Efficiency of Tunisian Commercial Banks According to the Intermediation Approach

This work focuses on the performance of the banking sector through the analysis of the efficiency of Tunisian commercial banks according to intermediation approach. To measure the cost inefficiencies in banks in Tunisia, we use a translog specification for single-product cost function. The estimated cost efficiency shows a strong correlation between positive and significant efficiency scores according to the approaches (SFA) and (SFAA). The empirical results show that the efficiency cost, depending on the approach SFAA has evolved mixed between 1993 and 2010. Regarding the estimation of efficiency depending on the size of banks, the results show that banks in small and medium size of our sample show medium cost efficiencies better than those made by banks in large sizes. While on the side of profit estimates show that banks of large and medium sizes are more efficient than banks of smaller sizes. Internal determinants of the productive performance of Tunisian banks show that the preponderance of credit activity in relation to other outputs represents a source of efficiency for banks.


Introduction
The banking system plays a crucial role in the functioning of the economy.For this reason, several studies have been conducted to determine the essential components of bank performance.One of these components is concerned with the notion of efficiency.The efficiency of financial institutions has long been investigated.The results have obvious implications for the management of banks seeking to improve operational performance and decision makers who are concerned about bank competition, bank security and soundness of banks.
There is a strong relationship between the efficiency and performance as they are interpreted in the same direction.Indeed, when a bank is efficient, it is obviously effective.According to Chaffai and Diesh (1998), bank profitability is dependent on the efficiency with which the bank organizes its production.Therefore, it seems that efficiency is a determinant of productivity and therefore the performance of the bank.The term performance is more comprehensive than the term efficiency.Moreover, according to the approach of performance in social management, performance covers the notion of productivity of men (efficiency) and social goals assigned to managers such as the development of skills and employee engagement (efficiency).Our goal is to test the hypothesis of positive correlation between efficiency and performance from some performance indicators and scoring efficiency.
Several studies on the efficiency of banks have adopted the intermediation approach (Florian 2012;Kablan 2012;Allen, Engert and Liu, 2006;Allen and Liu, 2005;Isik and Hassan, 2002;Burkart, Gonsard and Dietsch (1999) ;Chaffai 1998;Berger and Mester, 1997;Zaim, 1995).For example, the article by Burkart, Gonsard and Dietsch (1999) presented an analysis of the cost efficiency and profit efficiency of French banks.These authors showed that the average cost efficiency is around 12% and the dispersion of efficiency scores is quite low, which means that the French credit institutions are relatively close to each other in terms of productive performance.The average profit inefficiency is around 9% and it appears that the behaviors are more similar in terms of ability to extract profits.measurement of inputs and outputs of a bank.In general, each input definition and production carries with it a particular set of concepts that influence and limit bank analysis of the characteristics of the production industry.
The data of our work is mainly based on activity reports published by banks and databases of the Tunisian Professional Association of Banks and Financial Institutions (TPABFI).Our sample consists of all commercial banks with the exception of two banks which are the Tunisian Solidarity Bank (TSB) and Arab Banking Corporation (ABC) who had no regular activity during the period of our study.In addition, for reasons of statistical homogeneity, we excluded from our study another small bank which is Citibank (CB).
Our work is limited to the deposit banks because they occupy the most important place in the financing of the Tunisian economy.In fact, over 90% of the savings is collected by commercial banks and over 80% of loans are granted by banks.Our final sample includes observations starting from 1993 to 2010, 18 years, corresponding to the period of banking sector reforms.It refers to 11 Tunisian commercial banks namely:

Using the Intermediation Approach
Our methodological approach is based on the intermediation approach which seemed the most appropriate to address our concern, given the importance of interbank activity and weight of the interest costs.Tunisian banks use the funds at their disposal to give primarily loans.It follows that the liability of Tunisian banks tends to be regarded as an input rather than output.It should be noted that deposits are the financial capital of the bank and are considered an input.It should be noted that the financial costs are included in our work.

Definition of Variables
Variables that will be analyzed are two in number: the banking inputs and the banking outputs.

The Bank Outputs
The outputs provided by the Tunisian commercial banks are classified into two categories: Total Credits (TC) and portfolios of securities.These two types of outputs are measured in monetary units, that is to say dinar directly from the bank's balance sheet.

