APPENDIX A. DATA DESCRIPTION This study uses banks' balance sheet and income statement data for an unbalanced panel of 403 Italian CBs over the period 2006-2013, obtained from the Bilbank-Italian Banking Association database. The analysis also incorporates data on environmental variables that could affect bank efficiency. Information on the number of firms that went bankrupt over the total number of registered ones (DEFAULT_RATE) at the province level 1 are provided by Istituto Tagliacarne. Data on population, capital, patents, labour force and urban typology are obtained by the Italian Bureau of Statistics - ISTAT. The number of branches for each bank at the municipal level as well as data on deposits and loans at a provincial detail are taken from the Bank of Italy reports. Table A1 reports the sample coverage; Table A2 presents some descriptive statistics on the inputs and outputs used in estimating the cost frontier function. Descriptive statistics on variables used in estimating the growth model are shown in Table A3. 1 From an administrative point of view, Italy is divided into 20 regions, 110 provinces and 8,047 municipalities.
Table A1. Sample coverage 2006 2007 2008 2009 2010 2011 2012 2013 Number Cooperative banks: sample 398 403 403 402 400 399 379 367 Cooperative banks: national total 436 440 432 421 415 411 394 385 Italian banks: national total 793 806 799 788 760 740 706 684 Cooperative banks: sample coverage 91.28% 91.59% 93.29% 95.49% 96.39% 97.08% 96.19% 95.32% Cooperative banks: population percentage 54.98% 54.59% 54.07% 53.43% 54.61% 55.54% 55.81% 56.29% Branches Cooperative banks: sample 3,601 3,777 4,061 4,187 4,327 4,327 4,429 4,431 Cooperative banks: national total 3,752 3,922 4,109 4,243 4,373 4,427 4,445 4,449 Italian banks: national total 32,337 33,225 34,139 34,036 33,663 33,607 32,881 31,761 Cooperative banks: sample coverage 95.98% 96.30% 98.83% 98.68% 98.95% 97.74% 99.64% 99.60% Cooperative banks: population percentage 11.60% 11.80% 12.04% 12.47% 12.99% 13.17% 13.52% 14.01% Total assets (billions) Cooperative banks: sample 131 144 159 171 177 186 203 209 Cooperative banks: national total 135 148 163 173 181 187 204 210 Italian banks: national total 2,461 2,729 3,337 3,316 3,136 3,235 3,286 3,115 Cooperative banks: sample coverage 97% 97% 98% 99% 98% 100% 100% 99% Cooperative banks: population percentage 5.47% 5.43% 4.88% 5.21% 5.76% 5.77% 6.21% 6.74% This table reports the number of banks, branches and total assets for the cooperative bank group, both in the sample and population, and the whole Italian banking system for each calendar year. Sample coverage of the cooperative bank group and quota of the Italian cooperative banks on the whole Italian banking system are also provided. Source: Bank of Italy Annual Report and Bilbank (various years) - ABI data set. 2
Table A2. Descriptive statistics of input and output variables for the frontier model (2006-2013) Variable 1 Variable definition Obs Mean Std. Dev. Min Max Total production cost (i.e. 3,151 8,710.31 10,066.83 171.21 147,977.00 Total production cost personnel and other administrative fees) 2 Personnel expenses over the 3,142 69.90 7.67 16.26 126.27 Price of labour (p1) number of employees Interests expenses over total 3,147 1.04 0.42 0.04 3.99 Price of funding (p2) funding Administrative expenses 3,148 1.08 2.25 0.01 73.53 Price of fixed assets (p3) over total fixed assets Total amount of loans to 3,148 301,940.60 385,784.10 3,057.19 5,004,958.00 Total customer loans (q1) customers Total customer demand 3,150 208,185.20 300,997.90 1,702.80 5,452,095.00 Total customer deposits (q2) deposits Total of other earning assets 3,151 117,504.00 179,668.70 3,041.56 4,054,595.00 Total other earning assets (q3) (i.e. government bonds, corporate bonds and stocks) Total equity capital Total equity capital 3 3,046 45,641.84 52,777.27 1,552.63 659,389.90 Notes: 1 All values - except for the prices of funding and fixed assets - are in thousands of euros and deflated by using the national consumer price index provided by ISTAT (base: 2010); 2 We use the definition of operating expenses as shown in the income statement excluding net provisions for risks and charges, net adjustments to/recoveries of property and equipment, net adjustments to/recoveries of intangible assets, and other operating expenses; 3 Total capital is defined as the sum of reimbursable shares, equity instruments, reserves, share premium reserves, share capital, treasury shares, and net income (loss). 3
