Background
From a financial point of view, the objective of a business is to create value; which means being able to carry out an investment that reached a profitability rate exceeding the rate of profitability required considering risks. The literature on the choice of the financial structure actually began in 1958, a date on which Modigliani and Miller (MM) have published a first founder article. The financial structure of a company is determined by "the relative proportion of the debt and equity" in the liabilities of its balance sheet (Berk and DeMarzo
2008). This relative level is usually the subject of a decision on the part of the company. How is it influenced? This question is the starting point of the research of Modigliani and Miller that was carried out in 1950. Based on some assumptions, they have demonstrated in their first article that the financial structure of a company had no impact on its total value. A few years later, in 1963 and in a second article, Modigliani and Miller complemented their analysis by introducing the taxation as one of the leading market imperfections. When the financial difficulty is without cost, they suggest to browse to the maximum of indebtedness in order to take advantage of the benefits of tax saving. Miller (
1977) expands the framework defined by Modigliani and Miller in integrating the taxes on the income of natural persons. In such a context, where we take into account both the corporate and personal taxation, Miller concludes that the debt has no impact on the value of the firm. For this last research, there is no optimal structure of the capital. In contrast, since the series of work of Modigliani and Miller, the theoretical reflection on the behavior of capital’s structure of firms was significantly extended. In fact, three main theories have been developed to explain the financial structure of firms. The first theory called “
tradeoff”, the second is referred to as the theory of hierarchical funding or “
pecking order”. These first two theories are considered, for more than long, as two theoretical frameworks of reference for researchers while studying the capital’s structure of firms. The work of Becker and Wurgler (
2002) leads to the emergence of a third theoretical analysis, known under the name of the modern theory of the firm or still under the acronym of “Market
Timing Theory of Capital Structure”. It is thanks to their work that many searches are based on this new theory of “market
timing” in the study of the determinants of capital’s structure of enterprises. First, our goal is to adopt the main explanatory factors of the structural variations observed in the levels of debt. Then, we will attempt to test the various models arising from each theory to determine which one that best matches the context of Tunisia. We have analyzed the structure of the capital of a sample of Tunisian listed companies in the light of different financial theories (Adedeji
1998, Bacha and Attia
2016).
The purpose of this article is to shed light on the determinants of capital structure for companies listed on the Tunisian Stock Exchange. We use panel data models to compare the theoretical and empirical results in the literature with the Tunisian market, study the dynamic behavior of debt and test the existence of an adjustment process towards a target debt level (Shleifer and Vishny
1997; Soufeljil et al.
2016b).
Our contribution is thus twofold. First, we add to the existent literature by focusing on the determinant factors of the capital structure of an emerging country, namely Tunisia.
In fact, studying the Tunisian case may be interesting in terms of policy recommendations for this country and other emerging countries presenting similar features such as bankoriented financial systems and relatively small capital markets. Second, this paper contributes to the relatively limited literature on the dynamics of the capital structure. It checks out the existence of an adjustment process towards a target leverage in the Tunisian market.
The article is structured as follows. The second section presents the literature review and research hypotheses. The third section deals with the methodological choice. The fourth section presents the results and their discussions. The last section presents the conclusion.
Data & Methodology
In the case of Tunisia, Dr. Soufeljil and al (
2016a) showed that Tunisian companies first resort to debts that are mainly shortterm, then selffinancing and rarely to the issuance of new shares. Aware of the insufficient work on the determinants of the financial structure of Tunisian companies, we will try, in this work, to make a contribution in this direction.
Our objective is to understand the determinants of the financial structure of the Tunisian listed companies, in particular the enterprises of the two pillars of the national economy namely: Industry and trade. To do this, we propose to test empirically the theory of compromise and that of the hierarchical preferences of financing by trying to see:

If companies go into debt because, they are seeking to achieve an optimum debt ratio; or

