This section presents an empirical application using the proposed semiparametric estimation techniques to examine a sample of 340 commercial banks over the period 1993-2004. The sample banks come from 14 East European countries and are compiled from the unconsolidated accounting data of Bankscope database. After precluding all missing and zero observations, this unbalanced panel data contain a total of 1399 bank-year observations. Since 1993, East European countries have started to transit from socialism to capitalism gradually. During the transition period, these countries liberalize the financial institutions by privatizing state-owned banks and allowing for new entry of private and foreign banks. These nations are in the long process of building a stable, reliable, independent, and transparent banking system.
The enforcement of the deregulation policy should have intensified the competition in the banking industries. Consequently, an essential issue to regulators, financial institution managers, industry consultants, and potential investors is whether the policy does improve banks’ performance.
In the past two decades, the academic research on the performance of financial services has mainly utilized the frontier approaches to evaluate a firm’s TE. Measure TE reflects the ability of a firm to employ a minimal input mix to produce a given level of outputs or to produce maximum outputs using a given input combination. It has been widely studied by previous works. The economic efficiency of a firm combines TE with AE, i.e., EE=TE*AE. Measure AE reflects the capability of a firm to hire inputs in an optimal proportion, given their respective prices, so as to achieve a minimum cost for a certain level of outputs. In contrast to TE, AE draws relatively less attention of researchers. However, the investigation of allocative efficiency is crucial particularly in transition economies. As the input and/or the output prices are frequently under the control of the governments of the transition countries, these prices may be slowly adjusted in response to market conditions. Possible allocative distortion may play a pivotal role in transition countries. Therefore, we apply a cost function to estimate TE and AE scores simultaneously.
Following the intermediation approach, which regards a bank as an intermediary between depositors and borrowers, we identify two outputs: total loans (y ) and 1
investments (y ). Bank inputs are defined as labor (2 x ), physical capital (1 x ), and 2 borrowed funds (x ). The price of labor (3 w ) is equal to the total personnel expenses 1 divided by total assets.6 The price of physical capital (w ) is measured as the ratio of 2 non-interest expenses and total depreciation to total fixed assets, and the price of funds (w ) represents total interest expenses divided by total funds. Note that we also 3 estimate the translog cost function for the purpose of comparison.
In the first step, we simultaneously estimate two input share equations by the NISUR to obtain the estimates of country-specific allocative parameters, i.e.,H3l /H 1l and H2l/H (1l l1,…,14). Herein, labor input is arbitrarily chosen as the numeraire.
According to the allocative parameter estimates in Table 15, all but three such estimates are less than unity. This implies that banks in Poland are inclined to under-utilize input capital, while banks in Kazakhstan and FYR Macedonia are apt to under-utilize funds, relative to the numeraire labor as their corresponding allocative parameter estimates exceed unity. This may happens because banks in those three countries have lower input prices or use fewer input quantities. It can be seen on Table 16 that banks in Lithuania has the lowest average W W (11.1825), followed by 2 / 1 banks in Poland (16.2563) in ascending order. Although banks in Lithuania has the lowest input prices, they utilize more input quantities (X2/X1) than banks in Poland. That might cause the allocative parameter estimates in Poland to exceed unity, implying that banks in Poland are inclined to under-utilize input capital, relative to the numeraire labor. It also reveals that banks operating in Kazakhstan reach the lowest average W W (0.7066), followed by banks operating in 3/ 1 Estonia (1.2313), Lithuania (1.5370), Latvia (1.5395), and FYR Macedonia (2.0040) in ascending order. Even though banks in Estonia, Lithuania and Latvia have the lower input prices than banks in FYR Macedonia, they employ more input levels (X3/X1) than banks in FYR Macedonia.
This could be the reason that banks in Kazakhstan and FYR Macedonia are apt to under-utilize funds, relative to the numeraire labor.
6 Since data on the number of employees are missing for many banks, we instead use total assets to calculate the price of labor. Altunbas et al. (2001), Weill (2004), Bos and Schmiedel (2007) and others have utilized similar definitions.
To cut the production costs and eliminate the AI in the three nations, sample banks should decrease the employment of labor and/or increase the employment of the other two inputs. On the contrary, banks of the remaining countries tend to over-utilize both capital and funds relative to labor. Therefore, those banks should lower the employment of both capital and funds, along with hiring more labor.
Table 15. Estimates of the country-specific AE parameters
2/ 1
Note： ***: Significant at the 1% level. **: Significant at the 5% level. *: Significant at the 10%
level.
Table 16. average relative input prices and relative input quantities
2/ 1
W W X2/X1 W W3/ 1 X3/X1 Croatia 34.6558 0.0204 2.5953 0.8556 Czech Republic 108.5914 0.0247 17.3226 0.8785 Estonia 157.5801 0.0412 1.2313 0.8050 Hungary 196.5227 0.0226 4.4943 0.8825 Kazakhstan 23.3571 0.0380 0.7066 0.8264 Latvia 47.1168 0.0526 1.5395 0.8147 Lithuania 11.1825 0.0908 1.5370 0.8721 FYR Macedonia 37.0202 0.0527 2.0040 0.7695
Poland 16.2563 0.0251 4.6080 0.8089 Romania 37.1894 0.0477 5.9356 0.8458 Russia 133.5765 0.0374 3.9529 0.8000 Slovak Republic 497.3563 0.0440 8.6207 0.9338 Slovenia 76.7797 0.0324 2.7684 0.8447 Ukraine 78.9257 0.1121 3.4675 0.8156
Table 17 presents both the first and the third step parameter estimates for Models A and B. All of the first step estimates are significantly different from zero and their magnitudes and signs deviate from those of the third step estimates. Recall that the first step estimates are recommended to be exploited to calculate the AE measures shown in Table 19 as AE1.