Table A3. Descriptive statistics of variables used in the growth model (2006-2013) Std. Obs Mean Variables 1 Description Dev. Min Max FQ Cost Efficiency 806 0.80 0.12 0.02 1.00 Credit volume of minor banks FV relative to GDP (for millions of euros) 856 110.38 102.13 0.00 735.58 GPP Gross Provincial Product (millions of euros) 868 13,160.05 19,514.68 807.34 17,0598.5 OCCUPATION_GROWTH Growth rate of labour force 880 0.00 0.03-0.10 0.11 CAPITAL Stock of capital (millions of euros) 880 22,362.52 18,526.48 1,068.22 76,209.02 PATENTS Number of patents 824 94.23 279.01 0.00 2,542.00 ATM_BRANCHES Number of ATMs per branch 856 1.26 0.25 0.03 2.29 BRANCHESpc Number of branches for 1,000 inhabitants 868 0.60 0.22 0.10 1.66 % of CB's branches over the national CBs_BRANCHES total 880 11.51 9.73 0.00 60.53 Deposit per inhabitant (thousands of DEPOSITpc 860 13.65 5.88 2.43 42.33 euros) Notes: 1 All values are deflated by using the national consumer price index provided by ISTAT (base: 2010). 4
APPENDIX B. MODEL ESTIMATION AND LR TESTS Table B1 reports the parameter estimates of the model M4 (Table 4), that includes internal and territorial banking characteristics. Table B2 presents the results of the various null hypotheses tests associated with the inefficiency term of the model. 5
Table B1. ML estimates for the model M4 Coefficient Estimate Standard Error t-ratio Stochastic Frontier -3.818 0.024-156.239 0 1.403 0.173 8.089 k1-1.165 0.116-10.077 k 2-1.037 0.293-3.538 k3 0.003 0.072 0.041 p1 p 2-0.232 0.057-4.089 j1k 1-0.437 0.102-4.274 j1k 2 1.205 0.019 64.749 j1k 3 1.659 0.183 9.058 j 2k 2-0.059 0.021-2.809 j 2k 3 0.155 0.031 4.953 j3k 3-0.374 0.093-4.014 m1 p1 5.295 0.328 16.134 m1p2-0.027 0.025-1.080 m2 p1-0.070 0.026-2.661 k1 p1-0.202 0.223-0.908 k1 p 2-0.056 0.048-1.148 k 2 p1-0.151 0.063-2.396 k 2 p2 0.029 0.018 1.634 k 3 p1-0.030 0.018-1.640 k 3 p 2-0.041 0.007-5.602 t 0.033 0.006 5.568-0.005 0.001-7.156 t2 E 0.002 0.002 1.018 Inefficiency Model 0 3.448 0.144 23.967 0.727 0.110 6.633 SIZE -0.293 0.027-10.999 DIV REV LLP -0.047 0.018-2.600 0.019 0.006 2.970 NPL 0.155 0.004 36.265 BRANCHES 0.049 0.016 3.013 HQ_DISTANCE 0.329 0.026 12.637 H _ STAT POP 0.013 0.014 0.889 0.083 0.014 6.027 DEFAULT_ RATE Loglikelihood Function LL 1,505.133 LR test of the one sided error 1,399.315 6
Table B2. Hypotheses testing for the functional form of the stochastic production function H0 Hypothesis tested Log_ λ Number of 2 Decision whit 0.05 likelihood restrictions respect to H0 jk mp kp 0 j, k, p, m Cobb-Douglas specification 1027.49 955.29 15 24.996 Rejected t 2 t 0 Only biased technical change 881.60 1247.06 2 5.991 Rejected 0 DIV REV SIZE LLP NPL BRANCHES Inefficiency factors have 0 zero influence 850.65 1308.96 HQ _ DISTANCE H _ STAT POP DEFAULT _ RATE 10 18.307 Rejected Environment, territorial BRANCHES HQ _ DISTANCE H _ STAT POP DEFAULT _ R features and branches 1055.48 899.30 have zero influence 5 11.070 Rejected 0 Environment and HQ _ DISTANCE H _ STAT POP DEFAULT _ RATE territorial features have 928.81 1152.64 zero influence 4 9.488 Rejected 0 POP DEFAULT Environment has zero _ RATE influence 830.40 1349.47 2 5.991 Rejected 7
APPENDIX C. SOME INSIGHTS ON COST EFFICIENCY DISTRIBUTIONS Cost efficiency scores (CE), representing the relative distance from the cost frontier of the bestpractice banks, are computed for all banks over the period 2006-2013. The average CE over the sample period and across the banks sample is 0.84, indicating that on average banks may further reduce their actual costs by 16% to be fully efficient. The average CE per year slightly increases over time. In Figure C1, the CE downturn starting in 2008 is consistent with the Lehman bankruptcy; a further but less intense fall can be seen in 2011 when the Italian banks faced a more severe sovereign debt crisis. The Kernel density plot of CE (Figure C2) shows an asymmetric distribution around the mean value; and the left tail of the distribution is thin. The distribution of the CE suggest there is a small number of banks operating at a low level of cost inefficiency: only 3% of banks show a score lower than 0.65. Figure C1. Cost Efficiency dynamics 8
Figure C2. Kernel density of Cost Efficiency 0 1 2 3 4.4.6.8 1 9