If companies go into debt because, they having tiered funding preferences; or

If companies go into debt both because, they are seeking to reach a target level of debt and because they prioritize their funding resources.
The dynamic model of the theory of the “TradeOff.”
The dynamic model of theory of target ratio is as follows:
$$ {\mathbf{D}}_{\mathbf{it}}={\boldsymbol{\updelta}}_0+\left(1\boldsymbol{\upalpha} \right){\mathbf{D}}_{\mathbf{it}1}+{\boldsymbol{\updelta}}_1{\mathbf{TANG}}_{\mathbf{t}}+{\boldsymbol{\updelta}}_2{\mathbf{RENT}}_{\mathbf{t}}+{\boldsymbol{\updelta}}_3{\mathbf{TAILLE}}_{\mathbf{t}}+{\boldsymbol{\updelta}}_4{\mathbf{OG}}_{\mathbf{t}}+{\boldsymbol{\updelta}}_5{\mathbf{EINLD}}_{\mathbf{t}}+{\boldsymbol{\updelta}}_6\ {\mathbf{LIQ}}_{\mathbf{t}}+{\boldsymbol{\updelta}}_7\ {\mathbf{RD}}_{\mathbf{t}}+{\boldsymbol{\upvarepsilon}}_{\mathbf{it}} $$
Sample and data source
To validate the models of “TradeOff” and of “pecking order”, we have chosen a sample that is decomposed into 26 Tunisian firms listed on the stock exchange of Tunis. These firms belong to industrial, commercial and services sectors. In fact, financial firms are excluded from our sample because there is a risk of homogenization of data. In this regard, these companies have a financial structure that is different from that of the nonfinancial businesses. The data used in this research are derived primarily from the stock exchange of Tunis. The study period extends from 2005 to 2010 (6 years).
The industrial, commercial and service sectors are represented on the stock exchange of Tunis by 26 companies. Fourteen are large and listed on the main market, 11 are mediumsized companies listed on the market development and one is mediumsized listed on the growing market.
The choice of these companies is justified by the availability of accounting information on listed companies. There is no accessible database that centralizes the accounting information of unlisted Tunisian companies; and by the fact that this work is part of the research project on the evaluation of the reform process and the efficient upgrading of Tunisian companies by 2020. Only companies in the industrial, and services are retained. We chose only those companies for which we have at least five fiscal years. No sampling methodology was used due to the limited number of listed companies in the industrial, commercial and service sectors and for which we have at least five financial years.
The dependent variables
The choice of debt ratio has been the object of some theoretical studies. In addition, several authors argue that this ratio must be calculated from the book value of the debt. It is equal to the total value of assets less the book value of own funds. While others observe that it is the market value of debt which must be taken into account and that is obtained by substituting the accounting value of own funds by a market value. Several authors use these two measures at the same time as the studies developed by Becker and Wurgler (
2002), Fama and French (
2002), Kayhan and Titman (
2006) and Frank and Goyal (
2008). As well as other authors DeMiguel and Pindado (
2001), Chen (
2004) and Delcoure (
2006) have limited their works to the book value of debt ratio as a result of the shortcomings of data. In fact, concerning our empirical study, we will work on the book value of debt ratio because there is also no sufficient data and we will use throughout this research two types of ratios to know: the ratio of total indebtedness which is equal to the carrying value of debt divided by the total assets and the debt ratio in the long term that is equal to the longterm borrowing on the total assets. We will adopt the same dependent variable for the two models to know the theory of “
TradeOff” and that of the theory of “
pecking order”.
The independent variables
The explanatory variables of the model of the theory of “tradeoff “
The embodiment of assets:
According to Harris and Raviv (
1991), Rajan and Zingales (
1995) and Delcoure (
2006), the embodiment is equal to the ratio between the sum of the tangible capital assets and stocks, divided by the total assets.
Profitability
This variable is measured by:
$$ \mathbf{RENT}\kern0.5em =\kern0.5em \frac{\mathrm{Profits}\ \mathrm{before}\ \mathrm{interests}\ \mathrm{and}\ \mathrm{taxes}\ }{\mathrm{Total}\ \mathrm{Assets}} $$
Size
To approximate this variable, we will use the natural logarithm of total assets.
Opportunities for growth
The ratio between the tangible assets and the total of assets are used for the approximation of this variable.
The economies of taxes not related to the debt
This variable is measured by:
$$ \mathbf{EINLD}=\kern0.5em \frac{\mathrm{Endowments}\ \mathrm{to}\ \mathrm{amortization}\ \mathrm{and}\ \mathrm{to}\ \mathrm{provisions}\ }{\mathrm{Total}\ \mathrm{assets}} $$
Liquidity
This variable is introduced by Ozkan (
2001), the two signs are expected signs.
$$ \mathbf{LIQ}=\frac{\mathrm{Current}\ \mathrm{Assets}\ }{\mathrm{Current}\ \mathrm{liabilities}} $$
The risk of default
This variable is measured by:
$$ \mathbf{RD}=\frac{\mathrm{Financial}\ \mathrm{Charges}\ \left(\mathrm{expenses}\right)\ }{\mathrm{Profits}\ \mathrm{before}\ \mathrm{taxes}\ \mathrm{and}\ \mathrm{interests}} $$
Methods
The explanatory variables of the mode of theory of “pecking order”
Following the theory of the “
pecking order”, the coefficient of this variable is equal to the unit. This variable s calculated as follows:
$$ \mathbf{DEF}=\mathbf{I}+\mathbf{D}+\varDelta \mathbf{FR}\mathbf{CF} $$
With:
I: investment of the company during the year t;
D: dividends paid by the company during the year t;
∆FR: variation in the bottom of the bearing (FR = Current Assets  Current Liabilities);
CF: net cash flow is equal to results of holdings after the payment of taxes and interests.
The independent variables
The model discussed during this section contains four new variables: the ratio of MTB, the weighted average of the MTB ratios, the index of performance of securities in Tunis and rate of the financial market (Huang and Ritter
2005,
2006).
The ratio MTB
This variable is approximated as follows:
$$ \mathbf{MTB}=\kern0.5em \frac{\mathrm{Market}\ \mathrm{capitalization}+\mathrm{debts}}{\mathrm{Net}\ \mathrm{Assets}} $$
The weighted average of the ratio of MTB (EFWMB)
According to Becker and Wurgler (
2002), this variable is regarded as being the best approximation of the persistence of attempts of timing. This weighted average is used to test the persistence of the effect of attempts of timing in the structures of financing of businesses.
This ratio is calculated as follows:
$$ \mathbf{EFWMB}=\sum_{\mathbf{s}=1}^{\mathbf{t}1}\frac{{\mathbf{e}}_{\mathbf{s}}+\kern0.5em {\mathbf{d}}_{\mathbf{s}}}{\sum_{\mathbf{r}=1}^{\mathbf{t}1}\left({\mathbf{e}}_{\mathbf{r}}+{\mathbf{d}}_{\mathbf{r}}\right)}\times {\mathbf{MTB}}_{\mathbf{s}} $$
EFWMB: The weighted average of ratio of MTB for the period t  1;
Es: net emissions of shares during the period s;
Ds: net emissions of debts during the period s;
MTBs: The ratio of Market to book relating to the year s.
