Institutional determinants of capital structure choice; evidence from world-wide panel data

Institutional determinants of capital structure

choice; evidence from world-wide panel data

Author

P.J.J. Melching

Supervisor

Dr. A. Plantinga

July, 2008

UNIVERSITY OF GRONINGEN

Faculty of Economics & Management and Organization

Business administration MSc Finance

ABSTRACT

In this paper the influence of institutional variables on leverage ratio on a world level as well as on country level has been analyzed. Using panel data with a sample of 532 firms over six continents, multiple regression analyses have been performed. In total 13 variables were examined including six institutional variables and seven control variables based on the pecking order and trade-off theory. The evidence indicates that on a world level the legal system, the level of political stability, the degree of contract enforcement and the development of equity markets significantly affect capital structure choice. Moreover it can be concluded that the relevant institutional variables differ between developed and developing countries both in explaining power and in relevant institutional factors.

Pieter J. J. Melching

P.J.J.Melching@rug.nl

Preface

The last 6 months I have been writing my master thesis as a final assignment of my Msc Business Administration Finance at the University of Groningen. With a break of 3 months in which I have done an intern at Rabo Securities I have been studying the influence of institutional variables on capital structure choice.

There are a few people who I especially want to thank for their help and support during the writing process. Firstly, I have to thank my supervisor, dr. A. Plantinga which provided me with some very useful tips, comments and insights. Secondly, I would like to thank my family, friends and my girlfriend for their support.

I very much enjoyed the period in which I have written my thesis. The process was very enriching and a useful experience for my future career.

Table of contents

2.2 Pecking order theory………....12

2.3 Trade-off theory……….. 13

2.4 Pecking order, trade-off theory mutually exclusive………...15

3. Variable construction and justification……….... 16

3.1 Defining institutional variables……… 17

3.2 Defining control variables………... 16

3.3 Defining dependent variables……….. 20

3.4 Coefficient signs hypothesized……….21

4. Data………. 22

4.1 Sample design and availability………..22

4.2 Outlier analysis………22

4.3 Institutional variables………...23

4.4 Company specific variables………..24

4.5 Correlating variables………... 27

4.6 Summary data………..27

5. Methodology………...30

5.1 Hypotheses………..30

5.2 Panel data regression………... 30

5.3 Regression analysis……….. 33

6. Empirical results……… 35

6.1 Outlier analysis………35

6.2 Main results cross-firm regression with world sample………... 35

6.3 Standard error clustering………..37

6.4 Sub-sample analyses……… 40

6.5 Discussion of the results………... 43

7. Conclusion……….. 47

7.1 Caveats, limitations and recommendations……… 48

8. References………...50

List of figures and tables

Table 1. Predicted signs……….21

Table 2. Descriptive statistics institutional variables………... 23

Table 3. Descriptive statistics dummy variables………. 24

Table 4. Descriptive statistics company specific variables……….25

Table 5. Excluded firms from sample……….. 28

Table 6. Geographical spread sample……….. 29

Table 7. Definition variables………. 32

Table 8. Calculation method for standard errors……….33

Table 9. Pooled cross-firm regression……….. 36

Table 10. Pooled cross-firm regression with standard errors clustered………. 38

Table 11. General least squares regression with random firm effects……… 39

Table 12. Pooled regression across sub-samples……….41

Table 13. Significance levels for hypothesized coefficient signs……… 46

Appendix 1. Literature overview……….. 56

Appendix 2. Correlation matrix……….59

Appendix 3. Relative country weights in sample……… 60

Appendix 4. Outlier analysis………. 61

1. I

ntroduction

The capital structure debate has had extensive attention over the last 50 years. Starting in year 1958 with Nobel Prize winners Miller & Modigliani, who argued that (under strong assumptions) the way a company is financed does not affect the firm value. However many researchers in the decades after this publication have provided evidence that capital structure choice does affect the value of a firm. Two different theories evolved through time: the pecking order theory and the static trade-off theory. The trade-off theory assumes that capital structure choice is based on a trade off between the present value of tax benefits of debt and the present value of the cost of financial distress. The pecking-order theory states that firms prefer some finance opportunities above others due to asymmetric information between insiders and outsiders of the company. Furthermore this theory suggests that there is no optimal debt level for a company. These theories both use different variables in explaining capital structure. The trade-off theory includes (among others) the degree of tangibility of assets, product uniqueness, size, profitability and non-debt tax shields, (Titman et al, 1988), (Kale et al, 1991), (Sayillan, 2002). Whereas the pecking order theory focuses on the flows of cash variables like capital expenditures, dividend payout and operating cash flow Shyam-Sunder (1999).

Both theories have been proven to be valuable in explaining capital structure choice. However there is an extensive discussion about which theory has the most explanatory power, e.g. Fama et al (2002). In this research the focus is not on the company specific determinants of capital structure, but on the institutional environment. For a long time the company specific variables were seen as the only determinants of capital structure, however since the start of the 90’s the institutional environment –as a determinant of capital structure- has been investigated as well.

behind their decisions are financial flexibility and the impact of capital structure on financial statements. Moreover he provided evidence that managerial opinions do differ when the legal and financial institutions vary and that financial flexibility is less of concern for managers in countries with a market-based system in comparison to managers in a bank-based system. In this thesis we investigate the world-wide institutional determinants of capital structure on firm level and across developed and developing countries. The main questions which arise therefore:

Is the institutional environment, in terms of: economic development, legal system and political stability, a determinant of optimal capital structure choice world-wide?

Are there persistent differences in the relevant institutional variables between developed and developing countries?

In order to answer the questions a 3-dimensional panel data set has been used, which allows for differences over time and cross-section. 532 Firms in 35 countries over six different continents have been included. The included variables are partly based on existing literature. Rule of law, measures the extent to which government is able to enforce contracts and how well court is functioning. Two economically related variables have been included: the development of the equity market in terms of turnover and total market capitalization. These indicators measure the extent to which companies have access to external equity capital markets in their home country. Lastly two dummies are included one for the legal system (common vs. civil law) and another for the banking system (market vs. bank based).

2.

Literature review

In the 50’s of the last century the debate about optimal capital structures began. Mogdiliani & Miller (1958) (hereafter: MM) investigated this phenomenon for the first time. They concluded that the way a firm is financed does not affect the value of a particular firm. However they created a model with strong simplifications. For example they assumed that there were no corporate/personnel taxes, no transaction costs, complete contracting, perfect information and symmetric information. Assumptions under which internal and external financing are perfect substitutes.

