國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
18
5. Empirical results
Table 5.1 shows parameter estimates of the translog function for group 2 since the result of group 1 is unfavorable. But still we will insert the parameter estimates of group 1 in the Appendix 3. There are 18 out of 23 parameter estimates (excluding environment variables), or 73.9% of the total independent variables, achieving at least the 10% significance level. Except for the input price and output, the 8 environment variables are all significant at least at the 5% significance level.
The translog function appears to describe the data quite well.
As far as the environmental variables are concerned, the variable of ETA is negatively related to the inefficiency term, implying that the higher the ETA, the more efficient a bank will be. The result is the as same as Huang (2011) who estimates the Fourier flexible cost function. As for net interest margin (NIM), the influential direction of inefficiency is opposite to ETA, meaning that the increase in NIM tends to lower bank efficiency. The result is consistent with our expectation.
Diversification can be another essential factor to affect efficiency. Banks might face with the overbanking situation; therefore, starting to take diversified strategy by sacrificing the lending rate. As a consequence, cost increases, due to inefficiency(Lin et. Al, 2012; Lepetit et al., 2005).
The variable of loan to asset ratio is found to have negative effect on cost inefficiency, due possibly to the scale effect. Once loans are lent out to certain scale, a bank can take the advantage of scale to hire professional employees to scrutinize potential loan customers so as to earn more interest revenue and to minimize loan defaults. The coefficient of ROA, an indicator of profitability, is negative, implying that it is positively associated with efficiency, as expected. Both CSR and its squared term are considered the translog cost function. Both coefficients are significantly estimated at the 5% level, indicating that the relationship between CSR and cost efficiency are not linear, consistent with Barnett and Salomon (2006). Similarly, Mullen (1997) and Yang (2015) suggest that we should expect CSR rather in a long-term period than short-run.
Putting our focus on macroeconomic variables of GDP growth rate and real GDP per capita, we can tell from the results that these two variables have different
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
19
impacts on efficiency. GDP growth rate negatively affects efficiency, while real GDP per capita has positive effect on efficiency. The reason might be that the GDP growth rate signifies overall economic conditions and its increase may not trigger people to have higher demand and supply for banking service; or the forces from the demand and supply sides offset each other. Instead, the increase in real GDP per capita raises the demand for an array of banking services and the supply of loanable funds fueled by savings. The effect of real GDP per capita is as expected and similar results are found, e.g., Huang (2011).
Table 5.1 Parameter estimates of the translog cost function
*Significant at the 10% level of significance
** Significant at the 5% level of significance
***Significant at the 2.5% level of significance
****Significant at the 1% level of significance
Here, we emphasize the effect of CSR in Table 5.2. As we stated in Chapter 4, the mode of CSR index is 20. Table 6 displays that the majority efficiency scores range from 0.6 to 0.8, where the corresponding sample banks have CSR indices
Variable Parameter
estimates Standard errors Variable Parameter estimates
Standard errors
Constant 1.60E+00*** 7.21E-01 ln(Y3)ln(W3/W1) 2.88E-02** 1.72E-02
ln(Y1) 4.55E-01**** 8.80E-02 t -9.16E-01**** 2.43E-01
ln(Y2) 1.68E-01* 1.19E-01 0.5t2 -5.74E-02 4.69E-02
ln(Y3) 2.54E-01**** 1.13E-01 t x ln(Y1) 1.99E-01**** 2.63E-02
0.5ln(Y1)ln(Y1) -8.37E-04 4.44E-03 t x ln(Y2) 9.32E-03 1.99E-02
0.5ln(Y2)ln(Y2) -2.49E-02*** 1.21E-02 Tt x ln(Y3) 8.16E-03 2.05E-02
0.5ln(Y3)ln(Y3) -4.93E-03 1.18E-02 t x ln(W2/W1) -4.81E-02**** 1.72E-02
ln(W2/W1) 1.05E+00**** 1.07E-01 t x ln(W3/W1) -3.44E-02*** 1.61E-02
ln(W3/W1) 3.06E-01**** 1.00E-01 Constant 8.77E+00**** 3.35E+00
0.5ln(W2/W1)ln(W2/W1) -1.40E-02**** 2.42E-03 ETA -1.04E+00**** 3.98E-01
0.5ln(W3/W1)ln(W3/W1) 3.69E-04**** 1.43E-04 NIM 4.27E+00** 2.24E+00
ln(Y1)ln(W2/W1) -8.32E-02**** 1.10E-02 Loan/Asset -5.46E+00**** 8.27E-01
ln(Y1)ln(W3/W1) 5.17E-02**** 1.01E-02 ROA -1.23E+00**** 2.07E-01
ln(Y2)ln(W2/W1) 6.99E-04 1.64E-02 CSR 1.42E-02*** 6.40E-03
ln(Y2)ln(W3/W1) -3.85E-02**** 1.57E-02 CSR2 -1.28E-04*** 6.11E-05
ln(Y3)ln(W2/W1) -5.25E-02**** 1.85E-02 GDP growth rate 2.56E-01**** 4.35E-02
Log-likelohood -3.32E+02 ln(real GDP per
capita) -6.71E-01*** 3.03E-01
‧
respect to CSR, we obtain the turning point around 11. If CSR index is below 11, then the efficiency measure is decreasing with the increase in CSR, while the reverse is true when CSR index surpasses 11, i.e., the efficiency score is increasing when CSR grows.The finding enlightens us that the involvement of CSR for a bank should engage more than 11 such that the beneficiary effect on efficiency occurs. The higher CSR index usually takes more time to achieve, implying that the effect of CSR tends to be long-term instead of short-term (Mullen, 1997; Yang, 2015).
