Financial Liberalization,
Political Regime Instability and Bank Efficiency
Vu Thi Hong Nhung
Research Master IE&B
Faculty of Economics, University of Groningen
Supervisors: Prof. Niels Hermes
Dr. Bart Los
Financial Liberalization,
Political Regime Instability
and Bank Efficiency
∗Vu Thi Hong Nhung
Abstract
This paper investigates the impact of financial liberalization on bank efficiency. Moreover, we analyze whether the impact of financial liberalization on bank efficiency is conditional on political regime instability. We apply so called Data Envelopment Analysis (DEA) to calculate bank efficiency at the individual bank level and use a new and detailed database that measures different aspects of financial liberalization. The dataset consists of a sample of over 4,000 bank-year observations from 10 emerging countries in Asia and Latin America for the period 1991-2000. Overall, we find evidence to support our hypothesis that the effect of financial liberalization on bank efficiency is positive in case of regime stability. The positive effect of financial liberalization on bank efficiency is mitigated or even ceases as political regime instability increases.
Keywords: financial liberalization, bank efficiency, DEA, regime instability
∗
Table of Content
1. Introduction... 4
2. Review of the Existing Literature and Theoretical Framework Discussion... 5
2.1 Treatment of Efficiency in Theory ...6
2.2 Review of the Existing Literature on Financial Liberalization and Bank Efficiency...8
2.3 Theoretical Framework of the Interrelationship between Financial Liberalization and Political Regime Instability on Bank Efficiency ...13
3. Measuring Bank Efficiency ...14
4. Measuring Political Regime Instability ...21
5. Data and Methodology ... 24
5.1 Data of Financial Liberalization...24
5.2 Data of Bank Efficiency...25
5.3 Econometric Framework...29
6. Empirical Results... 32
6.1 The Impact of Financial Liberalization on Bank Efficiency ...32
6.2 The Interrelationship between Political Regime Instability and Financial Liberalization on Bank Efficiency ...36
7. Robustness Analysis... 42
8. Conclusion ... 43
References... 46
1. Introduction
During the last three decades many countries have liberalized their financial markets. The main goals of financial liberalization aim at enhancing competition, improving resource allocation, and acquiring more administrative efficiency in financial institutions by making them less state-directed and exposing them to increased market competition (Barajas and Steiner, 2000). Despite a similar motivation behind the trend towards financial liberalization, empirical evidence provides mixed results on the impact of financial liberalization on bank efficiency. Some studies show that bank efficiency improves considerably after the implementation of financial liberalization policies. However, others find little or even a negative impact of financial deregulation on bank efficiency. The results in both theoretical and empirical literature raise the question why bank performance in different countries responds differently with financial liberalization policies. This motives us to examine more deeply the relationship between financial liberalization and bank efficiency.
It has been widely discussed in economic growth theory that macroeconomic stability and price stability play important roles in the success of financial liberalization policies. Moreover, many authors claim that economic performance also partly depends on political and institutional factors (Marquand, 1990; De Haan and Siermann, 1996). Political instability raises uncertainty with respect to future institutions and policymakers, thereby generating uncertainty to government policies. Using literature discussed in the economic growth theory, we argue that the effect of financial liberalization policies on bank efficiency is conditional not only on macroeconomic conditions but also political factors. The success of financial liberalization depends on stable macroeconomic conditions and political stability.
have looked into the efficiency effects of financial liberalization policies, most of these studies focus on just one country. Our study is one of the very few that analyses this issue using panel data regression analysis. In the analysis, we explicitly take into account the extent of financial liberalization policies and their effect on bank efficiency. It is hypothesized that financial liberalization programs have a positive impact on bank efficiency. We use a new dataset holding information on several different dimensions of financial reform policies. Using this dataset we are able to measure in more detail how financial liberalization at the country level may affect the efficiency of banks over time.
Moreover, this study improves the existing literature by investigating whether the relationship between financial liberalization and bank efficiency is conditional on the political regime instability. The political regime instability is calculated by Exploratory Factor Analysis. The so called multiplicative interaction model is employed to examine the interrelationship between financial liberalization and political regime instability on bank efficiency. We hypothesize that the positive effect of financial liberalization on bank efficiency is mitigated if political regime is instable. To the best of our knowledge, this is the first study that explicitly investigates the interrelationship between financial liberalization and political regime instability on bank efficiency.
The paper is organized as follows. Section 2 discusses previous literature about the efficiency theory, the relationship between financial liberalization and bank efficiency, and then we discuss the framework through which political regime instability influences the relationship between financial liberalization and bank efficiency. Section 3 describes the methodology used to measure bank efficiency. Section 4 explains the measuring of political regime instability. Section 5 discusses data and research methodology. The results of empirical analysis into the relationship between financial liberalization and bank efficiency, the interaction effects of political instability and liberalization polices on bank efficiency are discussed in section 6. Section 7 checks the robustness of our results by using a different measure of financial liberalization. Section 8 concludes the paper.
interrelations between financial liberalization and political regime instability on bank efficiency.
2.1 Treatment of Efficiency in Theory
The theory of production considers the production process in which a firm utilizes various resources (inputs) and produces goods and services (outputs) to satisfy the desires of its customers. Efficiency is an important characteristic of producer performance. Efficiency is considered as the degree to which the observed use of physical resources to produce a given quantity of outputs matches the optimal use of physical resources to produce outputs of a given quantity (Debreu, 1951). Conventional microeconomic theory explicitly assumes that the producers are successful in operating in an efficient manner. In particular, from a technical or engineering perspective, it is assumed that producers are optimizing their behavior i.e. they do not waste resources. From economic perspective, producers are assumed to optimize by successfully operating at minimum cost (Fare and Lovell, 1994). However, in recent years, an increasing number of writers have turned their attention to the possibility of inefficiency in production since they have realized that the existence of productive inefficiency can not be denied (Fare et al., 1985).
Farrell (1957), inspired by the work of Debreu (1951) and Koopmans (1951), is one of the most influential writers dealing with measuring firm efficiency. Farrell (1957) recommended that the efficiency of the firm consists of two components. Overall technical efficiency reflects a firm’s ability to achieve maximal outputs with a given of inputs, and allocative efficiency, which reflects a firm’s ability to use inputs in optimal proportions to produce a given quantity of output at minimal cost of given the input prices. Allocative efficiency can be combined with overall technical efficiency to measure economic efficiency. Some of Farrell’s terminology differ from recent literature terminology; for example he uses price efficiency instead of allocative efficiency and the term overall efficiency instead of economic efficiency.
inputs which a firm uses to produce a unit of output. This isoquant represents the production frontier of fully efficient firms, which permits the measurement of overall technical efficiency. For a given set of input prices, the isocost line AA’ represents the various combinations of inputs that have the same level of costs. Farrell considers the case where the firm utilizes the quantities of two inputs to produce one unit of output given by point P in the Figure 1. Then he recognizes there are two types of inefficiencies. First, the overall technical inefficiency is represented by QP, since moving from point P to point Q the firm could produce the same output with fewer inputs. The overall technical efficiency (OTE) is measured by the ratio OTE = OQ/OP. This ratio takes value between one and zero, and therefore the higher value of this ratio indicates the higher technical efficiency of the firm. The perfectly efficient firm gets the value unity. Second, the allocative inefficiency is identified. Since in this case, the optimal method of production is Q’ not Q. Moving from point Q to Q’, a firm which adjusts the given input prices could produce the same output at a lower cost. The allocation efficiency (AE) can be measures by the ratio AE = OM/OQ. Farrell also defines economic (or overall) efficiency (EE) by the ratio EE = OM/OP, which relates to overall technical efficiency and allocative efficiency (OQ/OP x OM/OQ). Thus, economic efficiency is also considered as the product of overall technical efficiency and allocative efficiency EE = OTE x AE. It is noted that all three types of efficiencies are bounded by zero and one.