However when relaxing these assumptions it might be possible that the way a company is financed indeed affects firm value. During the last three decades of the last century two dominant views tried to capture the determinants of capital structure: the pecking order theory and the trade-off theory. Those two will be discussed in the following section because they function as the main theories behind the control variables used. The main focus of the paper – the impact of the institutional environment on capital structure choice- will be discussed first.

2.1 Institutional differences

sector in a country does not have any influence on the capital structures of large firms. Booth et al (2001) examined capital structure decisions in 10 different developing countries. Their research focuses on the question whether it is relevant to know the home country when explaining capital structures. They concluded that there are significant differences between the relative relevance of the institutional effects across countries. Among others it seems that business risk (standard deviation of cash flows) is much larger in Brazil than in Thailand. Therefore they conclude that among countries there are persistent differences, which signals that different institutions play a significant role. This conclusion is very similar to the conclusions of Beattie (2006). Jensen et al (1976) investigated information asymmetries between shareholders and debt holders of firms. In a sample with different countries he concluded that the institutional context of a firm has a significant impact on the agency cost a firm faces. Lee et al (1988), investigated the difference in capital structures between American multinational firms and domestic firms, he expected differences due to the different institutional context in terms of political (in)stability, exchange risk and agency cost differences. He concluded that agency costs for multinationals were smaller and that they tended to be less leveraged than domestic firms. He also suggested that agency cost are higher in political instable countries.

found that the determinants of capital structure in China are largely the same as in Western countries and in line with the assumptions of the trade-off theory. However they found as well that Chinese firms hold less debt and more equity compared to their counterparts in the ’Western’ world. One possible explanation they found is the institutional context of China. Due to the infant debt market the only lending option is created by banks and equity market not through the bond market. Another explanation they suggested is that firms favour the equity market rather than the debt market due to the favourable high stock prices. The non-binding character of debt could be another incentive for the management as well. On the other hand Green et al (2004) concluded that there is no conclusive evidence for either the trade-off nor the pecking order theory in China.

Although there is no consensus about significance and magnitude the magnitude of institutional effects, it seems that political (in)stability, capital market development (in terms of size and turnover) and legal system might have an impact on capital structure choice, these explaining variables will therefore be used in the research of this paper.

2.2 Pecking order theory

The pecking order theory suggests that firms prefer some finance options above others due to asymmetric information between insiders and outsiders of a company. The basic difference with the trade-off theory is that the pecking order theory suggests that there no optimal capital structure and that possible structure differences arise when there are not enough internal funds to finance investments. The model uses different dependent variables which describe the flow of funds and investment opportunities, including: free cash flow, capital expenditures, dividend pay-out and past profitability.(Huang et al, 2006).

Meyers et al (1984) suggest that firms prefer internal over external finance and when projects are externally financed they prefer debt over equity. Moreover they argue that there is no optimal debt level, they acknowledge that variables describing the trade-off model are relevant, however they suggest that those are of less importance.

less this is possible to occur. More debt means a higher fixed claim on free cash flow, which makes the free cash flow available to invest suboptimal smaller. This is one of the benefits of debt financing, (Jensen et al, 1986).

The second conflict between debt holders and equity holders emerges due to the difference in debt- and equity contracts. Debt holders receive a fixed pay, not based on performance measures. On the other hand, equity holders receive a pay based on the profit of the company. If an investment yields a high return (above the face value of debt) most of this premium is captured by the equity holders. However if the investment fails the debt holders pay the price. In other words the equity holders capture the premium of risky investments while debt holders bear the risk. This phenomenon leads to sub-optimal investments by equity holders in the form of investing in risky projects even when they are value decreasing for the firm. Conversely the effect works the same; when a company is likely to go bankrupt, there is no incentive for equity holders to invest in value increasing projects, because the largest part of the earned profit will go to the debt holders (Meyers, 1977).

Another incentive for managers to choose debt above equity is called the signalling effect. Narayanan (1988) and Ross (1977), suggest that lower quality firms have a higher marginal bankruptcy risk and therefore lower debt levels. In contrary, high quality firms can therefore hold higher debt levels due to a lower bankruptcy risk, higher debt levels signal a higher quality of the firm. These higher debt levels cannot be imitated by low quality firms, therefore it is a powerful signal. Castanias (1983) provided somewhat similar evidence; he investigated the failure rates in different industries, he found that ‘lines of business’ with relatively higher failure rates have lower debt levels. He concluded that the ex ante bankruptcy cost is high enough to impose that there are enough incentives and prospects to hold an optimal mix of debt and equity. Analyses by Heinkel et al (1990) concluded that part of the optimal capital structure is preferred stock because this enhances a firm’s debt capacity. This is a favourable situation because more debt leads to more first optimal investment decisions. Lastly Harris et al (1991) contributes to this topic in another way, in his paper he concludes that more debt has a benefit due to the option of debt holders to liquidate investments.

2.3 Trade-off theory

of business risk on capital structure. They used two variables in their model: size and industry. For both variables they did not find conclusive evidence, in their cross country analysis equity was not significantly higher for larger companies. Moreover they found that for every country the debt level of smaller firms is larger than for larger firms. Balikrisham et al (1993), also concluded that debt ratios do not significantly differ across industries. They provided evidence that firm specific assets and skills are more important determinants of capital structure decisions.

Demirgüç-Kunt et al (1988) suggest that the character of the product market of a company influences the debt ratios. They concluded that companies active in product markets with elastic demand for products will have higher debt levels, additionally they concluded that firms which produce products that require service and firms for which the reputation of producing high quality products is important will tend to hold lower debt levels.

2.4 Pecking order – trade off mutually exclusive?

3.

Variable construction and justification

In this section the variables used in this thesis will be discussed based on previous work in this field of research. Their use will be explained and justified. Section 3.1 explains and justifies the institutional dependent variables used. The company specific variables are discussed in section 3.2, lastly section 3.3 contains the dependent variable used.

3.1 Defining institutional variables Development (activity) stock markets

Booth et al (2001), Beattie et al (2006) and Huang et al (2006), argue that economic institutional differences play an important role in capital structure decisions. They conclude that besides firm specific characteristics the institutional economic environment plays an important role. Demirgüç-Kunt et al (1999) included variables which characterize the stock markets in different countries. they use a development indicator defined as

t i t i GDP tcap Stockmarke , ,

(marketcapitalization_i,t). To measure the activity of the equity they use

the turnover ratio of the stock market defined as

t i t i tcap Stockmarke traded Totalvalue , , _{,( )} ,t i turnover . Although their results of the turnover ratio were not significant they will be included in this study.

