the set of independent variables and
illustrating the causal structure among them.
In keeping with common practice, we impose the following assumptions that are
regularly used in the regression setting- the business environment obstacles cause firm
growth, not the other way around, and the model contains all common causes of the
variables in the model. However, this being a partial equilibrium model of the causation
of growth, it is to be expected that some of the obstacles may be jointly determined by
macroeconomic factors, all of which may not be in the model. We perform robustness
18
tests on our assumptions later in the paper using instrumental variable estimation and by
including additional control variables.
Figure 1 illustrates the application of this algorithm to our full sample. The input
to the algorithm is the correlation matrix from the sample of 4197 firms consisting of
correlations between firm growth and the ten business environment obstacles19
Insert Figure 1 here
Figure 1 shows that the only business environment obstacles that have a direct
effect on firm growth are Financing, Crime and Political Instability. Financing in turn is
directly affected by the Taxes and Regulation obstacle which includes factors such as
taxes and tax administration, as well as regulations in the areas of business licensing,
labor, foreign exchange, environment, fire and safety20. Crime is directly affected by the
Corruption and Judicial Efficiency obstacles and Political Instability is affected by
Inflation21. Political Instability, Crime, Taxes and Regulation and Exchange rates all
influence each other although the direction of causality is unknown22.
19 In addition, we select the significance level for tests of conditional independence performed by TETRAD. Because the algorithm performs a complex sequence of statistical tests, each at the given significance level, the significance level is not an indication of error probabilities of the entire procedure. Spirtes, Glymour, and Sheines (1993) after exploring several versions of the algorithm on simulated data conclude that “in order for the method to converge to correct decisions with probability 1, the significance level used in making decisions should decrease as the sample size increases, and the use of higher significance levels may improve performance at small sample sizes.” For the results in this paper obtained from samples ranging from 2659-4197 observations, we use a significance level of 0.10. 20 Figure 1 also shows that Taxes and Regulation are in turn determined by Judicial Efficiency, which is consistent with Desai, Dyck, and Zingales (2004), who find a strong interaction between a country’s legal system, its corporate governance mechanisms and tax revenue. They find that across a panel of countries, the level of corporate governance in the country determines the amount of tax revenue. 21 We find the DAG analysis and the set of causal structures determined by the algorithm as being useful for an objective selection of variables, with the heuristic interpretation that that if DAG analysis shows that obstacle X causes obstacle Y, then firms’ reports of X as an obstacle is also likely to affect the probability that they report of Y as an obstacle). For details refer to formal definitions. 22 We also reproduced this figure after removing country averages from each firm response which roughly corresponds to including fixed effects in a regression setting. We still get the same split among first order and second order effects, i.e., only Financing, Political Instability and Crime have a direct effect on firm
19
The output also shows that relations between the obstacles themselves is quite
complex and there are multiple causal relations in the DAG between the various business
environment obstacles. Since the main focus of this paper is to determine the direct
causes of growth, we do not dwell on the interactions and common causes of the different
variables and leave it for future work. Most importantly, the DAG analysis also identifies
only Financing, Crime and Political Stability as having direct effects on firm growth, as
suggested by specification 12 of Table 3. As discussed in Section II, the analysis
identifies direct effects after conditioning on all subsets of the other variables. This
suggests that in regression analysis, Financing, Crime and Political Instability will always
have significant coefficients irrespective of the subsets of other obstacles included in the
regression. Thus, these are binding constraints, and policies that relax these constraints
can be expected to directly increase firm growth.
C. Binding Constraints and Firm Size and Level of Development
Next we explore if these relationships are different for firms of different sizes and
at different levels of development.23 The first three columns of Table IV include
specifications that interact the three obstacles with firm size, given by the Logarithm of
sales. The interaction term with the Financing obstacle is positive and significant at one
percent, suggesting that larger firms are less financially constrained, confirming the
findings of Beck, Demirguc-Kunt and Maksimovic (2005). The interaction terms with
Political Instability and Crime are also positive, but only significant at ten percent. Thus,
growth. Including GDP per capita in the DAG analysis as an additional factor does not change the main results. Financing, Political Instability and Crime remain the only obstacles with direct effects on growth. 23 Note that our results remain the same if we do not control for size in the firm growth regressions in Table III.
20
although there is also some indication that large firms are also affected less by Crime and
Political Instability, this evidence is much weaker.
Insert Table IV here
We also interact the three obstacles with country income dummies, Upper Middle
Income, Lower Middle Income and Low Income as defined in the data Appendix. The
excluded category is High Income. The results indicate that all three obstacles tend to be
more constraining for middle income countries. However, the F-tests for the hypotheses
that all the entered interactions are jointly equal to zero, are rejected at the one percent
level of significance for Crime and Political Instability obstacles, but not for the
Financing obstacle. This suggests that firms in countries in all income groups are
similarly affected by the Financing obstacle.
D. Checking for Reverse Causality
So far we have identified Financing, Crime, and Political Instability as first order
constraints, significantly affecting firm growth. However, as already noted above, the
relations we observe may also be due to reverse causality if underperforming firms
systematically blame Financing, Crime and Political Instability instead of taking
responsibility for their poor performance. This is most likely to bias the Financing
obstacle results since it is easy to imagine that entrepreneurs might complain about
restricted access to external finance even in cases where access is restricted due to their
own deficiencies.
To assess the robustness of our results, we use instrumental variable (IV)
regressions to extract the exogenous component of the three obstacles. In selecting
21
instrumental variables for Financing, Crime and Political Instability, we use average
value of these obstacles for the three size groups in each country. While it is likely that
individual firms may blame the different obstacles for their poor performance, it is less
likely for all firms in a given size group to engage in such blame shifting. By
instrumenting the obstacles with the average obstacle for each size group in the country,
we are isolating the exogenous part of the possibly endogenous obstacle reports and using
that to predict growth. When we consider the obstacles at the country-size level of
aggregation, the causality is likely to run from the average obstacles to individual firms,
not vice versa.
Insert Table V here
Table V provides the results of the IV estimation. Instrumenting all three
obstacles individually, we see that their coefficients remain negative and significant. As
an additional robustness test, we also use the average value of each obstacle at the
country level as instruments. It is even less likely for the growth of individual firms to
affect obstacles at the country level. The results remain unchanged with this
specification. Overall, these results suggest that there are exogenous components of
Financing, Crime and Political Instability obstacles that predict firm growth and the
results we obtain are not due to reverse causality.
E. Growth Opportunities
The observed association between obstacles and firm growth might occur because
firms that face higher obstacles are also those that face limited growth opportunities. This
is a form of reverse causality, and does not explain our results since we find (in
22
subsection D above) that our binding obstacles affect firm growth when we control for
reverse causality using instrumental variables. We cannot directly control for growth
opportunities at the firm level because the WBES database does not contain appropriate
firm level control variables. However, following Fisman and Love (2004), we construct
two variables to proxy growth opportunities using average industry growth and firm level
dependence on external finance. First, we include average industry growth, which is the
growth rate averaged across all firms in each industry in each country. Second, we
include the proportion of investment financed externally by each firm as an indicator of
growth opportunities.
Insert Table VI here
In the regressions reported in Table VI, our specification is augmented by these
proxies for growth opportunities. Average industry growth is significant and positive in
all specifications as expected, and leaves the results unchanged. External financing of
investment does not enter the regressions significantly. When we use this proxy for
growth opportunities, only Political Instability loses significance in the last specification
where all obstacles are included.
F. Outlier Tests
We next investigate whether our results are driven by a few countries or firms. In
particular, we investigate two sets of countries: African and Transition economies.
Chandra et al. (2001) suggest that firms in African countries may exhibit different
responses than the other firms in the sample. A report by the United States General
Accounting Office, GAO-04-506 (2004) analyzes several firm level surveys on Africa,
23
including the WBES, and concludes that perceptions of corruption levels vary greatly for
African countries, proving a challenge for broad-based US Anticorruption Programs.
Ayyagari, Demirguc-Kunt and Maksimovic (2005) argue that Transition economies are
fundamentally different from other countries in their perceptions of protection of property
rights.
Insert Table VII here
In the first four columns of Table VII we run our preferred specification on
different samples eliminating Transition and African countries. We find that while
Financing and Crime are binding constraints as before, Political Instability loses
significance if we do not include these countries in the sample. These results suggest that
the type of Political Instability present in Transition and African economies is particularly
damaging to firm expansion.
We also noted that high inflation rates may be responsible for the very high firm
growth rates we observe in some countries, particularly in Uzbekistan, Estonia and
Bosnia-Herzegovina. 24 However, constructing real firm growth rates and replicating all
the analyses in this paper does not change the main results.
To check whether our results are driven by specific outlier firms, we re-run
specifications 1-4 of Table VII after eliminating all firms with very high growth rates.25
The fastest growing firms, reporting growth rates in excess of 100%, are typically from
the Transition and African countries and it is conceivable that these firms achieve these
high growth rates because of political connections and are not impacted by general
24 Uzbekistan, Estonia and Bosnia-Herzegovina appear to be outlier countries in that they have average firm growth rates above 60%, but their inflation rates are also in excess of 100%. 25 We eliminate 184 firms which have growth rates outside the range of +/- 100% (only 2 firms have growth rates <-100%).
24
business environment obstacles. Thus, the experience of these firms may differ from that
of the typical firm. We find that Financing remains the most binding constraint to firm
growth in our reduced sample, confirming that our results are not driven by the fastest
growing firms in the sample. The impact of Crime on firm growth is less robust to
eliminating high growth rate firms, however. It is also possible that young firms in the
sample are affected differently by business environment obstacles. Excluding all firms
younger than five years old, from the sample, leaves Financing and Crime results
unchanged, while Political Instability loses significance. This suggests that political
stability is quite important to ensure growth of younger firms. 26
In unreported regressions, we also investigated whether firm ownership drives our
results. The sample includes 203 firms with government ownership. Excluding these
firms does not change our results. The sample also includes 1340 firms with over 50
percent foreign ownership. Excluding these foreign firms from the analysis reduces the
significance of the Political Instability obstacle in some specifications. This suggests that
foreign owned firms are particularly sensitive to Political Instability. Including dummy
variables to control for government and foreign firms also leads to similar results in that
Political Instability loses significance.
G. Perception Biases
Survey responses are subject to several types of perception biases which may
affect our results. For example, some respondents may be more likely to exaggerate or
26 Financing is still the main binding constraint to growth when we use robust regression analysis to control for the presence of possible influential outliers. Robust regressions use iteratively re-weighted least squares to estimate regression coefficients and the standard errors by underweighting influential outliers. We do not report these results because robust regression does not country random effects.
25
underestimate all obstacles and respond accordingly. Kaufman and Wei (1999) discuss
this possibility and construct “kvetch” variables to control for such differences in
perceptions across respondents. Following their work, we construct two kvetch variables,
Kvetch1 and Kvetch2, which are deviations of each firm’s response from the mean
country response to two general survey questions. Kvetch1 uses the responses to the
question “How helpful do you find the central government today towards businesses like
yours?” and Kvetch2 is constructed using the responses to “How predictable are changes
in economic and financial policies?” Since higher values correspond to unfavorable
responses, positive deviations from the country mean indicate pessimism whereas
negative deviations indicate optimism. As reported in Table VIII, controlling for such
differences in perceptions leave the results unchanged.27
Insert Table VIII here
Throughout the analysis we use ordinal variables as the obstacles take the value
one to four and make the assumption that a one unit change in the variable has a similar
value regardless of the initial value of the obstacle. Finally, for the Financing variable we
relax that assumption and construct a dummy variable for each value of the obstacle,
entering the three that correspond to values 2-4 instead of the ordinal Financing variable
in our preferred specification. The results indicate that those firms that complain the
most are the ones that grow significantly slower.
27 Another type of perception bias can be due to the “Halo effect” which occurs when survey respondents respond more favorably to questions about richer countries, as explained in Glaeser, La Porta, Lopez de Silanes and Shleifer (2004). This type of perception bias is likely to be more problematic in the case of country level expert surveys rather than firm level surveys. However, even at the country level, Kaufmann, Kraay and Mastruzzi (2005) argue that for this to be a significant issue, the correlation between the perception error and income should be very high and the variance of the error should also be large relative to the variance of the indicator being measured.
26
H. Individual Financing Obstacles
Our results indicate that Financing is one of the most important obstacles that
directly constrain firm growth. We would like to get a better understanding of exactly
what type of obstacles related to financing are constraining firm growth. Fortunately, our
survey data also includes more detailed questions regarding the Financing obstacles.
To get at specific issues, the entrepreneurs were asked to rate the extent to which
the following financing factors represent an obstacle to their growth: (i) collateral
requirements of banks and financial institutions, (ii) bank paperwork and bureaucracy,
(iii) high interest rates, (iv) need for special connections with banks and financial
institutions, (v) banks lacking money to lend, (vi) access to foreign banks, (vii) access to
non-bank equity, (viii) access to export finance, (ix) access to financing for leasing
equipment, (x) inadequate credit and financial information on customers, and (xi) access
to long-term loans. The ratings are again on a scale of one to four, increasing in the
severity of obstacles.
Insert Table IX here
Table IX reports the regressions that parallel those in Table III, but this time
focusing on specific financing obstacles. In addition to the individual financing
obstacles, we also include a residual, the component of the general financing obstacle not
explained by the individual obstacles. The results indicate that not all financing obstacles
reported by firms are constraining. Only the coefficients of collateral, paperwork, high
interest rates, special connections, banks’ lack of money to lend, lease finance and the
residual are significant when entered individually. In terms of economic impact high
interest rates has the highest impact at 10.3 percent. Unlike the different obstacles we
27
examined above, specific financing obstacles are highly correlated with each other. When
we include all obstacles that are significant when entered individually in a regression, we
see that only high interest rates coefficient is significant. If we also include the residual,
only the residual remains significant. The residual is likely to summarize how different
firms are affected differently by the structure and ownership of the financial system, the
level of competition and other factors which are not fully captured by the specific
financial obstacles28.
We address the issue of correlations among obstacles further with DAG analysis.
Figure 2 illustrates the results of DAG analysis for specific financing obstacles. The
graph suggests that high interest rates is the only financial obstacle directly constraining
firm growth. The residual financing variable also constrains growth directly, but it is not
correlated with other obstacles by construction. Again, the results of DAG analysis are
consistent with regressions since the graph suggests that only these two variables have a
direct impact on firm growth. It may be noted that while we restrict the direction of
causation to be from the various financing obstacles to growth, we impose no ordering
amongst the individual financing obstacles themselves.
The finding that high interest rates constrains firm growth is not surprising since
the high interest rate obstacle captures the cost of financing and is itself an endogenous
variable that depends on the ability of the financial system to satisfy the demand for
capital. It can be expected to constrain all firms in all countries. Collectively, specific
28 The residual remains significant if in addition to the residual and the significant individual financing obstacles (collateral, paperwork, high interest rates, special connections, lack money to lend and lease finance), we were to include all the general obstacles (Political Instability, Crime, Inflation, Exchange rates, Taxes & Regulation, Anti-Competitive Behavior, Judicial Efficiency, Corruption and Infrastructure).
