Cardiff Economics Working Papers



Similar documents
Optimal Pricing Scheme for Information Services

Multi-Market Trading and Liquidity: Theory and Evidence

Incorporating Negative Values in AHP Using Rule- Based Scoring Methodology for Ranking of Sustainable Chemical Process Design Options

Lecture 3: Force of Interest, Real Interest Rate, Annuity

ALABAMA ASSOCIATION of EMERGENCY MANAGERS

WiMAX DBA Algorithm Using a 2-Tier Max-Min Fair Sharing Policy

Newton-Raphson Method of Solving a Nonlinear Equation Autar Kaw

Faculdade de Economia da Universidade de Coimbra

Answer: A). There is a flatter IS curve in the high MPC economy. Original LM LM after increase in M. IS curve for low MPC economy

YOU FINALLY FINISHED YOUR FILM. NOW WHAT? Distributor...? Sales agent...? GOT IT: SELF DISTRIBUTION

Irregular Repeat Accumulate Codes 1

Joint Opaque booking systems for online travel agencies

Loyalty Program and Customer Retention of Bank Credit Cards --an Logistic Regression Analysis based on Questionnaires

Section 5.3 Annuities, Future Value, and Sinking Funds

A Hadoop Job Scheduling Model Based on Uncategorized Slot

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ).

ORIGIN DESTINATION DISAGGREGATION USING FRATAR BIPROPORTIONAL LEAST SQUARES ESTIMATION FOR TRUCK FORECASTING

An Alternative Way to Measure Private Equity Performance

7.5. Present Value of an Annuity. Investigate

Can Auto Liability Insurance Purchases Signal Risk Attitude?

The CAT model: Predicting air temperature in city streets on the basis of measured reference data

Graphs on Logarithmic and Semilogarithmic Paper

Boolean Algebra. ECE 152A Winter 2012

Addendum to: Importing Skill-Biased Technology

Driver Attitudes and Choices: Speed Limits, Seat Belt Use, and Drinking-and-Driving

Fixed income risk attribution

Section 5.4 Annuities, Present Value, and Amortization

Resistive Network Analysis. The Node Voltage Method - 1

Chapter 7: Answers to Questions and Problems

The OC Curve of Attribute Acceptance Plans

Strategic Labor Supply

On the System Dynamics of the Logistic Risk of Marketing Channels

DlNBVRGH + Sickness Absence Monitoring Report. Executive of the Council. Purpose of report

BERNSTEIN POLYNOMIALS

Using Series to Analyze Financial Situations: Present Value

Treatment Spring Late Summer Fall Mean = 1.33 Mean = 4.88 Mean = 3.

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic

Small pots lump sum payment instruction

Lesson 28 Psychrometric Processes

Vector Geometry for Computer Graphics

Chapter 4 ECONOMIC DISPATCH AND UNIT COMMITMENT

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis

Financial Mathemetics

Construction Rules for Morningstar Canada Target Dividend Index SM

Simple Interest Loans (Section 5.1) :

Solution: Let i = 10% and d = 5%. By definition, the respective forces of interest on funds A and B are. i 1 + it. S A (t) = d (1 dt) 2 1. = d 1 dt.

Lecture 3 Gaussian Probability Distribution

1.1 The University may award Higher Doctorate degrees as specified from time-to-time in UPR AS11 1.

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek

Reporting Forms ARF 113.0A, ARF 113.0B, ARF 113.0C and ARF 113.0D FIRB Corporate (including SME Corporate), Sovereign and Bank Instruction Guide

ASP On-Demand versus MOTS In-House Software Solutions *

On the Optimal Control of a Cascade of Hydro-Electric Power Stations

Research on performance evaluation in logistics service supply chain based unascertained measure

Nordea G10 Alpha Carry Index

LIFETIME INCOME OPTIONS

1. Math 210 Finite Mathematics

High Correlation between Net Promoter Score and the Development of Consumers' Willingness to Pay (Empirical Evidence from European Mobile Markets)

S&P/CITIC CHINA BOND INDICES

How To Calculate The Accountng Perod Of Nequalty

Trade Adjustment and Productivity in Large Crises. Online Appendix May Appendix A: Derivation of Equations for Productivity

How To Get A Tax Refund On A Retirement Account

FINANCIAL MATHEMATICS. A Practical Guide for Actuaries. and other Business Professionals

The impact of hard discount control mechanism on the discount volatility of UK closed-end funds

10.2 Future Value and Present Value of an Ordinary Simple Annuity

DEFINING %COMPLETE IN MICROSOFT PROJECT

Fuzzy Clustering for TV Program Classification

Course outline. Financial Time Series Analysis. Overview. Data analysis. Predictive signal. Trading strategy

COMPARISON OF SOME METHODS TO FIT A MULTIPLICATIVE TARIFF STRUCTURE TO OBSERVED RISK DATA BY B. AJNE. Skandza, Stockholm ABSTRACT

Problem Set 3. a) We are asked how people will react, if the interest rate i on bonds is negative.