Although the concepts of the different types of efficiencies are not too difficult to understand, researchers encounter substantial difficulties to measure them. The majority challenge is that efficiency measure is based on the best practice frontier which includes all fully efficient firms. However, the best practice frontier is not known in practice. Therefore the main task of estimating efficiency is to estimate the best practice frontier. In practice, the best practice frontier is defined in each case based on a set of observations and the purposes of each research. In particular, the alternative of production, cost, profit or other frontiers can be used. First, the production frontier is associated with the maximum attainable level of output from given a level of inputs, or the minimum level of inputs required to produce a given output. Second, cost frontier refers to the minimization level of cost (input prices) needed to produce a particular level of output. Thirdly, the profit frontier is associated with the maximum level of profits that can be obtained from a given set of output and input prices. These three types of frontiers have the same characteristic of optimality, derived from maximum or minimum of technology and prices. Efficiency levels are measured as the distance from each observation to such a frontier. There are several techniques to estimate those frontiers (discussed in detail in the next part).
We have just investigated the concept of efficiency in the production theory. The theory of the firm considers that the managers’ main aim is to manage firms to operate in an efficient manner by maximizing firms’ profits and shareholders’ wealth (Isik and Hassan, 2003). The theory of production for financial firms emphasizes that financial firms, including commercial and saving banks, savings and loans associations, are also profit maximizing entities. They involve the production of intermediation services between borrowers and lenders (Hancock, 1990).
Efficiency and productivity of financial institutions have become more and more important in growing competitive business environments. Performance evaluation is considered as a necessary tool for financial firms to improve their business in order to survive and prosper.
2.2 Review of the Existing Literature on Financial Liberalization and Bank
Efficiency
2003). However, banking systems of many developing countries exhibited poor operations. Under-performance was seen as a result of many restriction regulations. Particularly, banking industry in many developing countries was controlled by government. Consequently, banks were instructed by the state in allocating credit to specific sectors, setting interest rates on deposits and loans, establishing deposit insurance, etc. Moreover, licensing market entry and exit and expanding the number of braches were also heavily regulated by government (Kumbhakar and Sarkar, 2003, Isik and Hassan, 2003). In this environment, banks had little motivation to improve their performance by reducing operating costs, mobilizing deposits and making loans in creditworthiness. The importance of financial sector and its influence on other sectors demanded that liberalization programs were undertaken. Starting in the 1980s, a number of countries implemented liberalization in banking sectors to aim at improving their efficiency and productivity. For instance, the United States ended a long trend toward stricter regulation in the late 1970s and began a deregulation era in the early 1980s (Brooks and Oh, 1999). Most European countries agreed to harmonize and unify the banking market under the objectives of the European Union in the 1990s. Asian and Latin American countries carried out financial liberalization during the early 1990s.
low interest rates on deposits discourage household savings. Interest rate liberalization leads to increase interest rates, hence increasing opportunities for banks to mobilize deposits. Moreover, the lifting of barriers for foreign entry in liberalization programs attracts an increase a number of foreign banks in the banking system. Foreign banks will foster competition and thereby influence the efficiency of domestic banks (Hermes and Lensink, 2001). These effects may be explained by the fact that the presence of foreign banks puts pressure on domestic banks to become more efficient by reducing costs. Moreover, domestic banks are likely to import technology from foreign banks, so technology spill-overs may help the formers to lower costs (Claessens et al., 2001).
However, critics have pointed out that financial liberalization has adverse effects to bank efficiency. For instance, the Post-Keynesian model shows that a rise in the interest rate will lead to an increasing in the cost of borrowing. The increase interest rates encourages risky projects, since the ‘safe” projects will not be feasible. Under high interest rate circumstances, banks now encounter problems of declining of average quality of their borrower pool. This dilemma results from the increase of loan interest rates rather than asymmetric information (Grable, 1995). In this case, the probability that banks allocate credit to risky projects may increase following financial liberalization, and hence the allocation of funds by banks becomes less efficient. Additionally, Gibson and Tsakalotos (1994) argue that with changes of induced expectations in the aftermath of financial liberalization, not only investors but also financial institutions may be compelled to pursue speculative activities. Therefore, these financial institutions may likely reduce financing real sector investments. As a result, banks are likely inclined to fund speculative investments, and may implement maturity transformation by borrowing short-term and lending long-term. In this case, if deposit interest rates increase as a result of financial liberalization, banks may suffer large losses since they cannot increase lending interest rates. Due to these problems discussed above, financial liberalization may threaten economic growth by financial crises, which leads to bank collapses and disruptive spillover effects in the real sector (Grable, 1995; Demirgüç-Kunt and Detragiache, 1998). Obviously, under these circumstances, bank efficiency declines.
studies using data from the same countries come to opposite conclusions. Below, we review the empirical research focusing on a number of emerging economies.
Gilbert and Wilson (1998) analyze the impacts of privatization and deregulation on the productivity of Korean banks in the period from 1980 to 1994. They use Malmquist indexes of productivity, allowing them to decompose changes in productivity into changes in technical efficiency and changes in technology. They find that privatization and deregulation improve productivity and potential output among Korean banks. However, extending the research of Gilbert and Wilson (1998), Hao et al. (2001) use the stochastic cost frontier approach to investigate the productive efficiency of private Korean banks from 1985 to 1995. They conclude that there is little or no positive relationship between the financial deregulation of 1991 and measured bank efficiency.
Isik and Hassan (2003) analyze total factor productivity changes in Turkish banks during the deregulation of the financial market 1981-1990 by employing the DEA-type Malmquist Index. Their results indicate that all types of Turkish banks improve their performance considerably after the implementation of financial liberalization. However, Denizer et al. (2007) examining banking efficiency before and after liberalization in the period 1970 – 1994 in Turkey, observe that bank efficiency declines after undertaking liberalization programs.