28
financing obstacles still do not capture everything measured by the general financing
obstacle, as illustrated by the effect of the residual.
Insert Figure 2 here
The DAG analysis also suggests that perceptions of high collateral requirements
and inability to access long term loans influence the perceptions of high interest rates.
High interest rates also influence perceptions of greater paperwork requirements, as well
as the need for special connections in banking29.
Insert Table X here
In Table X, we run regressions of high interest rate obstacle on individual
financing obstacles. Specific financing obstacles are all individually correlated with high
interest rates. When we consider all financing obstacles together, only collateral,
paperwork, special connections, lack of money to lend, and access to long term loans are
correlated with high interest rates, as in the DAG analysis.
V. Conclusions and Policy Implications We use firm level survey data for 80 countries to investigate which features of the
business environment are the most constraining for firm growth. Specifically, we
examine factors such as taxes and regulations, judicial efficiency, infrastructure
weaknesses, and financing issues that have attracted significant attention in the literature.
Although firms report many obstacles to their growth, not all of them are equally
constraining – in that they either affect firm growth only indirectly through their
influence on other factors, or not at all. Using regressions as well as Directed Acyclic
29 In unreported results, the TETRAD output indicates that high interest rates and lack of money to lend may be jointly determined by common causes, which is to be expected since these obstacles are likely to be determined by macroeconomic policies.
29
Graph (DAG) methodology, we find that only Finance, Crime and Political Instability are
binding constraints, which directly affect the growth rate of firms. Thus, while other
obstacles we study in this paper also affect firm growth, through their impact on each
other and on direct obstacles, maintaining political stability, keeping crime under control,
and undertaking financial sector reforms to relax financing constraints are likely to be the
most effective routes to promote firm growth. We also find that the Financing obstacle’s
impact on growth is robust to varying samples of countries while the Political Instability
and Crime results are less robust to the exclusion of Transition and African countries
where they might be the most problematic for business growth. We subject our results to
a battery of robustness tests; controlling for reverse causality, growth opportunities and
potential perception biases in survey responses do not alter our main results.
Further investigation of the Financing obstacles reveals the importance of high
interest rates in constraining firm growth. This result highlights the importance of
macroeconomic policies in influencing growth at the firm level as indicated by the
correlation between high interest rates and banks’ lack of money to lend variables.
Furthermore, high interest rates is also correlated with high collateral and paperwork
requirements, the need for special connections with banks and unavailability of long term
loans. These results suggest that bureaucracy and corruption in banking, greater
collateral requirements and lack of long term loans are common in high interest rate
environments. We leave further investigation of country and firm level determinants of
financing obstacles to future work.
30
References:
Alesina, A., Ozler, S., Roubini, N., Swagel, P., 1996. Political instability and economic growth. Journal of Economic Growth 1 (2), 188–211. Alesina, A., Perotti, R., 1996. Income Distribution, Political Instability, and Investment. European Economic Review 40 (6), 1203-1228. Allen, F., Qian, J., Qian, M., 2005. Law, finance, and economic growth in China, Journal of Financial Economics, forthcoming. Awokuse, T. O., Besler, D., 2003. Vector Autoregressions, Policy Analysis, and Directed Acyclic Graphs: An Application to the US Economy. Journal of Applied Economics 6(1), 1-24. Ayyagari, M., Demirgüç-Kunt, A., Maksimovic, V., 2005. How well do Institutional theories Explain Firms’ Perceptions of Property Rights? Mimeo Ayyagari, M., Demirgüç-Kunt, A., Maksimovic, V., 2005b. What determines protection of property rights? Analysis of direct and indirect effects using DAG methodology. Mimeo Batra, G., Kaufmann, D., Stone, A., 2003. The Firms Speak: What the World Business Environment Survey Tells Us about Constraints on Private Sector Development. World Bank mimeo. Beck, T., Demirgüç-Kunt, A., Maksimovic, V., 2005. Financial and Legal Constraints to Firm Growth: Does Firm Size Matter. Journal of Finance 60(1), 137-177. Beck, T., Demirgüç-Kunt, A., Laeven, L. Levine, R., 2005. Finance, Firm Size, and Growth. World Bank Mimeo. Beck, T., Levine, R., Loayza, N., 2000. Finance and the Sources of Growth. Journal of Financial Economics 58, 261-300. Bessler, D., Yang, J., 2003. The Stucture of Interdependence in International Stock Markets. Journal of International Money and Finance, 22. Carlin, W., Fries, S., Schaffer, M., Seabright, P., 2001. Competition and enterprise performance in transition economies: evidence from a cross-country survey. CEPR Discussion Paper No. 2840 Chandra, V., Moorty, L., Nganou, J., Rajaratnam, B., Schaefer, K., 2001. Constraints to Growth and Employment in South Africa. Discussion Paper, World Bank Southern Africa Department.
31
Cull, R., Xu, C. L., 2005. Institutions, ownership and finance: The determinants of profit reinvestment and Chinese firms. Forthcoming Journal of Financial Economics. Demirgüç-Kunt, A., Maksimovic, V., 1998. Law, finance, and firm growth, Journal of Finance 53, 2107-2137. Desai, M., Foley, C.F., Forbes, K. J., 2004. Financial Constraints and Growth: Multinational and Local Firm Responses to Currency Depreciations, SSRN Working Paper Series Desai, M., Dyck, A., Zingales, L., 2004. Theft and Taxes. SSRN Working Paper Series. Dollar, D., Hallward-Driemeier, M., Mengistae, T., 2004. Investment Climate and Firm Performance in Developing Countries, World Bank mimeo. Eifert, B., Gelb, A., Ramachandran, V., 2005. Business Environment and Comparative Advantage in Africa: Evidence from the Investment Climate Data, World Bank mimeo. Fisman, R., Love, I., 2004. Financial Development and Intersectoral Allocation: A New Approach. Journal of Finance 59(6), 2785-2807. Fisman, R., Svensson, J., 2004. Are Corruption and Taxation Really Harmful to Growth? Firm-Level Evidence. World Bank Policy Research Working Paper # 2485 Fleisig, H., 1996. Secured Transactions: The Power of Collateral. Finance and Development, 44-46. Friedman, E., Johnson, S., Kaufmann, D., Zoido-Lobatón, P., 2000. Dodging the grabbing hand: The determinants of unofficial activity in 69 countries. Journal of Public Economics 76,459-93. GAO-04-506, 2004. U.S. Anticorruption Programs in Sub-Saharan Africa Will Require Time and Commitment, United States General Accounting Office (GAO) Report to the Subcommittee on African Affairs, Committee on Foreign Relations, U.S. Senate. Gaviria, A., 2002. Assessing the Effects of Corruption and Crime on Firm Performance : Evidence from Latin America. Emerging Markets Review 3(3). Glaeser, E; R. La Porta, F. Lopez de Silanes and A. Shleifer, 2004, Do institutions cause growth? Journal of Economic Growth 9(3), 271-303. Haigh, M. S., Bessler, D. A., 2004. Causality and Price Discovery: An Application of Directed Acyclic Graphs, Journal of Business, 77 (4), 1099-1021. Hausmann, R., Rodrik, D., Velasco, A., 2004. Growth Diagnostics. Harvard University working paper.
32
Johnson, S., Mitton, T., 2003. Cronyism and capital controls: evidence from Malaysia, Journal of Financial Economics 67, 351-382. Johnson, S., McMillan, J., Woodruff, C., 2000. Entrepreneurs and the Ordering of Institutional Reform,” Economics of Transition, Vol. 8(1), 1–36. Kaufmann, D. and S. Wei, 1999, Does grease money speed up the wheels of commerce? World Bank Policy Working Paper No 2254, Washington DC. Kaufmann, D; A. Kraay and M. Mastruzzi, 2005, Governance Matters IV: Governance Indicators for 1996-2004, The World Bank working paper. King, R.G., Levine, R.,1993, Finance and Growth: Schumpeter Might Be Right, Quarterly Journal of Economics, 108, 717-38. Klapper, L., Laeven, L, Rajan, R., 2005. Barriers to Entrepreneurship. World Bank Mimeo. Knill, A., Maksimovic, V., 2005. How do institutions affect financing across countries: Analyzing the causal structure. University of Maryland Mimeo. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1997. Legal determinants of external finance. Journal of Finance 52, 1131-1150. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1998. Law and finance. Journal of Political Economy 106, 1113-1155 Levine, R., 2005. Finance and Growth: Theory and Evidence forthcoming in Philippe Aghion and Steven Durlauf, eds. Handbook of Economic Growth. The Netherlands: Elsevier Science. Levine, R., Loayza, N., Beck, T. 2000. Financial Intermediation and Growth: Causality and Causes. Journal of Monetary Economics 46, 31-77. Mauro, P.,1995. Corruption and growth. Quarterly Journal of Economics 110, 681-712. Pearl, J.,1988. Probabilistic Reasoning in Intelligent Systems. Morgan and Kaufman, San Mateo, CA. Pearl, J., 2000. Causality: Models, Reasoning, and Inference. Cambridge University Press, Cambridge, UK. Pissarides, F., Singer, M., Svejnar, J., 2003. Objectives and Constraints of Entrepreneurs: Evidence from Small and Medium-Sized Enterprises in Russia and Bulgaria. Journal of Comparative Economics, Vol. 31(3), 503–531.
33
Rajan, R., Zingales, L., 1998. Financial dependence and growth, American Economic Review 88, 559-587. Scheines, R., Spirtes, P., Glymour, C., Meek, C., 1994. TETRAD II: Users Manual, Lawrence Erlbaum Associates, Hillsdale, NJ. Shleifer, A., Vishny, R.W., 1993. Corruption. Quarterly Journal of Economics, 108: 599-617. Sleuwaegen, L.,Goedhuys, M., 2002. ‘Growth of firms in developing countries, evidence from Côte d’Ivoire’. Journal of Development Economics. 68 117-135. Spirtes, P., C. Glymour, Scheines, R., 2000. Causation, Prediction and Search 2nd Edition. MIT Press, Cambridge, MA Wei, S., 1997. How taxing is corruption on international investors. NBER Working Paper 6030.
34
Tax
es a
nd R
egul
atio
n
Ant
icom
peti
tive
B
ehav
ior
Judi
cial
Eff
icie
ncy
Infl
atio
n
Exc
hang
e R
ates
Infr
astr
uctu
re
Cor
rupt
ion
Stre
et C
rim
e
Pol
itic
al I
nsta
bilit
y
Fin
anci
ng
Sale
s G
row
th
Var
iabl
es w
ith C
omm
on L
aten
t Cau
ses
T
ax R
egul
atio
n <
> A
nti-
Com
peti
tive
Beh
avio
r;
Tax
Reg
ulat
ion
<>
Inf
rast
ruct
ure;
T
ax R
egul
atio
n <
> I
nfla
tion
; S
tree
t Cri
me
<>
Inf
lati
on;
Str
eet C
rim
e <
> I
nfra
stru
ctur
e
Fig
ure
1: I
mpa
ct o
f G
ener
al O
bsta
cles
on
Fir
m G
row
th
35
Hig
h In
tere
st r
ates
Col
late
ral
Spec
ial C
onne
ctio
ns
Pap
erw
ork
Sale
s G
row
th
Lac
k M
oney
Acc
ess
to fo
reig
n b
anks
Acc
ess
to n
on-b
ank
E
quit
y
Exp
ort F
inan
ce
Lea
se F
inan
ce
Cre
dit
Lon
g T
erm
Loa
ns
Var
iabl
es w
ith C
omm
on L
aten
t Cau
ses
H
igh
inte
rest
rat
es<
> L
ack
mon
ey to
lend
N
o C
onsi
sten
t Ori
enta
tion
Cre
dit #
Col
late
ral/P
aper
wor
k/Sp
ecia
l Con
nect
ions
/Lea
se f
inan
ce/L
ack
mon
ey to
lend
GC
F R
esid
ual
Fig
ure
2: I
mpa
ct o
f Sp
ecif
ic F
inan
cing
Obs
tacl
es o
n F
irm
Gro
wth
36
Tab
le I
: E
cono
mic
Ind
icat
ors
and
Gen
eral
Obs
tacl
es
The
var
iabl
es a
re d
escr
ibed
as
follo
ws:
GD
P pe
r ca
pita
is r
eal G
DP
per
capi
ta in
US
dolla
rs a
vera
ged
over
199
5-19
99. F
irm
gro
wth
is th
e pe
rcen
tage
cha
nge
in f
irm
sal
es o
ver
the
past
thre
e ye
ars
(199
6-99
). F
inan
cing
, Pol
itic
al I
nsta
bilit
y, I
nfla
tion
, Exc
hang
e R
ate,
Jud
icia
l Eff
icie
ncy,
Str
eet C
rim
e, C
orru
ptio
n, T
axes
and
Reg
ulat
ion,
Ant
i-co
mpe
titiv
e B
ehav
ior
and
Infr
astr
uctu
re a
re g
ener
al o
bsta
cles
as
indi
cate
d in
the
fir
m q
uest
ionn
aire
. T
hey
take
val
ues
1 to
4,
wit
h hi
gher
val
ues
indi
catin
g gr
eate
r ob
stac
les.
In
Pane
l A
, fi
rm v
aria
bles
are
ave
rage
d ov
er a
ll fi
rms
in e
ach
coun
try.
In
Pane
l B
, th
e co
untr
ies
are
clas
sifi
ed i
nto
Hig
h In
com
e, U
pper
Mid
dle
Inco
me,
Low
er M
iddl
e In
com
e an
d L
ow I
ncom
e gr
oup
coun
trie
s ac
cord
ing
to W
DI
and
firm
var
iabl
es a
re a
vera
ged
over
all
firm
s in
the
pa
rtic
ular
gro
up o
f co
untr
ies.
In
Pane
l C
, th
e co
untr
ies
are
clas
sifi
ed i
nto
five
geo
grap
hic
regi
ons
and
firm
var
iabl
es a
re a
vera
ged
over
all
firm
s in
the
par
ticul
ar g
eogr
aphi
c re
gion
. In
Pan
el D
, fi
rm
vari
able
s ar
e av
erag
ed a
cros
s th
ree
size
gro
ups-
Smal
l, M
ediu
m a
nd L
arge
Fir
ms.
Det
aile
d va
riab
le d
efin
ition
s an
d so
urce
s ar
e gi
ven
in th
e ap
pend
ix.
Pan
el A
: A
vera
ged
Acr
oss
Cou
ntri
es
G
ener
al O
bsta
cles
Nat
ion
GD
P p
er
capi
ta
Fir
m
Gro
wth
F
inan
cing
P
olit
ical
In
stab
ility
In
flat
ion
Exc
hang
e R
ate
Judi
cial
E
ffic
ienc
y St
reet
C
rim
e C
orru
ptio
n T
axes
and
R
egul
atio
n
Ant
i-co
mpe
titi
ve
Beh
avio
r In
fras
truc
ture
Alb
ania
80
6.78
0.
