Abstract
This paper proposes a semiparametric smooth-varying coefficient input distance frontier model with multiple outputs and multiple inputs, panel data, and determinants of technical inefficiency for the Indonesian banking industry during the period 2000 to 2015. The technology parameters are unknown functions of a set of environmental factors that shift the input distance frontier non-neutrally. The computationally simple constraint weighted bootstrapping method is employed to impose the regularity constraints on the distance function. As a by-product, total factor productivity (TFP) growth is estimated and decomposed into technical change, scale component, and efficiency change. The distance elasticities, marginal effects of the environmental factors on the distance elasticities, temporal behavior of technical efficiency, and also TFP growth and its components are investigated.
Similar content being viewed by others
Notes
We would like to thank an anonymous referee for this observation.
Although the semiparametric model is not as flexible as a fully nonparametric model, it suffers less from the “curse of dimensionality” (Li and Racine 2007).
The cost shares of inputs (i.e., the input elasticities) change over time because of reallocation of resources over time, ceteris paribus. If banks allocate more resources in credit and portfolio analysis, then banks’ output quality will improve over time. The marginal costs of outputs (i.e., the output elasticities) change over time when bank managers are risk-averse and decide to increase costs to reduce risk over time.
As NPLs increase, a risk premium is paid to depositors who receive a higher interest rate as compensation for tolerating the extra loan risk (Hannan and Hanweck 1988; Hughes and Mester 1993). This risk premium, along with risk-averse preference of bank managers who become more reliant on deposit or short-term funding from interbank market to generate income, would affect the cost shares of inputs (i.e., the input elasticities). NPLs also trigger extra operating tasks, including additional monitoring and handling of these NPLs (Berger and DeYoung 1997). These additional tasks would divert managers’ attention away from normal daily monitoring of financial transactions, and therefore have implications for banks’ income-generating capabilities and their marginal costs of outputs (i.e., the output elasticities).
Computationally speaking, the CWB requires quadratic programming techniques which are widely available in econometric software packages.
Although the two-step estimation procedure is not efficient, it is easy for practitioners to implement, and the constrained smooth coefficients estimated in the first step do not depend on distributional assumptions. One-step estimation of the constrained local-linear SPSC model with determinants of technical inefficiency is saved for future research.
Financial integration that facilitates cross-border financial transactions and international capital flows allows more risk diversification, but at the same time, intensifies bank competition. As the number of loans increases, the quality of these loans might decrease, which makes banks more fragile when confronted with financial crisis. While risk exposure provides banks with higher expected return, the risk management requires increased costs—banks sometimes have to trade profit for reduced risk (Hughes and Mester 1998).
In this paper, we focus on the input-oriented technical inefficiency; therefore, the actual outputs equal the maximum feasible outputs.
Unlike traditional fully parametric stochastic frontier models in which Zit only appears in the inefficiency function rather than the frontier function, the SPSC stochastic frontier models include Zit in the inefficiency and frontier functions at the same time. Therefore, the estimating equation, \(E({\varepsilon }_{it}^{* * }| \cdot )\) must include Zit, in addition to \({\mathrm{ln}}\,{Y}_{it}\) and \({\mathrm{ln}}\,{\widetilde{X}}_{it}\).
That is, the E(uit∣Zit) is absorbed by the intercept, \(\widetilde{\alpha }({Z}_{it})\).
The bandwidth selection method for estimating ρ(⋅) is the least-squares cross-validation (LSCV). Appendix A explains in detail how we select the bandwidths to estimate \(\tilde{\rho }(\cdot )\).
It is not necessary to test for the validity of the economic constraints given in (7)–(9) that are dictated by economic theory. For example, we do not want to test if the non-negative marginal cost constraint is valid when estimating a cost function. This is because the non-negative marginal cost is one of the properties of the cost function dictated by microeconomic theory. Data points that violate such a property would not be of any interest to policy makers. Test for the validity of constraints is warranted when some non-economic constraints are imposed—see Du et al. (2013) for testing for the validity of constraints under the CWB framework via bootstrapping.
The R codes for imposing these constraints are available from the authors upon request.