Appendix 4 exhibits detailed distribution of CSR and efficiency scores across different regions. As far as the three hypotheses mentioned in Section 3.2, our results fail to support the greenwashing hypothesis, but partially support the altruistic and strategic hypothesis. In particular, a bank is said to be altruistic if it engages CSR activities with the score less than 11, while a bank is said to be strategic if it actively conducts CSR with the score above 11.
Besides the hypothesis we get enlightened from Baron (2001), Dam at al (2009) for three assumptions, we also try to figure out where benefits come from after banks involve CSR more than 11. Because CSR investments lead to higher levels of credibility (Lin, Chen, Chiu, & Lee, 2011), improved image or reputation (Tewari, 2011), higher employee retention (Kim and Park, 2011) and build customer relationships (Peloza & Shang, 2011; Matute, Bravo, & Pina, 2010; Brown &
Dacin, 1997), we think the benefits may come from the above mentioned reasons.
Once banks engage in CSR more than 11, customers are more willing to pay higher price to get the services and at the same time banks can hire much more qualified workers to enhance the working efficiency and thus save costs. Perhaps, the benefits may reflect on higher profits for those banks with CSR scores in excess of 11. Since here we estimate the cost frontier, instead of the profit frontier, it’s difficult to relate profit efficiency to CSR activities.
Table 5.2 CSR index cluster with Efficiency Score CSR
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
21
Table 5.3 Average cost efficiency across in different regions
0.406-0.506 1 6 6 1 1 3 2 1 2 1 24
0.506-0.606 1 14 15 4 6 7 6 3 4 2 62
0.606-0.706 2 15 12 9 17 11 6 2 6 3 1 84
0.706-0.806 9 31 20 9 13 12 13 12 8 3 130
0.806-0.906 2 10 12 9 7 9 8 6 6 69
0.906-1 1 1 2 4
Total 15 92 75 39 59 52 45 34 31 14 1 457
What’s more, the efficiency is also found different from distinct regions. In Table 5.3 presents the average efficiency across five years in different regions. Due to the missing data, there are some cells have no data filled in. However, Table 5.3 says the most stable of cost efficiency is in Asia, primarily Japan. And the most volatile cost efficiency is in North Europe. And there is no obvious trait between West Europe and South Europe. Highest cost efficiency appears in 2011.
Regions 2010 2011 2012 2013 2014 Standard
deviation Asia 0.61159 0.64560 0.61412 0.629247 0.62853 1.224%
North
America 0.61505 0.56901 0.59465 0.62632 0.59331 1.976%
North Europe n.a 0.77081 0.68723 0.46879 0.61713 11.081%
West Europe 0.67605 0.62470 0.57055 0.60412 0.51329 5.435%
South Europe 0.63556 0.71264 0.74605 0.61189 n.a 5.476%
Total 0.63456 0.66455 0.64252 0.58807 0.58807
In order to test the fitness of the cost function we estimated, we calculate the cost share of labor (S1), cost share of capital (S2), and cost share of funds (S3), respectively in Table 5.4. Theoretically, the cost share of those three inputs should be positive, otherwise it is meaningless. There are 301 out of 457 samples, 66%
which the cost share of labor, capital and funds are positive. The results describe that the cost of funds accounts 43.17% of total costs, and cost of labor is 31.73% of total cost. However, the cost of capital is the least of three inputs. Thus, here we provide the insights for banks to reconsider the input price of funds which contributes over 40% to total costs when running a bank. Nonetheless, since banks run by loans lending out and deposits saving in, it’s reasonable that the majority
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
22
costs come from funds. What we have to notice is that are those expenses worthy to make enough profits.
Table 5.4 Cost share of labor (S1), capital (S2), and funds (S3)
S1a S2b S3c
31.73% 25.10% 43.17%
a W1X1/TC b W2X2/TC C W3X3/TC
Recall that SE < (>) 1 means decreasing (increasing) returns to scale.
However, as the time passes, the value of SE increases and achieves increasing returns to scale in 2014. From table 9, we can tell that the current production scale is closet to optimal When it comes to the standard deviation of five years, the volatility of estimated economies of scale is not too volatile across five years, whereas each year has the mild volatile of estimated economies of scale.
The phenomenon may be articulated by the heterogeneity of the sample banks.
Table 5.3 Average Economies of Scale over Time
2010 2011 2012 2013 2014 Total
SE 0.0673 0.4468 0.6481 0.8566 1.0515 0.6141
Standard deviation 0.1814 0.1739 0.1847 0.1539 0.1674 0.3750
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
23