Time Value of Money. Types of Interest. Compounding and Discounting Single Sums. Page 1. Ch. 6 - The Time Value of Money. The Time Value of Money

Recurrence. 1 Definitions and main statements

Why is the NSW prison population falling?

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12

REQUIRED FOR YEAR END 31 MARCH Your business information

Forecasting the Direction and Strength of Stock Market Movement

Econ 4721 Money and Banking Problem Set 2 Answer Key

Structural Estimation of Variety Gains from Trade Integration in a Heterogeneous Firms Framework

Efficient Project Portfolio as a tool for Enterprise Risk Management

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

Factoring Polynomials

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

MULTI-CRITERIA DECISION AIDING IN PROJECT MANAGEMENT OUTRANKING APPROACH AND VERBAL DECISION ANALYSIS

Polynomial Functions. Polynomial functions in one variable can be written in expanded form as ( )

reduce competition increase market power cost savings economies of scale and scope cost savings Oliver Williamson: the efficiency defense

International Automotive Production Networks: How the web comes together. serie. Leticia Blázquez and Belén González-Díaz WP-EC

Tuition Fee Loan application notes

Experiment 6: Friction

Vasicek s Model of Distribution of Losses in a Large, Homogeneous Portfolio

How To Set Up A Network For Your Business

Causal, Explanatory Forecasting. Analysis. Regression Analysis. Simple Linear Regression. Which is Independent? Forecasting

Small Businesses Decisions to Offer Health Insurance to Employees

Transcription:

Crdff Economcs Workng Ppers Workng Pper No. E204/4 Reforms, Incentves nd Bnkng Sector Productvty: A Cse of Nepl Kul B Luntel, Shekh Selm nd Pushkr Bjrchry August 204 Crdff Busness School Aberconwy Buldng Colum Drve Crdff CF0 EU Unted Kngdom t: +44 (0)29 2087 4000 f: +44 (0)29 2087 449 busness.crdff.c.uk Ths workng pper s produced for dscusson purpose only. These workng ppers re expected to be publshedn due course, n revsed form, nd should not be quoted or cted wthout the uthor s wrtten permsson. Crdff Economcs Workng Ppers re vlble onlne from: econppers.repec.org/pper/cdfwpper/ nd busness.crdff.c.uk/reserch/cdemc-sectons/economcs/workng-ppers Enqures: EconWP@crdff.c.uk

Reforms, Incentves nd Bnkng Sector Productvty: A Cse of Nepl Kul B Luntel *# Crdff Unversty LuntelK@crdff.c.uk Shekh Selm # Unversty of Westmnster selms@westmnster.c.uk Pushkr Bjrchry # Trbhuvn Unversty pushkrb2000@gml.com Abstrct: We model bnks s proft-cum-utlty mxmzng frms nd study, nter l, bnkers ncentves (optml effort) nd ncentve drven productvty followng deregultons. Our model puts to test pnel of Neplese commercl bnks whch went through deep fnncl reforms n the recent pst. We fnd tht () bnkers efforts nd productvty hve notbly mproved n Nepl, () bnkers efforts sgnfcntly expln the bnkng sector s productvty, () the proporton of nonperformng lons hs consderbly declned, nd (v) bnkng servces hve become costly, lthough the bnk spred hs modertely declned. Our pproch s dfferent from the wdely used dt envelopment nlyss (DEA) of bnk productvty, hence complements the lterture. It lso nforms the current polcy debte n Nepl where the Centrl Bnk s seen to be gered towrds regultng the fnncl system nd mcro-mngng the bnkng nsttutons. JEL Codes: G2, G28, O4, O5. Keywords: Reforms; ncentves; productvty; pnel ntegrton; contegrton; smulton. *Correspondng uthor. # We re thnkful for the constructve suggestons receved from the hgh rnkng offcls of the Centrl Bnk nd Commercl Bnks of Nepl t the Neplese Bnkers Assocton semnr, Kthmndu, December 20. We re lso thnkful to the semnr prtcpnts t Crdff Busness School, Southmpton Mngement School, Unversdd de Alcnte nd those t the CFE (20) conference, London. The usul dsclmer pples.