We use the fact that \(\mathop{\sum }\nolimits_{j = 1}^{J}{S}_{j}=1\) and \({\dot{\widetilde{X}}}_{1}=0\). See Appendix B for derivation details for the decomposition. Kumbhakar and Wang (2005) showed a similar decomposition based on a parametric production function.
See Parmeter et al. (2014) for more details about imposing constraints on fully parametric models via the CWB approach.
Banks liquidated or closed down, or banks established during the sample period, are also excluded from our dataset.
The bootstrap standard error of the mean (quartile) of each quantity, say, γ1, is calculated as follows (Cameron and Trivedi 2005, Chapter 11). First, generate a bootstrap sample \({\gamma }_{1}^{b}\) by randomly selecting from γ1 with replacement. Second, compute the mean (quartile) of \({\gamma }_{1}^{b}\). Call it the bootstrap mean (quartile) estimate. Repeat these two steps 1000 times, and the standard error of the mean (quartile) is viewed as the standard deviation calculated using the bootstrap mean (quartile) estimates.
The asymptotic standard errors of the mean are calculated by dividing the sample standard deviation of each quantity, say, γ1, by the square root of the sample size, and those of the quartiles are calculated according to Koenker and Bassett (1978). The quantreg package of R is useful for this purpose.
Even if there is no violation for a particular quantity (e.g., γ3 and β4), it is still possible that the unconstrained and constrained densities significantly differ from each other. This is because a large number of violations of any other quantities (e.g., β2) would require a significant amount of perturbation of the uniform weights, and hence the dependent variable. In summary, it is not that a smaller percentage of violations must produce a larger p-value, it is how the weights have to change to ensure that all the constraints are imposed simultaneously.
Recall that \({\beta }_{1}\equiv \partial \mathrm{ln}\,D/\partial \ \mathrm{ln}\,{X}_{1}=1-\mathop{\sum }\nolimits_{j = 2}^{J}{\beta }_{j}\).
This is because a 1% increase in Yk would cause X1 to increase by −γk%, and in order to hold everything else constant, including the input ratios in log (i.e., \({\mathrm{ln}}\,{X}_{j}-{\mathrm{ln}}\,{X}_{1}\)), Xj must also increase by the same percentage, i.e., −γk%, ∀ j = 2, …, J.
We would like to thank an anonymous referee for pointing this out.
It is worth noting that the bootstrap and asymptotic standard errors are the same to the fourth decimal place for all the mean values, but are quite different for some quartile values. Generally speaking, the asymptotic standard error of the mean does not require the quantity of interest to follow the normal (i.e., Gaussian) distribution, but those of the quartiles assume that the quantity of interest follows the normal distribution. Although the local-linear SPSC estimator is asymptotically normal (Geng and Sun 2019; Li and Racine 2007), the local-linear smooth coefficients and their derivative estimates from a finite sample may follow non-normal distributions. It turns out that the finite sample distributions of the marginal effect estimates are far away from the normal distribution—kernel density plots of the marginal effect estimates are omitted to save space, but are available upon request, and therefore the asymptotic standard errors of the quartiles would be misleading, and it is recommended that we use the bootstrap to compute the standard errors of the quartiles.
Since \(RTS=-1/\mathop{\sum }\nolimits_{k = 1}^{K}{\gamma }_{k}\left(Z\right)\), then \(\partial RTS/\partial {Z}_{s}={\left(\mathop{\sum }\nolimits_{k = 1}^{K}\gamma \left(Z\right)\right)}^{-2}\cdot \mathop{\sum }\nolimits_{k = 1}^{K}\partial \gamma \left(Z\right)/\partial {Z}_{s}\), ∀ s = 1, 2, where s = 1 for NPLs in log; s = 2 for time. Since \({\left(\mathop{\sum }\nolimits_{k = 1}^{K}\gamma \left(Z\right)\right)}^{-2}\) is positive, the sign of ∂RTS/∂Zs depends on the sign of the sum of the derivatives of all the γ’s with respect to a particular Z.