Kumbhaka et al. (2001) employ flexible variable profit functions to panel data from 1986-1995 to examine the impact of deregulation reform on the performance of Spanish savings banks. They conclude that regulatory reforms lead to slightly better banking performance. In particular, they find evidence that despite declining technical efficiency, the productivity growth rate increases in the post-liberalization period. In contrast, Grifell-Tatje and Lovell (1996), examining productive efficiency of Spanish savings banks during post-deregulation periods from the 1986-1991 by conducting Malmquist productivity index, suggest that deregulation programs are followed by an observed decline in productivity.
the beginning of the period but later on their performance improves, reaching levels close to those of the state-owned banks. The privately-owned banks are the least efficient. Ataullah and Le (2006) focus on India and the impact of a broader set of reforms (i.e. financial reforms, fiscal reforms and private investment liberalization) and find that efficiency of banks increased due to increased competition. Ataullah et al. (2004) utilizing the DEA method also find evidence to support the positive impact on bank efficiency in India and Pakistan. During the period 1988-1998, they find that the overall technical efficiency of the banking sector increases following the liberalization programs especially after 1995-1996. In case of Pakistan banks, this efficiency is primarily generated from scale efficiency. Hardy and Patti (2001) investigate the effects of financial reforms on profitability, cost and revenue efficiency of Pakistan banks during 1981-1998 by employing the distribution free approach. They also observe that financial liberalization programs have a positive impact on banking sector performance. Specifically, the cost and revenue efficiency of banks increases. They conclude that increasing in revenue efficiency transfers these benefits to customers including borrowers and depositors. Moreover, they also find that the overall profitability efficiency does not increase after implementing reform programs. Patti and Hardy (2005) employing various techniques OLS, GLS and Least Absolute Deviation (LAD) examine the cost and profit efficiency pre- and post financial liberalization in Pakistan from 1981-2002. They find that financial liberalization, consisting of the privatization of major banks, leads to increases in profits in the first round of financial reform 1991-1992. However, in subsequent years reform does not seem to have positive effects on banks. They show that profitability tends to decline after 1997 due to the deterioration of business conditions.
related to foreign acquisition. Consequently, they suggest that the benefits of foreign governance may take longer time to be realized.
Hermes and Vu (2009) employ DEA and panel least square fixed effects model to investigate the impact of financial liberalization on bank efficiency in Latin American and Asian countries from 1991-2000. They find strong evidence to support the positive relationship between financial liberalization and bank performance.
The above discussions of both theoretical and empirical literature give us unclear answer about the relationship between financial liberalization and bank efficiency. Moreover, most of these studies assume a linear effect of financial liberalization on bank efficiency. This gives us the incentives to examine more deeply which factors can alter the relationship between financial liberalization and bank efficiency. In the next part, we discuss the framework to what extent political regime instability can influence the effect of financial liberalization on bank efficiency.
2.3 Theoretical Framework of the Interrelationship between Financial
Liberalization and Political Regime Instability on Bank Efficiency
In this study, we define political regime instability as the propensity of a government change which reflects changes in political leaders, cabinets or relates to instable government. We study whether regime instability will decrease the beneficial effect of financial liberalization on bank performance. Below, we discuss the mechanisms through which regime instability may reduce the favorable effect of financial liberalization.
because they are afraid of the increased risk of capital loss. This leads to the so-called capital flight. Pindyck (1988)also argues that the irremediable nature of most investment makes investors extremely careful to changes in policies and related risks. Ingersoll and Ross (1992) also claim that the uncertainty of interest rate policy affects the timing of investment.Therefore, bank efficiency following financial liberalization may be reduced under instable regime conditions due to uncertainty of policies.
In addition, bank efficiency is expected to improve following financial liberalization through the increasing numbers of foreign banks. However, previous studies find that the motivation to invest or disinvest depends on the likelihood of regime stability in a predictable future (Benhabib and Spiegel, 1992; Alesina and Perotti, 1996; Benhabib and Rustichini, 1996). Foreign banks are unlikely to invest their capital in instable regime environment. Hence, political instability reduces the capital flow from foreign banks due to the uncertainties related to changing regime.
Furthermore, Ali (2001) argues that under instable regimes, the current regimes given an expected short term in office have tendency to maximize their rents for instance by taxing productive capital. Under these circumstances, savers, bankers and investors may find more advantageous to invest abroad rather than domestically. This induces capital flight. Therefore, the expected benefits of financial liberalization on bank efficiency under regime instability are reduced1.
3. Measuring Bank Efficiency
In the literature on bank efficiency various approaches have been used to measure efficiency. Basically, as discussed in section 2.1 these approaches come down to estimating a specific form of the so-called best practice frontier such as the (maximum) production frontier, the (minimum) cost frontier, or the (maximum) profit frontier. The efficiency level of an individual bank is then defined as the distance of this individual
1
bank’s production, costs, or profits to the frontier. Coelli et al. (1999) provide a comprehensive overview of the measurement of efficiency and productivity. In this paper, we focus on measuring so-called technical efficiency. A bank is considered to be technically efficient if it produces optimal quantities of output given the amount of inputs, or alternatively, if it produces given amounts of output with minimum quantities of inputs. This also means that when measuring efficiency, we focus on production, instead of costs or profits. This choice is driven by data availability; data on input prices and/or profits of banking services are more difficult to obtain compared to data on production of these services. Technically efficient banks operate on the best practice production frontier, whereas technically inefficient banks perform below this frontier. Put differently, technical efficiency is measured as the difference between the observed output-to-input ratio of a bank and the same ratio achieved by those banks operating on the production frontier.
present during the entire period. This makes it more difficult to measure cost, revenue or profit functions, ruling out the use of parametric approaches (Bhattacharyya et al., 1997).
Following a number of other studies in the bank efficiency literature, e.g., among others, Aly and Grabowski (1990), Ferrier and Lovell (1990), Berg et al. (1993), Bhattacharyya et al. (1997), and Wheelock and Wilson (1995), we employ the Data Envelopment Analysis (DEA) to calculate technical efficiency of banks. DEA uses linear programming methods to construct a nonparametric piece-wise surface (or frontier) over the selected sample of banks based upon measures of bank output. Efficiency of a bank is measured as the distance from each individual bank’s output to this surface.
The DEA model
The following DEA model includes the assumptions of constant returns to scale and accounts for the objective of minimizing inputs for a given level of output (an input-orientated version of DEA). It proceeds by solving a linear programming model:
Subject to :
We assume there are N banks in the sample producing I different outputs (yin denotes the observed amount of output i for bank n) and using K different inputs (xkn denotes the observed amount of input k for bank n). The λj is weights applied across the N banks. θ is scalar (0 ≤θ ≤ 1).
It should be emphasized that the linear program must be solved N times, once for each bank in the sample. When the nth linear program (equation 1) is solved, the
efficiency score for the nth bank, n
θ
*, is the smallest number ofθ
n that satisfies the three sets of constraints listed above (equations 2, 3 and 4).θ
n will satisfy 0 ≤θ
n ≤ 1, with a value of one indicating that this bank is on the production frontier and hence it is a technically efficient bank. 1-θ
n gives the proportion by which input can reduce to move the bank from the interior of the production frontier to the piece-wise linear boundary of the production set, with the quantity of output is held constant. In Figure 1,θ
ncorresponds with the ratio OQ/OP.