It is a price index (simple arithmetic average of courses).
The rate of monetary market
The rate of borrowing on the market is indexed by the TMM. It is the average of the last twelve monthly average rates of the monetary market of Tunisia before the date of payment of interest. After quoting different theoretical journals on the capital structure as well as the recent empirical works, this part focuses on the choice of funding of Tunisian firms and their abilities to return to the target ratio of debt.
Results and discussions
Tests that will be filed; deposited are based on estimates of panel data. All these estimates will be carried out thanks to the Software”
STATA 11 “A study made by Kennedy (
1985), affirms the existence of a problem of multicollinearity if and only if, the Pearson correlation coefficients exceeds (0,8). In fact, the matrix of Pearson is given by the following Table
1.
Table 1
Correlation coefficients of the explanatory variables
Tang  1.000       
Rent  −0.569  1.000      
Size  −0.403  0.284  1.000     
Og  0.237  −0.293  −0.084  1.000    
Liq  0.289  0.150  −0.391  −0.228  1.000   
Rd  −0.079  0.093  0.101  −0.074  −0.013  1.000  
Einld  −0.595  0.274  −0.023  −0.134  0.082  0.168  1.000 
We note well, after this correlation matrix of explanatory variables, that there is no problem of multicollinearity between the variables studied. (All the coefficients do not exceed 0,8).
The empirical validation of models of “tradeoff” and the” pecking order”
In this first section, we prove the models of the theory of “TradeOff” and those of “pecking order”. Then, we will present three models. The first is devoted to the static version of theory of “TradeOff” while the second one is used to test the dynamic version of this theory of “TradeOff”. Finally, the third allows you to check the theory of “pecking order”.
The static model of theory of “Tradeoff”
After performing the stationarity tests for all the series included in the model, we found that most of the series are stationary in level; this explains the use of stationary panel data. The first interest is to determine the tests of specification or tests of homogeneity of the data.We are going to go up if the considered model is entirely identical to all firms in the sample or if there are specific features of each firm. The results based on the statistics of Fisher show the rejection of the hypothesis of global homogeneity knowing that there are common coefficients for all firms as well as the presence of individual specificities for each firm (
Pvalue < 10%). We have used the method of panel data. A problem of multicollinearity is present in a regression when an explanatory variable is close to a linear combination of at least one other explanatory variable. When the multicollinearity is perfect and we cannot obtain an estimate of the parameters. For this, audits are required to be ranging from the simple correlation matrix of the explanatory variables to other statistics such as the inflation factor of the variance (VIF), the most used indicator by software (Joeveer
2013; Kötter et al.
2009).
Variance Inflation Factors (VIF) provide a measure of the increase in the variance of the estimated regression coefficients with respect to a situation where the prediction variables do not have a linear relationship. They allow describing the importance of multicollinearity (correlation between predictors) in a regression analysis. Multicollinearity is problematic because it can increase the variance of the regression coefficients, making them regressive and difficult to interpret.
We can use the VIF command after the regression to check for multicollinearity. VIF stands for the variance inflation factor. As a rule of thumb, a variable whose VIF values are greater than ten may merit further investigation. Tolerance, defined as 1/VIF, is used by many researchers to check on the degree of collinearity. A tolerance value lower than 0.1 is comparable to a VIF of 10. It means that the variable could be considered as a linear combination of other independent variables.
The investigation of the results indicates that the problem of multicollinearity is not serious in our case since the VIF is lower than 3. However, it seems that there is a collinearity that is relatively disturbing between variables Tang and einld with keen above 2. By eliminating one of the two variables in the regression, the collinearity issue will be resolved. By eliminating the variable einld, the regression indicates that the variable einld is not significant to thresholds of the usual risks. The estimation of the model by data from Panel allows you to control the heterogeneity of the variables in their individual quantities. The estimation of fixed effects uses the differences to the individual averages and eliminates the consistent differences between the companies. This method allows identifying and measuring the effects which are not directly identifiable. Thus, the random effects model assumes the independence between terms of errors and explanatory variables (Dang et al.
2014) (Table
2).
Table 2
Coefficients of the variables correlation of the variables
Tang  1.000       
Rent  −0.569  1.000      
Size  −0.403  0.284  1.000     
Mtb  0.665  −0.501  −0.277  1.000    
Efwmb  0.352  −0.385  −0.195  0.496  1.000   
IPSET  −0.038  0.118   .002  −0.047  −0.018  1.000  
Tmm  0.014  −0.059  −0.066  −0.076  0.000  −0.033  1.000 
The test of Hausman (
1978) allows you to validate the exogeneity of the specific effect in relation to the independent variables. For the two models, the estimates by the ordinary least squares (OLS) are biased and the probability of the test specification of Hausman is greater than the threshold of 10%. Nevertheless, the test of Hausman does not differentiate between the fixedeffects model and the random effects model. Besides, (R2 within <R2 between) has led us to promote the random effects model for the two ratios of indebtedness. The same is considered for the two ratios of debt, the probability of Statistics of the test of BreushPagan (
1979) shows that there is no problem of heteroscedasticity. Furthermore, the test of Wooldridge (
2002) indicates that there is no problem of the firstorder autocorrelation in tailings for both ratios of debt. We can note that the model used explains well the levels of indebtedness of Tunisian businesses listed on the stock exchange of Tunis and this is for the two versions of indebtedness designated. In fact, for the ratio of total indebtedness the coefficient of determination is 0.9151, while for longterm debt ratio this coefficient is 0.8915.
The tangibility variable is significant at the threshold of 1% in all of the estimated models. This variable is positively correlated with the debt ratios. This result is affirmed by Harris and Raviv (
1991), Rajan and Zingales (
1995) and Delcoure (
2006). In other words, the Tunisian businesses with a large mass of tangible capital assets have a high liquidation value. This explains their increased use of services of debt.