22
3.04
3.
48
2.75
2.
61
2.69
3.
42
3.34
3.
15
2.72
3
Arg
enti
na
8000
.15
0.08
3.
01
3.07
1.
77
1.73
2.
27
2.39
2.
58
3.34
2.
41
1.93
Arm
enia
84
4.11
-0
.2
2.45
2.
87
2.73
2.
69
1.5
1.85
1.
96
3.39
1.
9 1.
77
Aze
rbai
jan
407.
75
-0.2
3.
11
2.55
2.
9 2.
61
2.59
2.
39
3 3.
17
2.96
2.
43
Ban
glad
esh
338.
66
0.13
2.
6 3.
08
2.86
3.
09
2.38
3.
07
3.61
3.
03
2.4
Bel
arus
22
34.9
1 0.
1 3.
33
2.95
3.
63
3.16
1.
55
2.17
1.
88
3.34
1.
99
1.7
Bel
ize
2737
.7
0.12
2.
81
2.38
2.
04
1.73
1.
56
2.12
1.
96
2.77
1.
96
2.19
Bol
ivia
93
8.55
0.
04
3.03
3.
1 2.
58
2.46
2.
78
2.76
3.
56
3.15
2.
71
2.63
Bos
nia
& H
erze
govi
na
1177
.75
0.63
3.
09
3.19
1.
33
1.25
2.
54
1.86
2.
56
3.16
2.
58
2.65
Bot
swan
a 35
92.8
6 0.
32
2.24
1.
55
1.93
1.
33
1.
88
1.65
1.
89
2.
16
Bra
zil
4491
.67
0.03
2.
67
3.53
2.
8 2.
94
2.56
2.
83
2.53
3.
66
2.49
2.
18
Bul
gari
a 14
14.6
1 0.
15
3.16
3.
03
2.76
2.
37
2.26
2.
64
2.64
3.
1 2.
34
2.23
Cam
bodi
a 28
1.61
0.
07
2.04
2.
9 2.
61
2.32
2
3.29
2.23
2.
21
2.33
Cam
eroo
n 63
0.92
0.
12
3.14
2.
03
2.03
2.
28
2.
94
3.36
2.
7
3.44
Can
ada
2054
8.97
0.
17
2.1
2.18
2.
15
2.16
1.
47
1.32
1.
4 2.
59
1.62
1.
41
Chi
le
5002
.7
0.09
2.
36
2.58
2.
16
2.59
1.
97
2.4
1.86
2.
36
1.91
1.
86
Chi
na
676.
76
0.05
3.
36
2.1
2.23
1.
83
1.5
1.83
1.
94
2.03
2.
13
1.89
Col
ombi
a 23
81.1
9 0.
06
2.67
3.
49
3.01
3.
34
2.4
3.37
2.
87
3.17
2.
33
2.46
Cos
ta R
ica
3692
.47
0.25
2.
62
2.67
2.
93
2.75
2.
2 2.
89
2.52
2.
8 2.
44
2.63
Cot
e d'
Ivoi
re
762.
8 0.
05
2.78
2.
85
2.37
1.
97
3.
29
3.24
2.
49
2.
29
Cro
atia
38
45.2
7 0.
1 3.
26
3.11
2.
47
2.86
2.
74
2.09
2.
59
3.34
2.
04
1.94
Cze
ch R
epub
lic
5158
.04
0.1
3.18
2.
95
3 2.
46
2.18
2.
09
2.1
3.44
2.
16
2.5
Dom
inic
an R
epub
lic
1712
.31
0.21
2.
63
3.02
2.
85
2.88
2.
43
3.22
3
3.96
2.
75
2.63
37
G
ener
al O
bsta
cles
Nat
ion
GD
P p
er
capi
ta
Fir
m
Gro
wth
F
inan
cing
P
olit
ical
In
stab
ility
In
flat
ion
Exc
hang
e R
ate
Judi
cial
E
ffic
ienc
y St
reet
C
rim
e C
orru
ptio
n T
axes
and
R
egul
atio
n
Ant
i-co
mpe
titi
ve
Beh
avio
r In
fras
truc
ture
Ecu
ador
15
38.4
8 -0
.06
3.27
3.
6 3.
76
3.78
3.
04
3.49
3.
53
3.07
2.
55
2.67
Egy
pt, A
rab
Rep
. 11
07.5
4 0.
16
2.91
3.
14
2.68
2.
9
2.24
3.
14
3.43
3.23
El S
alva
dor
1705
.79
-0.0
2 2.
93
2.97
3.
16
2.55
2.
65
3.67
3.
06
2.93
2.
36
2.52
Est
onia
36
63.4
9 0.
63
2.47
2.
62
2.41
1.
89
1.72
2.
09
1.88
2.
67
1.85
1.
64
Eth
iopi
a 10
9.01
0.
26
3.02
2.
38
2.26
2.
47
1.
51
2.46
2.
33
3.
04
Fran
ce
2771
9.92
0.
2 2.
61
2.2
2.03
1.
82
1.79
1.
77
1.62
3.
13
2.02
1.
81
Geo
rgia
41
1.39
0.
14
3.29
2.
84
3.29
2.
94
1.86
2.
32
3.04
3.
22
2.18
2.
14
Ger
man
y 30
794.
03
0.11
2.
59
1.63
1.
87
1.64
2.
12
1.56
1.
88
3.17
2.
3 1.
71
Gha
na
393.
44
0.19
3.
1 2.
37
3.43
2.
58
2.
37
2.78
2.
83
2.
74
Gua
tem
ala
1503
.25
0.18
2.
99
3.16
3.
32
3.6
2.5
3.22
2.
7 2.
75
2.28
2.
52
Hai
ti 36
8.73
0
3.28
3.
18
2.92
2.
9 2.
35
3.81
3.
08
2.73
3.
1 3.
89
Hon
dura
s 70
7.52
0.
1 2.
97
2.53
3.
41
3.3
2.41
3.
23
2.9
2.83
2.
79
2.56
Hun
gary
47
05.6
5 0.
28
2.6
2.61
2.
59
1.6
1.32
1.
76
1.95
3.
01
2.14
1.
53
Indi
a 41
4.15
0.
15
2.59
2.
81
2.77
2.
42
2.02
1.
98
2.8
2.43
2.8
Indo
nesi
a 10
45.0
4 -0
.05
2.83
3.
14
3.21
3.
4 2.
26
2.69
2.
69
2.59
2.
96
2.37
Ital
y 19
645.
96
0.16
1.
97
2.97
2.
23
1.83
2.
22
2.22
1.
76
3.25
2.
19
2.24
Kaz
akhs
tan
1315
.1
0.1
3.29
2.
88
3.62
3.
48
2.08
2.
6 2.
7 3.
37
2.55
2.
1
Ken
ya
339.
17
0.03
2.
76
3.03
2.
8 1.
75
3.
27
3.56
2.
53
3.
64
Kyr
gyz
Rep
ublic
80
0.34
0
3.47
3.
23
3.78
3.
48
2.13
3.
26
3.19
3.
59
3 1.
98
Lit
huan
ia
1907
.93
0.08
3.
03
2.27
2.
3 1.
91
2.25
2.
52
2.44
3.
26
2.31
1.
82
Mad
agas
car
237.
95
0.16
3.
08
2.67
3.
32
2.3
2.
79
3.44
2.
75
3.
03
Mal
awi
153.
79
0.64
2.
81
2.2
3.56
2.
54
3.
08
2.65
2.
37
3.
76
Mal
aysi
a 45
36.2
3 0.
01
2.57
2.
14
2.44
1.
94
1.63
1.
78
2 2.
03
1.91
1.
92
Mex
ico
3394
.75
0.24
3.
24
3.27
3.
48
3.13
2.
77
3.37
3.
31
3.21
2.
75
2.23
Mol
dova
66
7.74
-0
.15
3.42
3.
6 3.
86
3.51
2.
51
3.11
2.
93
3.58
2.
93
2.64
Nam
ibia
23
24.9
4 0.
3 2
1.66
2.
08
2.08
1.96
1.
71
1.98
1.63
Nic
arag
ua
434.
69
0.21
3.
05
2.91
3.
39
3.07
2.
33
2.8
2.88
2.
96
2.42
2.
71
Nig
eria
25
3.62
0.
26
3.11
3.
43
3.21
2.
92
3.
3 3.
37
3.1
3.
68
Paki
stan
50
5.59
0.
05
3.28
3.
64
3.21
2.
87
2.56
3.
03
3.54
3.
2 2.
67
3.08
Pana
ma
3123
.95
0.09
2.
06
2.72
2.
04
1.42
2.
4 2.
98
2.8
2.38
2.
44
2.19
38
G
ener
al O
bsta
cles
Nat
ion
GD
P p
er
capi
ta
Fir
m
Gro
wth
F
inan
cing
P
olit
ical
In
stab
ility
In
flat
ion
Exc
hang
e R
ate
Judi
cial
E
ffic
ienc
y St
reet
C
rim
e C
orru
ptio
n T
axes
and
R
egul
atio
n
Ant
i-co
mpe
titi
ve
Beh
avio
r In
fras
truc
ture
Peru
23
34.9
4 -0
.02
3.09
3.
21
2.85
2.
99
2.55
2.
81
2.83
3.
35
2.68
2.
27
Phili
ppin
es
1125
.81
0.07
2.
69
2.85
3.
36
3.43
2.
24
2.8
3.13
3.
08
2.9
2.88
Pola
nd
3216
.04
0.33
2.
47
2.75
2.
58
2.27
2.
3 2.
37
2.27
3.
08
2.23
1.
67
Port
ugal
11
582.
33
0.12
1.
8 2.
08
2.1
1.74
1.
88
1.64
1.
73
2.15
2.
18
1.75
Rom
ania
13
72.0
2 0.
07
3.26
3.
44
3.75
3.
19
2.59
2.
45
2.88
3.
57
2.52
2.
44
Rus
sian
Fed
erat
ion
2223
.57
0.29
3.
2 3.
49
3.53
3.
15
2.17
2.
65
2.62
3.
58
2.67
2.
12
Sene
gal
563.
39
0.15
3
2.21
2.
56
2
2.61
3.
04
2.97
2.88
Sing
apor
e 24
948.
09
0.12
1.
97
1.5
1.61
1.
88
1.32
1.
22
1.28
1.
55
1.58
1.
42
Slov
ak R
epub
lic
3805
.41
0.14
3.
34
1.53
3.
13
2.43
2.
13
2.49
2.
47
3.25
2.
26
1.98
Slov
enia
10
232.
73
0.29
2.
3 2.
6 2.
23
2.21
2.
29
1.68
1.
64
2.91
2.
43
1.74
Sout
h A
fric
a 39
25.0
6 0.
26
2.34
1.
97
2.45
2.
39
3.
58
2.58
2.
64
1.
83
Spai
n 15
858.
03
0.25
2.
21
2.17
2.
27
1.93
1.
97
1.92
2.
08
2.65
2.
25
1.94
Swed
en
2825
8.28
0.
23
1.83
2.
46
1.66
1.
78
1.46
1.
54
1.18
2.
67
1.97
1.
52
Tan
zani
a 18
2.29
0.
25
2.85
2.
48
2.65
2.
07
1.
96
2.88
2.
7
3.21
Tha
iland
28
36.2
9 -0
.02
3.1
3.49
3.
4 3.
62
2.13
3.
48
3.47
3.
54
3.6
2.76
Tri
nida
d an
d T
obag
o 45
26.2
8 0.
18
3.03
1.
81
2.49
2.
41
1.45
2.
18
1.68
2.
78
1.79
2.
1
Tun
isia
21
99.7
8 0.
14
1.79
1.
94
1.7
1.94
1.55
2.
11
2.12
2.1
Tur
key
2993
.89
0.1
3.12
3.
55
3.61
2.
83
2.3
2.09
2.
89
3.16
2.
79
2.22
Uga
nda
324.
37
0.18
3.
17
2.47
2.
68
1.78
2.27
2.
93
2.48
2.81
Ukr
aine
86
6.52
0.
03
3.45
3.
22
3.43
3.
05
2.16
2.
49
2.51
3.
7 2.
86
2.22
Uni
ted
Kin
gdom
20
186.
56
0.27
2.
33
2.19
2.
16
2.28
1.
5 1.
95
1.24
2.
87
1.72
1.
69
Uni
ted
Stat
es
2925
0.32
0.
16
2.38
2.
05
2.12
1.
71
1.84
2.
14
1.88
2.
39
1.7
1.83
Uru
guay
61
13.6
0
2.73
2.
61
2.03
2.
39
1.91
2.
07
2 3.
21
1.71
1.
9
Uzb
ekis
tan
448.
02
0.64
2.
77
2.03
3.
04
2.6
1.68
1.
77
2.22
2.
66
2.28
1.
95
Ven
ezue
la
3482
.51
-0.0
2 2.
62
3.64
3.
48
3.12
2.
65
3.18
3
3.1
2.63
2.
31
Zam
bia
393.
56
0.18
2.
95
2.57
3.
45
1.88
3.18
2.
78
2.39
3.07
Zim
babw
e 69
3.13
0.
47
3.05
2.
73
3.83
2.
93
2.
57
2.87
2.
87
2.
53
Ave
rage
46
43.1
3 0.
15
2.80
2.
72
2.76
2.
49
2.15
2.
51
2.56
2.
90
2.37
2.
34
39
G
ener
al O
bsta
cles
Nat
ion
GD
P p
er
capi
ta
Fir
m
Gro
wth
F
inan
cing
P
olit
ical
In
stab
ility
In
flat
ion
Exc
hang
e R
ate
Judi
cial
E
ffic
ienc
y St
reet
C
rim
e C
orru
ptio
n T
axes
and
R
egul
atio
n
Ant
i-co
mpe
titi
ve
Beh
avio
r In
fras
truc
ture
Pan
el B
: A
vera
ged
Acr
oss
Cou
ntry
Inc
ome
Gro
ups
Hig
h (N
=11)
21
376.
34
0.19
2.
19
2.2
2.04
1.
93
1.81
1.
71
1.59
2.
67
2 1.
72
Upp
er-M
iddl
e (N
=18
) 41
31.8
17
0.19
2.
75
2.62
2.
54
2.27
2.
13
2.38
2.
29
2.93
2.
18
1.99
Low
er-M
iddl
e (N
=26
) 19
84.8
52
0.11
3
3.14
3.
1 2.
94
2.31
2.
72
2.73
3.
24
2.59
2.
31
Low
Inc
ome
(N=
25)
435.
3 0.
14
2.85
2.
84
3.02
2.
61
2.15
2.
78
2.98
2.
73
2.53
2.