Reforms, Incentves nd Bnkng Sector Productvty: A Cse of Nepl. Introducton The world hs seen sustned fnncl lberlzton, ncresng prvtzton nd grdul loosenng of cptl controls snce the md-990s. The economc thnkng behnd ll ths s tht the fnncl enttes, functonng under lberlzed fnncl regmes, operte t hgher levels of effcency nd productvty. Productvty mprovements my ensue from dfferent sources yet the noton tht reforms-led prvte.e. the ndvdul nsttuton s ncentve s vtl n enhncng productvty nd growth s fundmentl one. Put dfferently, deregulted fnncl system s vewed s ncentvzng nsttutons (n ths nstnce bnks) for hgher levels of effort, productvty nd proftblty. A lrge body of lterture (Fre et l., 994; Humphrey nd Pulley, 997; Wheelock nd Wlson, 999; Mukherjee et l., 200; Ihsn nd Hssn, 200; Trtroglu, 2005; Psours, 2008; Dels et l., 20; to nme but few) exmnes bnk effcency nd productvty followng reforms nd regultory chnges. They re pnel s well s country-specfc studes whch mostly employ non-prmetrc dt envelopment nlyss (DEA) to compute vrous effcency decompostons techncl effcency, scle effcency, effcency chnge (ctchng up or fllng behnd) nd productvty growth. 2 Whlst vluble, DEA nlyss s mesure of reltve effcency wth respect to the smple t hnd, one cnnot be confdent tht the bnks on the effcent fronter re ndeed opertng optmlly; nsted, they my merely represent the best prctces contned n the dt. Whether the best prctces re ndeed the optml ones re n entrely dfferent mtter. Furthermore, Wheelock nd Wlson (999; p 2) pont out tht effcency evluton through DEA could gve msledng pcture, especlly when the bnkng sector s gong through mjor chnges. Ths pper ms to contrbute to the lterture by nlysng, mong other thngs, bnkers optml efforts (ncentve) nd ncentve drven productvty followng deregultons nd reforms. Our pproch dffers from DEA n tht we model bnks s proft-cum-utlty mxmzng frms. We drectly model bnkers optml level of productvty rther thn reltve productvty, s s done under DEA. Reforms nd lberlzton vl opportuntes nd ncentvze bnks to optmze. It s ntcpted tht bnkers wll ncrese ther efforts nd productvty. 2

Our theoretcl model combnes bnks output technology wth ther optmzng behvour. Bnks techncl producton functon s Cobb-Dougls whch s stndrd n the lterture (Clrk, 984, 988; Humphrey, 99). We derve bnks nsttutonl producton functon, whch embeds bnks proft-cum-utlty mxmzng optml levels of effort, thereby cpturng bnkers optml response to nsttutonl chnges. In ths setup, bnkng sector productvty becomes endogenous to bnkers optml level of effort, reltve nput-output prces nd some techncl prmeters. Often, bnk proftblty surges followng lberlzton but ths surge could lso be relsed through ncresed volume (quntty) of bnk ctvtes.e. ncresed volumes of deposts nd credts nd bnk spred (cost of bnk servces) rther thn through the productvty gns. Therefore, the effects of fnncl deregulton nd reforms on the cost of bnk servces nd bnk productvty remn mportnt publc polcy ssues. Our nlytcl model s put to exmne Neplese fnncl lberlzton nd reforms. Nepl s one of the lest developed nd poorest countres of the world whch hs wtnessed Most nsurgency nd the overthrow of the Monrchy n 2008. Nepl embrked on deep nd fr rechng bnkng nd fnncl sector reforms through two epsodes of lberlzton mplemented durng 986-89 nd 992-994. The ltter epsode, n prtculr, opened up the country s fnncl sector nd brought bout profound structurl chnges to t. Although fnncl reforms were concluded n 994, the ctvtes were lcklustre untl fter 2000 due to the lunch of rmed nsurgency (People s Wr) by the Mosts n 996. However, snce 2002 Nepl s fnncl system hs been through fundmentl chnges (see Secton ). In our vew, Neplese bnkng sector mkes n nterestng test cse nd ts scrutny would provde mportnt polcy nsghts s to whether reforms hve produced ntcpted productvty mprovements n Nepl s bnkng sector. Ths nlyss s lso useful n nformng the current polcy debte n Nepl where the Centrl Bnk s wdely perceved s ntrudng the free functonng of fnncl mrkets through new regultons nd ttemptng to retrct from lberl polces n order to mcro-mnge the bnkng nsttutons. We employ cuttng-edge econometrc methods n our emprcl estmtons. To prevew our mn results, bnkers optml level of effort, optml bnk productvty nd bnk proftblty hve consderbly mproved n Nepl followng fnncl lberlzton. The rto of non-performng to totl lon hs consderbly