The variable profitability is also significant at the threshold of 1% for the model of the ratio of total indebtedness and significant contribution to the 10% threshold for the model of debt ratio in the long the term. This variable is negatively correlated with the debt ratios. This result is confirmed by several authors (Shyam and Myers
1999; Fama and French
2002; Chen
2004). The negative correlation between these two variables means that the more firms are profitable the more their debt levels are falling. Thus, this variable holds significant and high coefficients (for the ratio of total indebtedness these coefficients vary between −1.161 and −1.149 and between −0,224 and −0,218 for debt ratio in long term). Then, levels of indebtedness of Tunisian firms are influenced negatively by their profitability.
The variable size is positively correlated with the dependent variable. This latter variable is consistent with the assumptions of several journals of literature (Booth et al.
2001; Fama and French
2002; Delcoure
2006). This variable is significant at the 5% threshold in all the estimated models. The positive relationship means that the more the size of firms is high their probabilities of bankruptcy is low. But, this variable affects more debt ratios totals than those in the long term. In effect, their coefficients are between 0,057 and 0,058 for the ratio of total debt and between 0,044 and 0,046 for debt ratio in long term.
The variable relative to the liquidity is significant at the threshold of 1% for all the estimated models. This variable is negatively correlated with both ratios of debt. This negative relationship means that the Tunisian firms having several liquid assets, have tended to use it for their financial investments. In fact, the coefficient is equal to −0,026 for the ratio of total debt and equal also to −0,031 for debt ratio in the long term. As regards to the other variables, they are not significant. It is to say that the Tunisian firms do not give a particular thought to these factors.
The dynamic model of the theory of the “TradeOff”
A dynamic model is a model in which one or several delays of the dependent variable are included as explanatory variables. When econometric techniques standards as the OLS do not allow obtaining efficient estimates of such a model, it is to estimate on Panel Data dynamic. This form of estimation is based on the estimator of Arellano and Bond (
1991) which is the most effective thanks to the dependent delayed variable located to the right of the equation.
The Estimator by the method of generalized time (GMM) in first difference of Arellano and Bond (
1991) is to take for each period the first difference of the estimated equation. To eliminate the specific effects of countries, we instrumented the explanatory variables of the equation in first difference by their values at the delayed level of a period or more (Chirinko and Singha
2000) (Table
3).
Table 3
The coefficients of correlation of variables
Tang  1.000          
Rent  −0.569  1.000         
Size  −0.403  0.284  1.000        
Og  0.237  −0.293  −0.084  1.000       
Liq  0.289  0.150  −0.391  −0.228  1.000      
Rd  −0.079  0.093  0.101  −0.074  −0.013  1.000     
Def  0.098  −0.164  −0.428  From 0.099  0.089  −0.115  1.000    
Mtb  0.665  −0.501  −0.277  0.363  0.022  −0.152  0.026  1.000   
Efwmb  0.352  −0.385  −0.195  0.168  −0.005  −0.097  0.061  0.496  1.000  
IPSET  −0.038  0.118   .002  0.000  −0.012  −0.025  −0.155  −0.191  0.000  1.000 
Although it is asymptotically more efficient, the GMM estimator in two steps leads to biased results. In effect, for the two indebtedness ratios the GMM estimator in the two stages is not effective since the coefficient associated with the lagged dependent variable is not significant for debt ratio in the long term. In this regard, this coefficient gives a higher value to the unit for ratio of total debt. Similarly, the majority of explanatory variables are not significant. Then, we can choose the estimator of a step for both ratios of debt.
Two tests are associated to the estimator of GMM in the dynamic panel. The first test is the test of overidentification of Sargan/Hansen that allows you to test the validity of lagged variables as an instrument. The second test C is also the test of the autocorrelation of Arellano and Bond (
1991) where the null hypothesis is the absence of autocorrelation of a second errors’ order of the equation in difference. In fact, the test of Hansen (
p = 1.000 for the two ratios of debt), the test of Sargan (for the ratio of total indebtedness
p = 0.019 and for debt to equity ratio of longterm
P = 0.035) and the test of autocorrelation of second order of Arellano and Bond (
1991) (for debt ratio total
P = 0.004, for the debt to equity ratio of longterm
p = 0.017) do not allow to reject the hypothesis of validity of the lagged variables as an instrument, and the hypothesis of the absence of autocorrelation of a second order.
The positive coefficients are always significant at the threshold of 1% of the delayed variable. This confirms the existence of adjustment costs on the Tunisian market. The coefficient for adjustment of target ratio being equal to (
1α) with
α as the coefficient of the delayed variable of debt. In fact, these adjustment costs are equal to 0.003 for the ratio of total indebtedness and 0.031 for a debt ratio in the long term. These values demonstrate that the Tunisian firms adjust their indebtedness ratios in a low way in order to achieve their target levels. This result is contradictory to the result obtained in the Studies of DeMiguel and Pinaldo (
2001) (whose speed is equal to 0.79), Shyam and Myer (
1999) (this speed is equal to 0.41) and Ozkan (
2001) (whose speed is equal to 0.47).
Hertig (
1998) has analyzed the variations in the financial structures of Swiss companies. This speed is due to the shortcomings in procedures of internal and external control of credit institutions. In fact, we also note that the credit policies of Swiss banks are focused on personal relationships.
The representative variables of embodiment, profitability and size are significant at the threshold of 1% for two dependent variables. This is confirmed by the results obtained in the static model. In another word, the liquidity variable is significant at the threshold of 1% for both ratios of debt. In addition, the variable risk of default is significant at 5% threshold for the two ratios of debt. Nevertheless, the influence of these variables is very limited due to the low value of the associated coefficient (Barclay and Smith
1995; Benito
2003).
The model of theory of “pecking order”
The model of the theory of tiered funding is as the follows:
$$ {\mathbf{D}}_{\mathbf{it}}{\mathbf{D}}_{\mathbf{it}1}={\boldsymbol{\upalpha}}_0+{\boldsymbol{\upalpha}}_1{\mathbf{D}\mathbf{EF}}_{\mathbf{it}}+{\boldsymbol{\upvarepsilon}}_{\mathbf{it}} $$
The study of this model will be executed similarly to the one developed for a static model of the theory of “
TradeOff” (Table
4).
Table 4
Summary of the results of regressions carried out in the static model of TradeOff
Tang  0.343 (0.023)^{a}
 0.340 (0.024)^{a}
 0.343 (0.023)^{a}
 0.270 (0.020)^{a}
 0.267 (0.020)^{a}
 0.270 (0.020)^{a}