Bank-based versus market based system

Large differences in financial systems among countries occur. Although there is no consensus about which is optimal it might have some effect on the type of external financing companies choose. The key distinction between the two systems is that market-based systems especially focus on the well functioning of security markets, whereas bank-based systems especial especially focus on financial intermediation, (Levine, 2000). It is therefore rather straightforward that the different systems might impose an effect on capital structure choice. A dummy is included which is 1 in case of a bank-based system and 0 otherwise,

t i

banksystem

D( )_, .

Legal environment

correlation between long term debt holdings and the efficiency index. In this study we do not use such an index due to data unavailability. We will use Rule of law (Kaufmann et al 2006) which especially measures the extent to which agents have confidence in the quality of contract enforcement and court. It will be defined as ruleoflaw_i,t.

Common versus civil law

Different countries have diverse legal systems, either more or less developed. LLSV (1997, 1998, 1999) investigated different legal systems in several countries. They identified several legal systems: common law (e.g. U.K.) which provides better legal protection to shareholders and creditors compared to civil law countries (e.g. France) which provide significantly less protection. A dummy variable will be included which describes the state of the legal system in a particular country, which is 1 in case of a common law countries and 0 otherwise,

t i m legalsyste D( )_, . Political stability

Already mentioned in section 2, the political (in)stability can have a significant impact on capital structure choice. Lee et al (1988) argue that increased political risk might lead to a higher variability of cash flow (risk). As a result, higher cash flow variability leads to an increased risk for debt holders and therefore it is possible that creditors want better terms. This might lead to more equity financing. Krainer (1969) refers to it as a ’’repatriation risk’’. Philips Patrick (1991) argues that firms with more assets in place rather than intangible assets are more vulnerable for political instability. This can have consequences for the chosen capital structure policy the variable is defined as: Political_i,t. And measures the perceptions and likelihood that the government will be destabilized or overthrown by unconstitutional or violent means, (Kaufmann, 2006).

3.2 Defining control variables

Size

The variable size is included because several papers suggest that size is an important determinant of the amount of leverage, (Archer et al, 1966). On the one hand firms with a larger value seem to be less vulnerable to bankruptcy due to a more diversified sales portfolio. Moreover when the value of a firm decreases, the bankruptcy cost tends to become a larger part of that firms value. Therefore one might expect that larger firms will have more leverage due to relatively lower bankruptcy cost. On the other hand issuing equity is much more expensive for smaller firms and therefore it is possible that smaller firms use more leverage. Smith (1977) shows that the size and the relative cost of an equity issue are inversely related. The variable size will be defined in the model as the natural logarithm sales LN(sales)_i,t. Tangibility of assets

When a firms’ assets have collateral value it is possible to issue secured debt. This option is less expensive than unsecured debt. The cost associated with an unsecured debt issue can be related to an asymmetric information problem Meyers et al (1984), when (like explained in previous section) managers have more information than investors. There are cost associated with that information problem, however when debt is secured, these cost are lower. When debt is secured the manager is expected to use the money for a specified project, this guarantee cannot be given with non-secured debt. Therefore creditors might want better terms when such a guarantee cannot be given, this creates another incentive to use equity in stead of debt, (Jensen et al, 1976). On the other hand managers have a tendency to use perquisites (see section 2) Jensen (1986) states that when equity levels are high (debt low), due to close monitoring by bondholders this problem can be resolved. Therefore also companies with fewer options to secure debt might want to hold more debt to resolve the perquisite problem by the manager. The tangibility of assets is measured as property, plant and equipment (hereafter: PP&E) as a function of total assets

Totassets E PP &

and will be defined in the model as

t i

gibilty_,

tan . Profitability

external. The profitability is defined as a function of total assets t i t i s Totalasset Ebitda , ) 1 ( ,  the definition

in the model is profitability_i,(t1). The reason to choose for Ebitda rather than net income lies in accounting difference between countries and the large amount of different industries in the sample. Firstly, due to accounting differences among countries, some firms in the sample can expense more debt and assets in the form of depreciation and amortization than others. Which leads to different net income numbers, with the use of Ebitda this problem is overcome, because it measures operating income before depreciation and amortization. Secondly, different industries have different depreciation ratios depending on the amount of fixed assets. Taking operational income before depreciation beats this problem

Non-debt tax shields

Non-debt tax shields (hereafter: Ndt) can be defined as every tax deductible cost except those associated with debt (e.g. depreciation), investment tax credit and tax loss carry forwards). Bowen et al (1982) conclude that the level of Ndt shields are an important determinant of capital structure. The more non-debt tax shields the lower the relative value of debt-tax shields. Therefore one might expect a negative relation between debt and Ndt shields. However the value of Ndt’s is many times undervalued in previous work on this topic, in many studies the calculation of the Ndt’s is restricted to the sum of depreciation, amortization and selling expenses, (e.g. Huang et al 2006, Sayiligan et al, 2002). With this calculation many tax deductible costs are ignored. Therefore this paper will define this variable broader, in the form of Sales-Ebit. In this way all the tax deductible expenses are captured. As a result the independent variable will be defined as

t i t i t i s Totalasset Ebit Sales , , ,  , in the model Ndt_i,t.

Operating cash flow, capital expenditures

in operating cash flow before interest and taxes (as a measure of internal finance possibilities) is measured as 1 , 1 , ,    t i t i t i Cashflow Casflow Cashflow t i cashflow_,

 and capital expenditures (as a measure of investment opportunities) as t i t i s Totalasset Capex , , _, t i

Capex_, are included in the research.

Financial slack

Financial slack –liquid assets or reserve borrowing power- is valuable for companies. It is valuable because it gives a company the possibility to invest in positive NPV projects when they need to, moreover it prevents underinvestment. Therefore companies may issue equity to acquire a ‘good’ financial slack position and do not always issue for direct investment, Meyers (1984). The relation between leverage and financial slack is measured as the difference in working capital Wc$_i,t measured in Dollars and as a function of total assets

t i Wc_,  . Wc_i,t WC_i,(t1), t i t i s Totalasset Wc , , _.