declned from 24.6% n 2004 to.4% n 202. Forml tests show tht bnkers efforts sgnfcntly expln bnk productvty. Evdence lso suggests tht n recent yers the bnk spred hs slghtly reduced, ndctng compettve pressure, yet bnkng servces hve become more costly. On the whole, fnncl reforms nd lberlzton hve been frutful experence n Nepl. The rest of the pper s orgnzed s follows. We present our nlytcl model n the followng secton; Secton brefly outlnes the fnncl regmes of Nepl nd rgues why Nepl s n nterestng test cse; econometrc specfcton nd dt re dscussed n Secton 4; emprcl methodologes re dscussed n Secton 5; emprcl results re presented n Secton 6; clbrtons nd smulton of optml effort nd productvty re dscussed n Secton 7; nd Secton 8 concludes the pper. 2. Model Fnncl lberlzton, mong other thngs, frees prces. Interest (depost nd lendng) rtes, bnkers wges, CEOs py nd other ncentves such s bonuses, re compettvely determned but there re lwys entry nd ext restrctons n the bnkng ndustry. These restrctons re mntned by the Centrl Bnk whch my be motvted by ts concerns over fnncl frglty nd/or some notonl optml sze of the bnkng ndustry n the economy. Restrctons to entry nd ext hve mportnt mplctons becuse they llow bnks to ern postve economc profts even n the long run. We construct prtl equlbrum model where representtve bnker, followng lberlzton, opertes wth compettve envronment nd mnmzes ts cost functon to cheve the mxmum fesble level of proft. 4 The representtve bnker s decson mkng unt (DMU) nd hs full flexblty nd freedom n decson mkng. Ths cn be vewed s the CEO of lrge bnk mkng decsons whch re fully mplemented by hs stff. In order to model the ncentve drven productvty, we need to specfy bnks producton functon. There re lterntve wys to mesure bnk nputs nd output n the lterture. Promnent ones re the producton pproch (Berger nd Humphrey, 992) nd the ntermedton pproch (Sely nd Lndley, 997; Aly et l., 990; Dels et l., 20). As n Wheelock nd Wlson (999), mutully exclusve dstncton between nputs nd output s vtl for modellng productvty hence we follow the ntermedton pproch. We specfy tht bnks use three nputs, nmely, 4

lbour (bnk stff, N), totl fxed sset (F), nd totl deposts (D) to produce ther output (Q). Lbour s mesured by the number of full-tme equvlent bnk employees; totl fxed sset s the book vlue of premses nd other fxed ssets, whch s equvlent to physcl cptl stock; nd deposts re the totl depost lbltes. Bnk output s mesured by totl credts nd nvestment. 5 The techncl constnt returns to scle (CRTS) Cobb-Dougls producton functon for bnk output s: Q Where 0, 0 2 N F D () 0 ; [0,],,2, re the shre prmeters such tht. N s the ncentve (effort) ugmented lbour (N) nd 0 cptures the bnkng technology. The prmeter denotes the level of effort or brodly defned ncentve level of typcl (representtve) bnker whch my chnge n response to chnges n polces nd nsttutons. The de s tht followng lberlzton bnks become more ncentvzed nd ncrese ther level of effort (qulty of work). Ths s precsely the rson d être of fnncl lberlzton nd reforms. Such efforts my nclude the bnker s wllngness to embrk on new types of lendng, nvestments nd servces whch mght hve been restrcted nd/or brred htherto. It s plusble to thnk tht the qulty of N nd F my lso mprove due to the hrng of more qulfed stff, computerzton nd the openng up of new brnches n more strtegc loctons. We ssume the qulty of nputs nd the level of effort to be postvely correlted. In other words, bnks nvest n nputs qulty n order to ncrese ther effort level (or mke ther effort more effectve). Incresed bnker s effort ( ) n deregulted envronment s lkely to mpct on bnks levels of rsk. The generl percepton s tht deregulton ncreses bnks rsks. However, conceptully the level of rsk could go ether wy n ggressve lendng by the bnker my ncrese the level of rsk wheres prudent lendng my do just the opposte. Snce our focus s on the outcome of lberlzton nd deregultons.e. whether reforms ncentvzed bnkers nd ncresed bnkng sector productvty the nlyss s essentlly n ex post one. Hence, we cn convenently sdestep the ssue of uncertnty nd cpture the level of bnk rsk exposure through the rto of performng to totl lons ( ). The hgher the the lower tends to be the rsk exposure nd vce vers. Gven tht the producton functon 5