Rent  −1.149 (0.135)^{a}
 −1.161 (time: 0.140)^{a}
 −1.149 (0.134)^{a}
 −0.218 (0.116)^{c}
 −0.224 (0.131)^{c}
 −0.218 (0.116)^{c}

Size  0.057 (0.022)^{b}
 0.058 (0.022)^{b}
 0.057 (0.022)^{b}
 0.044 (0.019)^{b}
 0.046 (0.019)^{b}
 0.044 (0.019)^{b}

Oc  2.804 (2.277)  2.881 (2304)  2.804 (2.277)  −0.506 (1954)  −0.384 (1.975)  −0.506 (1954) 
Liq  −0.026 (0.009)^{a}
 −0.025 (0.009)^{a}
 −0.026 (0.009)^{a}
 −0.031 (0.008)^{a}
 −0.030 (0.008)^{a}
 −0.031 (0.008)^{a}

Rd  −0.025 (0.017)  −0.027 (0.017)  −0.025 (0.017)  −0.028 (0.014)^{c}
 −0.029 (0.015)^{c}
 −0.028 (0.015)^{c}

_Cons  −1.006 (0.416)^{b}
 −1.025 (0.421)^{b}
 −1.005 (0.416)^{b}
 −0.746 (0.357)^{b}
 −0.784 (0.362)^{b}
 −0.746 (0.357)^{b}

R2 adjusted  0.8364  0.8364  0.8364  0.7096  0.7095  0.7096 
R2 within  –  0.8349  0.8348  –  0.7068  0.7067 
R2 between  –  0.9129  0.9151  –  0.8869  0.8915 
F  121.85^{a}
 116.27^{a}
  58.23^{a}
 55.44^{a}
 
Wald  –  –  731.13^{a}
 –  –  349.36^{a}

H  –  0.60  –  1.18 
LM  –  –  1.18  –  –  0.98 
M1  –  –  9.121  –  –  0.939 
N  150  150  150  150  150  150 
In summary, the results do not allow us to reject the hypothesis of the presence of a tiered funding under its form semistrong. The model is broadly significant at a 5% threshold for the majority of tested regressions. As well, we can note that the model used is unable to explain the levels of indebtedness of Tunisian firms listed on the stock exchange of Tunis. In fact, the coefficients of determination vary between 0.007 and 0.0421 for the ratio of total debt and between 0.0182 and 0.0751 for the debt ratio in the long term.
For the two models, the estimates by OLS are biased and the probability of test specification of Hausman is greater than the threshold of 10%. However, the test of Hausman does not differentiate between the fixedeffects model and the random effects one. Thus, the ratio of total indebtedness (R2 within > R2 between) has led us to promote the fixedeffects model. However, the debt ratio in the long term (R2 within < r2 between) has led us to promote the random effects model. Indeed, even for the two ratios of debt, the probability of Statistics of the test of BreushPagan shows that there is a problem of heteroscedasticity. Similarly, the test of Wooldridge proves the existence of a problem of the firstorder autocorrelation in tailings for both ratios of debt.
The variable of a deficit of funding is significant at 5% threshold for the ratio of total indebtedness, but it is not significant only at the threshold of 10% for the debt ratio in the long term. These results are similar to those of Shyam and Myers (
1999), Frank and Goyal (
2003), Kayhan and Titman (
2006).
Finally, we can conclude that the weakness of coefficients of determination which are very close to 0 allowed to check the lack of relevance in the theory of hierarchical funding in the context of Tunisia (Bias et al.
1995; Borisova et al.
2015; Boyd and Solarino
2016).
The empirical validation of model of “market timing”
The Table
2 shows that the correlation between the variables is average. Nevertheless, the MMR which reflects the cost of debt and the SET index which measures the performance of the stock market are weakly correlated. What has led us to exclude the variable TMM. As we have already seen in the previous section, all the tests have been carried out by data from the panel. Thus, the software used is the “
STATA 11”. All the models that have been analyzed previously are static models (Table
5).
Table 5
Estimation of adjustment toward a debt level target
Rdette_1  0.997 (0.029)^{a}
 2.519 (0.854)^{a}
 –  – 
Rdlt_1  –  –  0.969 (0.042)^{a}
 0.441 (0.727) 
Tang  −0.323 (0.020)^{a}
 0.054 (0.296)  −0.274 (0.018)^{a}
 −0.672 (0.359)^{c}