3.3 Defining dependent variables

Leverage ratio

In this paper total debt as a function of total assets is used as leverage ratio,

t i t i s Totalasset Totaldebt , , _, t i

leverage_,. This is a good proxy for changes in leverage ratios, for example Haung et al (2006) use this as a dependent variable and concluded that it is a good alternative for measuring leverage changes within firms.

3.4 Coefficient signs hypothesized

Signs based on the relation with the dependent variable; debt over total assets

Table 1

Predicted signs

Development stock market Turnover stock market Bank system dummy (1) Rule of law

Dummy civil law dummy (1) Political stability

Trade-off theory Pecking order theory

Size +

-Profitability +

-Non-debt tax shields

-Tangibility +

Operating cash flow -

4.

Data

4.1 Sample design/availability

To investigate whether the chosen institutional variables have an impact on capital structure choice we used a world sample. Due to the limited data availability in Thomson Datastream we could (only) include 47 countries2. In order to construct the company sample similar to a total world sample we randomly picked listed companies from the available constituency lists for every available country, with a random generator. The amount of companies included per country depended on the number of listed companies in that particular country, relative to the total amount of listed companies in the sampled countries:

% 100 * % ompsample Totlistedc Listedcomp ry firmscount x x 

The number of listed firms is based on the number of firms per country listed on the constituency lists of Datastream. Finally 1,000 firms were selected however due to the difference in structure and presentation of company specific items in the financial sector all 157 financials (i.e. Banks, insurance companies) were excluded from the sample. To be included in the sample all data (institutional as well as company specific) had to be available on firm and country level.

4.2 Outlier methodology

An intensive debate about outliers has been part of literature. The problem with outliers lies in the fact that they can increase error variance and reduce the power of statistical tests (Zimmerman, 1994). In general there are two forms of potential outliers; data errors and outliers which are legitimate (outliers which have extreme values and affect the mean significantly but which are a legitimate part of the dataset). The first type should always be removed from the dataset, but with the second type simple removal is not always the best option. When a data point is a legitimate part of the sample simple exclusion could lead to the loss of valuable data.

methodology; the extreme data points at both ends of the sample are removed until the mean does not significantly differ with one residual removal, (Anscombe et al, 1960). However like discussed above this can lead to the exclusion of valuable data points. Therefore two datasets will be analyzed one with the inclusion of the legitimate outliers (n=596) and one without (n=532). In this way it is possible to see if the outliers have a substantial effect on the coefficients in the regression equation using Cook’s Distance (Cook, 1977). This measures the change in the regression coefficient with and without legitimate outliers.

4.3 Institutional variables

Institutional variables were obtained on country level via the databases of the worldbank and Datastream. Those variables are dynamic over time and again countries with no complete dataset were excluded from the sample. The fact that rule of law and political stability are measured dynamic over time in stead of static is very important, due to the volatility of these indicators over time in some countries, (Kaufmann et al, 2006). The other institutional variables were obtained via Thomson Datastream, the descriptive statistics of those are shown below.

Descriptive statistics institutional variables with inclusion of all countries (n=43); in parentheses descriptive statistics for data corrected for data errors and legitimate outliers on country level (n=35)

Market capitalization is a measure of the capital market development in a particular country. The median and average are pretty close, the maximum is Switzerland and the minimum is Poland the removal of those has no significant impact on the average. The volume traded indicator has one large outlier of 451.106 from Turkey. This outlier is obviously a data error

Table 2

Descriptive statistics institutional variables

Variable Median Average Quartile 1 Quartile 3 Maximum Minimum

Market cap 0,404 0,505 0,196 0,730 2,495 0,015

Turnover 0,023 7,625 0,002 0,072 451,106 0,000

(0.022) (0.053) (0.002) (0.067) (0.473) (0.000)

Rule of law 0,792 0,739 0,011 1,732 2,025 0,005

type and therefore the two remaining Turkish firms have been removed from the dataset. The other two variables show no important statistics which should be discussed.

The descriptive statistics of the dummy variables which indicate the legal and economic system of a country are presented below.

The sample includes significantly more civil than common law countries, one possible reason is that in Europe (which covers almost half of the sample) civil law is dominant. For the economic dummies the different systems are more evenly distributed 53.7% bank system against 46.3% market system.

4.4 Company specific variables

The sample contains 596 (532) firms of which there are data in 5 consecutive years. In the table below the descriptive statistics of the company specific variables are presented. The table includes the median, mean, quartile 1 & 3, the maximum and the minimum. In this way it is possible to see how the data are spread over quartiles and how outliers affect the average. Data errors are removed and with the trimmed mean methodology legitimate outliers are excluded. In parentheses are the new statistics, (without legitimate outliers and data errors) with the exclusions based on the bases of the above discussed criterions.

Table 3

Descriptive statisitics dummy variables

No of countries % of total sample

Table 4

Descriptive statistics company specific variables

Variable Median Average Quartile 1 Quartile 3 Maximum Minimum

Leverage ratio 24,7% 26,1% 10,7% 38,3% 388,3% 0%. (24.8%) (25.9%) (11.3%) (38.1%) (93.6%) (0.0%) Net sales $1,3b $5,9 b $437m $4,5b $201b $34k ($1,4b) ($6,1b) ($489m) ($4,7b) ($201b) ($7,5m) 21,07 21,11 19,95 22,24 26,03 10,43 (21.1) (21.1) (20.0) (22.4) (26.0) (15.8) Profitability 11,0% 11,6% 7,1% 15,7% 84,9% -72,2% (11.4%) (12.1%) (7.6%) (16.2%) (63.3%) (-72.2%) Tangibility 32,6% 35,8% 18,0% 49,7% 98,9% 0,4% (32.5%) (36.9%) (19.3%) (50.6%) (98.9%) (0.6%) Ndt 84,5% 95,5% 50,1% 120,5% 611,7% -61,0% (84.7%) (95.8%) (51.6%) (120.4%) (611.7%) (-1.3%) Δ Cash flow -4,1% 12,2% -25,3% 38,5% 72865,0% -29711,0% (5.8%) (10.9%) (19.7%) (37.7%) (785.5%) (-931.8%) Capex 4,2% 5,6% 2,4% 7,1% 57,5% 0,0% (4.5%) (5.8%) (2.6%) (7.5%) (57.5%) (0.0%) Δ Wc 1.2% 1.4% -2.4% 5.0% 85.7% -90.3% (1.1%) (1.3%) (-2.4%) (4.8%) (85.7%) (-90.3%) Δ Wc$ $10,3m $635m $-27,4m $89,8m $266b $-86b ($9,6m) ($60,9m) ($-27,9m) ($87,5m) ($16,9m) ($-16,9b) Ln(sales)