() s homogeneous of degree one, ncluson of smply scles bnk productvty. Any ncrese n performng (non-performng) lons scles up (down) the level of totl fctor productvty of bnks nd vce vers; t does not lter our nlytcl results whch we show n the ppendx. The producton functon n per-bnker terms cn be expressed s: q 0 2 f d (2) Where q, f, nd d represent output, totl fxed ssets nd totl depost per bnker. The bnker chooses the level of nputs such tht the totl cost TC of rentng these nputs (.e. the opportunty cost of ownng these fctors) s mnmzed. Let us denote the unt cost of these three nputs by,,2,. The bnker chooses the nput set ( N, F nd D ) n order to mnmze: TC w w N w F w () 2 D subject to (). The consoldted frst order condton ssocted wth ths cost mnmzton problem s: N w F w D w Q 2 Q 2 Q Equton (4) s the stndrd result n cost mnmzton whch sttes tht the rto of mrgnl cost to mrgnl revenue or ts recprocl should be the sme cross ll the nputs employed. Substtutng (4) nto () nd mposng the CRTS condton, wth some lgebrc mnpulton, the optml totl cost functon s gven by: where * TC w Q (5) 0 2 2, nd (4) * TC s the mnmum vlue functon of TC n () subject to () whch depends on unt nput costs, the level of output nd shre prmeters. The optmum cost of producton per-bnker s therefore: * tc w q { } (6) Let p denote the mrket clerng prce per unt of credt. Then, the totl revenue per bnker s: tr pq (7) 6

Input nd output prces tht bnks fce tend to dffer nd they re lkely to chnge dfferently followng lberlzton. We cpture ths by w p, the rto of the observed weghted nput to output prces. From (6) nd (7) nd utlzng, the proft per bnker s gven by: p q (8) The bnker lkes proft but dslkes effort nd s mndful of the trde-off between proft nd the utlty cost of effort. The bnkers utlty from workng n the bnk s defned over profts nd effort levels s: 6 U Where prmeter 0,, (9) s the elstcty of substtuton between proft nd effort. The mrgnl utlty of proft s constnt but the mrgnl dsutlty of effort s ncresng wth the hgher level of effort. The reltve rsk verson equvlent coeffcent of ths utlty functon s gven by. The s the dsutlty prmeter; ts vlue ensures tht the utlty functon s jontly strctly qus-concve. The representtve bnker chooses effort level,, n order to mxmze (9) subject to (8) nd (2). The representtve bnker s utlty mxmzton problem s: mx U, s. t. p q 0 2 q f d The soluton to ths problem gves us the optml level of effort s: * p 2 d f (0) 0 The vrble * s the optml level of effort of representtve bnker followng lberlzton. It depends on nput nd output prces, substtuton prmeter, techncl prmeters nd deposts nd fxed ssets per bnker. When the bnkng sector goes through reforms, the reltve prce of nput nd output.e. the spred chnges, whch ffects bnkers optml effort (ncentve) level. Substtutng the optml level of effort (0) nto the techncl producton functon (), we get: Q BN F D 2 () 7