Total Credits (TC)
The total of credits includes the Customer loans (CC) and interbank loans (PI).Indeed, customer loans portfolio is formed by discount accounts receivable from customers, loans on special resources and other loans to customers.While for interbank loans, they gather loans to banks and agencies, cash, Central Bank of Tunisia (CBT), certificates of deposit, treasury bills purchased and giro.

Securities Portfolios (PT)
Portfolios securities represent securities portfolios and investments considered services provided by the bank.Securities portfolios included as a line item in the balance sheet (PT = trading securities portfolios + investment portfolios).To measure banking, we have taken as a proxy aggregate output (Q) obtained by aggregation method proposed by BHH (1982)  m: The number of banks in the sample; : A geometric average of bank outputs defined as: .

Banking Inputs
The outputs listed above are produced by the combination of factors of production, namely labor input (L), physical capital input (K) and the financial capital factor (F). Various forms of deposits which constitute the financial capital input are considered, as stipulated supporters of intermediation approach.These inputs are measured as follows:  L: The number of employees;  K: Net Fixed Assets;  F: Borrowing from (CBT), money market and specialized agencies + Demand deposits Customer deposits + Savings + Coupon accounts and other financial terms + Other amounts due to customers + Special resources + Bonds and other loans.

Bank Production Costs
Endogenous variable is defined by the Total Cost (TC), which includes all the operating and financial costs.Financial costs are mainly interest expenses.Operating costs correspond to the costs of labor and capital, that is to say, the payroll and general operating expenses.

Bank Input Prices
After determining the cost of each input bank, we can evaluate the prices of these inputs.In fact, the price of each factor of production is measured by the ratio between its cost and quantity.Bank input prices are determined as follows:  The price of labor "P L " is obtained by dividing personnel expenses by the number of bank employees:  The price of physical capital "P K " is obtained by dividing the general operating expenses by property:  The price of financial capital "P F " is obtained by dividing the financial burden on borrowed resources: Table 1 shows the descriptive statistics (mean, standard deviation, minimum and maximum) of the total cost, total profit, aggregate output, inputs and their prices.We emphasize that public banks (TBC, ANB and HB) are ranked the first in almost all inputs and outputs.Thus, they have larger amounts of deposits, capital, labor, total credit.Only (IABT) ranks sometimes with these three banks.
But on the prices of inputs, we note that private banks offer higher salaries.The price of physical capital and financial capital are relatively lower compared to the price of labor.The (IABT) offers the highest salary about one and a half times higher than the average TBC.This could be explained by the different recruitment policies between banks (in terms of qualifications, for example). Source: Calculated from data provided by the (TPABFI).We note that these data are in thousands of dinars, with the exception of labor (L) which is measured by the number of employees.

Analysis of the Econometric Method Used
The method of stochastic frontier when panel data assumes that the specific inefficiency of the company varies over the time, which is the major advantage of this method of efficiency estimation.For our work, we use the cost function translogaritmique.

The Translog Cost Function
To measure the inefficiencies cost in banks in Tunisia, we use a translog specification for single-product cost function: CT it : The cost function to estimate the bank i in year t.
Q it : The aggregate output of bank i at time t.
PL it : The price of labor bank i at time t.
PK it : The price of physical capital factor of bank i at time t.
PF it : The price factor financial capital of bank i at time t.
v it : The term random error identically and independently distributed according to a normal law N(0, σ v 2 ).
u it : The term asymmetric error measuring inefficiency: u it → N(m it , σ u 2 ).
Distribution of random term (u it ) on the extent of inefficiency is that of a truncated normal distribution with zero variance (σ u

2
) and expectation (m it ) defined by: m it = Z it δ Where (δ) is a vector of (p) parameters to be estimated, and (Z it ) a vector of (p) variables can affect the efficiency of bank (i) at time t.However, the Hessein the cost function is symmetric, the equality must be satisfied for any pair of variables and .
Symmetry leads to the following restrictions: ; , , , In addition, any cost function must be homogeneous of degree one in input prices.Thus a proportional increase in prices increases the total cost in the same proportion without demand factors being affected.This homogeneity condition implies other constraints expressed as follows: Homogeneity constraint is taken into account by normalizing the Total Cost (TC), the price of physical capital (P K ) and the price of financial capital (P F ) by the price of capital work (P L ).A choice that does not affect the results since the estimators are obtained by the method of maximum likelihood.
Within these constraints, we obtain the following transformed model:

Test of Homogeneity Total
We start first by testing complete homogeneity: the latter is noted as Fisher's test A; B = A i , B i in the result file of TSP.The letter A is the constant here, while the letter B denotes the vector of coefficients of the explanatory variable.
As part of our sample, the realization of the Fisher statistic associated with the test , denoted F 1 is 0.7385.
The software also shows the number of degrees of freedom for this statistic (Note that F1 followed a Fischer with [(N-1)(K+ 1)] and [NT-N(K +1)] degrees of freedom).
Given the size of our sample and the number of explanatory variables (K = 9), we must compare the value of this achievement to the threshold of a Fischer F (100, 88).The software gives us directly the p-value associated with this test.In this instance, the p-value indicates here that until the 75% threshold, the null hypothesis can not be rejected.So our panel represented by equation ( 2) is completely homogeneous (pooled).This homogeneity is explained by the fact that banks in our sample are of the same nature (commercial banks and we excluded development banks and offshore banks).In addition, these banks are indirectly controlled and supervised by the State and apply the same banking reforms.
We can regain our panel structure (pooled model) performing the test that is to test the equality of N individual constants : This test is denoted as Fisher's test A; B = A i , B in the result file of TSP.This test is to confirm or refute the test results , given the fact that reducing the number of linear restrictions can increase the power of the test of Fischer (Note that with F 3 followed a Fischer [(N-1) and N( T-1)-K] degrees of freedom, ie by an F (10; 178)).As part of our sample, the realization of the Fisher statistic indicates a value of 1.5786.The p-value (0.5536) is well above the 5% threshold.So we can not reject H 0 equality of constants .

Determination of the Level of Inefficiency
Added to the estimation of the efficiency of Tunisian banks, our goal is to highlight the link between performance and productive variables in banking supervision on the one hand and external variables on the other.Since each bank has its own characteristics, this study uses the internal characteristics of the bank to assess managerial efficiency.The first relates to the financial characteristics of banks and the second concerns its organizational characteristics.

The Financial Characteristics
Financial characteristics are assessed by three variables:

Variable Trade Policy
To examine the relationship between efficiency and trade policy of banks, we used the following ratio: total active credit (RC).Given the nature of the debt economy, causing economic agents in Tunisia to be funded primarily through the banking system, it is tempting to equate this indicator with an enviable achievement, which is a performance as it relates to deposits collected, the ability to refinance the money market and the importance of equity.A bank is successful in this regard when this ratio is high.Thus a positive impact ratio (RC) on bank efficiency is expected.

Variable Regulatory
Among the variables related to the regulation, it holds the weight of equity to total assets (CAP), variable reflecting the state of regulatory constraints on capital: its orientation depends on the degree of risk aversion.

Variable PPC
To account for the influence of non-performing loans on efficiency in the banking sector, we introduce the variable (PPC) measured by the ratio of provisions for credit losses (as a proxy of non-performing loans) to total loans: a negative impact is expected.

The Organizational Characteristics
Organizational characteristics are assessed by two variables: the variable logarithm of total assets (Ln(TA)) and the variable Trend (T).

Variable Logarithm of Total Assets (Ln (TA))
The final impact of this variable on expected efficiency depends on the critical size.Indeed, Aly et al (1990), Berger et al (1993) found a positive relationship between size and efficiency of large U.S. banks, and Isik et al (2005) for banks in Jordan.On the other hand, Hermalin and Wallace (1994), Kaparakis et al. (1994), De Young andNolle (1996), Isik and Hassan (2002a) found a negative relationship.However, other studies did not find any significant relationship between the size and the level of efficiency, for example the work of Aly and al. (1990), Cebenoyan and al. (1993), Mester (1993), Pi and Timme (1993), Mester (1996), Berger and Hannan (1995), Berger and Mester (1997) and Chang and al. (1998).