Descriptive statistics company specific variables with inclusion of all firms, (n=615), in parentheses; descriptive statistics for data corrected for data errors and legitimate outliers (n=532)

The dependent variable in this research shows no extraordinary statistics, except for one data error. The maximum value of debt over assets is 388.32%, which is more than 100% this is clearly not possible. As a result the extreme value has been removed from the sample. The tangibility of assets measured as property, plant & equipment over total assets, shows a median and an average between 32% - 35% which is rather normal. The fact that the median and average are close to one another means that the data are not affected by many large outliers in the same direction. The maximum of 98,88 % is oil and gas firm with a high level of tangibility due to e.g. oil platforms The minimum of 0.33% is from Immobiliere de Belgique, a real estate development company with less own fixed assets. No data errors were found and moreover the removal of the extreme values did have a significant effect on the mean.

data points did not sort out a significant effect on the overall average. The Ndt shields have a quite larger difference between the median and the average which could be an indication of outlier influence. There is also a much larger quartile distance than previously noticed. However the trimmed mean methodology did not result in any influence on the average. Therefore the large spread in the data can be accepted. The large spread explanation is rather straightforward the sample consists 16 different industries with all different profit margins and related to that different expenses therefore the numbers in the sample are more extremely distributed and the quartile distance larger. However the minimum value is rather extraordinary due to the fact that it is negative. Where (Sales-EBIT) as well as total assets can never be negative the non-debt tax shield can never be negative as well. This indicates that there are some data errors in the sample. Data with negative values for Ndt shields are useless, 6 companies had to be removed on this basis. The descriptive statistics of investment in fixed assets in terms of capital expenditures over total assets have a small quartile distance, and a mean and median within 1.5%. The difference in capex between firms is related to for instance acquisitions by a firm in a particular year. For example the maximum of 57.5% is due to an acquisition by Lucky cement of a new textile mill in 2005, no exclusions were done because it sorted out no significant effect on the overall average. The independent variable Δ cashflow has substantial data error outliers and legitimate outliers which affect the mean, therefore these outliers were removed, 50 firms were excluded from the sample. The last two variables Δ wc$ & Sales are absolute values unlike the other variables which were calculated as ratio’s. These absolute values were obtained via Datastream in their respective local currencies. To make these variables cross-country comparable all the local currencies were diverted to US Dollars, with middle exchange rates (exchange rates which hold the midpoint between the bid rate and the offered rate)3. The Netsales have a rather large mean-median and quartile distance which like discussed before could indicate the presence of substantial outliers. However the maximum net Sales is from one of the world’s largest firms General Electric, after double checking the total revenue it became clear that this was no data error, nor a legitimate outlier. On the other hand the minimum value of 34K is a data error and the firm is therefore excluded. Based on the trimmed mean methodology the exclusion of residual extreme values did not have a significant effect on the average. as a result no remaining data were excluded from the sample. Contrary to Net sales the absolute variable Δ

WC$ does have 14 legitimate outliers all 14 were expelled from the sample on the basis of the trimmed mean methodology.

4.5 Correlating variables

In a model where the goal is to understand how the different independent variables impact the dependent variables, multicollinearity is a problem. Multicollinearity can affect the outcomes; it can lead to larger standard errors for the involved variables. This might lead to type II errors; a rejection of H , while the coefficients are significant. The correlation matrix for the ₁ variables used is included in appendix 2. Some of the institutional variables are highly correlated, banking system with legal system have a correlation of 0.675, which is rather high. The reason however is straightforward and in line with LLSV (1998). When a country has for example common law, shareholders have better protection than under civil law. In such an environment, equity market development is high and a focus on the well functioning of equity markets (market-system) rather than on financial intermediaries (bank-system) is a logical consequence. For the same reason also market capital development and turnover of the equity market are negatively correlated to the banking and legal system. Another striking correlation is observable between political and the rule of law variable. Again a straightforward relation the more stable the political environment is the more inhabitants of a certain country will have confidence in the degree of contract enforcement. The company specific variables show low correlation levels with the institutional variables. Among one another and the dependent variable there is some. Clearly this is caused by the similar denominators of certain variables, for example capex and tangibility are highly correlated due to their common denominator total assets.

The correlations are in some instances high, this problem is addressed by testing the institutional variables alone as well as jointly.

4.6 Summary data

criterion, again 2 countries were lost, Taiwan and Singapore due to the unavailability of their total market capitalization of listed firms. Moreover a number of 83 firms were excluded due to data errors and outliers which affected the mean significant. A schematic outline of the number of firms dropped out based on which criterion is given below.

A number of 532 firms in 35 countries met all the criterions. The requirement that all firms need to have all variables available might bias our sample towards larger firms, because smaller firms would easier drop out due to data unavailability. However there is no possibility this bias can be avoided in the sample selection. The final sample used in this research is much larger and geographically more spread than in comparable studies (eg, Bancel et al, 2001; Titman et al 1988). These studies used significantly less firms and their sample did not cover all continents in the world, because this sample is more diversified and larger the results can be expected to be better and more reliable than those in comparable research. A schematic outline of the geographical spread is presented below.

Table 5

Excluded firms from sample

Criterion No of firms excluded No of firms left Countries left

Exclusion financials 157 843 47

Not all company specific 205 637 43

Not all institutional variables 23 615 35

Data errors 19 596 35

Outliers sample 64 532 35

Table 6

Geographical spread sample

Continent No of countries % of total sample

The largest part of our sample consists of European countries (U.K., Sweden, Portugal, Poland, Netherlands, Luxembourg, Italy, Ireland, France, Finland, Czech Republic, Belgium, Germany, Austria, Denmark, Hungary, and Switzerland), 43,9%, followed by the Asian countries (Thailand, Pakistan, Malaysia, Japan, Israel, Indonesia, India, China, Korea and the Philippines). In South America 5 countries remain (Peru, Mexico, Chile, Brazil and Argentina). From North America, Oceania and Africa a number of 5 countries were included (United States, Canada, Australia, New Zealand, South Africa).

5.

Methodology

5.1 Hypotheses

The hypotheses based on the main questions are:

:

0

H The institutional variables have no impact on leverage ratios on a world-wide firm-level, while having company specific variables in the equation.