7
Pan
el C
: A
vera
ged
Acr
oss
5 G
eogr
aphi
c R
egio
ns
Eur
ope
and
Nor
th
Am
eric
a (N
=9)
2286
3.72
0.
19
2.2
2.22
2.
06
1.89
1.
79
1.78
1.
63
2.77
1.
98
1.76
Lat
in A
mer
ica
(N=
20)
3022
.2
0.09
2.
83
3.02
2.
84
2.8
2.39
2.
95
2.74
3.
01
2.43
2.
4
Asi
a (N
=10
) 27
72.5
2 0.
05
2.59
2.
82
2.74
2.
66
1.99
2.
62
2.71
2.
51
2.44
2.
43
Tra
nsiti
on (
N=2
3)
2417
.02
0.19
3.
05
2.99
3.
06
2.7
2.17
2.
39
2.5
3.28
2.
44
2.09
Afr
ica
(N=1
8)
1115
.81
0.23
2.
77
2.43
2.
75
2.21
2.64
2.
80
2.32
2.75
Pan
el D
: Ave
rage
d A
cros
s 3
Size
Gro
ups
Smal
l 37
59.3
3 0.
13
2.89
2.
84
2.90
2.
59
2.13
2.
64
2.62
2.
94
2.43
2.
24
Med
ium
43
77.9
8 0.
16
2.86
2.
87
2.84
2.
60
2.18
2.
46
2.53
3.
00
2.41
2.
26
Lar
ge
4365
.68
0.17
2.
54
2.75
2.
65
2.55
2.
19
2.49
2.
43
2.70
2.
23
2.36
40
Table II: Summary Statistics and Correlations Panel A presents the summary statistics and Panel B presents the correlations. The variables are described as follows: GDP/capita is real GDP per capita in US$. Firm Growth is the percentage increase in firm sales over the past three years. Firm Size is the Log of Sales. Financing, Political Instability, Inflation, Exchange Rate, Judicial Efficiency, Street Crime, Corruption, Taxes and Regulation, Anti-competitive Behavior and Infrastructure are general obstacles as indicated in the firm questionnaire. They take values 1 to 4, with higher values indicating greater obstacles. Firm variables are averaged over all firms in each country. Detailed variable definitions and sources are given in the appendix.
Panel A:
Variable N Mean Standard Deviation Minimum Maximum
GDP/capita 80 4643.13 7487.42 109.01 30794.03
Firm Growth 6938 0.15 0.57 -3 9
Firm Size 6625 9.85 7.82 -2.12 25.33
Financing 6481 2.81 1.12 1 4
Political Instability 6370 2.83 1.08 1 4
Inflation 6415 2.83 1.07 1 4 Exchange Rate 6299 2.59 1.15 1 4
Judicial Efficiency 5206 2.16 1.05 1 4
Street Crime 6201 2.54 1.15 1 4
Corruption 5838 2.54 1.15 1 4 Taxes and Regulation 6938 2.97 0.99 1 4
Anti-competitive Behavior 5152 2.39 1.13 1 4
Infrastructure 6417 2.27 1.07 1 4
41
Pan
el B
:
G
DP
/cap
ita
Fir
m
Gro
wth
F
irm
Siz
e F
inan
cing
P
olit
ical
In
stab
ility
In
flat
ion
Exc
hang
e R
ate
Judi
cial
E
ffic
ienc
y St
reet
C
rim
e C
orru
ptio
n T
axes
and
R
egul
atio
n
Ant
i-co
mpe
titi
ve
Beh
avio
r
Fir
m G
row
th
0.02
11*
Fir
m S
ize
0.26
88**
* -0
.053
3***
Fin
anci
ng
-0.1
853*
**
-0.0
692*
**
-0.1
745*
**
Pol
itic
al I
nsta
bilit
y -0
.222
5***
-0
.073
7***
-0
.101
0***
0.
2261
***
Infl
atio
n -0
.296
1***
-0
.052
7***
-0
.180
5***
0.
2619
***
0.42
66**
*
Exc
hang
e R
ate
-0.2
192*
**
-0.0
475*
**
-0.0
591*
**
0.23
01**
* 0.
3696
***
0.57
82**
*
Judi
cial
Eff
icie
ncy
-0.1
486*
**
-0.0
527*
**
0.01
04
0.17
74**
* 0.
3815
***
0.25
48**
* 0.
2947
***
Stre
et C
rim
e -0
.269
3***
-0
.086
2***
0.
0728
***
0.12
79**
* 0.
3391
***
0.29
22**
* 0.
2735
***
0.40
25**
*
Cor
rupt
ion
-0.3
33**
* -0
.064
1***
-0
.010
5 0.
2613
***
0.40
65**
* 0.
3525
***
0.31
48**
* 0.
5595
***
0.55
41**
*
Tax
es a
nd R
egul
atio
n -0
.077
9***
-0
.060
1***
-0
.232
9***
0.
3275
***
0.40
25**
* 0.
3131
***
0.30
67**
* 0.
2908
***
0.18
95**
* 0.
277*
**
Ant
i-co
mpe
titi
ve B
ehav
ior
-0.1
674*
**
-0.0
297*
* -0
.040
7***
0.
2213
***
0.34
78**
* 0.
3***
0.
2695
***
0.35
47**
* 0.
3698
***
0.47
35**
* 0.
2837
***
Infr
astr
uctu
re
-0.2
253*
**
-0.0
322*
**
0.09
72**
* 0.
2561
***
0.26
96**
* 0.
212*
**
0.18
06**
* 0.
2853
***
0.28
26**
* 0.
3406
***
0.19
48**
* 0.
2221
***
*,
**,
and
***
indi
cate
sig
nifi
canc
e le
vels
of
10, 5
, and
1 p
erce
nt r
espe
ctiv
ely
42
Tab
le I
II:
Fir
m G
row
th-I
mpa
ct o
f O
bsta
cles
The
reg
ress
ion
equa
tion
esti
mat
ed is
: Fir
m G
row
th =
α +
β1
GD
P/ca
pita
+ β
2 Si
ze+
β3
Fina
ncin
g +
β4P
olit
ical
Ins
tabi
lity
+ β
5Inf
lati
on +
β6
Exc
hang
e R
ates
+ β
7Jud
icia
l Eff
icie
ncy
+ β
8 St
reet
Cri
me+
β9
Cor
rupt
ion
+ β
10 T
axes
and
Reg
ulat
ion+
β11
Ant
i-co
mpe
titiv
e B
ehav
ior
+ β 1
2 In
fras
truc
ture
. The
var
iabl
es a
re d
escr
ibed
as
follo
ws:
Fir
m G
row
th is
the
perc
enta
ge in
crea
se in
fir
m s
ales
ove
r th
e pa
st th
ree
year
s. G
DP/
capi
ta is
log
of r
eal G
DP
per
capi
ta in
US$
. Fir
m S
ize
is th
e L
og o
f fi
rm s
ales
. Fin
anci
ng, P
oliti
cal I
nsta
bilit
y, I
nfla
tion
, Exc
hang
e R
ate,
Jud
icia
l Eff
icie
ncy,
Str
eet C
rim
e, C
orru
ptio
n, T
axes
an
d R
egul
atio
n, A
nti-
com
petit
ive
Beh
avio
r an
d In
fras
truc
ture
are
gen
eral
obs
tacl
es a
s in
dica
ted
in t
he f
irm
que
stio
nnai
re.
The
y ta
ke v
alue
s 1
to 4
, w
ith
whe
re 1
ind
icat
es n
o ob
stac
le a
nd 4
ind
icat
es
maj
or o
bsta
cle.
In
spec
ific
atio
ns (
1) t
o (1
0),
each
of
the
obst
acle
var
iabl
es i
s in
clud
ed i
ndiv
idua
lly.
Spec
ific
atio
n 11
inc
lude
s fi
nanc
ing,
pol
itica
l in
stab
ility
and
str
eet
crim
e ob
stac
les
only
whi
le
spec
ific
atio
n 12
incl
udes
the
full
mod
el. I
n sp
ecif
icat
ions
1-1
2, th
e fi
rst r
ow r
epre
sent
s th
e pa
ram
eter
est
imat
e of
the
obst
acle
, the
sec
ond
row
rep
orts
rob
ust s
tand
ard
erro
rs a
nd th
e th
ird
row
rep
orts
the
econ
omic
impa
ct o
f th
e ob
stac
le. A
ll re
gres
sion
s ar
e es
tim
ated
usi
ng c
ount
ry r
ando
m e
ffec
ts. D
etai
led
vari
able
def
initi
ons
and
sour
ces
are
give
n in
the
appe
ndix
.
1
2 3
4 5
6 7
8 9
10
11
12
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Con
stan
t 0.
250*
* 0.
209*
* 0.
253*
* 0.
214*
0.
147
-0.0
14
0.22
8**
0.19
1*
-0.0
53
0.16
0 0.
266
0.39
0***
[0
.106
] [0
.105
] [0
.108
] [0
.109
] [0
.106
] [0
.141
] [0
.114
] [0
.102
] [0
.137
] [0
.113
] [0
.172
] [0
.109
]
GD
P/C
apita
0.
000
0.00
1 -0
.003
0.
000
0.00
3 0.
025
-0.0
04
0.00
6 0.
028
0.00
0 0.
011
-0.0
06
[0
.013
] [0
.013
] [0
.014
] [0
.014
] [0
.014
] [0
.018
] [0
.014
] [0
.013
] [0
.017
] [0
.014
] [0
.021
] [0
.013
] Si
ze
0.00
0 0.
000
0.00
0 0.
000
0.00
0 -0
.001
0.
000
0.00
0 -0
.002
0.
001
-0.0
02
0.00
0
[0
.002
] [0
.002
] [0
.002
] [0
.002
] [0
.002
] [0
.002
] [0
.002
] [0
.002
] [0
.002
] [0
.002
] [0
.003
] [0
.002
]
Fi
nanc
ing
-0.0
34**
* [0
.008
] -0
.096
-0.0
36**
* [0
.010
] -0
.101
-0.0
29**
* [0
.008
] -0
.082
Po
litic
al
Inst
abili
ty
-0.0
28**
* [0
.009
] -0
.079
-0.0
26**
[0
.013
] -0
.074
-0.0
16**
[0
.009
] -0
.045
St
reet
Cri
me
-0.0
33**
* [0
.013
] -0
.084
-0.0
38**
* [0
.015
] -0
.096
-0.0
27**
* [0
.013
] -0
.068
Infl
atio
n
-0.0
23**
* [0
.008
] -0
.065
-0.0
11
[0.0
10]
-0.0
31
Exc
hang
e R
ates
-0.0
06
[0.0
08]
-0.0
16
0.01
6 [0
.011
] 0.
041
Judi
cial
E
ffic
ienc
y
-0.0
21**
* [0
.001
] -0
.045
-0.0
05
[0.0
10]
-0.0
11
C
orru
ptio
n
-0.0
20**
* [0
.011
]
0.00
5 [0
.013
]
43
1
2 3
4 5
6 7
8 9
10
11
12
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
-0
.051
0.01
3
Tax
es a
nd
Reg
ulat
ion
-0.0
30**
* [0
.012
] -0
.088
-0.0
02
[0.0
13]
0.00
0
Ant
i-C
ompe
titiv
e B
ehav
ior
-0.0
06
[0.0
07]
-0.0
14
0.01
8*
[0.0
07]
0.04
3
-0.0
09
[0.0
07]
-0.0
20
0.00
8 [0
.009
] 0.
018
In
fras
truc
ture
N
6235
61
33
5964
61
75
6068
51
42
5620
63
43
5091
62
05
4187
57
78
NC
ount
ries
79
79
79
79
79
61
78
79
60
79
58
78
R
2 (w
ithi
n)
0.00
36
0.00
16
0.00
28
0.00
10
0.00
00
0.00
11
0.00
09
0.00
17
0.00
01
0.00
04
0.01
00
0.00
66
R2 (
betw
een)
0.
0438
0.
0922
0.
0880
0.
0411
0.
0454
0.
0618
0.
0812
0.
0631
0.
0979
0.
0033
0.
1131
0.
1104
R2 (
all)
0.
0055
0.
0047
0.
0064
0.
0030
0.
0040
0.
0066
0.
0030
0.
0044
0.
0080
0.
0000
0.
0170
0.
0112
*,
**,
and
***
indi
cate
sig
nifi
canc
e le
vels
of
10, 5
, and
1 p
erce
nt r
espe
ctiv
ely
44
Tab
le I
V:
Fir
m G
row
th-I
nter
acti
on E
ffec
ts
T
he r
egre
ssio
n eq
uati
on e
stim
ated
is: F
irm
Gro
wth
= α
+ β
1 G
DP/
capi
ta +
β2 Si
ze +
β3
Fina
ncin
g +
β4 Po
litic
al I
nsta
bilit
y +
β5
Stre
et C
rim
e+ β
6 Fi
nanc
ing
x In
com
e D
umm
ies
+ β
7 Fi
nanc
ing
x Si
ze +
β8
Polit
ical
Ins
tabi
lity
x In
com
e D
umm
ies
+ β
9 Po
litic
al I
nsta
bilit
y x
Size
+ β
10 S
tree
t C
rim
e x
Inco
me
Dum
mie
s +
β11
Str
eet
Cri
me
x Si
ze.
The
var
iabl
es a
re d
escr
ibed
as
follo
ws:
Fir
m G
row
th i
s th
e pe
rcen
tage
inc
reas
e in
fir
m s
ales
ove
r th
e pa
st t
hree
yea
rs.
GD
P/ca
pita
is
log
of r
eal
GD
P pe
r ca
pita
in
US$
. Fi
rm S
ize
is t
he l
og o
f sa
les.
Fin
anci
ng,
Polit
ical
Ins
tabi
lity
and
Stre
et C
rim
e ar
e ge
nera
l ob
stac
les
as in
dica
ted
in th
e fi
rm q
uest
ionn
aire
. The
y ta
ke v
alue
s 1
to 4
, with
whe
re 1
indi
cate
s no
obs
tacl
e an
d 4
indi
cate
s m
ajor
obs
tacl
e. I
ncom
e du
mm
ies
are
coun
try
dum
mie
s cr
eate
d on
the
basi
s of
in
com
e le
vel
of t
he c
ount
ry. H
igh
Inco
me
Dum
mie
s ta
kes
the
valu
e 1
for
coun
trie
s be
long
ing
to t
he h
igh
inco
me
grou
p an
d 0
othe
rwis
e, U
pper
Mid
dle
Inco
me
dum
my
take
s th
e va
lue
1 fo
r co
untr
ies
belo
ngin
g to
the
uppe
r m
iddl
e in
com
e gr
oup
and
0 ot
herw
ise,
Low
er M
iddl
e In
com
e du
mm
y ta
kes
the
valu
e 1
for
coun
trie
s be
long
ing
to th
e lo
wer
mid
dle
inco
me
grou
p an
d 0
othe
r w
ise,
Low
Inc
ome
dum
mie
s ta
kes
the
valu
e 1
for
low
inco
me
grou
p co
untr
ies
and
0 ot
herw
ise.