In order to minimize
θ
n, we base on the various changes of the weights λj and the scoreθ
n itself. The weights λj describe the percentage of the other best practice banks to construct the virtual bank (also lie on the frontier) that will be compared with the analyzed bank. The first constraint (equation 2) forces the weighted average of the other banks (the best practices) to produce at least as much of each output, as does analyzed bank nth. The second constraint (equation 3) finds out how much less input the best practice banks would need. The third set of constraint (equation 4) simply limits the weights to being either zero or positive.overall technical efficiency can be further decomposed into its scale and pure technical efficiency components.
Figure 2. Pure technical efficiency and Scale efficiency
This is illustrated in Figure 2 for the one input (x) and one output (y) case, and three banks A, B and C. A constant return to scale production frontier is illustrated by Og which measures the optimal level of output produced with a given level of input. Banks can lie on or below this frontier. For example for bank C, the overall technical efficiency (OTE) is measured by the ratio FE/FC, which corresponds to the ratio OQ/OP in Farrell’s definition in Figure 1. In order to calculate scale efficiency, the constant returns to scale is not appropriate; the variable returns to scale is developed; this is illustrated by hBAi in Figure 2. Under the variable returns to scale frontier, the pure technical efficiency (PTE) of bank C is determined by the ratio FD/FC. The difference between OTE and PTE, which is illustrated by the distance ED, is due to scale inefficiency. Thus, the scale efficiency is derived by the ratio FE/FD. The scale efficiency measure can be interpreted as the proportional reduction of input use to be obtained if the bank operates at the optimal scale (constant returns to scale).
In order to determine pure technical efficiency (PTE) the above linear programming problem is solved with the following additional constraint:
This additional constraint allows for variable returns to scale (VRS) production technology. This constraint ensures that the inefficient bank is only benchmarked against banks with similar size. This restriction is not imposed in the CRS case. Thus, in the CRS DEA model, a bank may be benchmarked against banks which are substantially larger or smaller than it is itself. The value PTE ranges from zero to one, one implying pure
technical efficiency. The degree of scale efficiency ( SE ) is found by dividing the efficiency measure under the CRS assumption by the efficiency measure under the VRS assumption:
To conclude, calculations of efficiency based upon this method leads to a decomposition of overall technical efficiency (OTE) into scale (SE) and pure technical efficiency (PTE) components. Scale efficiency can be interpreted as the proportional reduction of input use to be obtained if the bank operates at the optimal scale (constant returns to scale). PTE refers to the bank’s managerial and marketing skills in using its inputs in order to maximize outputs. This relates to skills such as controlling operating expenses, effective screening and monitoring of borrowers, marketing activities focusing on attracting depositors, efficient risk management techniques. Therefore, OTE is determined by economies of scale due to the size of the bank (SE) and managerial efficiency (PTE).
In order to be able to calculate efficiency, we need to select input and output measures of bank activities. In the literature five common approaches are used: the production approach, the intermediation approach, the asset approach, the user-cost approach and the value added approach. Appendix A provides a summary of input and output approaches used in previous research. Of these five approaches the production approach and the intermediation approach are the most widely used (e.g., by Ferrier and Lovell, 1990; Aly and Grabowski, 1990; Berger and Humphey, 1997; and Hunter and Timme, 1995). According to the production approach, banks produce services to depositors and borrowers. The approach uses traditional production factors such as
PTE OTE
land, capital and labor as inputs to produce outputs specified by the number of accounts serviced and/or transactions processed.
According to the intermediation approach, banks are intermediaries, transforming and transferring financial resources they borrow from depositors into the credit lent to borrowers. The alternative of the intermediation approach is the so called asset approach which also reflects banks as intermediaries between fund beneficiaries and liability holders. The intermediation approach uses deposits collected and funds borrowed from financial markets (i.e. bank liabilities) as inputs, whereas loans and other assets are considered to be the bank’s outputs. One limitation of this approach is that it may not consider all activities provided by banks, e.g. intermediation of corporate and government bonds, investment banking activities, underwriting activities, etc. (Favero and Papi, 1995).
According to the user-cost approach, in order to determine a particular asset or liability item is an input or an output, we have to consider its net contribution to bank revenue. Hancock (1990) who is among the first applying the user cost approach to banking shows that if for a certain asset the financial returns exceed the opportunity cost of funds (or if for a liability the financial costs are less than the opportunity cost), the instrument is considered a financial output, otherwise it is identified as an input. Under Hancock’s rule, demand deposits are specified as outputs while time deposits are considered as inputs. Favero and Papi (1995) and Grigorian and Manole (2002) also reveal the shortcomings of this approach. They state that the difficulties stem from collecting accurate data since interest rate fluctuations lead to changes in the user cost. Moreover, marginal revenues and costs for each individual liability are difficult to measure.
Finally, according to the value added approach, both liability and assets that have substantial value added are considered as outputs, while the others are treated as inputs or intermediate products. This approach has been used in Berg et al. (1993), Grigorian and Manole (2002).
deposits and certificates of deposits. Moreover, we use two output measures, i.e. total demand deposits and total net loans. Net loans are defined as total loans net of loan loss reserves.
4. Measuring Political Regime Instability
Political instability is not one dimensional indicator. There is much evidence from political science that political instability is multidimensional (Rummel, 1966; Morrison and Stevenson, 1971; Hibbs, 1973). However, Jong-A-Pin (2009) argues that although all these studies find that political instability has more dimensions, their solutions may still suffer from measurement error or some aspects of political instability are ignored. He suggests using Exploratory Factor Analysis (EFA, hereafter) to examine multidimensionality of political instability. Hence, we replicate this method to re-assess how different political instability indicators can be used to measure regime instability. We then use political regime instability scores in each country to examine the existence of interrelationship between political regime instability and financial liberalization on bank efficiency.
EFA is based on a model and identifies the common factors (separates from the specific factors) and explains their relationship to observed data (Lattin et al. 2003). We briefly discuss factor analysis model as follows. With factor analysis, our assumption is that the observed political instability indicators can be described as a function of a small number of underlying common factors and a set of specific factors (one specific factor for each political instability indicator). The factor analysis model with multiple factor can be described as following:
j j j
X =λξ+δ (7)
where X denotes the observed indicator for country j; ζ is a vector of common factors (latent variables), λ is a matrix of factor loadings, δ is a specific factor (random error term) assumed to be mutually uncorrelated and uncorrelated with underlying common factor ζ.
in Banks dataset (2005) and examine which indicators will be attributable to common factor for regime instability. These political instability indicators and their definitions are listed in Appendix B.
In order to check whether data is suitable to implement factor analysis, we conduct the following tests. The pair-wise correlation matrix of the political instability can be found in Appendix C. KMO (Kaiser-Meyer-Olkin) measure of sampling adequacy equals approximately 60 percent for the overall test and all each individual variable has measure of sampling adequacy larger 50 percent (see Appendix D). Furthermore, the Bartlett test of sphericity is statistically significant at the 1 percent level. These test results show that the use of factor analysis is appropriate (Hair et al., 2006).