Rent  0. 623 (0.147)^{a}
 −1.225 (5.836)  0 and 1.138 (0.267)^{c}
 −4.537 (4.199) 
Size  −0.062 (0.019)^{a}
 0.118 (0.442)  −0.062 (0.018)^{a}
 −0.010 (0.137) 
Og  −2.126 (1937)  –  −0.398 (1.812)  – 
Liq  0.025 (0.008)^{a}
 0.016 (0.186)  0.025 (0.007)^{a}
 0.267 (0.3) 
Rd  0.026 (0.015)^{b}
 −0.081 (0.221)  −0.028 (0.013)^{b}
 1.243 (0.628)^{b}

Wald  1618.14^{a}
 300.69^{a}
 867.15^{a}
 261.51^{a}

Sargan  160.14^{b}
 160.14^{b}
 155.10^{b}
 155.10^{b}

Hansen  0.00  0.00  0.00  0.00 
M2  −2.87^{b}
 −1.05  −2.38^{b}
 1.30 
N  132  132  132  132 
The explanatory power of the tested models varies between 0,8663 and 0,9516 for the ratio of total debt and between 0,6887 and 0,8782 for the debt ratio in the long term. In comparison with the work of Becker and Wurgler (
2002) (r2 = 0,20) and Hovakimian (
2006) (r2 = 0.267) (Table
6).
Table 6
Summary of the results of regressions carried out to the funding model hierarchy
Def   3.949 (1.677)^{b}
  4.157 (1.752)^{b}
 −3.949 (1.677)^{b}
 −3.399 (1.477)^{b}
 −2.218 (1.573)  −2.698 (1.437)^{c}

_Cons  −1.84 e + 07 (2.38E + 07)  −1.94 e + 07 (2.79 E + 07)  −1.84E + 07 (2.38E + 07)  −1.58E + 07 (2.42 E + 07)  −1.03E + 07 (2.15E + 07)  −1.34E + 07 (3.59E + 07) 
R2 adjusted  0.04  0.0400  0.0400  0.0389  0.0389  0.0389 
R2 within  –  0.0421  0.0421  –  0.0182  0.0182 
R2 between  –  0.0070  0.0070  –  0.0751  0.0751 
F  5.54^{b}
 5.63^{b}
 –  5.30^{b}
 1.99  – 
Wald  –  –  5.54^{b}
 –  –  3.53^{c}

H  –  0.17  –  0.56 
LM  –  1.41^{c}
 –  –  –  12.12^{a}

M1  –  123.82^{a}
 –  –  –  89650^{a}

N  135  135  135  133  133  133 
On the other hand, a probability of the test specification of Hausman is greater than the threshold of 10% for the two models. Therefore, the test of Hausman does not differentiate between the fixedeffects model and the random effects one. As well, for the two models (R2 within < r2 between), we will promote the random effects model. In fact, the likelihood of statistics of the test of BreushPagan shows that there is a problem of heteroscedasticity for both ratios of debt. Similarly, the test of Wooldridge shows that there is a problem of the firstorder autocorrelation in tailings for both ratios of debt.
The variables relating to embodiment, profitability and size always retain their significance at the threshold of 1% in all the estimated models. This significance is well used in the analysis of structures of the capital of Tunisian businesses. Similarly, the ratio of MTB has kept its importance at the threshold of 1% for both ratios of debt. In this regard, for the debt ratio in the long term, the negative sign of the coefficient confirms its negative impact on the levels of indebtedness of Tunisian firms. But for the ratio of total indebtedness this coefficient has a positive but unexpected sign (Lenz and Al
2015).
The weighted average of MTB ratios is not significant and does not present even the expected sign. These results coincide with those of Leary and Roberts (
2005) that suggest that the American firms tend to readjust their structures of capital towards the target levels of indebtedness. Similarly for Alti (
2005), Flannery and Rangan (
2006), Kayhan and Titman (
2006) and Hovakimian (
2006) who consider that the attempted effect of “timing” is limited to short term. These results are in contradiction with those obtained by Becker and Wurgler (
2002) and those by Huang and Ritter (
2006). The latter consider that assessment stock prices of companies have a persistent effect and a significant impact on structures of the capital of firms.
Empirical validation of the combination of the various theories
The majority of recent studies are devoted to the analysis of variations of structures of the capital of the companies. These studies consider that such a combination is necessary considering the complexity that surrounds the decisions related to the structures of capital. Hence, this last section will deal with the combination of various theoretical frameworks that we have already presented throughout this work (Table
7).
Table 7
Summary of the results of the regressions carried out the model of “market timing”
Tang  0.249 (0.024)^{a}
 0.246 (0.024)^{a}
 0.249 (0.024)^{a}
 0.215 (0.023)^{a}
 0.213 (0.023)^{a}
 0.215 (0.023)^{a}