In
spec
ific
atio
ns (
1) to
(3)
in e
ach
pane
l, th
e ob
stac
le v
aria
bles
and
its
inte
ract
ions
is in
clud
ed in
divi
dual
ly. S
peci
fica
tion
4 in
bo
th p
anel
s in
clud
es th
e fu
ll m
odel
. A
ll re
gres
sion
s ar
e es
tim
ated
usi
ng c
ount
ry r
ando
m e
ffec
ts. E
ach
spec
ific
atio
n al
so r
epor
ts th
e p-
valu
e of
the
join
t sig
nifi
canc
e te
st o
f th
e in
tera
ctio
n te
rms.
Det
aile
d va
riab
le d
efin
ition
s an
d so
urce
s ar
e gi
ven
in th
e ap
pend
ix.
1 2
3 4
1 2
3 4
Fi
rm G
row
th
Firm
Gro
wth
Fi
rm G
row
th
Firm
Gro
wth
Fi
rm G
row
th
Firm
Gro
wth
Fi
rm G
row
th
Firm
Gro
wth
C
onst
ant
0.31
0***
0.
247*
* 0.
272*
* 0.
484*
**
0.20
5***
0.
169*
* 0.
204*
**
0.26
4***
[0.1
07]
[0.1
07]
[0.1
07]
[0.1
12]
[0.0
72]
[0.0
71]
[0.0
69]
[0.0
54]
GD
P/C
apita
0.
002
0.00
3 -0
.001
-0
.002
[0.0
13]
[0.0
13]
[0.0
14]
[0.0
13]
Firm
Siz
e -0
.008
**
-0.0
05
-0.0
04
-0.0
13**
* -0
.001
0
0 0
[0
.003
] [0
.003
] [0
.003
] [0
.004
] [0
.002
] [0
.002
] [0
.002
] [0
.002
] U
pper
Mid
dle
Inco
me
Dum
my
0.08
9 0.
128
0.04
-0
.087
**
[0
.085
] [0
.085
] [0
.081
] [0
.034
] L
ower
Mid
dle
Inco
me
Dum
my
0.04
1 0.
088
0.07
7 -0
.044
**
[0
.079
] [0
.082
] [0
.077
] [0
.021
] L
ow I
ncom
e D
umm
y
0.
012
-0.0
26
-0.0
98
0.03
7
[0.0
85]
[0.0
84]
[0.0
84]
[0.0
25]
Fina
ncin
g -0
.060
***
-0.0
54**
* -0
.002
-0
.013
[0.0
11]
[0.0
12]
[0.0
19]
[0.0
19]
Fina
ncin
g*Si
ze
0.00
3***
0.
003*
**
[0
.001
]
[0
.001
]
Fi
nanc
ing*
Upp
er M
iddl
e
-0
.042
*
-0
.026
[0.0
24]
[0.0
23]
Fina
ncin
g*L
ower
Mid
dle
-0.0
41*
-0.0
12
[0
.022
]
[0
.021
] Fi
nanc
ing*
Low
Inc
ome
-0.0
18
-0.0
16
[0
.024
]
[0
.022
] Po
litic
al In
stab
ility
-0.0
44**
*
-0.0
25*
0.
011
0.
007
[0.0
12]
[0
.013
]
[0.0
21]
[0
.021
] Po
litic
al In
stab
ility
*Siz
e
0.00
2*
0.
001
[0.0
01]
[0
.001
]
45
1
2 3
4 1
2 3
4
Firm
Gro
wth
Fi
rm G
row
th
Firm
Gro
wth
Fi
rm G
row
th
Firm
Gro
wth
Fi
rm G
row
th
Firm
Gro
wth
Fi
rm G
row
th
Polit
ical
Inst
abili
ty*U
pper
Mid
dle
-0
.059
**
-0
.042
[0
.026
]
[0.0
26]
Polit
ical
Inst
abili
ty*L
ower
Mid
dle
-0
.059
**
-0
.03
[0.0
24]
[0
.024
] Po
litic
al In
stab
ility
*Low
Inc
ome
-0
.009
-0.0
11
[0.0
25]
[0
.026
] St
reet
Cri
me
-0.0
46**
* -0
.038
***
-0.0
09
-0.0
2
[0.0
11]
[0.0
11]
[0.0
22]
[0.0
23]
Stre
et C
rim
e*Si
ze
0.00
2*
0.00
1
[0.0
01]
[0.0
01]
Stre
et C
rim
e*U
pper
Mid
dle
-0.0
23
-0.0
05
[0
.026
] [0
.027
] St
reet
Cri
me*
Low
er M
iddl
e
-0
.053
**
-0.0
27
[0
.025
] [0
.025
] St
reet
Cri
me*
Low
Inc
ome
0.02
5 0.
035
[0
.027
] [0
.027
] N
62
35
6133
59
64
5778
62
35
6133
59
64
5778
N
Cou
ntri
es
79
79
79
78
79
79
79
78
R2 (
wit
hin)
0.
0051
0.
0023
0.
0033
0.
0089
0.
0044
0.
0038
0.
0064
0.
0115
R
2 (be
twee
n)
0.05
83
0.07
07
0.10
91
0.13
31
0.04
61
0.04
75
0.02
84
0.00
84
R2 (
all)
0.
0076
0.
0043
0.
0074
0.
0141
0.
0078
0.
0068
0.
0066
0.
0011
F-
Tes
t of
Inte
ract
ions
0.
0016
0.
068
0.08
89
0.00
25
0.18
98
0.00
72
0.00
04
0.00
17
*, *
*, a
nd *
** in
dica
te s
igni
fica
nce
leve
ls o
f 10
, 5, a
nd 1
per
cent
res
pect
ivel
y
46
Tab
le V
: R
obus
tnes
s T
est-
Inst
rum
enta
l Var
iabl
es
T
wo
stag
e in
stru
men
tal v
aria
ble
regr
essi
ons
are
used
. The
fir
st s
tage
reg
ress
ion
is F
inan
cing
(or
Pol
itic
al I
nsta
bilit
y or
Str
eet C
rim
e) =
α +
γ1
GD
P/ca
pita
+ γ
2 F
irm
Siz
e +
γ 3 (
Ave
rage
val
ue o
f)
Fina
ncin
g (o
r Po
litic
al I
nsta
bilit
y or
Str
eet C
rim
e). T
he s
econ
d st
age
regr
essi
on e
quat
ion
esti
mat
ed is
: Fir
m G
row
th =
α +
β1
GD
P/ca
pita
+ β
2 Fi
rm S
ize+
β3
Fina
ncin
g (p
redi
cted
val
ue f
rom
fir
st s
tage
) +
β 4
Polit
ical
Ins
tabi
lity
(pre
dict
ed v
alue
fro
m f
irst
sta
ge)
+ β
5 St
reet
Cri
me
(pre
dict
ed v
alue
fro
m f
irst
sta
ge).
The
var
iabl
es a
re d
escr
ibed
as
follo
ws:
Fir
m G
row
th is
the
perc
enta
ge in
crea
se in
fir
m s
ales
ov
er th
e pa
st th
ree
year
s. G
DP/
capi
ta is
log
of r
eal G
DP
per
capi
ta in
US$
. Fir
m S
ize
is th
e L
og o
f Sa
les.
Fin
anci
ng, P
olit
ical
Inst
abili
ty, a
nd S
tree
t Cri
me
are
gene
ral o
bsta
cles
as
indi
cate
d in
the
firm
qu
estio
nnai
re. T
hey
take
val
ues
1 to
4, w
ith
whe
re 1
indi
cate
s no
obs
tacl
e an
d 4
indi
cate
s m
ajor
obs
tacl
e. I
n sp
ecif
icat
ions
(1)
-(3)
, the
Ave
rage
val
ue o
f th
e ge
nera
l obs
tacl
e is
cal
cula
ted
by a
vera
ging
th
e ob
stac
le a
cros
s co
untr
y an
d fi
rm s
ize
cate
gori
es a
nd in
spe
cifi
cati
ons
(4)-
(6),
the
Ave
rage
val
ue o
f th
e ge
nera
l ob
stac
le is
cal
cula
ted
by a
vera
ging
the
obst
acle
acr
oss
coun
trie
s. S
peci
fica
tion
s (1
) an
d (4
) in
stru
men
t the
Fin
anci
ng o
bsta
cle
wit
h its
ave
rage
val
ue, s
peci
fica
tions
(2)
and
(5)
inst
rum
ent t
he P
olit
ical
Ins
tabi
lity
obst
acle
wit
h its
ave
rage
val
ue a
nd s
peci
fica
tion
s (3
) an
d (6
) in
stru
men
t the
St
reet
Cri
me
obst
acle
wit
h it
s av
erag
e va
lue.
Eac
h sp
ecif
icat
ion
repo
rts
the
adju
sted
R-s
quar
es f
rom
the
firs
t sta
ge r
egre
ssio
n an
d th
e p-
valu
es o
f th
e F-
test
for
the
inst
rum
ents
use
d. D
etai
led
vari
able
de
fini
tion
s an
d so
urce
s ar
e gi
ven
in th
e ap
pend
ix.
1
2 3
4 5
6
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Con
stan
t 0.
304*
**
0.38
0***
0.
343*
**
0.35
9***
0.
387*
**
0.31
8***
[0
.061
] [0
.070
] [0
.076
] [0
.072
] [0
.077
] [0
.081
]
GD
P/ca
pita
0.
013*
**
0.01
1**
0.00
2 0.
011*
* 0.
010*
0.
004
[0
.005
] [0
.005
] [0
.006
] [0
.005
] [0
.005
] [0
.007
]
Firm
Siz
e -0
.006
***
-0.0
06**
* -0
.003
***
-0.0
06**
* -0
.006
***
-0.0
03**
*
[0
.001
] [0
.001
] [0
.001
] [0
.001
] [0
.001
] [0
.001
]
Fina
ncin
g -0
.070
***
-0.0
84**
*
[0
.014
]
[0
.017
]
Polit
ical
Inst
abili
ty
-0
.091
***
-0.0
93**
*
[0.0
14]
[0.0
16]
Stre
et C
rim
e
-0
.073
***
-0.0
66**
*
[0
.013
]
[0
.015
]
Firs
t Sta
ge A
dj R
2 (F
inan
cing
) 0.
191
0.15
59
Firs
t Sta
ge A
dj R
2 (P
olit
ical
Inst
abili
ty)
0.
2813
0.
2552
Firs
t Sta
ge A
dj R
2 (S
tree
t Cri
me)
0.
2955
0.
2703
F-T
est o
f In
stru
men
ts
0 0
0 0
0 0
N
6235
61
33
5964
62
35
6133
59
64
*, *
*, a
nd *
** in
dica
te s
igni
fica
nce
leve
ls o
f 10
, 5, a
nd 1
per
cent
res
pect
ivel
y
47
Tab
le V
I: R
obus
tnes
s T
est-
Con
trol
ling
for
Gro
wth
Opp
ortu
niti
es
T
he r
egre
ssio
n eq
uati
on e
stim
ated
is:
Fir
m G
row
th =
α +
β1
GD
P/ca
pita
+ β
2 Si
ze+
β3
Fina
ncin
g +
β4P
oliti
cal
Inst
abili
ty +
β5
Stre
et C
rim
e +
β6
Ave
rage
Sec
tor
Gro
wth
/Ext
erna
l Fi
nanc
e. T
he v
aria
bles
ar
e de
scri
bed
as f
ollo
ws:
Fir
m G
row
th is
the
perc
enta
ge in
crea
se in
fir
m s
ales
ove
r th
e pa
st th
ree
year
s. G
DP/
capi
ta is
log
of r
eal G
DP
per
capi
ta in
US$
. Fir
m S
ize
is th
e L
og o
f fi
rm s
ales
. Tw
o pr
oxie
s fo
r gr
owth
opp
ortu
niti
es a
re u
sed:
Ave
rage
Sec
tor
Gro
wth
is
the
grow
th r
ate
aver
aged
acr
oss
all
firm
s in
eac
h se
ctor
in
each
ind
ustr
y an
d E
xter
nal
Fina
nce
stan
ds f
or t
he p
ropo
rtio
n of
inv
estm
ent
fina
nced
ext
erna
lly.
Fina
ncin
g, P
olit
ical
Ins
tabi
lity,
and
Str
eet
are
gene
ral
obst
acle
s as
ind
icat
ed i
n th
e fi
rm q
uest
ionn
aire
. T
hey
take
val
ues
1 to
4,
wit
h w
here
1 i
ndic
ates
no
obst
acle
and
4 i
ndic
ates
m
ajor
obs
tacl
e. I
n sp
ecif
icat
ions
(1)
-(3)
and
(5)
-(7)
, Fin
anci
ng, P
oliti
cal
Inst
abili
ty a
nd S
tree
t C
rim
e ar
e en
tere
d in
divi
dual
ly.
In s
peci
fica
tion
(4)
and
(8),
the
y ar
e en
tere
d to
geth
er.
All
regr
essi
ons
are
esti
mat
ed u
sing
cou
ntry
ran
dom
eff
ects
. Det
aile
d va
riab
le d
efin
ition
s an
d so
urce
s ar
e gi
ven
in th
e ap
pend
ix.
1
2 3
4 5
6 7
8
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Con
stan
t 0.
085*
0.
052
0.10
0*
0.18
6***
0.
023
-0.0
22
0.05
8 0.
239*
[0
.051
] [0
.052
] [0
.053
] [0
.061
] [0
.128
] [0
.131
] [0
.134
] [0
.141
]
GD
P/C
apita
-0
.001
0
-0.0
05
-0.0
06
0.03
2**
0.03
3**
0.02
5 0.
017
[0
.006
] [0
.006
] [0
.006
] [0
.006
] [0
.016
] [0
.016
] [0
.017
] [0
.017
]
Firm
Siz
e 0
0 0
0 -0
.003
-0
.002
-0
.002
-0
.002
[0
.001
] [0
.001
] [0
.001
] [0
.001
] [0
.002
] [0
.002
] [0
.002
] [0
.002
]
Ave
rage
Sec
tor
Gro
wth
0.
974*
**
0.99
6***
0.
983*
**
0.96
2***
[0
.037
] [0
.038
] [0
.038
] [0
.039
]
Ext
erna
l Fin
ance
-0
.008
-0
.013
-0
.02
-0.0
1
[0
.025
] [0
.025
] [0
.025
] [0
.026
]
Fina
ncin
g -0
.026
***
-0.0
22**
* -0
.039
***
-0.0
35**
*
[0
.007
]
[0
.007
] [0
.009
]
[0
.009
]
Polit
ical
Inst
abili
ty
-0
.018
**
-0
.007
-0.0
26**
*
-0.0
11
[0.0
07]
[0
.008
]
[0.0
10]
[0
.010
]
Stre
et C
rim
e
-0
.025
***
-0.0
20**
*
-0
.038
***
-0.0
33**
*
[0
.007
] [0
.007
]
[0
.009
] [0
.010
]
N
5814
57
14
5546
53
69
4251
42
17
4150
40
81
NC
ount
ries
79
79
79
78
58
58
58
58
R
2 (w
ithi
n)
0.04
04
0.03
85
0.03
85
0.04
03
0.00
38
0.00
1 0.