Next, following Hair et al. (2006) and Lattin et al. (2003), we apply Catell’s scree test to define the appropriate number of factors. The graph below in the Catell’s scree test suggests four factors can be extracted. However, Hair et al. (2006) argue that the extracted number of factors should be based not only on statistical test but also on a conceptual foundation and existing empirical evidence. Jong-A-Pin (2009) defines four factors when he combines data from different sources such as Databanks International (2005), International Country Risk Guide (2005), Polity IV, International Peace Research Institute Oslo and Database of Political Institutions. Based on his study, we find that most of political instability indicators in Banks dataset (2005) belonging to three factors. Therefore, in our study, we find that it is likely more appropriate to extract three factors instead of four.
Figure 3. Catell’s scree plot
estimation results of the Varimax orthogonal rotated factor solution are shown in Table 1. The indicators with high factor loadings can be used to interpret the factors.
Based on the result of factor loadings in Table 1, the first factor has high loadings for indicators related to the instability of government, cabinet changes, elections and political leaders. Therefore, we label this factor as “political regime instability”. The second factor has high factor loadings on indicators such as demonstrations, riots and strikes which are associated with “civil protest”, thus we use this phrase to label this factor. Revolution, assassination and warfare indicators have high factor loadings on the third factor. Therefore, we label this factor as “political violence”. The results in factor loadings confirm our expectations of the multidimensional of political instability. In this study, we only use the factor scores in political regime instability to examine our hypothesis about the conditional effect of instable government and financial liberalization on bank efficiency as discussed on section 2. The occurrences of political violence or civil protest are also good indicators to measure political instability. In our view, however, these indicators measure the general nature of political conditions and they may affect many aspects. It remains unclear how such political instability affects financial liberalization policies and bank efficiency.
Table 1. Rotated factor loading matrix
Indicators Regime Protest Violence
Government Crises .422 .251 -.057
Number of Cabinet Changes .669 .131 -.101 Changes in Effective Executive .994 .086 -.056 Number of Legislative Elections .466 -.048 -.018 Number of Major Constitutional Changes .273 -.058 -.037
Number of Coups d'Etat .112 -.057 .055
Riots .005 .694 -.044 General Strikes -.024 .301 -.037 Anti-Government Demonstrations .067 .953 .132 Assassinations .033 .011 .762 Guerrilla Warfare -.098 -.047 .548 Revolutions -.057 .006 .778
5. Data and Methodology
5.1 Data of Financial Liberalization
In this study, we use financial liberalization dataset, created by Abiad et al. (2008). This dataset distinguishes between seven different dimensions of financial sector policy. The seven dimensions of financial market policies considered are:
− Credit controls and reserve requirements: focuses on directed credit towards favoured sectors or industries, ceilings on credit toward sectors, and high reserve requirements;
− Interest rate controls: deals with direct interest rate controls by the government, or interest rate controls through the use of floors, ceilings and interest rate bands;
− Entry barriers: deals with licensing requirements for newly established domestic financial institutions, entry barriers for foreign banks, and restrictions on certain types of banking practices, such as specialized bank services or establishing universal banks; − State ownership in the banking sector: refers to the share of the banking sector assets
controlled by state-owned banks;
− Restrictions on international financial transactions: refers to capital current account controls and restrictions, transaction taxes and the use of multiple exchange rates;
− Securities market policies: relates to policies with respect to the auctioning of government securities, the establishment of securities markets, tax incentives related to investments in securities and policies regarding the openness of markets to foreign investors; and − Prudential regulations and supervision of the banking sector: refers to (1) the adoption of
risk-based capital adequacy ratios risk-based on the Basel standard; (2) the independency of banking supervisory agency; and (3) the coverage of supervisory oversight; note that for this dimension more government intervention is interpreted as a positive reform policy (i.e. contributing to improving the efficiency of the banking system).
For each of the above dimensions, a country is assigned a score that runs from zero to three. For the first six dimensions reflecting financial liberalization, the meaning of the scores is as follows: 0 means that for a particular dimension of financial market policies, the country is fully repressed; 1 means partial repression; 2 means largely liberalized; and 3 means fully liberalized.2 Repression or liberalization of a certain dimension thus refers to the extent to which the government interferes with financial markets. For the last dimension describing regulation and supervision of the banking
sector the interpretation of the scores is different, however. Here, 0 means that the country has put in place none of three dimensions of regulation and supervision (i.e. capital adequacy ratios, independency of the central bank and a wide coverage of supervisory oversight); 1 means that the country has put in place one of the three dimensions of regulation and supervision; 2 means that the country has put in place two of the three dimensions of regulation and supervision; and 3 means that the country has put in place all three dimensions of regulation and supervision.
Following Abiad and Mody (2005), we use a financial liberalization index (FL) reflecting the aggregate scores of all seven dimensions discussed above.
5.2 Data of Bank Efficiency
Table 2. Number of sample bank observations by countries, 1991-2000 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 Total Argentina 18 27 18 18 12 17 19 67 76 69 341 Brazil 13 23 16 98 105 145 68 144 118 135 865 India 36 53 54 65 67 74 72 71 71 62 625 Indonesia 16 45 64 78 82 81 52 55 60 50 583 Korea, Rep. 6 17 12 29 33 35 36 27 25 23 243 Mexico 8 10 6 22 34 35 26 43 43 37 264 Pakistan 5 15 9 23 24 29 25 29 28 25 212 Peru 5 16 9 23 26 27 26 29 22 21 204 Philippines 14 20 16 31 36 42 46 43 43 36 327 Thailand 14 27 23 41 45 46 32 37 36 37 338 Total 135 253 227 428 464 531 402 545 522 495 4002
As discussed in section 3, we use three different measures of bank efficiency: OTE, PTE and SE. Table 4 provides detailed information on these three measures and their changes over the period 1991-2000 for the ten countries in our sample. We present the result of bank efficiency in Table 4 as annual average efficiency scores of all banks at the country level. However, for the econometric analysis, we keep measures of bank efficiency at bank level. We use separate annual frontiers for each country (which in the literature is known as the national frontier) to calculate efficiency scores, rather than one common frontier for all countries. Both the common frontier and the national frontier have been used in the literature. The common frontier approach assumes the same technology among countries; it does not capture cross-country differences (Weill, 2004; Berger and Humphrey, 1997). The national frontier approach does not follow this assumption (Coelli et al., 1999; Berger and Humphrey, 1997). Therefore, the national frontier is likely more appropriate in this paper since the developing countries in our sample may not have the same technology and we want to capture the differences of bank efficiency across countries.
The data in Table 4 show that the patterns of the three measures of efficiency are similar when looking at the individual country level. Moreover, Table 4 also shows that for all countries in the sample the measures fluctuate across countries during 1991-2000. In general, however, for most countries the efficiency measures are higher at the beginning of the period as compared to the value of these measures at the end of the period. This finding may be partly due to the fact that the number of banks per country for which data are available at the beginning of the period is relatively low. Thus, the figures for the efficiency measures may suffer from a sample bias in the early 1990s. When looking at trends for individual countries Table 4 indicates that for Argentina, Brazil, Indonesia, Pakistan and Thailand efficiency seems to go down during most of the period, whereas for India, Korea, Malaysia and Peru efficiency goes up from the mid-1990s, after an initial decline during the first years of the decade. For the Philippines no clear trend is observed.
in our sample. Therefore, in order to improve efficiency, these banks should focus more on controlling operating expenses, investing rationally in assets, using risk management techniques, and having strategically planning. In addition, bank managers should have loan policies that evaluate prudently creditworthiness of borrowers.