:

1

H The institutional variables have a significant impact on a world-wide firm-level, while having company specific variables in the equation.

:

0

H The relevant institutional variables are the same for developed and developing countries.

:

1

H The relevant institutional variables vary between developed and developing countries. Several analyses hve been performed to test these hypotheses, including OLS, and a battery of robustness tests to examine if the obtained results are robust, cross time and cross section. For the analysis of hypothesis 2 the sample will be divided in sub-samples for developed/developing countries. In this section the performed tests will be described, explained and justified.

5.2 Panel data regression

Some of the benefits of panel data regression include that this method is particularly suitable to distinguish inter-individual differences form intra-individual differences, it reduces the estimation bias and it reduces the problems with data multicollinearity.

A regression analysis, which is limited to just cross-section or times-series data are more likely to reflect inter-individual rather than intra-individual differences and are therefore not very suitable, especially not when the discrimination between two hypotheses is based on social demographic or institutional differences which are varying over time and cross-section, which is the case in this research. Just single time series data disregard inter-individual differences, because it does not take into account cross-section variances. Cross-section data do include these differences, however it ignores their dynamics over time.

The richness of the dataset with panel data is a rather large benefit. Not only the possibility to hold either ior t constant to look just at times-series or cross-section effects but also because panel data offer a larger amount of freedom. This combined with the richness of the information provided over time/cross-section, can reduce the gap between information requirements of a sample and information provided by a dataset, (Hsiao, 1985).

5.3 Regression analysis

In order to test the above-described hypothesis, an OLS regression will be done. The regression will be performed using the following equation.

If the covariances between the dependent variables are zero, the  error term is evenly _i,t distributed over i,t with zero mean and constant variance. Then the OLS will have an unbiased and consistent outcome for all its coefficients, (homoskedasticity). However when there is heteroskadasticity, the standard error of the different variables is underestimated. As a result it could seriously underrate the confidence interval of the tested variables, which can induce type II errors to occur. In normal times-series/cross-section regression this problem could be addressed with a residual analysis using a general White’s test. However in our panel data regression the residuals can suffer from heteroskedasticity over time and cross-section. When there is correlation of the error terms cross-section (firm-effect), standard errors can be biased while using just the White’s test.

0 ) , (X_{i,t i,t} X_{itk itk}  Cov   (3) Table 7 Defenition variables

Leverage ratio Total debt over total assets

Market capitalization Total stock market capitalization over gross domestic product

Turnover Total value traded in domestic market over stock market cap

Bank system Dummy, which is (1) for bank-based and (0) otherwise

Rule of Law Measure of confidence of contract enforcement on a scale of 1-5

Legal system Dummy, which is (1) for civil law countries and (0) otherwise

Political Measure of political stability in a country on a scale of 1-5

Sales Natural logarithm of sales in Dollars

Tangibility Property, plant & equipment over total assets

Profitability Ebitda (t-1) over total assets

Ndt Non-debt tax shields (Sales -EBIT) over total assets

Δ Cash flow % Change yearly change in operational cash flow

Capex Capital expenditures over total assets

Δ Wc Working capital change over assets

Therefore another way of calculating standard errors must be used. When the firm effect is temporary, a calculation of standard errors clustered by firm leads to unbiased estimates. Yet, when the firm effect is permanent a fixed and random effects model must be used, (Peterson, 2007). When there is correlation across time,

0 ) ,

(X_{i,t i,t} X_{kt kt} 

Cov   (4)

the Fama et al (1973) approach for calculating standard errors is the best, using:

) ) 1 ( 1 ( 2 2 s x x T NT Var        (5)

This method yields the best results for standard errors –variances- when there is correlation over time between the residuals in the model.

Lastly there is the possibility that there is correlation among error terms through time and cross-section (across firms) in this case dummy variables have to be included, holding either cross-section variations or cross-time variations constant. The standard error analysis and the best practices to calculate them are summarized in the table below.

5.3 Detection firm & time residual correlation

In order to determine which model has to be applied to calculate the right standard error, we have to identify which effect is present in the dataset. Due to the small time frame of just 5 years it is not possible to determine whether the time or firm effect is temporary or permanent. Therefore the assumption is made that the effect is temporary in line with

Table 8

Calculation methods for standard errors

Effect Temporary Permanent

Firm effect Standard errors clustered Fixed-effect model

Time effect Fama/Macbeth model Fama/Macbeth model

temporary because residuals in succeeding years can be severely high however their correlation is not expected to remain persistent. He suggests for example 2001 to 2003 residual correlation is much lower than 2003 to 2004 residual correlation.

Firm-effects can be found with the use of year dummies, while holding the time constant, correlation among firms within the same year can be investigated. With the use of a time dummy (which absorbs the time-effect) the correlation between firms over the same period is completely removed and only a firm effect is left in the data, (Sapienza, 2004). For time effects, firm dummies will be included, which leaves just time effects in the sample. By taking the White’s standard error calculation as a starting point we can compare the standard error with and without the firm/time dummy and when the standard errors significantly differ, a firm/time effect at hand.

5.4 Additional & sub-sample tests

When error terms are correlated either across time or cross-section another possible bias may occur. The estimation of the coefficients can be incorrect when using OLS. The use of just simple OLS causes the estimations of the coefficients to be incorrect because not all the information available in the dataset is utilized properly. To adress this problem a Generalized Least Squares (hereafter: GLS) with random effects is employed, (Chapman et al, 2004). The coefficient estimations in this model are much more efficient in comparison to the OLS estimations when error term correlation is present.

Another possible bias might occur due to the fact that the regression analyses are performed on firm-level. As a result it is possible that the regression coefficients are affected by countries with many firms in the sample. To overcome this problem two additional tests will be done. A cross-country analysis where the country means of all variables are regressed and a test with the exclusion of the three largest countries in the sample, namely the U.K., U.S. and Japan.

6.

Empirical results

In this section the results will be discussed, in the different subsections several tests will be interpreted. In section 6.1 the sample used is explained with outlier analysis, section 6.2 & 6.3 include the main findings based on cross-firm OLS/GLS analyses with different error term definitions and random effects. In section 6.4, hypothesis 2 has been tested for developing and developed countries. Lastly section 6.5 includes the interpretation of the results in the light of current literature.