Rent   1.142 (0.130)^{a}
  1.167 (0.121)^{a}
  1.143 (0.118)^{a}
  0.332 (0.115)^{a}
  0.349 (0.118)^{a}
 −0.332 (0.115)^{a}

Size  0.082 (0.018)^{a}
 0.082 (0.019)^{a}
 0.082 (0.018)^{a}
 0.071 (0.018)^{a}
 0.073 (0.018)^{a}
 Execution time: 0.066 (0.018)^{a}

Mtb  0.097 (0.014)^{a}
 0.097 (0.015)^{a}
 0.096 (0.015)^{a}
 0.048 (0.014)^{a}
 0.047 (0.015)^{a}
 −0.506 (1954)^{a}

Efwmb  0.005 (0.009)  0.004 (0.009)  0.005 (0.009)  −0.007 (0.008)  −0.007 (0.009)  −0.031 (0.008) 
Ibvmt  −4.37E06 (0.00004)  −3.54E06 (0.0004)  −4.37E06 (0.0004)  9.38E06 (0.0004)  8.42E06 (0.0004)  9.38E06 (0.0004) 
_Cons  −1.570 (0.347)^{a}
 −1.569 (0.302)^{a}
 −1.570 (0.347)^{a}
 −1.358 (0.338)^{b}
 −1.374 (0.345)^{a}
 −1.358 (0.338)^{a}

R2 adjusted  0.8664  0.8663  0.8664  0.6935  0.6934  0.6935 
R2 within  –  0.8648  0.8648  –  0.6888  0.6887 
R2 between  –  0.9503  0.9516  –  0.8724  0.8782 
F  155.58^{a}
 148.20^{a}
 –  54.31^{a}
 51.28^{a}
 – 
Wald  –  –  933.47^{a}
 –  –  325.83^{a}

H  –  0.89  –  0.84 
LM  –  –  1.40  –  –  1.47 
M1  –  –  0.789  –  –  0.897 
N  151  151  151  151  151  151 
This correlation matrix of independent variables shows that there is no problem of multicollinearity between the studied variables,(all the coefficients do not exceed 0.8). The tests are developed using panel data through the software “STATA 11”. All the models analyzed are static models. They use the methodology adopted in the first section.
These models explain moderately the levels of indebtedness of Tunisian businesses listed on the stock exchange. In fact, the coefficients of determination are high for both ratios of debt. For the two models, the probability of the test specification of Hausman is greater than the threshold of 10%. In this case this test does not differentiate between the fixedeffects model and the random effects one. Of this fact, the two models (R2 within < r2 between) have led us to favor the random effects model. In addition, the probability of statistics of the test of BreushPagan shows that there is a problem of heteroscedasticity for both ratios of debt. Similarly, the test of Wooldridge shows that there is a problem of the firstorder autocorrelation in tailings for both ratios of debt (Table
8).
Table 8
Summary of results of carried out regressions
Tang  0.263 (0.024)^{a}
 0.261 (0.024)^{a}
 0.263 (0.024)^{a}
 0.235 (0.023)^{a}
 0.234 (0.023)^{a}
 0.235 (0.023)^{a}

Rent  −1.037 (0.123)^{a}
 −1.044 (0.126)^{a}
 −1.035 (0.123)^{a}
 −0.177 (0.130)   0.191 (0.120)  −0.177 (0.130) 
Size  0.056 (0.022)^{b}
 0.058 (0.022)^{b}
 0.056 (0.022)^{b}
 0.049 (0.021)^{b}
 0.049 (0.021)^{b}
 0.049 (0.021)^{b}

Oc  −0.697 (2.090)  −0.671 (2.120)  −0.697 (2.090)  −2.203 (1.989)  −2.109 (2.018)  −2.202 (1.989) 
Liq  −0.025 (0.008)^{a}
 −0.025 (0.008)^{a}
 −0.026 (0.008)^{a}
 −0.031 (0.008)^{a}
 −0.030 (0.008)^{a}
 −0.031 (0.008)^{a}

Rd  −0.012 (0.015)  −0.011 (0.016)  −0.013 (0.015)  −0.022 (0.0145)  −0.022 (0.015)  −0.022 (0.015) 
Def  −1.83E − 11 (6.49E10)  1.99e11 (6.72E10)  1.83E11 (6.49E10)  4.60e10 (6.18E10)  4.24e10 (6.39E10)  4.60e10 (6.18E10) 
Mtb  0.094 (0.016)^{a}
 0.094 (0.016)^{a}
 0.094 (0.016)^{a}
 0.047 (0.015)^{a}
 0.046 (0.015)^{a}
 0.047 (0.014)^{a}