Table 17. Parameter estimates of the Semiparametric regression
Model A Model B
Table 18 summarizes the estimates of the distribution parameters for the three models. All of the three parameters are significantly estimated at the 1% level. It is important to note that the estimates of Model A are intimately close to those of Model B as expected, while they substantially deviate from those of Model C. As the estimate of is positive, the TE of the sample banks is found to improve over time during the sample period. However, Model C tends to overestimate the rate of
enhancement in production efficiency.
Table 18. Estimates of the distribution parameters of the three models
Model A Model B Model C
log-likelihood -2700.90 -2698.70 -2878.38
We use the parameter estimates to evaluate the TE and AE scores. Table 19 reports the outcomes. The average AE1 is equal to 89.21%, meaning that the potential percent of cost savings for an average bank that achieves AE alone is 10.79%. We ignore measure AE2 since the simulation results fail to support it as a suitable measure. The average TE score of Model A attains 79.39%, which is indistinguishable from the TE measure of Model B. This verifies the finding of the previous section, i.e., Model B is compatible with Model A particularly for panel data with long time periods. Evidence is found that measure TI dominates measure AI for the sample states. Sample banks are advised to elevate their managerial capability, followed by optimizing the input mix.
Table 19. Average TE and AE scores of the three models
Model A Model B Model C
Mean St. Dev. Mean St. Dev. Mean St. Dev.
AE1 (%) 89.21 0.1013 89.21 0.1013 89.21 0.1013 AE2 (%) 85.85 0.0950 94.11 0.0816 N/A N/A
TE (%) 79.39 0.1209 79.37 0.1102 77.30 0.1185
Since both Models A and B have similar TE and AE measures, Table 20 merely reports the country-specific efficiency scores of Model A. Banks in Poland reach the highest average AE score (99.25%), followed by Czech Republic (98.66%), Slovak Republic (92.10%), and Slovenia (91.35%). Banks in FYR Macedonia are the most technically efficient (90.47%), followed by Slovenia (84.02%), Croatia (83.78%), and
Poland (83.77%). The correlation coefficient between the TE and AE measures is equal to 0.19. Both measures are somewhat positively correlated, indicating that a more technically efficient bank tends to be more or less allocatively efficient.
Table 20. Country-specific TE and AE measures of Model A
AE (%) TE (%)
Mean St. Dev. Mean St. Dev.
Croatia 88.11 0.0314 83.78 0.1096
Czech Republic 98.66 0.0162 77.65 0.1050
Estonia 77.91 0.0601 77.72 0.0920
Hungary 90.08 0.0599 83.36 0.0813
Kazakhstan 76.63 0.0399 79.76 0.1264
Latvia 72.07 0.0642 71.97 0.1234
Lithuania 82.00 0.0498 67.46 0.1158
FYR Macedonia 75.96 0.0668 90.47 0.0626
Poland 99.25 0.0045 83.77 0.0856
Romania 90.55 0.0806 78.88 0.0705
Russia 83.60 0.0913 71.61 0.1469
Slovak Republic 92.10 0.0488 73.93 0.0597
Slovenia 91.35 0.0244 84.02 0.0571
Ukraine 83.75 0.0558 72.61 0.1677
( , )
Corr AE TE 0.19
Figure 1 depicts the various efficiency measures over the sample period of 1993 to 2004. The average AE scores vary in a narrow range between 86.92% and 91.59%, while the TE scores of the three models show a gradually increasing trend over the transition period. It is evident that the enforcement of financial liberalization and privatization by the governments of the sample countries is indeed effective in prompting the TE of their financial industries across time. Models A and B provide similar trends on the TE scores, while Model C underestimates the average TE measure for the first half of the sample period.
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
year
TE_model A TE_model B TE_model C TE_Translog AE
Figure 1. Average efficiency by year
We estimate the translog cost function as well. Its parameter estimates and the TE and AE measures are shown in Appendices XXII and XXIII. A vast majority of the parameters are significantly estimated. As far as the three distribution parameters are concerned, although the translog estimate of is fairly near those of Models A and B, the remaining two translog estimates of 2 and differ considerably from those of Table 18. If we feel comfortable with the semiparametric model, then the inferences based on a purely parametric specification are likely to be doubtful.
Consequently, the average TE scores obtained from the translog cost function tend to be undervalued to a large extent, as shown in Figure 1 and Table XXIII. The introduction of a more flexible semiparametric model likely avoids the possibility of an inconsistent estimation arising from incorrect parameterization and potential confounding of specification error with inefficiency.