6.1 Outlier analysis

As previously stated legitimate outliers can substantially influence the estimated coefficients of a regression analysis. Therefore two OLS regressions have been performed, one with the legitimate outliers and one without. The coefficient estimates are in appendix 4, the magnitudes of the coefficients differ dramatically. For example the magnitude of the coefficient of the cash flow is more than 5 times larger without legitimate outliers. Although potentially valuable data points are lost, such large changes in coefficients -caused by less than 10% of the data- are unacceptable. Therefore the regression analyses will be done with the sample without legitimate outliers (n=532).

6.2 Main results cross-firm regressions for world sample

Table 9 contains the results based on the OLS regression analysis. All institutional variables have been tested while allowing for company specific effects. Model II, introduces two institutional variables, market capitalization ratio and the turnover of the stock market as a measure of stock market activity. The variable market capitalization has a significant negative coefficient, which suggests that a higher development of the stock market results in an increased use of equity relative to leverage. On the other hand turnover seems to be not significantly related with leverage ratio.

Model IV & V illustrate that the legal system is a better way to explain variations in leverage ratios. Companies in civil law countries hold significantly more debt. Moreover model V suggests that the use of equity increases when there is better legal protection in terms of contract enforcement.

Dependent variable is total debt over total assets, bank dummy is 1 for bank-based systems and 0 for market-based systems. The legal system dummy has a value equal to 1 for civil law and 0 for common law countries. Rule of law increases with better contract enforcement. Variable political is higher when political stability increases.

Table 9

Pooled cross-firm regression

Variable I II III IV V VI VII

Market cap -0,028 -0,022 (0.001)*** (0.064)* Turnover 0,024 0,034 (0.407) (0.287) Bank system 0,006 -0,015 (0.381) (0.140) Rule of law -0,013 -0,023 (0.002)*** (0.011)** Legal system 0,015 0,043 (0.023)** (0.000)*** Political -0,019 -0,040 (0.000}*** (0.000)*** Sales 0,018 0,019 0,018 0,019 0,018 0,019 0,019 (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** Profitability -0,394 -0,387 -0,388 -0,394 -0,381 -0,410 -0,398 (0.000]*** (0.000]*** (0.000]*** (0.000]*** (0.000]*** (0.000]*** (0.000]*** Ndt -0,036 -0,035 -0,036 -0,035 -0,036 -0,034 -0,034 (0.000)*** [0.000]*** [0.000)*** [0.000)*** [0.000)*** [0.000)*** [0.000)*** Tangibility 0,247 0,240 0,247 0,239 0,249 0,243 0,250 (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** Δ Cash flow -0,007 -0,007 -0,007 -0,007 -0,007 -0,007 -0,007 (0.011)** (0.011)** (0.011)** (0.010)** (0.011)** (0.010)** (0.010)** Capex -0,386 -0,379 -0,383 -0,376 -0,374 -0,394 -0,390 (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** Δ Wc -0,154 -0,157 -0,154 -0,159 -0,154 -0,159 -0,159 (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)***

Δ Wc $ 3,19E-12 3,18E-12 3,19E-12 3,17E-12 3,16E-12 3,24E-12 0,000

(0.323) (0.323) (0.323) (0.325) (0.327) (0.314) (0.314)

c -0,100 -0,110 -0,109 -0,112 -0,124 -0,105 -0,145

(0.014)** (0.008)*** (0.010)** (0.006)*** (0.003)*** (0.010)** (0.001)***

R squared 0,157 0,160 0,157 0,159 0,158 0,159 0,168

In Model VI it appears that political stability has a significant negative impact on the amount of leverage in a company on a 1% significance level

Having looked at the institutional variables we turn to the company specific variables. Worth mentioning is that every parameter shows a high level of significance in favor of both of dominant theories. Only the absolute change of working capital is insignificant, a possible reason lies in the definition of the variable. Whereas all the variables are related to the size or asset base of the company this variable is not. Therefore the absolute value of working capital might not well reflect the actual working capital position of the company.

The R² of the model varies between 15.8% and 16.8% which means that, between 15.8% and 16.8% of the variation in leverage ratio is explained by the variables in the sample. Regression VII has the highest explaining power which is straightforward because all expected relevant variables have been included.

6.3 Standard error clustering

Dependent variable is total debt over total assets, bank dummy is 1 for bank-based systems and 0 for market-based systems. The legal system dummy has a value equal to 1 for civil law and 0 for common law countries. Rule of law increases with better contract enforcement. Variable political is higher when political stability increases. ***significant at 1%, **significant at 5%, *significant at 10%. In parentheses are the p-values.

As expected did the significance levels change somewhat in some cases (e.g. market capitalization, legal system & political). However due to the fact that they were already highly significant under OLS assumptions the newly defined error term did not substantially

Table 10

Pooled cross-firm regression with errors clustered by firm

Variable I II III IV V VI VII

on a 1% and 5% level. Although the significance levels remained more or less the same, the detection of a firm effect might impose an impact on the coefficients. According to Wooldridge (2007) it is possible that an OLS analyses does not exploit all the available information efficiently when there is a firm effect present.

Dependent variable is total debt over total assets, bank dummy is 1 for bank-based systems and 0 for market-based systems. The legal system dummy has a value equal to 1 for civil law and 0 for common law countries. Rule of law increases with better contract enforcement. Variable political is higher when political stability increases. ***significant at 1%, **significant

Table 11

General least square regression with random firm effects

Variable I II III IV V VI VII

To overcome this problem a General Least Squares Regression, with random firm effects has been performed, (table 11). Some coefficients substantially changed after allowing random intercepts between firms. Namely the coefficients of market capitalization, political and rule of law. In model II the magnitude of the coefficient market capitalization (in comparison with no random effects) rises with more than 37%, whereas the coefficient almost doubles in magnitude for model VII. For rule of law the magnitude drops by 7.5. Lastly the coefficient of the political parameter substantially changes. In model VI the coefficient is halved in comparison to table 9, for model VII this change is approximately 30%. For the company specific variables used the magnitude differs with random intercepts as well, the magnitude of coefficients size, profitability, tangibility, cash flow, capital expenditures and working capital decline substantially.