Trend Variable (T)
To take account of technological developments in the Tunisian banking sector during the period of our study, a positive impact on productive efficiency is expected.To isolate the effects of the internal characteristics of the bank on efficiency, it is necessary to control by other factors that were used as determinants of bank efficiency.We also examine how the efficiency of the banking sector is related to the structure of the banking market.Therefore, we analyze the competition by Tunisian banks index approach, and more specifically by the index of concentration of Herfindahl Hirshman Index (HHI).
Several studies have highlighted a negative relationship between concentration and efficiency (Fecher and Pestieu, 1993;Berger and Hannan 1997), while others have identified a positive relationship.Indeed, the Cournot model of oligopolistic behavior, predicts that market power is positively related to the profitability and efficiency, as banks with large market shares can charge higher prices, discourage competition and be more productive (Berger 1995;Berger and Mester 1997;Weill, 1998, Isik andHassan 2003;Isik and al. 2005 for banks in Jordan).
The final impact of the expected concentration on efficiency is therefore ambiguous.
Thus, the expected random term on the extent of inefficiency is defined by: (3) Table 2 shows the main statistical characteristics of the variables of inefficiency term.

Interpretation of Model Results
To assess the relevance of the results obtained in this analysis, on the one hand we have based our arguments on the significance of the explanatory variables used in this study and on the other hand, we have based on apparent sign of the coefficient.The estimated parameters of the cost frontier function (II) are shown in the following table.
Coefficients and degrees of efficiency of each bank are estimated by the method of maximum likelihood using the software FRONTIER 4.1 (Coelli, Prasada Rao and Battese, 1998).It uses alternative parameterization of the likelihood function, which substitutes: In particular, a value of zero indicates that the deviations around the border are entirely due to noise, while a value of unity indicates that all deviations are due to inefficiency.The quality of the estimates is satisfactory and the coefficients are significant at the threshold of 1%.
From Table 3, the first result that can be drawn concerns the test of maximum likelihood.In fact, this test checks whether a model is generally explanatory.When the empirical ratio in question (L.R) is greater than the theoretical value of chi-square at the 1% level, we conclude that the fit is generally considered explanatory.In our case, the overall explanatory model, the theoretical value of chi-square, 21.67, at the 1% and 9 degrees of freedom is less than the respective empirical ratio (the degree of freedom is the number of exogenous variables model).
The second result shows that the parameter is significantly different from zero.This result rejects the hypothesis that the variance of the efficiency is zero.Therefore the term can not be excluded from the regression and parameter estimation by ordinary least squares method is inadequate.

Hypothesis Testing for the Model Parameters
Before interpreting the results of the function of stochastic cost frontier we make various specification tests.The following table presents a number of statistical tests based on the likelihood ratio, which are expected to examine our model in comparison to more limited forms.
The first hypothesis examines the extent to which it is possible to assume a more limited form of the translog specification of the cost function.Indeed, this first hypothesis examines whether the cost function can be represented by a Cobb-Douglas.This hypothesis is rejected at a significance level of 99%.Other additional hypotheses examine alternative features of the equation that determines the cost inefficiency, because the cost function remains in its translog form.
The second hypothesis is to examine whether it is possible to assume a model in which the cost inefficiency does not exist, ie, if it is possible to work with a model of ordinary least squares (OLS).This hypothesis is strongly rejected at a significance level of 99%.The third hypothesis examines the case in which the inefficiency is not a linear function of the independent variables.This hypothesis is strongly rejected at the 1% level.In the latter case, we performed the reasonableness test to test the hypothesis of technological change in the Tunisian banking sector.
The null hypothesis of no technological change was strongly rejected at the 1% level.
In conclusion, the proposed model represents an improvement when compared to functional characteristics and particularly restrictive in comparison with these characteristics in which the inefficiency component is not based on financial and managerial variables that determine the activity of the bank.