nd the shre prmeters re 2 2 In lne wth McMlln et l. (989) we lbel () s the nsttutonl producton functon of bnk whch s dfferent from the techncl producton functon (). Wheres equton () s the techncl reltonshps between nputs nd output, equton () ddtonlly cptures bnkers optml responses (efforts) to regultory chnges. Although s re emprclly dfferent from s nonetheless 8 (2) () (4), whch contnues to preserve the CRTS ssumpton. The prmeters of the techncl producton functon,,,2, nd nsttutonl producton functon,,,2, re relted through the substtuton prmeter. 7 The techncl producton functon () contns n unobservble nput, N, wheres the nsttutonl producton functon () s defned over ll observble nputs (N, F nd D). The effort prmeter s now embedded nto the optml productvty prmeter, B, whch cptures the bnkers optml response to reforms nd lberlzton. The bnkng sector s optml productvty prmeter, B, s gven by: B 0 p (5) In equton (5) the bnkers ncentve drven productvty ( B term: Bnc p (6) nc ) s cptured by the B nc s drectly ffected by the reform-nduced chnges n the nput nd output prces tht shpe the ncentve structure of the bnkng sector. The rest of the expresson on the RHS of equton (5) s constnt term whch only hs scle effect on B. Emprclly, ths sclng effect s close to unty (see below). If the bnkng sector productvty mproves followng lberlzton, then we expect n evdent postve trend n both B nc nd B. nc

. Fnncl Regmes nd Reforms n Nepl Nepl s one of the poorest countres of the world wth per cpt ncome of current US$ 69. Nepl s lndlocked nd sndwched between two gnts of As, vz. Chn nd Ind wth over bllon popultons ech. Nepl hs populton of 26.40 mllon. 8 Nepl hs bnkng hstory of over three-qurters of century the country s frst ever commercl bnk ws estblshed n 97 followed by the estblshment of the Centrl Bnk n 956. However, the fnncl sector ws under the frm grp of the uthortes untl the reforms tht concluded n 994. Only two commercl bnks nd one development bnk operted untl 984 nd the fnncl sector ws lrgely dormnt. The ppontment of government bureucrts, often retred ones, s the bnks CEOs ws the norm. Bnkng nd fnncl sector polces were domnted by soclst bnkng phlosophy, smlrly to those n Ind (Burgess nd Pnde, 2005). 9 Nepl Rstr Bnk (the Centrl Bnk of Nepl, henceforth NRB) operted hghly controlled regme of nterest rte mngement: there were bout 20 controlled bnk rtes dfferentted between sectors, use of funds nd types of collterls (NRB, 996; p 50). The term structures of nterest rtes were fully controlled. A lqudty requrement of t lest 25% comprsng mnmum of 5% of totl depost n government securtes nd further 20% of other lqud ssets ncludng reserves t the Centrl Bnk ws n operton. Commercl bnks were brred from tkng foregn currency deposts. A regme of drected credt progrmmes exsted whch mde t mndtory for bnks to chnnel s hgh s 25% of ther totl lendng to the Stte-defned Prorty Sectors, encompssng grculture, cottge ndustres, exports etc. Interest rtes on Prorty Sector lendng were lwys set t low levels nd commercl bnks were penlsed f they dd not meet the trget of 25%. Nepl ntted n mportnt step towrds fnncl openness n984. Foregn bnks, for the frst tme ever, were llowed to open jont venture bnks (n jont nvestment wth Neplese nvestors) n the country. Ths led to the estblshment of three foregn jont venture bnks under foregn mngement Nepl Arb Bnk Lmted, Nepl Indo Swss Bnk, Nepl ANZ Grndlys Bnk mkng totl of fve commercl bnks. They were the sole commercl bnks n operton untl 990. The openness of 984 ws followed by the frst phse of fnncl lberlzton (986-989) whch mnly focussed on nterest rte lberlzton. Intlly, NRB relnqushed the mcro mngement of nterest rtes; t only set the mnmum depost 9