6.4 Sub-sample analysis

A severe concern arises when testing on firm level. It might be possible that the obtained results are caused by observations from large countries which are rather dominant in our sample, (see: appendix 3). For example 38.2% of the firms have their origin in Japan and the United States. Since the impact on the overall test results can be severely high a regression analysis with country means is applied. In this test every country has equal weight and as a result large countries in the sample cannot have a substantial impact. The results show that there is indeed a large difference when the countries are weighted equally. The expected negative effect of stock market capitalization is suddenly not significant, neither is the variable turnover, however the latter was already insignificant in our initial sample. The coefficient of the banking dummy flips sign and is suddenly negative and significant, which is rather strange because this means that countries with a banking system have lower leverage ratios. However the statistic relevance might be doubted. Looking at tables 9 through 11 the variable bank system coefficient shows large changes when tested jointly compared to testing with just the inclusion of the control variables. This might be due to the high correlation between the legal and economic system4 . There is a large correlation between the legal system and banking system (67,5%, see appendix: 2) this high correlation leads to a serious bias in the coefficients and significance level when tested jointly. Tests without inclusion of

the legal system confirm this suggestion, no significance for the banking system in the different models (Model II through V) neither without the legal system nor when tested alone.

Dependent variable is total debt over total assets, bank dummy is 1 for bank-based systems and 0 for market-based systems. The legal system dummy has a value equal to 1 for civil law and 0 for common law countries. Rule of law increases with better contract enforcement. Variable political is higher when political stability increases. ***significant at 1%, **significant at 5%, *significant at 10%. In parentheses are the p-values.

Table 12

Pooled regression across sub-samples

Variable World sample Country means Without UK/U.S/JapanDeveloped countriesDeveloping countries

The coefficients for legal system and rule of law are both similar to the initial results, however the political coefficient has been doubled in magnitude.

To investigate what causes these inconsistencies, the sample has been divided in sub-samples (table 12). Model III is a regression where the 3 most influential countries (U.S., U.K. & Japan) were excluded, to examine how these 3 influenced the coefficients and significance levels.

Comparing model I with model III there is one obvious change, the variable market capitalization becomes insignificant. So there is suddenly no significant relation between leverage and the market capitalization in the sample without the big three. For the remaining part the significance levels and magnitudes of the coefficients are somewhat larger in model III, which indicates that the institutional effects in Japan, U.K. & U.S. are smaller than in the rest of the sampled countries.

Model IV & V illustrate that the differences between the world sample and country mean sample is for the largest part a result of the differences between institutional coefficients and significance levels between developed and developing countries. It appears that in developed countries stock market development leads to the use of more equity (less debt), (significant on 1% level), whereas in developing countries there is no relation. Another difference occurs in banking system, in developing countries there is no relation between capital structure and the economic system. In developed countries there is however the sign is contrary to what was initially expected. However like previously discussed the statistical relevance of this finding might be doubted. Furthermore in developing countries there is a negative relation between political stability and leverage with a quite large magnitude (0,100) and high significance level of 1%. The effect of higher agency cost of equity (leverage more attractive) when the political environment is instable is only present in developing countries and leads to quite large variations in leverage ratios. Rule of law shows opposite coefficients when comparing developing and developed countries. Whereas higher quality of contract enforcement and court leads to significantly more equity financing in developing countries it has the opposite effect in developing countries, however not significant. Lastly the legal system is consistent over all 5 models however its magnitude in developing countries is almost three times larger compared to developed countries.

For developing countries there is a slight difference; capital expenditures and working capital change are not significant.

A striking finding is the R² difference between model IV & V, it seems that much more variation in leverage ratios is determined by the included 14 variables in developing countries compared to developed countries. The independent variables are for 41% responsible for variations in leverage in developing countries versus 14,3% in developed countries.

6.5 Discussion of the results

The results as shown in the previous section show that there are institutional factors influencing capital structure choice. In this section those results will be discussed in the perspective of current literature.

Economic environment

Two variables have been used to characterize the equity market development of a country, namely market capitalization and turnover of the stock market. Market capitalization has been found significant for the world sample tested with OLS and GLS regressions. However the relation is not present over all the sub-samples. It seems that only for developed countries there is a substantial positive relation between the use of equity and the development of the stock market. This finding is contrary to Demirgüç-Kunt et al (1998) who conclude that in developed stock markets further development leads to a substitution of equity for debt, where in countries with a developing stock market (as in most developing countries) this substitution effect does not take place. The variable turnover seems to be insignificant across all sub-samples in line with the findings of Demirgüç-Kunt et al (1999). A possible explanation might be the fact that the parameter turnover is highly sensitive to currency rates and investor climate. This might induce a bias in the data.

largest three countries and the sample with only developing countries. However like previously stated this result might be biased by high correlation between the legal dummy and the dummy representing the economic system.

Legal environment

The rule of law is significantly positive related to leverage ratio in our world sample, sample with country means and the sample without the US, UK and Japan. No pronounced sign had been hypothesized because better contract enforcement both stimulates debt and equity holdings (due to lower transaction cost). As a result it could affect the leverage ratio of a particular firm, however in which direction is not clear, Levine (2000). The model shows that in our world sample better and more efficient contract enforcement leads to more equity and therefore less leverage. However when the sample is divided in developed and developing countries rule of law shows a negative relation for developed countries and positive for developing countries, however not significant.

The only variable which is positively significant across all 5 sub-samples is the dummy legal system. Which can be explained due to the fact that in civil law countries shareholeder protection is weak. As a result the development of capital markets is slower. This leads to an environment where it is easier for firms in civil law countries to raise money via debt markets than equity markets. This finding is in line with the findings of LLSV et al (1997).

Political environment

The political system seems to have a significant negative relation with leverage in every sample used except the sample with developed countries. This is again conform expectations due to the fact that in political stable countries agency cost are lower, the use of equity is less expensive (and vice-versa) and therefore less debt will be used to finance projects, (Lee et al, 1988). However looking to only developed countries there is no significant relation. A possible explanation is that in the developed regions of the world the political system is relatively stable and as a result a factor which not per se has to be evaluated when making capital structure decisions.

Company specific variables

countries. However profitability is significantly negative and its magnitude is more than twice as high compared to developed countries.

***significant at 1%, **significant at 5%, *significant at 10%

Table 13

Significance levels for hypothesized signes

Variables World Developed Developing

sample countries countries

Development stock market -* -***

Turnover stock market

Bank system dummy (1) -***

Rule of law +**

Dummy civil law dummy (1) +*** +* +**

Political stability -*** -***

Trade-off Pecking order theory theory

Size + - +*** +*** +***

Profitability + - -*** -*** -***

Non-debt tax shields - -*** -*** -***

Tangibility + +*** +*** +***

Operating cash flow - - -** -* -***