0039
0.
0081
R
2 (be
twee
n)
0.97
09
0.96
97
0.97
2 0.
9644
0.
1359
0.
1327
0.
0977
0.
1158
R
2 (al
l)
0.11
41
0.11
76
0.11
4 0.
1133
0.
0142
0.
011
0.01
16
0.01
72
*, *
*, a
nd *
** in
dica
te s
igni
fica
nce
leve
ls o
f 10
, 5, a
nd 1
per
cent
res
pect
ivel
y
48
Tab
le V
II:
Rob
ustn
ess
Tes
t-V
aryi
ng S
ampl
es
T
he r
egre
ssio
n eq
uatio
n es
tim
ated
is:
Fir
m G
row
th =
α +
β1
GD
P/ca
pita
+ β
2 Si
ze+
β3
Fina
ncin
g +
β4P
olit
ical
Ins
tabi
lity
+ β
5 St
reet
Cri
me.
The
var
iabl
es a
re d
escr
ibed
as
follo
ws:
Fir
m G
row
th i
s th
e pe
rcen
tage
inc
reas
e in
fir
m s
ales
ove
r th
e pa
st t
hree
yea
rs.
GD
P/ca
pita
is
log
of r
eal
GD
P pe
r ca
pita
in
US$
. Fi
rm S
ize
is t
he L
og o
f fi
rm s
ales
. Fi
nanc
ing,
Pol
itic
al I
nsta
bilit
y, a
nd S
tree
t ar
e ge
nera
l ob
stac
les
as in
dica
ted
in th
e fi
rm q
uest
ionn
aire
. The
y ta
ke v
alue
s 1
to 4
, with
whe
re 1
indi
cate
s no
obs
tacl
e an
d 4
indi
cate
s m
ajor
obs
tacl
e. S
peci
fica
tion
s (1
) to
(4)
exc
lude
cer
tain
cou
ntri
es f
rom
the
full
sam
ple
of f
irm
s w
hile
spe
cifi
catio
ns (
5) t
o (9
) ex
clud
e th
e co
untr
ies
from
a r
educ
ed s
ampl
e w
hich
doe
s no
t in
clud
e fi
rms
repo
rtin
g ve
ry h
igh(
/low
) gr
owth
rat
es (
>+
/-10
0%).
All
regr
essi
ons
are
esti
mat
ed u
sing
cou
ntry
ran
dom
eff
ects
. Sta
ndar
d er
rors
rep
orte
d ar
e ad
just
ed f
or c
lust
erin
g at
the
coun
try
leve
l. D
etai
led
vari
able
def
initi
ons
and
sour
ces
are
give
n in
the
appe
ndix
.
Hig
h G
row
th F
irm
s In
clud
ed
Hig
h G
row
th F
irm
s E
xclu
ded
Cou
ntri
es E
xclu
ded
Cou
ntri
es E
xclu
ded
Tra
nsiti
on
Eco
nom
ies
Afr
ican
ec
onom
ies
Afr
ican
an
d T
rans
ition
E
cono
mie
s
Uzb
ekis
tan,
B
osni
a &
H
erze
govi
na,
Est
onia
N
one
Tra
nsiti
on
Eco
nom
ies
Afr
ican
ec
onom
ies
Afr
ican
an
d T
rans
ition
E
cono
mie
s
Uzb
ekis
tan,
B
osni
a &
H
erze
govi
na,
Est
onia
1 2
3 4
5 6
7 8
9
Firm
G
row
th
Firm
G
row
th
Firm
G
row
th
Firm
G
row
th
Firm
G
row
th
Firm
G
row
th
Firm
G
row
th
Firm
G
row
th
Firm
G
row
th
Con
stan
t 0.
350*
**
0.17
7 0.
148
0.22
5**
0.13
6*
0.27
1***
0.
01
0.14
2 0.
108
[0
.092
] [0
.118
] [0
.101
] [0
.092
] [0
.072
] [0
.074
] [0
.090
] [0
.101
] [0
.071
] G
DP/
Cap
ita
-0.0
12
0.01
6 0.
013
0.00
1 0.
005
-0.0
04
0.02
2**
0.01
2 0.
006
[0
.010
] [0
.014
] [0
.011
] [0
.011
] [0
.009
] [0
.008
] [0
.011
] [0
.011
] [0
.009
] Fi
rm S
ize
0 -0
.001
-0
.001
0.
002
0.00
4***
0.
001
0.00
2 0
0.00
4***
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
01]
[0.0
02]
[0.0
01]
[0.0
02]
[0.0
01]
-0.0
13**
-0
.027
***
-0.0
20**
* -0
.019
***
-0.0
18**
* -0
.017
***
-0.0
21**
* -0
.022
***
-0.0
16**
* Fi
nanc
ing
[0.0
06]
[0.0
07]
[0.0
06]
[0.0
06]
[0.0
05]
[0.0
06]
[0.0
05]
[0.0
06]
[0.0
05]
-0.0
13*
-0.0
09
-0.0
11
-0.0
09
-0.0
17**
* -0
.015
**
-0.0
15**
* -0
.012
* -0
.017
***
Polit
ical
Inst
abili
ty
[0.0
07]
[0.0
08]
[0.0
08]
[0.0
07]
[0.0
05]
[0.0
06]
[0.0
06]
[0.0
07]
[0.0
05]
-0.0
17**
-0
.026
***
-0.0
19**
* -0
.020
***
-0.0
08
-0.0
18**
* -0
.007
-0
.019
***
-0.0
09*
Stre
et C
rim
e
[0.0
07]
[0.0
07]
[0.0
07]
[0.0
07]
[0.0
05]
[0.0
06]
[0.0
05]
[0.0
07]
[0.0
05]
N
3224
52
30
2682
55
32
5631
32
02
5107
26
78
5421
N
Cou
ntri
es
54
62
38
75
78
54
62
38
75
R2 (
wit
hin)
0.
0039
0.
0066
0.
0092
0.
0044
0.
0065
0.
0083
0.
0076
0.
0114
0.
0063
R
2 (be
twee
n)
0.19
53
0.12
33
0.19
21
0.07
26
0.18
66
0.17
39
0.21
44
0.19
19
0.24
75
R2 (
all)
0
.012
2 0
.014
7 0
.019
5 0
.005
5 0.
0226
0.
0195
0
.024
8 0
.023
3 0
.026
2 *,
**,
and
***
indi
cate
sig
nifi
canc
e le
vels
of
10, 5
, and
1 p
erce
nt r
espe
ctiv
ely
Tab
le V
III:
Rob
ustn
ess
Tes
t-D
ealin
g w
ith
Per
cept
ion
Bia
ses
T
he r
egre
ssio
n eq
uatio
n es
timat
ed is
: Fir
m G
row
th =
α +
β1
GD
P/ca
pita
+ β
2 Si
ze+
β3
Fina
ncin
g +
β4P
oliti
cal I
nsta
bilit
y +
β5 St
reet
Cri
me
+ β
6 K
vetc
h1/K
vetc
h2. T
he v
aria
bles
are
des
crib
ed a
s fo
llow
s:
Firm
Gro
wth
is
the
perc
enta
ge i
ncre
ase
in f
irm
sal
es o
ver
the
past
thr
ee y
ears
. GD
P/ca
pita
is
log
of r
eal
GD
P pe
r ca
pita
in
US$
. Fir
m S
ize
is t
he L
og o
f fi
rm s
ales
. Fi
nanc
ing,
Pol
itica
l In
stab
ility
, and
St
reet
are
gen
eral
obs
tacl
es a
s in
dica
ted
in t
he f
irm
que
stio
nnai
re.
The
y ta
ke v
alue
s 1
to 4
, w
ith
whe
re 1
ind
icat
es n
o ob
stac
le a
nd 4
ind
icat
es m
ajor
obs
tacl
e. K
vetc
h1 i
s th
e de
viat
ion
of e
ach
firm
’s
resp
onse
fro
m th
e m
ean
resp
onse
for
the
cou
ntry
to th
e qu
estio
n “H
ow h
elpf
ul d
o yo
u fi
nd th
e ce
ntra
l/nat
iona
l gov
ernm
ent t
oday
tow
ards
bus
ines
ses
like
you
rs”.
Kve
tch2
is th
e de
viat
ion
of e
ach
firm
’s
resp
onse
for
m t
he m
ean
resp
onse
for
the
cou
ntry
to
the
ques
tion
“How
pre
dict
able
are
cha
nges
in
econ
omic
and
fin
anci
al p
olic
ies”
. In
spe
cifi
cati
ons
(1)-
(3)
and
(5)-
(7),
Fin
anci
ng, P
olit
ical
Ins
tabi
lity
and
Stre
et C
rim
e ar
e en
tere
d in
divi
dual
ly. I
n sp
ecif
icat
ion
(4)
and
(8),
they
are
ent
ered
toge
ther
alo
ng w
ith
the
othe
r ob
stac
les.
All
regr
essi
ons
are
esti
mat
ed u
sing
cou
ntry
ran
dom
eff
ects
. Sta
ndar
d er
rors
re
port
ed a
re a
djus
ted
for
clus
teri
ng a
t the
cou
ntry
leve
l. D
etai
led
vari
able
def
initi
ons
and
sour
ces
are
give
n in
the
appe
ndix
.
1 2
3 4
5 6
7 8
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Con
stan
t 0.
249*
* 0.
210*
* 0.
249*
* 0.
383*
**
0.11
2 0.
067
0.11
0.
313*
*
[0
.106
] [0
.104
] [0
.108
] [0
.111
] [0
.137
] [0
.141
] [0
.142
] [0
.147
]
GD
P/C
apita
-0
.001
0.
001
-0.0
03
0 -0
.002
-0
.001
-0
.001
-0
.001
[0
.013
] [0
.013
] [0
.014
] [0
.002
] [0
.002
] [0
.002
] [0
.002
] [0
.002
]
Firm
Siz
e 0
0 0
-0.0
06
0.02
0.
02
0.01
6 0.
007
[0
.002
] [0
.002
] [0
.002
] [0
.014
] [0
.017
] [0
.018
] [0
.018
] [0
.018
]
Kve
tch1
-0
.014
**
-0.0
12*
-0.0
15**
-0
.012
*
[0
.007
] [0
.007
] [0
.007
] [0
.007
]
Kve
tch2
-0
.014
**
-0.0
11*
-0.0
14**
-0
.01
[0
.007
] [0
.007
] [0
.007
] [0
.007
]
Fina
ncin
g -0
.034
***
-0.0
28**
* -0
.041
***
-0.0
36**
*
[0
.007
]
[0
.007
] [0
.008
]
[0
.008
]
Polit
ical
Inst
abili
ty
-0
.028
***
-0
.016
*
-0.0
30**
*
-0.0
16
[0.0
08]
[0
.008
]
[0.0
09]
[0
.010
]
Stre
et C
rim
e
-0
.032
***
-0.0
25**
*
-0
.038
***
-0.0
31**
*
[0
.007
] [0
.008
]
[0
.008
] [0
.009
]
N
6030
59
35
5777
56
02
5241
51
61
4992
48
76
Nco
untr
ies
79
79
79
78
60
60
60
59
R2 (
wit
hin)
0.
0045
0.
0024
0.
0036
0.
0071
0.
0056
0.
0026
0.
0046
0.
0098
R
2 (be
twee
n)
0.04
02
0.08
76
0.07
84
0.10
27
0.11
01
0.09
53
0.11
59
0.11
87
R2 (
all)
0.
0062
0.
0054
0.
0068
0.
0113
0.
0132
0.
0093
0.
0117
0.
0171
*,
**,
and
***
indi
cate
sig
nifi
canc
e le
vels
of
10, 5
, and
1 p
erce
nt r
espe
ctiv
ely
50
Tab
le I
X:
Fir
m G
row
th-I
mpa
ct o
f In
divi
dual
Fin
anci
ng O
bsta
cles
T
he r
egre
ssio
n eq
uatio
n es
timat
ed is
: Fir
m G
row
th =
α +
β1
GD
P/ca
pita
+ β
2 Si
ze+
β3
Col
late
ral +
β4P
aper
wor
k +
β5
Hig
h In
tere
st R
ates
+ β
6 Sp
ecia
l Con
nect
ions
+ β
7 L
ack
mon
ey to
len
d+ β
8 A
cces
s to
fo
reig
n ba
nks
+ β 9
Acc
ess
to n
on-b
ank
equi
ty+
β10
Exp
ort
fina
nce
+ β
11 L
ease
fin
ance
+ β
12 C
redi
t +
β13
Lon
g T
erm
Loa
ns +
β14
(Res
idua
l).
The
var
iabl
es a
re d
escr
ibed
as
follo
ws:
Fir
m G
row
th i
s th
e pe
rcen
tage
inc
reas
e in
fir
m s
ales
ove
r th
e pa
st t
hree
yea
rs.
GD
P/ca
pita
is
log
of r
eal
GD
P pe
r ca
pita
in
US$
. Fi
rm S
ize
is t
he L
og o
f Sa
les.
Col
late
ral,
Pape
rwor
k, H
igh
Inte
rest
Rat
es,
Spec
ial
Con
nect
ions
, Lac
k m
oney
to le
nd, A
cces
s to
for
eign
ban
ks, A
cces
s to
non
-ban
k eq
uity
, Exp
ort f
inan
ce, L
ease
fin
ance
, Cre
dit,
Lon
g T
erm
Loa
ns a
re in
divi
dual
fin
anci
ng o
bsta
cles
as
indi
cate
d in
the
firm
qu
estio
nnai
re. T
hey
take
val
ues
1 to
4, w
ith
whe
re 1
indi
cate
s no
obs
tacl
e an
d 4
indi
cate
s m
ajor
obs
tacl
e. I
n sp
ecif
icat
ions
(1)
to (
11),
eac
h of
the
obst
acle
var
iabl
es is
incl
uded
indi
vidu
ally
. Res
idua
l is
the
resi
dual
fro
m a
reg
ress
ion
of t
he G
ener
al F
inan
cing
Obs
tacl
e on
all
the
indi
vidu
al f
inan
cing
obs
tacl
es.
Spec
ific
atio
n 13
inc
lude
s C
olla
tera
l, Pa
perw
ork,
Hig
h In
tere
st R
ates
, Sp
ecia
l C
onne
ctio
ns,
Lac
k m
oney
to
lend
, and
Lea
se F
inan
ce. S
peci
fica
tions
12
and
14 i
nclu
de t
he F
inan
cing
Res
idua
l. In
spe
cifi
catio
ns 1
-14,
the
fir
st r
ow r
epre
sent
s th
e pa
ram
eter
est
imat
e of
the
obs
tacl
e, t
he s
econ
d ro
w
repo
rts
robu
st s
tand
ard
erro
rs a
nd t
he t
hird
row
rep
orts
the
eco
nom
ic i
mpa
ct o
f th
e ob
stac
le.