Negation of the sign of γ’s would cause the sign of the derivative of γ’s to change, with the same bootstrap and asymptotic standard errors.
See Hughes and Mester (1998) who expressed scale economy as a function of NPLs.
References
Aigner DJ, Lovell CAK, Schmidt P (1977) Formulation and estimation of stochastic frontier production functions. J Econ 6(1):21–37
Altunbas Y, Liu M-H, Molyneux P, Seth R (2000) Efficiency and risk in Japanese banking. J Bank Finance 24(10):1605–1628
Assaf AG, Matousek R, Tsionas EG (2013) Turkish bank efficiency: Bayesian estimation with undesirable outputs. J Bank Finance 37(2):506–517
Bank Indonesia (2012–2017) Indonesian banking statistics, Report.
Berger AN, DeYoung R (1997) Problem loans and cost efficiency in commercial banks. J Bank Finance 21(6):849–870
Berger AN, Humphrey DB (1997) Efficiency of financial institutions: international survey and directions for future research. Eur J Oper Res 98(2):175–212
Bhaumik SK, Das PK, Kumbhakar SC (2012) A stochastic frontier approach to modelling financial constraints in firms: an application to India. J Bank Finance 36(5):1311–1319
Bhaumik SK, Kumbhakar SC, Sun K (2015) A note on a semiparametric approach to estimating financing constraints in firms. Eur J Finance 21:992–1004
Cai Z, Fan J, Li R (2000) Efficient estimation and inferences for varying-coefficient models. J Am Stat Assoc 95(451):888–902
Cai Z, Fan J, Yao Q (2000) Functional-coefficient regression models for nonlinear time series. J Am Stat Assoc 95(451):941–956
Cameron, A. and Trivedi, K. (2005) Microeconometrics: methods and application, Cambridge University Press, chapter 11, pp. 376–377
Chen R, Tsay R (1993) Functional-coefficient autoregressive models. J Am Stat Assoc 88:298–308
Das A, Kumbhakar SC (2012) Productivity and efficiency dynamics in Indian banking: An input distance function approach incorporating quality of inputs and outputs. J Appl Econom 27(2):205–234
Defung F, Salim R, Bloch H (2016) Has regulatory reform had any impact on bank efficiency in Indonesia? A two-stage analysis. Appl Econ 48:5060–5074
Denizer C (2000) Foreign entry in Turkey’s banking sector, 1980–97, Policy Research Working Paper Series 2462, The World Bank, Washington DC
Diewert WE, Wales TJ (1987) Flexible functional forms and global curvature conditions. Econometrica 55(1):43–68
Du P, Parmeter C, Racine J (2013) Nonparametric kernel regression with multiple predictors and multiple shape constraints. Stat Sinica 23(3):1347–1371
Esho N (2001) The determinants of cost efficiency in cooperative financial institutions: Australian evidence. J Bank Finance 25(5):941–964
Fan J, Zhang W (1999) Statistical estimation in varying-coefficient models. Ann Stat 27:1491–1518
Fan Y, Li Q (1996) Consistent model specification tests: omitted variables and semiparametric functional forms. Econometrica 64:865–890
Feng G, Serletis A (2010) Efficiency, technical change, and returns to scale in large US banks: panel data evidence from an output distance function satisfying theoretical regularity. J Bank Finance 34(1):127–138
Fofack, H.L. (2005) Nonperforming loans in Sub-Saharan Africa: Causal analysis and macroeconomic implications, Policy Research Working Paper Series 3769, The World Bank, Washington DC
Fries S, Taci A (2005) Cost efficiency of banks in transition: evidence from 289 banks in 15 post-communist countries. J Bank Finance 29(1):55–81
Geng X, Sun K (2019) Gradient estimation of the local-constant semiparametric smooth coefficient model. Econ Lett 185:108684
Gunes H, Yildirim D (2016) Estimating cost efficiency of Turkish commercial banks under unobserved heterogeneity with stochastic frontier models. Central Bank Rev 16(4):127–136