When grouping the data for countries into two regions, i.e. Latin America and Asia, it appears that Asian banks have lower overall technical efficiency than banks in Latin America. The levels of pure technical efficiency of the banks in Asia and Latin America are not substantially different. This means that the major source of lower OTE in Asian banks is low levels of SE, i.e. Asian banks generally suffer from higher scale inefficiencies.
Table 4. Average bank efficiency by counties
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Panel A: Latin America
Table 4 (continued) 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 Panel B: Asia India OTE 0.66 0.44 0.58 0.55 0.52 0.60 0.67 0.52 0.59 0.67 PTE 0.75 0.52 0.69 0.70 0.71 0.72 0.75 0.66 0.68 0.73 SE 0.88 0.85 0.84 0.78 0.73 0.83 0.89 0.79 0.87 0.92 Indonesia OTE 0.80 0.77 0.65 0.72 0.61 0.48 0.68 0.58 0.63 0.60 PTE 0.85 0.86 0.76 0.82 0.77 0.67 0.83 0.77 0.79 0.79 SE 0.94 0.89 0.85 0.88 0.79 0.71 0.82 0.75 0.80 0.76 Korea, Rep. OTE 0.94 0.81 0.92 0.61 0.59 0.66 0.76 0.75 0.77 0.73 PTE 0.98 0.91 0.97 0.66 0.72 0.75 0.82 0.87 0.84 0.85 SE 0.97 0.89 0.95 0.92 0.83 0.88 0.93 0.87 0.92 0.86 Pakistan OTE 0.99 0.77 0.87 0.84 0.81 0.73 0.78 0.76 0.71 0.85 PTE 1.00 0.96 0.98 0.92 0.89 0.88 0.88 0.87 0.80 0.85 SE 0.99 0.80 0.89 0.92 0.91 0.83 0.88 0.88 0.89 1.00 Philippines OTE 0.84 0.83 0.81 0.69 0.80 0.73 0.67 0.71 0.75 0.70 PTE 0.94 0.89 0.84 0.79 0.89 0.86 0.80 0.79 0.81 0.81 SE 0.89 0.94 0.97 0.88 0.91 0.85 0.83 0.90 0.93 0.87 Thailand OTE 0.87 0.80 0.87 0.74 0.74 0.54 0.81 0.60 0.54 0.52 PTE 0.97 0.87 0.92 0.83 0.81 0.68 0.87 0.73 0.72 0.74 SE 0.90 0.91 0.94 0.89 0.91 0.79 0.93 0.83 0.75 0.71
5.3 Econometric Framework
We conduct two steps analysis in this study. In the first step, we investigate whether financial liberalization contributes to improving bank efficiency. We apply bank and time specific fixed effect panel model. While bank fixed effects take into account all characteristics specific to each individual bank in each country, time fixed effects are included to capture all variation in the data specific to some year. We construct the econometric model in this step as following:
In the second step, we aim to investigate whether the relationship between financial liberalization and bank efficiency is conditional on the political regime instability. We modify bank and time specific fixed effect panel model with a multiplicative interaction term. Our model specification is:
t j i t i t j t j t j t j t j t i t j i FL REGIME FL REGIME M S y, , =α +ϕ +β0+β1 , +β2 , +β3 , * , +γ , +τ , +ε, , (9)
where the dependent variable yijt is the efficiency index calculated by DEA at bank ith at county jth at time t; α
i refers to bank fixed effect of bank i, φt is a time fixed effect at specific time t; β0 is a constant term; FLjt is the aggregated index of seven dimensions of overall financial liberalization for county jth at time t; REGIME
jt is a vector of regime instability scores measured by EFA at county jth at time t; M
jt is a vector of control variables described the macroeconomic environment at country j at time t; Sjt is a vector of control variables describing the bank features at bank ith at county jth at time t; εijt is a random error term.
Regarding Mjt vector of control variables describing the macroeconomic environment, we choose a set of variables including GDP per capita and density of demand. Following Grigorian and Manole (2002), banks in higher per capita income countries are more efficient in terms of attracting more deposits and generating stronger cash flows than those in low income countries. Additionally, countries with higher per capita income are inclined to generate more savings, and thus more deposits. Therefore, bank efficiency and GDP per capita (LnGDP_CAP) are positively correlated.
Density of demand (DD) is defined as the ratio of total value of deposits per square kilometer. In the literature it is hypothesized that banks operating in a market with lower density of demand suffer from higher expenses in making loans and gathering deposits through their branches. Therefore, bank efficiency and demand density are positively correlated (Dietsch and Lozano-Vivas, 2000; Lozano-Vivas et al., 2001 and 2002). The macroeconomic data is obtained from World Development Indicators provided by the World Bank.
efficiency can be positive or negative. Berger and De Young (1997) suggest that a higher capital to asset ratio indicates lower bad loan problems, which reduces the additional costs to recover these bad loans. Dietsch and Lozano-Vivas (2000) and Lozano-Vivas et al. (2001) argue that a lower capital to asset ratio is associated with lower bank efficiency, since it involves higher risk taking. Moreover higher leverage ratios are also more costly to the bank. Higher levels of ETA are therefore associated with higher bank efficiency. In contrast, low capital ratios may encourage banks to undertake risky business by investing in highly profitable projects. This may help banks obtain higher efficiency at least in the short term (Lozano-Vivas et al., 2002).
The return on equity ratio (ROE) is used as a proxy of competitiveness in the banking industry. Assuming a competitive market environment, this ratio is expected to have a positive impact on efficiency (see, e.g., Lozano-Vivas et al., 2001 and 2002). Finally, the loan to deposit ratio (LTD) is a measure of the efficiency of banks in terms of the extent to which they are able to transform deposits into loans. The higher this ratio, the more efficient the process of financial intermediation provided by the bank (Dietsch and Lozano-Vivas, 2000 and Fries and Taci, 2005). Thus, higher levels of LTD are associated with higher levels of bank efficiency.
Table 5 shows the correlation matrix of the independent variables in the model. The matrix shows that in general correlation between the exogenous variables is low (except for the correlation between FL and LnGDP_CAP), which means that multicollinearity problems are not severe or non-existent. Table 6 provides descriptive statistics of all endogenous and exogenous variables used in the empirical investigation.