Efwmb  0.004 (0.009)  0.004 (0.009)  0.003 (0.009)  −0.008 (0.008)  −0.007 (0.009)  −0.008 (0.008) 
Ipset  0.00003 (0.0004)  0.00003 (0.0004)  0.00003 (0.0004)  −1.20E06 (0.0004)  0.00003 (0.0004)  −1.20E06 (0.0004) 
_Cons  −1.130 (0.405)^{a}
 −1.088 (0.405)^{a}
 −1.130 (0.406)^{a}
 −0.863 (0.386)^{b}
 −0.876 (0.385)^{b}
 −0.863 (0.386)^{b}

R2 adjusted  0.8766  0.8759  0.8766  0.7307  0.7306  0.7307 
R2 within  –  0.8744  0.8744  –  0.7253  0.7252 
R2 between  –  0.9572  0.9712  –  0.9385  0.9407 
F  98.78^{a}
 104.47^{a}
 –  37.71^{a}
 39.60^{a}
 – 
Wald  –  –  987.81^{a}
 –  –  377.11^{a}

H Pvalues  –  0.50 (0.0000)  –  0.52 
LM  –  –  2.13  –  –  2.09 
M1  –  –  9.144^{b}
 –  –  9.002^{b}

N  150  150  150  150  150  150 
The variable relative to the size is significant at 5% threshold for the two ratios of debt. From another point of view, the embodiment is significant at 1% threshold for all estimates. Similarly, the variable relative to profitability is significant. Also, it has a negative sign for a ratio of total debt. This result is consistent with the predictions of the theory of hierarchical funding.
The representative variable of liquidity is significant at the threshold of 1% and with a negative sign for the two ratios of debt. This negative relationship implies that the Tunisian businesses support current assets which are to their provisions in order to finance their investments. The risk of default is not significant for the two versions of debt ratio. This result indicates that this variable does not affect the decisions related to the use of debt.
The variable on the opportunities of growth is not significant for both ratios of debt. The negative sign associated with this variable is consistent with the predictions of theories of “TradeOff” and “pecking order”. Thus, the theoretical frameworks consider that this variable affects the debtequity ratios of companies negatively. Similarly, this negative relationship is contrary to that obtained in the developed works by Dalbor and Upneja (
2004).
The variable of the deficit of funding is not significant for both ratios of debt and aims a low coefficient inversely to the predictions of the theory of “pecking order” which attaches great importance to this variable. In fact, we can conclude that this variable affects weakly the decisions on the funding of Tunisian firms.
As regards to the ratio MTB, it is significant to the threshold of 1% and positive for both ratios of debt, while the weighted average of the ratio of MTB and the variable IPSET are still not significant for versions of debt ratios.
In fact, we can conclude that the assessments affect very weakly debt the ratios of Tunisian businesses listed on the stock exchange.
The study of El Amri and al (
2015) investigates five determinants of capital structure (leverage) in three subsectors of the Omani Industrial companies (food, construction and chemical) listed on Muscat Securities Market for the period 2008–2012.
The capital structure or leverage is measured by total debt ratio. In the industrial sector as a whole; the findings of the study indicate that there is a statistically positive association between risk and tangibility and leverage. Also, there is a statistically negative association between growth rate and profitability and leverage, while there is no association with size. Regression analysis indicates that size, tangibility and risk have a statistically significant effect on leverage.
Conclusion
In terms of this analysis, the theory of “TradeOff” brings a better explanatory power on the variations of structures of the capital of Tunisian companies listed on the stock exchange.
The results obtained have demonstrated that the introduction of all the variables from the various theoretical frameworks has helped to increase the explanatory power of the model. These results lead us to conclude that the confrontations and attempts that tend to promote a theoretical framework in relation to others do not seem to be justified. In addition, the combination of these various theories allows the improvement of the explanation and the analysis in terms of financial structures of firms.
The research focused on the financial structure of Tunisian companies. It was part of the work that attempted to validate the explanatory theories of the determinants of the financial structure of firms empirically.
A brief overview of the contributions to the debate on the financial structure was presented in the first section. In the last section, the compromise theory which asserts the existence of an optimal ratio of the debt that the companies seek to achieve. The theory of hierarchical financing preferences has been tested, where companies follow a hierarchy financed by the need for funds. For this purpose, the estimate was made on a panel of data from 26 listed Tunisian companies belonging to the industrial sectors. The estimation is carried out by the panel method of data for the period from 2005 to 2010. The empirical results showed that the indebtedness of the Tunisian companies is explained by the desire to reach a target ratio of debt and not by the need of external funds.
However, several limitations characterize our research: In the first place, this study concerned only 26 listed companies (the banks were eliminated) and the period of analysis used only extends over 6 years due to insufficient accounting and financial data for the listed companies that belong to Industrial sectors. In order to have more accounting information and extend the analysis over a longer period, a subsequent study will cover all listed companies in Tunisia. Second, the “Tradeoff” theory which states that the financial structure of firms depends on market conditions, has not been the subject of empirical verification. A subsequent study will carry out an empirical check in this direction. In addition, other variables that may influence the financial structure have not been identified. In the present study only variables from the compromise theory and that of the hierarchical preferences of financing were selected. Like any research work, this study has some limits. First, the Tunisian stock market is very modest in terms of number of listed companies which justifies our small sample. Secondly, we find that the coefficients of determination are low for the regression model which indicates a poor linear fit. Introducing aspects of behavioral finance such as over confidence and optimism could be a new insight into the theory of capital structure. Further studies are also suggested to examine the impact of national culture and religion on capital structure choice.