Hadad MD, Hall MJB, Santoso W, Satria R, Kenjegaliev K, Simper R (2008) Efficiency in Indonesian banking: Recent evidence, Discussion paper series, Department of Economics, Loughborough University
Hall P, Huang H (2001) Nonparametric kernel regression subject to monotonicity constraints. Ann Stat 29(3):624–647
Hannan T, Hanweck G (1988) Bank insolvency risk and the market for large certificates of deposit. J Money Credit Bank 20(2):203–211
Hastie T, Tibshirani R (1993) Varying-coefficient models. J R Stat Soc. Series B (Methodological) 55(4):757–796
Heshmati A, Kumbhakar SC, Sun K (2014) Estimation of productivity in Korean electric power plants: a semiparametric smooth coefficient model. Energy Econ 45:491–500
Hughes JP, Mester LJ (1993) A quality and risk-adjusted cost function for banks: Evidence on the “too-big-to-fail” doctrine. J Prod Anal 4(3):293–315
Hughes JP, Mester LJ (1998) Bank capitalization and cost: evidence of scale economies in risk management and signaling. Rev Econ Stat 80(2):314–325
Koenker RW, Bassett GW (1978) Regression quantiles. Econometrica 46:33–50
Kumbhakar SC (2013) Specification and estimation of multiple output technologies: a primal approach. Eur J Oper Res 231:465–473
Kumbhakar SC, Park BU, Simar L, Tsionas EG (2007) Nonparametric stochastic frontiers: a local maximum likelihood approach. J Econ 137(1):1–27
Kumbhakar SC, Sun K (2012) Estimation of TFP growth: a semiparametric smooth coefficient approach. Empir Econ 43:1–24
Kumbhakar SC, Wang H-J (2005) Estimation of growth convergence using a stochastic production frontier approach. Econ Lett 88(3):300–305
Li Q, Huang C, Li D, Fu T (2002) Semiparametric smooth coefficient models. J Bus Econ Stat 20(3):412–422
Li, Q. and Racine, J. (2007) Nonparametric Econometrics: Theory and Practice, Princeton University Press
Li Q, Racine J (2010) Smooth varying-coefficient estimation and inference for qualitative and quantitative data. Econ Theory 26:1–31
Longford NT (1994) Random Coefficient Models (Oxford Statistical Science Series) 1st Edition, Clarendon Press
Meeusen W, van den Broeck J (1977) Efficiency estimation from Cobb-Douglas production functions with composed error. Int Econ Rev 18(2):435–444
Mester LJ (1996) A study of bank efficiency taking into account risk-preferences. J Banking Finance 20(6):1025–1045
O’Donnell CJ, Coelli TJ (2005) A Bayesian approach to imposing curvature on distance functions. J Econom 126(2):493–523
O’Donnell CJ, Rambaldi AN, Doran HE (2001) Estimating economic relationships subject to firm- and time-varying equality and inequality constraints. J Appl Econom 16(4):709–726
Park KH, Weber WL (2006) A note on efficiency and productivity growth in the Korean banking industry, 1992–2002. J Bank Finance 30(8):2371–2386
Parmeter CF, Sun K, Henderson DJ, Kumbhakar SC (2014) Estimation and inference under economic restrictions. J Product Anal 41(1):111–129
Podpiera J, Weill L (2008) Bad luck or bad management? Emerging banking market experience. J Financ Stab 4(2):135–148
Resti A (1997) Evaluating the cost-efficiency of the Italian banking system: what can be learned from the joint application of parametric and non-parametric techniques. J Bank Finance 21(2):221–250
Sealey CW, Lindley JT (1977) Inputs, outputs, and a theory of production and cost at depository financial institutions. J Finance 32(4):1251–1266
Servin R, Lensink R, van den Berg M (2012) Ownership and technical efficiency of microfinance institutions: Empirical evidence from Latin America. J Bank Finance 36(7):2136–2144
Shephard R (1953) Cost and production functions. Princeton University Press, Princeton
Shephard R (1970) Theory of cost and production functions. Princeton University Press, Princeton