Table 5. Correlation matrix of the independent variables
FL REGIME LnGDPCAP DD ETA ROAE LTD
FL 1 REGIME -0.0744 1 LnGDP_CAP 0.5159 -0.1268 1 DD 0.216 -0.0504 0.3986 1 ETA 0.1223 -0.0644 0.1745 -0.1093 1 ROE 0.0079 0.0131 -0.015 -0.0381 0.0086 1 LTD -0.051 0.0072 -0.0288 -0.0037 0.0557 0.0042 1
Table 6. Summary statistics
Variables Observations Mean Std. Dev. Min Max
OTE 4002 0.6416 0.2628 0.0001 1.0000 PTE 4002 0.7528 0.2512 0.0148 1.0000 SE 4002 0.8511 0.1870 0.0001 1.0000 FL 4002 11.9832 3.6965 2.0000 19.0000 REGIME 4002 -0.0520 0.8156 -0.7438 5.9589 LnGDP_CAP 4002 7.4455 1.0645 5.7434 9.2956 DD 4002 535.0551 1799.3700 2.9978 9079.1750 ETA 4002 0.1129 0.1184 -2.9328 0.9457 ROE 3994 0.0138 3.2738 -112.9368 110.9932 LTD 3996 1.0149 2.8110 0.0000 120.8119 6. Empirical Results
As discussed above. we have two main parts focusing on this paper. In the first step, we examine whether financial liberalization has a positive effect on bank efficiency. In the second step, we focus on analyzing the impact of financial liberalization on efficiency is conditional on the political regime instability.
6.1 The Impact of Financial Liberalization on Bank Efficiency
The empirical model in the first step is estimated using the panel least square fixed effects methodology. The model is tested for each of the three measures of efficiency, i.e. OTE, PTE and SE. The research strategy follows the specific-to-general approach (Brooks, 2002). We start by investigating the relationship between the financial liberalization variable (FL) and efficiency. Next, we include the control variables one by one to test the stability of the main independent variable FL. The results of the estimations are presented Tables 7-9.
Table 7. The panel least square regression of OTE and FL Variables (1) (2) (3) (4) (5) (6) FL 0.0170*** 0.0155*** 0.0142*** 0.0145*** 0.0145*** 0.0149*** (0.0028) (0.0029) (0.0029) (0.0029) (0.0029) (0.0029) LnGDP_CAP 0.124* 0.169** 0.173** 0.171** 0.185** (0.0715) (0.0732) (0.0732) (0.0734) (0.0731)
DD -1.78e-05*** -1.80e-05*** -1.79e-05*** -1.81e-05***
(6.27E-06) (6.27E-06) (6.27E-06) (6.24E-06)
ETA -0.0557 -0.0567 -0.0665* (0.0353) (0.0353) (0.0352) ROE 0.000635 0.000557 (0.0009) (0.0009) LTD 0.00403*** (0.0013) Constant 0.649*** -0.239 -0.552 -0.579 -0.562 -0.67 (0.0264) (0.5140) (0.5250) (0.5250) (0.5260) (0.5240)
Time fixed effects Yes Yes Yes Yes Yes Yes
Bank specific effects Yes Yes Yes Yes Yes Yes
Observations 4002 4002 4002 4002 3994 3988
R-squared 0.094 0.095 0.097 0.098 0.098 0.102
Number of banks 727 727 727 727 727 727
Note: ***, **, *, indicates significance at the 1, 5 and 10 percent level, respectively. Robust standard errors are presented in parentheses.
Both country specific variables including GDP per capita and density of demand are statistically significant. The coefficient of GDP per capita has a positive sign. This supports our expectations since banks in higher per capita income countries are more efficient in terms of attracting more deposits and generating stronger cash flows than those in low income countries. It is widely supported in the literature that countries with higher per capita income are inclined to generate more savings, and thus more deposits. However, the coefficient of the DD variable is negative and statistically significant at the 1 percent level3. This is opposite to what we expected based on the theory that banks operating in markets with higher density of demand incur lower costs of mobilizing deposits and granting loans, resulting in higher bank efficiency.
Two of the three bank-specific variables are statistically significant. The capital to asset ratio (ETA) has a negative coefficient that is significant at the 10 percent level only in the final specification (column 10), supporting the idea that low capital ratios encourage banks to undertake risky business by investing in highly profitable projects.
3 In a previous study (Hermes and Vu, 2009), we find the same result with DD. Fries and Taci (2005) also
We find no evidence to support the significant effect of ROE. However, we find that LTD is statistically significant at the 1 percent level. This result support our expectation that the higher this ratio, the more efficient the process of financial intermediation provided by the bank. Most importantly, in addition, adding more control variables, FL is still statistically significant at the 1 percent level. This strongly supports our hypothesis that financial liberalization has the positive impact on bank efficiency.
Next, we use the pure technical efficiency (PTE) as the dependent variable. The results of the empirical analysis remain to be very similar to the ones we get when we use overall technical efficiency (OTE) as the dependent variable. The results for PTE are presented in Table 8. Again, we find that the coefficient of FL is positive and significant at the 1 per cent level, which supports the hypothesis that financial liberalization leads to improvements of bank efficiency. Adding control variables does not change the results for FL: the coefficient remains to be significant at the 1 per cent level. Again, the majority of the control variables are statistically significant and their coefficients do not change much in the different specifications of the model. The only major difference with the results in Table 7 is that when we use PTE as dependent variable ETA is statistical significant in all specifications while LnGDP_CAP is no longer statistically significant.
Table 8. The panel least square regression of PTE and FL Variables (1) (2) (3) (4) (5) (6) FL 0.0108*** 0.0109*** 0.00950*** 0.00925*** 0.00923*** 0.00945*** (0.0027) (0.0028) (0.0029) (0.0029) (0.0029) (0.0029) LnGDP_CAP -0.00912 0.0414 0.0363 0.0266 0.036 (0.0697) (0.0713) (0.0713) (0.0714) (0.0713) DD -2.01e-05*** -1.99e-05*** -1.96e-05*** -1.99e-05***
(6.10E-06) (6.10E-06) (6.10E-06) (6.09E-06)
ETA 0.0612* 0.0645* 0.0578* (0.0344) (0.0344) (0.0344) ROE 0.00112 0.00112 (0.0009) (0.0009) LTD 0.00268** (0.0013) Constant 0.758*** 0.824 0.472 0.502 0.572 0.499 (0.0257) (0.5010) (0.5110) (0.5120) (0.5120) (0.5120)
Time fixed effects Yes Yes Yes Yes Yes Yes
Bank specific effects Yes Yes Yes Yes Yes Yes
Observations 4002 4002 4002 4002 3994 3988
R-squared 0.046 0.046 0.049 0.05 0.05 0.052
Number of banks 727 727 727 727 727 727
Note: ***, **, *, indicates significance at the 1, 5 and 10 percent level, respectively. Robust standard errors are presented in parentheses.