All
regr
essi
ons
are
esti
mat
ed u
sing
cou
ntry
ran
dom
eff
ects
. St
anda
rd e
rror
s re
port
ed a
re a
djus
ted
for
clus
teri
ng a
t the
cou
ntry
leve
l. D
etai
led
vari
able
def
initi
ons
and
sour
ces
are
give
n in
the
appe
ndix
.
1
2 3
4 5
6 7
8 9
10
11
12
13
14
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
C
onst
ant
0.20
6*
0.18
2*
0.26
4**
0.18
2*
0.21
8**
0.18
4 0.
160
0.17
8 0.
106
0.13
7 -0
.025
-0
.077
0.
289*
**
0.08
[0.1
08]
[0.1
04]
[0.1
08]
[0.1
00]
[0.1
04]
[0.1
12]
[0.1
07]
[0.1
13]
[0.1
12]
[0.1
16]
[0.1
39]
[0.1
48]
[0.1
30]
[0.1
78]
GD
P/C
apita
0.
001
0.00
3 -0
.002
0.
000
-0.0
01
0.00
2 0.
001
0.00
0 0.
005
0.00
2 0.
026
0.02
9*
-0.0
02
0.02
4
[0.0
14]
[0.0
13]
[0.0
14]
[0.0
13]
[0.0
13]
[0.0
14]
[0.0
14]
[0.0
14]
[0.0
14]
[0.0
15]
[0.0
18]
[0.0
16]
[0.0
02]
[0.0
17]
Firm
Siz
e 0.
000
0.00
0 0.
000
0.00
0 -0
.001
-0
.001
0.
000
-0.0
01
0.00
0 0.
000
-0.0
02
-0.0
02
0.00
1 -0
.002
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
02]
[0.0
12]
[0.0
02]
-0.0
24**
*
-0.0
03
-0.0
09
[0.0
07]
[0
.012
] [0
.012
] C
olla
tera
l -0
.060
-0.0
07
-0.0
22
-0
.026
***
-0.0
07
-0.0
16
[0
.009
]
[0
.009
] [0
.010
] Pa
perw
ork
-0
.065
-0
.017
-0
.040
-0
.032
***
-0
.020
**
-0.0
15
[0.0
10]
[0
.010
] [0
.012
] H
igh
Inte
rest
R
ates
-0
.103
-0.0
65
-0.0
48
-0
.017
**
-0.0
06
-0.0
06
[0
.007
]
[0
.009
] [0
.013
] Sp
ecia
l C
onne
ctio
ns
-0
.037
-0
.013
-0
.013
-0
.026
***
-0
.006
-0
.009
[0
.008
]
[0.0
11]
[0.0
13]
Lac
k m
oney
to
lend
-0
.055
-0.0
13
-0.0
19
-0
.015
**
-0.0
04
0.01
4
[0.0
09]
[0.0
13]
[0.0
15]
Lea
se
Fina
nce
-0
.031
-0
.008
0.
029
-0.0
03
[0
.007
]
Acc
ess
to
fore
ign
bank
s
-0
.006
51
1
2 3
4 5
6 7
8 9
10
11
12
13
14
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
Fi
rm
Gro
wth
-0.0
06
[0
.008
]
A
cces
s to
no
n-ba
nk
equi
ty
-0
.013
0.
004
[0
.008
]
Exp
ort
Fina
nce
0.00
8
0.
001
[0
.007
]
C
redi
t
0.00
2
-0
.009
[0.0
08]
L
ong
Ter
m
Loa
ns
-0.0
24
-0.0
24**
-0.0
25**
Fi
nanc
ing
Res
idua
l
[0.0
11]
[0
.011
] N
60
24
6133
62
98
6002
58
08
5076
50
93
5037
44
40
5332
50
30
2988
45
79
2988
N
Cou
ntri
es
79
79
79
79
79
78
78
78
78
78
60
58
78
58
R2 (
wit
hin)
0.
0022
0.
0023
0.
0030
0.
0010
0.
0021
0.
0006
0.
0000
0.
0000
0.
0000
0.
0000
0.
0000
0.
0010
0.
0027
0.
0052
R
2 (bet
wee
n)
0.01
19
0.01
40
0.00
56
0.02
89
0.07
96
0.01
61
0.00
78
0.03
31
0.00
10
0.00
10
0.09
83
0.10
98
0.06
29
0.13
64
R2 (
all)
0.
0034
0.
0028
0.
0033
0.
0015
0.
0064
0.
0025
0.
0018
0.
0028
0.
0006
0.
0018
0.
0086
0.
0075
0.
0089
0.
0136
*, *
*, a
nd *
** in
dica
te s
igni
fica
nce
leve
ls o
f 10
, 5, a
nd 1
per
cent
res
pect
ivel
y
52
Tab
le X
: H
igh
Inte
rest
Rat
es-I
mpa
ct o
f In
divi
dual
Fin
anci
ng O
bsta
cles
T
he r
egre
ssio
n eq
uati
on e
stim
ated
is: H
igh
Inte
rest
Rat
es =
α +
β1
GD
P/ca
pita
+ β
2 Si
ze+
β3
Col
late
ral +
β4P
aper
wor
k +
β5 Sp
ecia
l Con
nect
ions
+ β 6
Lac
k m
oney
to le
nd+
β7
Acc
ess
to f
orei
gn b
anks
+ β
8
Acc
ess
to n
on-b
ank
equi
ty+
β9 E
xpor
t fin
ance
+ β
10 L
ease
fin
ance
+ β
11 C
redi
t + β
12L
ong
Ter
m L
oans
. The
var
iabl
es a
re d
escr
ibed
as
follo
ws:
GD
P/ca
pita
is lo
g of
rea
l GD
P pe
r ca
pita
in U
S$. F
irm
Siz
e is
the
Log
of
Sale
s. C
olla
tera
l, Pa
perw
ork,
Hig
h In
tere
st R
ates
, Spe
cial
Con
nect
ions
, Lac
k m
oney
to le
nd, A
cces
s to
for
eign
ban
ks, A
cces
s to
non
-ban
k eq
uity
, Exp
ort f
inan
ce, L
ease
fin
ance
, Cre
dit,
Lon
g T
erm
Loa
ns a
re i
ndiv
idua
l fin
anci
ng o
bsta
cles
as
indi
cate
d in
the
fir
m q
uest
ionn
aire
. The
y ta
ke v
alue
s 1
to 4
, with
whe
re 1
ind
icat
es n
o ob
stac
le a
nd 4
ind
icat
es m
ajor
obs
tacl
e. I
n sp
ecif
icat
ions
(1)
to
(10)
, eac
h of
the
obst
acle
var
iabl
es is
incl
uded
indi
vidu
ally
. Spe
cifi
cati
on (
11)
is th
e fu
ll m
odel
. The
reg
ress
ions
are
est
imat
ed u
sing
ord
ered
pro
bit w
ith c
lust
erin
g at
the
coun
try
leve
l. D
etai
led
vari
able
de
fini
tion
s an
d so
urce
s ar
e gi
ven
in th
e ap
pend
ix.
1
2 3
4 5
6 7
8 9
10
11
Hig
h In
tere
st
Rat
es
Hig
h In
tere
st
Rat
es
Hig
h In
tere
st
Rat
es
Hig
h In
tere
st
Rat
es
Hig
h In
tere
st
Rat
es
Hig
h In
tere
st
Rat
es
Hig
h In
tere
st
Rat
es
Hig
h In
tere
st
Rat
es
Hig
h In
tere
st
Rat
es
Hig
h In
tere
st
Rat
es
Hig
h In
tere
st
Rat
es
GD
P/C
apita
-0
.119
* -0
.140
**
-0.1
08
-0.0
68
-0.1
51**
* -0
.155
***
-0.1
68**
* -0
.162
***
-0.1
56**
* -0
.036
-0
.150
***
[0
.065
] [0
.059
] [0
.066
] [0
.074
] [0
.048
] [0
.047
] [0
.049
] [0
.047
] [0
.048
] [0
.084
] [0
.056
] Fi
rm S
ize
-0.0
08
-0.0
07
-0.0
1 -0
.005
-0
.004
-0
.003
-0
.003
-0
.006
-0
.007
-0
.005
-0
.008
[0.0
07]
[0.0
07]
[0.0
06]
[0.0
06]
[0.0
07]
[0.0
06]
[0.0
06]
[0.0
06]
[0.0
06]
[0.0
07]
[0.0
07]
Col
late
ral
0.46
1***
0.21
5***
[0.0
35]
[0
.038
] Pa
perw
ork
0.
514*
**
0.24
4***
[0
.036
]
[0
.037
] Sp
ecia
l Con
nect
ions
0.47
8***
0.11
7***
[0.0
37]
[0
.030
] L
ack
mon
ey to
lend
0.
352*
**
0.07
6**
[0.0
31]
[0.0
35]
Lea
se f
inan
ce
0.32
3***
0.04
3
[0.0
23]
[0
.034
] A
cces
s to
for
eign
ban
ks
0.28
2***
-0
.006
[0
.021
]
[0
.033
] A
cces
s to
non
-ban
k eq
uity
0.28
6***
-0.0
12
[0
.024
]
[0.0
39]
Exp
ort f
inan
ce
0.30
3***
0.
034
[0.0
26]
[0.0
40]
Cre
dit
0.29
6***
0.01
3
[0.0
22]
[0
.033
] L
ong
term
Loa
ns
0.44
2***
0.
168*
**
[0.0
40]
[0.0
38]
N
5933
60
36
5921
57
27
5003
50
24
4970
43
83
5252
49
58
2998
Ps
eudo
R2
0.09
43
0.09
9 0.
0836
0.
0595
0.
0601
0.
054
0.05
43
0.05
98
0.05
55
0.09
81
0.16
81
*, *
*, a
nd *
** in
dica
te s
igni
fica
nce
leve
ls o
f 10
, 5, a
nd 1
per
cent
res
pect
ivel
y
53
App
endi
x A
1: V
aria
ble
Def
init
ions
and
Sou
rces
Var
iabl
e D
efin
ition
Sour
ce
Firm
Gro
wth
E
stim
ate
of th
e fi
rm's
sal
es g
row
th o
ver
the
past
thre
e ye
ars
W
orld
Bus
ines
s E
nvir
onm
ent S
urve
y
Firm
Siz
e D
umm
ies
A f
irm
is d
efin
ed a
s sm
all i
f it
has
betw
een
5 an
d 50
em
ploy
ees,
med
ium
siz
e if
it h
as b
etw
een
51 a
nd 5
00
empl
oyee
s an
d la
rge
if it
has
mor
e th
an 5
00 e
mpl
oyee
s.
W
orld
Bus
ines
s E
nvir
onm
ent S
urve
y
GD
P/C
apita
R
eal G
DP
per
capi
ta in
US$
, ave
rage
199
5-99
Wor
ld D
evel
opm
ent I
ndic
ator
s
Bus
ines
s E
nvir
onm
ent O
bsta
cles
Fina
ncin
g H
ow p
robl
emat
ic is
fin
anci
ng f
or th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a m
inor
ob
stac
le (
2), a
mod
erat
e ob
stac
le (
3) o
r a
maj
or o
bsta
cle
(4)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Polit
ical
Inst
abili
ty
How
pro
blem
atic
is p
oliti
cal i
nsta
bilit
y fo
r th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a
min
or o
bsta
cle
(2),
a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
W
orld
Bus
ines
s E
nvir
onm
ent S
urve
y
Stre
et C
rim
e H
ow p
robl
emat
ic is
str
eet c
rim
e fo
r th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a m
inor
ob
stac
le (
2), a
mod
erat
e ob
stac
le (
3) o
r a
maj
or o
bsta
cle
(4)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Infl
atio
n H
ow p
robl
emat
ic is
infl
atio
n fo
r th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a m
inor
ob
stac
le (
2), a
mod
erat
e ob
stac
le (
3) o
r a
maj
or o
bsta
cle
(4)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Exc
hang
e R
ates
H
ow p
robl
emat
ic is
exc
hang
e ra
tes
for
the
oper
atio
n an
d gr
owth
of
your
bus
ines
s: n
o ob
stac
le (
1), a
min
or
obst
acle
(2)
, a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
W
orld
Bus
ines
s E
nvir
onm
ent S
urve
y
Judi
cial
Eff
icie
ncy
How
pro
blem
atic
is f
unct
ioni
ng o
f th
e ju
dici
ary
for
the
oper
atio
n an
d gr
owth
of
your
bus
ines
s: n
o ob
stac
le
(1),
a m
inor
obs
tacl
e (2
), a
mod
erat
e ob
stac
le (
3) o
r a
maj
or o
bsta
cle
(4)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Cor
rupt
ion
How
pro
blem
atic
is c
orru
ptio
n fo
r th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a m
inor
ob
stac
le (
2), a
mod
erat
e ob
stac
le (
3) o
r a
maj
or o
bsta
cle
(4)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Tax
es a
nd R
egul
atio
n H
ow p
robl
emat
ic a
re ta
xes
and
regu
lati
on f
or th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a
min
or o
bsta
cle
(2),
a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
W
orld
Bus
ines
s E
nvir
onm
ent S
urve
y
Ant
i-C
ompe
titiv
e B
ehav
ior
How
pro
blem
atic
is a
nti-
com
petit
ive
beha
vior
by
othe
r en
terp
rise
s or
the
gove
rnm
ent f
or th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a m
inor
obs
tacl
e (2
), a
mod
erat
e ob
stac
le (
3) o
r a
maj
or o
bsta
cle
(4)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Infr
astr
uctu
re
How
pro
blem
atic
is in
fras
truc
ture
for
the
oper
atio
n an
d gr
owth
of
your
bus
ines
s: n
o ob
stac
le (
1), a
min
or
obst
acle
(2)
, a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
W
orld
Bus
ines
s E
nvir
onm
ent S
urve
y
Indi
vidu
al F
inan
cing
Obs
tacl
es
Col
late
ral
How
pro
blem
atic
are
col
late
ral r
equi
rem
ents
of
bank
s an
d fi
nanc
ial i
nstit
utio
ns f
or th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a m
inor
obs
tacl
e (2
), a
mod
erat
e ob
stac
le (
3) o
r a
maj
or o
bsta
cle
(4)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
54
Var
iabl
e D
efin
ition
Sour
ce
Pape
rwor
k H
ow p
robl
emat
ic is
pap
erw
ork
for
the
oper
atio
n an
d gr
owth
of
your
bus
ines
s: n
o ob
stac
le (
1), a
min
or
obst
acle
(2)
, a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Hig
h In
tere
st R
ates
H
ow p
robl
emat
ic a
re h
igh
inte
rest
rat
es f
or th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a
min
or o
bsta
cle
(2),
a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Spec
ial C
onne
ctio
ns
How
pro
blem
atic
are
the
need
for
spe
cial
con
nect
ions
wit
h ba
nks,
for
the
oper
atio
n an
d gr
owth
of
your
bu
sine
ss: n
o ob
stac
le (
1), a
min
or o
bsta
cle
(2),
a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Lac
k m
oney
to le
nd
How
pro
blem
atic
is, b
anks
lack
ing
mon
ey to
lend
, for
the
oper
atio
n an
d gr
owth
of
your
bus
ines
s: n
o ob
stac
le (
1), a
min
or o
bsta
cle
(2),
a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Acc
ess
to f
orei
gn b
anks
H
ow p
robl
emat
ic is
lack
of
acce
ss to
for
eign
ban
ks f
or th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a m
inor
obs
tacl
e (2
), a
mod
erat
e ob
stac
le (
3) o
r a
maj
or o
bsta
cle
(4)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Acc
ess
to n
on-b
ank
equi
ty
How
pro
blem
atic
is la
ck o
f ac
cess
to e
quity
par
tner
s fo
r th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obst
acle
(1)
, a m
inor
obs
tacl
e (2
), a
mod
erat
e ob
stac
le (
3) o
r a
maj
or o
bsta
cle
(4)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Lea
se F
inan
ce
How
pro
blem
atic
is la
ck o
f ac
cess
to le
ase
fina
nce
for
the
oper
atio
n an
d gr
owth
of
your
bus
ines
s: n
o ob
stac
le (
1), a
min
or o
bsta
cle
(2),
a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Exp
ort F
inan
ce
How
pro
blem
atic
is la
ck o
f ac
cess
to e
xpor
t fin
ance
for
the
oper
atio
n an
d gr
owth
of
your
bus
ines
s: n
o ob
stac
le (
1), a
min
or o
bsta
cle
(2),
a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Cre
dit
How
pro
blem
atic
, is
inad
equa
te c
redi
t inf
orm
atio
n on
cus
tom
ers,
for
the
oper
atio
n an
d gr
owth
of
your
bu
sine
ss: n
o ob
stac
le (
1), a
min
or o
bsta
cle
(2),
a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Lon
g T
erm
Loa
ns
How
pro
blem
atic
is la
ck o
f ac
cess
to lo
ng te
rm b
ank
loan
s fo
r th
e op
erat
ion
and
grow
th o
f yo
ur b
usin
ess:
no
obs
tacl
e (1
), a
min
or o
bsta
cle
(2),
a m
oder
ate
obst
acle
(3)
or
a m
ajor
obs
tacl
e (4
)?