APPENDIX D. ROBUSTNESS CHECKS Some checks are carried out to further verify the robustness of results with respect to the estimation strategies. Model M4 is adopted as the baseline for robustness checks (Table D1). To support our modelling strategies, we consider different model estimators, a parsimonious cost function and a different cost function approach. Following the Greene approach to frontier function (Greene, 2005), we estimate model M4 by means of fixed and random effects estimators. Results show no significant differences in parameter estimates with respect to a ML approach. A Cobb Douglas function is estimated instead of the translog specification; the LR test strongly rejects the Cobb Douglas in favour of a flexible specification (see Table 2 in Appendix B). Berger and Humphrey (1992) identify two main approaches for the selection of inputs and outputs: the production approach and the intermediation approach. The first assumes that banks produce loans and deposits, using labour and capital as inputs, and that the number and type of transactions or documents processed as measures of outputs. The second approach perceives banks as financial intermediaries between savers and investors and may be a more appropriate measure for evaluating financing institutions as a whole. The other two versions of the intermediation approach (apart from the value added used in this analysis) are the asset and the user cost approaches. The asset approach is a reduced form modelling of the banking activity. Deposits and other liabilities, together with real resources (labour and capital) are defined as inputs, whereas the output set only includes bank assets. This approach was first suggested by Sealey and Lindley (1977). The user-cost approach determines whether a financial product is an input or an output on the basis of its net contribution to bank revenue. If the financial returns on an asset exceed the opportunity cost of the funds or if the financial costs of a liability are less than the opportunity cost, then they are considered as outputs. Otherwise, they are considered as inputs. Then, as a robustness check we suggest testing the intermediation approach proposed by Sealey and Lindley (1977) and positing deposits and other liabilities, together with real resources 10
(labour and capital) as inputs, whereas the output set only includes bank assets. Results largely confirm those obtained by using the valued added approach. Table D1. Parameter estimates of the inefficiency models with respect functional form and estimation approach Model 4 Model 6 Model 7 Model 8 Model 9 Full FE RE Cobb- Douglas Asset Approach SIZE -0.293*** -0.203*** -0.151*** -0.111*** -0.215*** (0.030) (0.007) (0.007) (0.005) (0.036) DIV REV 0.727*** 0.447*** 0.428*** 0.662*** 0.386** (0.110) (0.043) (0.042) (0.049) (0.164) LLP -0.047*** -0.401-0.721*** 0.046*** -0.014 (0.020) (0.323) (0.314) (0.010) (0.029) NPL 0.019*** 0.080*** 0.062*** 0.069*** 0.036** (0.010) (0.006) (0.006) (0.008) (0.016) BRANCHES 0.155*** 0.174*** 0.166*** 0.002 0.121*** (0.000) (0.010) (0.009) (0.007) (0.031) HQ_DISTANCE 0.049*** 0.004-0.003 0.044*** 0.020 (0.020) (0.005) (0.005) (0.004) (0.016) H_STAT 0.329*** 0.239*** 0.247*** 0.290*** 0.285*** (0.030) (0.013) (0.013) (0.014) (0.032) POP 0.013 0.012*** 0.008* -0.070*** -0.101 (0.010) (0.005) (0.004) (0.012) (0.070) DEFAULT_RATE 0.083*** 0.054*** 0.035*** 0.076*** 0.034 (0.010) (0.005) (0.005) (0.008) (0.029) CONST 3.448*** 3.909*** 4.839*** 1.447*** 15.341*** (0.140) (1.487) (1.445) (0.060) (0.546) Yearly sample size 403 403 403 403 403 LL 1,505.133.. 1,027.486 1,295.636 Note: The sample includes an unbalanced panel of cooperative banks observed over the period 2006-2013. The dependent variable for all the models is the inefficiency term it from Eq. (3). The independent variables are as follows: internal bank features (SIZE, DIV REV ), risks (LLP, NPL), spatial characteristics and market conditions (BRANCHES, HQ_DISTANCE, H_STAT), population and default ratio (POP and DEFAULT_RATE). Statistical significance is indicated by ***, ** and * for the 1%, 5% and 10% level, respectively. Standard errors are reported in parentheses. 11
REFERENCES Berger, A. N., & Humphrey, D. B. (1992). Measurement and efficiency issues in commercial banking, in Z. Griliches (Ed.) Output Measurement in the Service Sectors, pp. 245 279. The University of Chicago Press, Chicago. Greene, W. (2005). Reconsidering heterogeneity in panel data estimators of the stochastic frontier model. Journal of Econometrics, 126, 269 303. Sealey, C. W., & Lindley, J. T. (1977). Inputs, outputs, and a theory of production and cost at depository financial institutions. Journal of Finance, 32, 1251-1266. 12