Interpretation of Efficiency Scores
Our goal is to interpret the evolution of efficiency scores over time.Indeed, the following table shows annual estimates respectively of cost inefficiency scores obtained from the two approaches (SFAA) and (SFA).By definition, the method (DFA) can not give annual scores, but only an estimation of the average efficiency over the entire period (that is to say by bank).From this table, we see that the efficiency cost, depending on the approach (SFAA) has evolved mixed between 1993 and 2010.This allows to deduce that the efficiency of Tunisian commercial banks fluctuates.This result was verified by Chaffai and Dietsch (1998) and Cook, Hababou and Gordon (2000).It is from 2000 that the Tunisian banking sector marked a steady increase in efficiency, saving the cost efficiency scores large enough to reach, in 2009, its highest level of the period which is to the order of 85.67%.
In fact, the average cost efficiency has increased by 8.17% during the period 2000-2005.
Table 5.Average scores of cost efficiency in time according to the approach (SFAA) On the approach (SFA), we see that the average efficiency has increased during our study.Indeed, examining the evolution of cost efficiency over the study period, we note that efficiency levels increase from one year to another almost on a regular basis.Indeed, the efficiency scores have increased from 68.37% in 1993 to 74.62% in 2010.In fact, the average cost efficiency has increased by 6.25% during the period 1989-2006. www.ccsen

Estimated Cost Efficiency
Banks of small and medium size of our sample show better than average efficiencies realized by banks of large sizes.In this context it's important to emphasize that the results by size of banks, shown in the following table, draw a trend toward decreasing relationship between the size and the degree of efficiency cost.
The average efficiency reaches Indeed, 86.20% for small banks, 71.44% for medium-sized banks and 57.54% for large banks.This confirms the words of Berger and De Young (1997), suggesting that inefficiencies are primarily the result of a misallocation of financial resources (deposits and borrowed funds) and of miscalculating the risk of non-performing loans which is on a large scale in financial institutions and public and private banks of large sizes.The estimated parameters and the hope of inefficiency (III) are shown in Table 9.These estimates of the determinants of productive efficiency provide a first insight into the managerial assumptions that we would evaluate.Our results show that there is a negative relationship between the provision for credit losses (as a proxy for non-performing loans) and the level of cost inefficiency.This result can tell us that after a deterioration in the level of productive efficiency of Tunisian banks in a year, the amount of non-performing loans increases in the following year.This evidence was also confirmed by testing Spearman correlations and Kendall.Spearman's rho (which is equal to -0.4919 and significant at 1%) and Kendall's score (which is equal to -6861 and significant at 1%) allow us to reject the hypothesis H 0 that the non-performing loans and cost efficiency are independent.Negative signs tests show that the two variables are negatively correlated.
We also find a positive relationship between the variable capitalization and cost efficiency but not significant.
After the period has been marked by fluctuations in the cost efficiency Tunisian banking sector has improved considerably.Thus, even if the process of modernization and reorganization was launched in 1997, it is from the year 2000 that we could report a significant effect of this process on bank efficiency scores.Which justifies the negative sign and significant coefficient on Trend (T), reflecting technological developments in the banking industry.
In fact, investments in new technologies at the interbank clearing or development of the network atmosphere (ATM), were allowed to develop banking with a lower cost.But some time was needed so that the predicted effects of these investments appear because these new technologies are accompanied by a reorganization of the business and a new allocation of functions to the bank.
Technological advance has a positive effect on the development of the banking sector mainly in the field of information technology and communication.In fact, this technological advance has created new channels of distribution of banking products and services such as online banking, Internet, electronic funds transfer and others.These changes reduce transaction costs and increase competition in the banking sector.Thanks to technological progress, we can properly perform several operations with minimum cost such that the operation of collection or analysis and provision of information.
In addition, new technologies have contributed greatly to the transparency of information which was often hidden for strategic reasons.The information is obtained quickly and with less cost.Technology is a favorable asset to improve the profitability of banks.It is present at all stages of the bidding process.Innovation in products can also be a competitive advantage.Each bank has to offer a new product or service that will take advantage and increase its visibility.
In addition, we find a positive relationship between cost efficiency and total assets ratio credits (RC), which is a measure of credit risk.This result suggests that the most active banks in the credit segment tend to be more efficient.
As logarithm of total assets (Ln (TA)), its coefficient is significantly positive for the banking sector.This source of inefficiency appears to be related to the critical size, that is to say the presence of diseconomies of scale.Thus, the size of the bank affects negatively the level of efficiency.The more the size increases, the more the efficiency level decreases.
This result shows that Tunisian banks do not have the managerial capacity to manage an important total assets.The more the size increases, the more the waste of resources increases.This result is consistent with the work of Hermalin and Wallace (1994), Kaparakis et al. (1994), DeYoung and Nolle (1996), Isik and Hassan (2002a) who found a negative relationship between efficiency and total assets.
Finally, concerning the structure of the banking market which is measured by the index of concentration of Herfindahl Hirshman Index (HHI), its impact on bank efficiency is negative.This result can be supported by the fact that financial liberalization in the banking sector in Tunisia was meant as it seems, to increase competition, to improve the productive efficiency of banks.
Which justifies the negative impact of the market share variable (PM) on the level of efficiency.In fact, the three major Tunisian banks (ANB, TBC and IABT) which are less efficient on average, share almost 50% of market share of the Tunisian banking sector.This shows that the concentration is less beneficial in terms of efficiency in Tunisian commercial banks than the competition.Therefore, these banks will be forced to diversify their activities.
We can account for these results by the fact that (ANB) is heavily involved in the agricultural sector subject to weather conditions and thereby erecting in a very risky sector, which implies that the bank has relatively high provision risks and therefore very low margins.The (TBC) continues its role as the centerpiece of the economic state policy in financing industrial firms which are in difficulty.In additional to that, (IABT) finances a large part of SMEs whose size is lower and whose risk of default is high.
The results of the efficiency side of profit show a positive and significant relationship between efficiency and profit capitalization ratio.This shows that an increase in bank capitalization leads to an improvement in the quality of capital and subsequently an increase for banks.In addition, this could be explained by the fact that, in light of the high standards of equity, banks may decide to replace loans by alternative forms of capital (see Vanhoose, 2007;Pasiouras and al. 2008).
However, it is hard to notice that our estimates do not imply causality between productive efficiency and managerial behavior.We simply find a correlation between non-performing loans and cost efficiency, which supports the view that most Tunisian banks are saddled with an overhang of non-performing loans which obviously affect their efficiency levels.