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Oth
er F
irm
Lev
el V
aria
bles
Ext
erna
l Fin
ance
Shar
e of
inve
stm
ent f
inan
ce c
omin
g fr
om e
xter
nal s
ourc
es c
alcu
late
d as
(1-
prop
orti
on o
f fi
rm’s
fix
ed
inve
stm
ent t
hat h
as b
een
fina
nced
fro
m in
tern
al f
unds
). T
his
is th
e sa
me
as p
ropo
rtio
n of
the
firm
’s f
ixed
in
vest
men
t tha
t has
bee
n fi
nanc
ed f
rom
equ
ity
+ lo
cal c
omm
erci
al b
anks
+ in
vest
men
t fun
ds +
for
eign
ba
nks
+ fa
mily
/fri
ends
+ M
oney
lend
ers
and
info
rmal
sou
rces
+ S
uppl
ier
Cre
dit +
Lea
sing
Arr
ange
men
t +
Stat
e +
Oth
er S
ourc
es.
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Ave
rage
Sec
tor
Gro
wth
G
row
th r
ate
of f
irm
s av
erag
ed a
cros
s ea
ch s
ecto
r in
eac
h co
untr
y.
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Kve
tch1
Thi
s is
the
devi
atio
n of
eac
h fi
rm’s
res
pons
e fr
om th
e co
untr
y m
ean
to th
e qu
estio
n “H
ow h
elpf
ul d
o yo
u fi
nd th
e ce
ntra
l./na
tion
al g
over
nmen
t tod
ay to
war
ds b
usin
esse
s lik
e yo
urs-
(1)
Ver
y he
lpfu
l (2)
Mild
ly
help
ful (
3) N
eutr
al (
4) M
ildly
unh
elpf
ul (
5) V
ery
unhe
lpfu
l (6)
Don
’t k
now
?”
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Kve
tch2
Thi
s is
the
devi
atio
n of
eac
h fi
rm’s
res
pons
e fr
om th
e co
untr
y m
ean
to th
e qu
estio
n “H
ow p
redi
ctab
le a
re
chan
ges
in e
cono
mic
and
fin
anci
al p
olic
ies-
(1)
Com
plet
ely
pred
icta
ble
(2)
Hig
hly
pred
icta
ble
(3)
Fair
ly
pred
icta
ble
(4)
Fair
ly u
npre
dict
able
(5)
Hig
hly
unpr
edic
tabl
e (6
) C
ompl
etel
y un
pred
icta
ble
(7)
Don
’t
know
”
Wor
ld B
usin
ess
Env
iron
men
t Sur
vey
Appendix A2 : Preliminaries on Directed Acyclic Graphs
Directed Acyclic Graphs (DAGs) help us in model selection especially in the absence of well defined theory that can point to regressors that need to be included in a multivariate linear regression. The model selected by DAGs can then be submitted to a standard regression analysis for parameter estimation. The DAGs themselves provide a compact representation of joint probability distributions with the nodes of the graphs representing the random variables and the (lack of) edges connecting the nodes representing conditional independence assumptions. We describe below the assumptions behind linking probability dependence/independence relations to causal inference and illustrate how the software program TETRAD produces a causal pattern from raw data and conclude with a specific example of how supplementing regression analysis with DAGs can be useful and provide more accurate results.
A directed acyclic graph (DAG) is a picture or a path diagram representing causal flow between or among a set of variables. For example, given a set of three vertices: {X1, X2, X3}, and a set of two edges among these vertices: {X1 X2, X2 X3}, the corresponding DAG would be:
X1 X2 X3. For the above DAG to be ascribed causal inference, we need the Causal Markov
Condition. Formally, the Causal Markov Condition states that for a variable X and any set of variables Y that does not include the effects of X, X is probabilistically independent of Y conditional on the direct causes of X. The intuition behind the Causal Markov assumption is that each variable is independent of all other variables that are not its effects, conditional on its immediate causes. So the above DAG implies that X3 is independent of X1 conditional on X2. (In graph theory, the equivalent of the Causal Markov Condition is referred to as d-separation Pearl (1988)). The Causal Markov Condition would also assert that if X and Y are related only as effects of a common cause Z, then X and Y are probabilistically independent conditional on Z.
How TETRAD Works The intuition behind discovering a causal pattern from observational data is that, under the Causal Markov condition, observed patterns of statistical independence limit the number of possible causal graphs compatible with observed data. Consider again the above example with variables X1, X2 and X3, where, say, we observe from the data that X1 and X3 are independent conditioning on X2. This observation implies that the causal graph
X1 X2X3 is incompatible with the data, since if X1 and X3 were both causes of X2, then conditioning on X2 would render X1 and X3 statistically dependent. The causal graphs that are compatible with the observed independence pattern include the one we saw before
X1 X2 X3 as well as
X1X2X3 and X1X2 X3. Software tools such as TETRAD take observed data, either in raw form or as correlations (and the independence conditions they embody) as input, and use algorithms to search for all compatible graphs. The number of graphs are significantly reduced, (maybe to even one) with added assumptions based on prior theory or knowledge of temporal order of the variables (in our case, we assumed all the business environment variables cause growth) or axiomatic assumptions such as:
(a) Faithfulness (or Stability): Assuming that a population is Faithful is to assume that whatever independencies occur in it arise not from incredible coincidence but rather from structure. If there are any independence relations in the population that are not a consequence of the Causal Markov condition, then the population is unfaithful. For
56
instance, if in the above example we had {X1 X2, X2 X3 and X1 X3}, applying the Causal Markov Condition gives no independence relations. However, by coincidence X1 could be independent of X3 (X1 has a negative direct effect on X3 but X1 has a positive effect on X2 which has a positive effect on X3. If the direct and indirect effects of X1 on X3 exactly cancel each other, then there will be no association between X1 and X3). In such a case, the population is said to be unfaithful to the causal graph that generated it.
(b) Causal Sufficiency: Causal Sufficiency is satisfied if we have measured all the common causes of the measured variables.
An Example
To illustrate the algorithm and the link between graphs and probability conditions, consider the possible interrelations between Firm Growth and three reported obstacles, Financing, Taxes and Regulation, and Judicial Efficiency. The undirected graph below shows all possible relations between these variables. An edge or path between two variables indicates that the two variables may be dependent.
The TETRAD search algorithm begins by assuming that all variables in the model are
dependent, corresponding to the graph shown above. It then checks for conditional independence relations between the variables and depending on the relations found in the data, the edges between the variables are oriented. Knowledge of temporal precedence (for example that the three obstacles affect growth and not the other way around) allows for limiting the number of tests for conditional independence. For the purpose of this example, we will not impose any temporal ordering. Suppose the following independence relations (and no other) are found in the data at a certain significance level: {Taxes & Regulation} and {Judicial Efficiency} are independent (1) {Growth}and{Taxes & Regulation} are independent conditional on the value of Financing (2) {Growth} and {Judicial Efficiency} are independent conditional on the value of Financing (3) Note that these are the only exhaustive set of independence relations found in the data and all other dependencies assumed originally remain. Under the assumption of the Causal Markov Condition and Faithfulness, the above independence relations imply the following:
Taxes and Regulation
Judicial Efficiency
Growth Financing
57
Equation (1) implies that there is no edge connecting {Taxes & Regulation} and {Judicial Efficiency}. Equation (2) implies that there is no edge between Growth and Taxes and any dependence between them is only through Financing. The following patterns would be consistent with equation (2): (i) Taxes & RegulationFinancing Growth OR (ii) Taxes & RegulationFinancingGrowth OR (iii) Taxes & Regulation Financing Growth The pattern that is inconsistent with equation (2) is Taxes & Regulation Financing Growth since it violates the Causal Markov Condition. Similarly Equation (3) implies that there is no edge between Growth and Judicial Efficiency and any dependence between them is only through Financing. The following patterns are consistent with Equation (3): (iv) Judicial EfficiencyFinancing Growth OR (v) Judicial EfficiencyFinancingGrowth OR (vi) Judicial Efficiency Financing Growth Again, the pattern that is inconsistent with equation (3) is Judicial Efficiency Financing Growth since it violates the Causal Markov Condition. Further, since (1)-(3) are the only set of independence relations found in the data, we know that none of the following patterns are possible between {Taxes & Regulation} and {Judicial Efficiency} with respect to a third variable since they all imply that {Taxes & Regulation} and {Judicial Efficiency} are independent conditional on Financing (or Growth), which is not one of the independence relations found in the data. (vii) Taxes& RegulationFinancingJudicial Efficiency OR (viii) Taxes& RegulationFinancing Judicial Efficiency OR (ix) Taxes& Regulation Financing Judicial Efficiency OR (x) Taxes& RegulationGrowthJudicial Efficiency OR (xi) Taxes& RegulationGrowth Judicial Efficiency OR (xii) Taxes& Regulation Growth Judicial Efficiency Working only with the set of compatible patterns implied by the independence relations ((i) to (vi), we have the following candidate graphs:
58
Case (a): Combination of (i) and (iv) Taxes & Regulation Financing Growth Judicial Efficiency Case (a) is not possible since it encompasses the incompatible case of (viii) Case (b): Combination of (i) and (vi) Taxes & Regulation Financing Growth Judicial Efficiency Case (b) is not possible since it encompasses the incompatible case of (vii) Case (c): Combination of (ii) and (v) Taxes & Regulation Financing Growth Judicial Efficiency Case (c) is not possible since it encompasses the incompatible case of (viii) Case (d): Combination of (iii) and (iv) Taxes & Regulation Financing Growth Judicial Efficiency Case (d) is not possible since it encompasses the incompatible case of (ix) Case (e): Combination of (iii) and (vi) Taxes & Regulation Financing Growth Judicial Efficiency Case (e) is the only combination that works and doesn’t include any of the incompatible cases in (vii) to (xii). Thus from the conditional independence relations found in the data, Tetrad has identified that Financing has a direct effect on Growth while Taxes and Regulation and Judicial Efficiency are indirect effects in that they affect Growth only through Financing.
59
DAGs versus Regression Analysis The strengths of Directed Acyclic Graph (DAG) methodology over regression analysis
lies in its ability (i) to distinguish genuine from spurious correlations in a set of data (ii) to identify which variables need to be included in a model to accurately measure one variable’s effect on another and (iii) to differentiate between direct and indirect effects of different variables. When two variables are correlated, it could mean one of four things: random variation or one variable causes the other or they have a prior common cause or that we have conditioned on a common effect of the two variables (which induces a correlation between the two causes). While identifying the source of correlation is not possible in an ordinary least squares regression analysis, it is possible through the Directed Acyclic Graph methodology that relies on the concept of the Markov Condition (or d-separation) to identify which variable is a true cause or predictor of the outcome variable.
To illustrate this, consider the following example from Spirtes et. al (2000) where we have the following linear structure (unknown to us) between the outcome variable Y and regressors X1, X2, X3, X4, and X5 where X3 and Y share a common unmeasured cause Z :
X1 X2 X3
X4 Y X5
The same can be represented by the following set of equations: Y = a1X1 + a2X5 + a3Z + eY (A1) X1 = a4X2 + a5X4 + e1 (A2) X3 = a6X2 + a7Z + e3 (A3) A linear multiple regression of Y on the X variables will give all variables in the set {X1, X2, X3, X5} non-zero coefficients, even though X2 has no direct influence on Y and X3 has no direct/indirect influence on Y. Linear regression takes a variable Xi to influence Y provided the partial correlation of Xi and Y controlling for all of the other X variables does not vanish. However, this is a sufficient condition only when there is no X variable that has a common unmeasured cause with Y (such as X3 in the example above) or is an effect of Y. Indeed, regressing Y on all subsets of the X’X matrix would show that X2 and X4 will not be significant for some combination(s) of other X variables entered as regressors.
However, application of the PC algorithm correctly selects {X1, X5} as the variables that directly influence Y. Again, comparing to the regression analysis, these are the only two variables that will have significant coefficients in regressions of Y on X, regardless of which X variables are entered as regressors. Further the Directed Acyclic Graph is also able to point that X3 and Y may be driven by a common cause and that X2 and X4 indirectly affect Y though their effect on X1. Thus, the Directed Acyclic Graph points to a possible pattern of causal influences between the different variables and hence can be very useful as a starting point in model selection.
Z
top related