Table 9. The panel least square regression of SE and FL
Variables (1) (2) (3) (4) (5) (6)
FL 0.00904*** 0.00654*** 0.00664*** 0.00701*** 0.00715*** 0.00744***
(0.0021) (0.0022) (0.0022) (0.0022) (0.0022) (0.0022)
LnGDP_CAP 0.207*** 0.204*** 0.211*** 0.217*** 0.227***
(0.0542) (0.0555) (0.0554) (0.0555) (0.0550)
DD 1.43E-06 1.18E-06 1.07E-06 9.57E-07
(4.75E-06) (4.74E-06) (4.74E-06) (4.70E-06)
ETA -0.0895*** -0.0911*** -0.101*** (0.0267) (0.0267) (0.0265) ROE -0.00077 -0.00083 (0.0007) (0.0007) LTD 0.00222** (0.0010) Constant 0.851*** -0.638 -0.613 -0.658* -0.695* -0.776** (0.0200) (0.3890) (0.3980) (0.3980) (0.3980) (0.3950)
Time fixed effects Yes Yes Yes Yes Yes Yes
Bank specific effects Yes Yes Yes Yes Yes Yes
Observations 4002 4002 4002 4002 3994 3988
R-squared 0.063 0.067 0.067 0.071 0.072 0.078
Number of banks 727 727 727 727 727 727
6.2 The Interrelationship between Political Regime Instability and Financial
Liberalization on Bank Efficiency
In the second step, we examine to what extent regime instability can influence the positive relationship between financial liberalization and bank efficiency by adding the regime change variable (REGIME) and interaction term (FL*REGIME). The results present in Table 10, 11 and 12 for OTE, PTE and SE as dependent variable, respectively.
Column [1] in Table 10 shows that the coefficients of financial liberalization, regime instability and interaction term are highly significant at the 1 percent level. This implies that the effect of financial liberalization and regime instability is non-linear and the interaction effects are present in the data. However, Brambor et al. (2006) claim that we should not interpret the constitutive elements of interaction terms as unconditional or average effects. Therefore, the estimated coefficients and their standard errors in multiplicative interaction models in Table 10 are meaningless to evaluate our hypothesis. We need to calculate the appropriate marginal effects of the financial liberalization variable and the uncertainty with which it is estimated (Brambor et al., 2006). Before we do that, we need to take into account a potential omitted variable bias by including different control variables that have been discussed above. In column [2]-[6] we add several control variables to capture both underlying macroeconomic conditions such as GDP per capita (LnGDP_CAP), density of demand (DD) and bank features including capital ratio (ETA), profitability ratios (ROE) and intermediation ratio (LTD) as we have discussed above in the first step to test the effect of financial liberalization. In general, all control variables have the same results as we have discussed in the first step. Most importantly, adding more control variables, FL, REGIME and the interaction term are still statistically significant at the 1 percent level. This strongly supports the presence of interaction effects in our data.
Table 10. The impact of FL and REGIME on OTE Variables (1) (2) (3) (4) (5) (6) FL 0.0175*** 0.0160*** 0.0148*** 0.0150*** 0.0150*** 0.0154*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) REGIME 0.0603*** 0.0597*** 0.0608*** 0.0607*** 0.0601*** 0.0600*** (0.011) (0.011) (0.011) (0.011) (0.011) (0.011) FL*REGIME -0.00403*** -0.00396*** -0.00411*** -0.00411*** -0.00405*** -0.00402*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) LnGDP_CAP 0.120* 0.164** 0.169** 0.167** 0.181** (0.071) (0.073) (0.073) (0.073) (0.073)
DD -1.80e-05*** -1.81e-05*** -1.80e-05*** -1.82e-05***
(6.26e-06) (6.26e-06) (6.26e-06) (6.26e-06)
ETA -0.053 -0.055 -0.0645* (0.035) (0.035) (0.035) ROE 0.000503 0.000427 (0.00088) (0.00087) LTD 0.00395*** (0.001) Constant 0.633*** -0.227 -0.537 -0.563 -0.551 -0.659 (0.027) (0.512) (0.523) (0.523) (0.524) (0.522)
Time fixed effects Yes Yes Yes Yes Yes Yes
Bank specific effects Yes Yes Yes Yes Yes Yes
Observations 4002 4002 4002 4002 3994 3988
R-squared 0.103 0.104 0.106 0.107 0.107 0.111
Number of banks 727 727 727 727 727 727
Note: ***, **, *, indicates significance at the 1, 5 and 10 percent level, respectively. Robust standard errors are presented in parentheses.
Figure 4. Marginal effect of financial liberalization on overall technical efficiency as regime instability changes
-. 1 -. 0 5 -. 0 2 5 0 .0 2 5 .0 5 .1 M a rg in a l E ff e c t o f F L 0 1 2 3 4 5 6 REGIME Marginal Effect of FL 95% Confidence Interval
Dependent Variable: OTE
Marginal Effect of FL on OTE As REGIME Changes
The results in column [6] in Table 10 indicate that financial liberalization has a positively significant effect on bank efficiency when there is regime stability (coefficient of FL is statistically positive and significant). However, regime instability happens in all countries in our sample at different levels. Therefore, it is necessary to examine how the marginal effect of financial liberalization changes across the calculated range of regime instability. The solid slopping line in Figure 4 indicates how the marginal effect of financial liberalization on bank efficiency changes with the level of regime instability. 95 percent confidence intervals around the line allow us to determine the conditions under which financial liberalization has a statistically significant effect on bank efficiency. Brambor et al. (2006) show that when including the interaction term, the financial liberalization variable has a statistically significant effect on bank performance whenever the upper and lower bounds of the confidence interval are both above (or below) the zero line.
Figure 4 shows that financial liberalization and regime instability are significantly related to bank efficiency. As predicted, financial liberalization has a strong positive effect on bank efficiency in case of less regime instability. However, this positive effect of financial liberalization on bank efficiency declines as the regime instability increases. Once the regime instability score reaches approximately 2.14 (factor scores calculated by EFA, the higher scores indicate the higher risk), financial liberalization no longer has the significantly positive impact on overall technical efficiency.
This result supports our hypothesis the higher propensity of regime instability the lower the positive effect of financial liberalization because of the higher uncertainty of government policies. We find the evidence to substantiate for the theoretical framework discussed above. The lack of confidence and skepticism about the stability economic policies due to regime changes force people who have potential benefits from financial liberalization policies such as domestic and foreign savers, investors and bankers to postpone capital investment. This explains why under instable regime conditions, the positive effect of financial liberalization policies may be reduced or even ceased.
on political regime instability. Previous literature may find biased results of the influence of financial liberalization on bank efficiency since most of them use linear additive models. In this case they assume that financial liberalization has a constant effect on bank efficiency. However, we show that this may not be the case in practice.
Next, we use pure technical efficiency (PTE) as the dependent variable. The results of the empirical analysis remain to be very akin to the ones we get when we use overall technical efficiency (OTE) as the dependent variable. The results for PTE are presented in Table 11. Again, based on Figure 5, we find that financial liberalization has a strong positive effect on pure technical efficiency in case of regime stability. However, this positive effect of financial liberalization on PTE declines as the regime instability increases. The effect of regime change and financial liberalization is more sensitive in case of PTE than when using OTE as the dependent variable. Once the regime instability score reaches 1.51, financial liberalization no longer has the impact on pure technical efficiency. Again, the results of all control variables are similar with the results we have discussed in the first step with PTE.