Conclusion
In this model, we have used an extension of the stochastic frontier approach which assumes a truncation parameter specific to each bank.This allowed us to assess changes in the level of cost efficiency of Tunisian commercial banks and to identify relationships between the productive performance and a combination of internal and external determinants.
The estimated cost efficiency shows a strong correlation between positive and significant efficiency scores according to the approaches (SFA) and (SFAA).The empirical results show that the efficiency cost, depending on the approach SFAA has evolved mixed between 1993 and 2010.
Regarding the estimation of efficiency depending on the size of banks, the results show that banks of small and medium size of our sample show medium cost efficiencies better than those made by banks of large sizes.While on the side of profit, estimates show that large and medium-sized banks are more efficient than smaller ones.
Focusing on the determinants of internal productive performance of Tunisian banks, our results show that the preponderance of credit activity in relation to other outputs represents a source of efficiency for banks.
Among the factors that negatively affect the level of efficiency, there is the size.Thus, the larger the size, the more the level of efficiency decreases.In addition, a negative relationship between the provision for credit losses (as a proxy for non-performing loans) and the level of cost inefficiency is clear.Finally, concerning the structure of the banking market, we find a negative relationship with bank efficiency.This shows that a competitive environment is more beneficial in terms of efficiency for Tunisian commercial banks.

Table 1 .
Descriptive statistics of sample variables

Table 2 .
Statistical characteristics of the explanatory variables of the inefficiency term

Table 3 .
Parameter estimates of the translog cost function (SFAA)

Table 4 .
Tests of hypothesis likelihood ratio on the parameters of the stochastic cost frontier approach according to (SFAA) : represent the logarithms of the likelihood of the estimated models under the null and alternative hypothesis.This statistic is asymptotically follows chi-square with two degrees of freedom as the number of restrictions under the null hypothesis.(**) Value likelihood ratio test given by the table.1.The cost function can be represented by a Cobb-Douglas.2. The term inefficiency has no effect.3. The inefficiency is not a linear function of the independent variables.4. No technological change in the Tunisian banking sector.

Table 8 .
Average scores of cost efficiency in Tunisian commercial banks according to their size

Table 9 .
Explaining the level of cost inefficiency