Stevenson R (1980) Measuring technological bias. Am Econ Rev 70:162–173
Sun K (2015) Constrained nonparametric estimation of the input distance function. J Product Anal 43(1):85–97
Sun K, Kumbhakar SC (2013) Semiparametric smooth-coefficient stochastic frontier model. Econ Lett 120(2):305–309
Sun K, Kumbhakar SC, Tveterås R (2015) Productivity and efficiency estimation: a semiparametric stochastic cost frontier approach. Eur J Oper Res 245(1):194–202
Terrell D (1996) Incorporating monotonicity and concavity conditions in flexible functional forms. J Appl Econ 11(2):179–194
Wang H-J, Ho C-W (2010) Estimating fixed-effect panel stochastic frontier models by model transformation. J Econ 157:286–296
Wang H-J, Schmidt P (2002) One-step and two-step estimation of the effects of exogenous variables on technical efficiency levels. J Product Anal 18:129–144
Yao F, Zhang F, Kumbhakar SC (2019) Semiparametric smooth coefficient stochastic frontier model with panel data. J Bus Econ Stat 37(3):556–572
Acknowledgements
This research is funded by the National Natural Science Foundation of China (Grant ID number: 71801146). The authors would like to thank two anonymous referees for helpful comments, and remain responsible for all remaining errors.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A
This appendix explains how we select the bandwidths to estimate the smooth coefficients and their derivatives in (11). Following Li and Racine (2010), we employ the most commonly used least-squares cross-validation (LSCV) method, which is a fully automatic data-driven approach, to select the bandwidth vector h; that is,
where CVll(h) determines the cross-validation bandwidth vector h for local-linear estimator, \({W}_{it}^{\prime}{\hat{\rho }}_{-it}({Z}_{it})\) is the leave-one-out local-linear kernel conditional mean, and 0 ≤ M(⋅) ≤ 1 is a weight function that serves to avoid difficulties caused by dividing by zero. The same bandwidth vector is used to estimate the constrained smooth coefficients.
Appendix B
This appendix details the TFP growth decomposition given in Section 2.3.
First, take the time derivatives of both sides of (4) (ignoring the noise term), and we would have:
The TC component measures the non-neutral shift of the IDF over time; therefore,
where \(TC=\frac{\partial \alpha (Z)}{\partial t}+\mathop{\sum }\nolimits_{k = 1}^{K}\frac{\partial {\gamma }_{k}(Z)}{\partial t}{\mathrm{ln}}\,{Y}_{k}+\mathop{\sum }\nolimits_{j = 2}^{J}\frac{\partial {\beta }_{j}(Z)}{\partial t}{\mathrm{ln}}\,{\widetilde{X}}_{j}\), and \(EC=-\frac{u}{t}\) measures the efficiency change.
Add \(\dot{TFP}\) to both sides of (35) and rearrange to obtain:
where \(SC=\mathop{\sum }\nolimits_{k = 1}^{K}{R}_{k}{\dot{Y}}_{k}+\mathop{\sum }\nolimits_{k = 1}^{K}{\gamma }_{k}(Z){\dot{Y}}_{k}=\mathop{\sum }\nolimits_{k = 1}^{K}({R}_{k}+{\gamma }_{k}(Z)){\dot{Y}}_{k}\). To further decompose the SC into one sub-component related to RTS and the other related to market power:
given that \(RTS=-1/\mathop{\sum }\nolimits_{k = 1}^{K}{\gamma }_{k}(Z)\). Finally, it can be shown that:
using the fact that \(\mathop{\sum }\nolimits_{j = 1}^{J}{S}_{j}=1\). Furthermore,
using the fact that \({\dot{\widetilde{X}}}_{1}=0\). Therefore, plug (39) into (38) to obtain \(AC=\mathop{\sum }\nolimits_{j = 2}^{J}({\beta }_{j}(Z)-{S}_{j}){\dot{\widetilde{X}}}_{j}\). QED.
Rights and permissions
About this article
Cite this article
Sun, K., Salim, R. A semiparametric stochastic input distance frontier model with application to the Indonesian banking industry. J Prod Anal 54, 139–156 (2020). https://doi.org/10.1007/s11123-020-00589-3
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11123-020-00589-3