rte. Bnks were lso gven freedom to set ther lendng rtes, subject to n upper lmt of 5% on the prorty sector lendng only. NRB completely deregulted ts nterest rte polcy n August 989 bnks nd fnncl nsttutons were gven free hnd n settng ther depost nd lendng rtes t ther own dscreton. The sttutory lqudty rto nd credt celngs were bolshed nd the sle of government securtes through open mrket opertons ws ntted. The frst phse of lberlzton ws qute mportnt n settng nterest rtes free to mrket forces, but the controls on foregn exchnge nd entry nto nd ext from the fnncl sector were tghtly mntned. The second phse of fnncl lberlzton, whch strted n 992, focussed on foregn exchnge lberlzton nd further openng up of the fnncl sector. Neplese currency ws mde fully convertble nto current ccounts n 99 nd mesures of cptl ccount lberlzton were dopted. Exporters could retn 00% of ther export ernngs n foregn currences nd mntn bnk ccounts n convertble currences. Commercl bnks were lso uthorzed to ssue credt n foregn currences; foregn nvestors could exptrte 00% proft to ther hbtt. The prvte sector could enter nto the bnkng nd fnncl sector wth ese nd wth or wthout foregn prtcpton. Although the second phse of lberlzton concluded n 994, whch fully opened up the fnncl sector, these polcy reforms remned lrgely dormnt there ws no zest n fnncl sector ctvtes untl fter 2000. Ths ws due to the poltcl nstblty trggered by the lunch of Mosts nsurgency n 996. However, post 2000 Mosts entered nto dlogue wth poltcl prtes whch sw the end of nsurgency n 2005. Followng the prospects of the end of Most nsurgency nd the ensung pece nd securty, the second phse of lberlston begn to tke effect snce 2002. 0 Ths quckly led to deep structurl chnges nd restructurng of the Neplese fnncl sector. The number of commercl bnks more thn doubled from n 2000 to n 20. The number of development bnks hs reched 88 from only seven n 2000. Furthermore, whole host of new types of fnncl nsttutons hve prolferted whch ether dd not exst or hd no sgnfcnt presence pre-994 reform. They nclude 69 Fnnce Compnes, 24 Mcrofnnce Development Bnks, 6 Svngs nd Credt Co-opertves nd 6 NGOs (fnncl ntermedres). The old nd lrge bnks lso went through deep restructurng. Nepl Bnk Lmted nd Rstry Bnjy Bnk, the two oldest nd lrgest 0

commercl bnks of the country, respectvely, hd s hgh s 56% nd 60% of ther totl lon portfolo clssed s non-performng n 2002. Both bnks hd reduced ther non-performng lon to round 6% by 202. Gven the scle of structurl trnsformton nd the restructurng of the bnkng sector followng the lst epsode of lberlzton, Nepl mkes n nterestng test cse for the fnncl reforms-led productvty growth n the bnkng sector nd we exmne ths through our nlytcl model presented n secton 2. 4. Econometrc Specfcton nd Dt The nlytcl model presented n Secton 2 derves the optml level of * ncentvzed effort ( ) of bnker whch s embedded n the nsttutonl producton functon (). In order to compute bnkers ncentvzed optml productvty, we need to estmte the structurl prmeters (, 2 nd ) of the nsttutonl producton functon. The log-lnerzed uxlry regresson of the nsttutonl producton functon, for pnel of bnks, tkes the followng form: log Q log N log F log D e (7) t t t 2 t t t (=,,M; nd t=,,t). Specfcton (7) s fxed effects pnel model. The subscrpts nd t denote the cross-sectonl nd tme seres dmensons, respectvely; cptures the bnk-specfc fxed effects nd t cptures the tme effects. Snce the regresson s specfed n logrthms, the prmeters re elstctes. Equton (7) specfes prmeters s bnk (pnel unt) specfc. In the estmton we llow both for the heterogenety (bnkspecfc) nd the homogenety (ndustry-wde) of prmeters cross pnel unts. All prmeters re expected to resume postve sgns pror nd one would expect the pont estmte (elstcty) of totl depost lbltes to be by fr the lrgest n bnk s producton functon. We hve collected dt on Neplese ndvdul commercl bnks totl deposts (D), totl lons nd dvnces (L), nvestments (I), fxed ssets (F), nterest expenses on deposts (RE), nterest ncome (RY), bnk stff (N), stff expenses (NE), other opertng expenses (OE) nd opertng proft ( ). As of 20, commercl bnks re n operton n Nepl but 4 of them re new: they cme nto operton n 2007 or lter. These 4 new bnks, due to ther very short dt length, could not be

consdered for ny credble econometrc nlyss. Furthermore, most of these new entrnts hve yet to consoldte ther bnkng ctvtes. Of the remnng 7 bnks, we hve obtned complete nd consstent qurterly dt for 2 bnks coverng smple perod of yers 2002() to 202(). 2 The choce of ths smple perod s delberte to focus on the most ntense perod of bnkng ctvtes followng lberlzton. As stted bove, polcy reforms remned lrgely dormnt untl the cesston of Mosts nsurgency tht begn n 2002 nd the nsurgency ended n 2005 hence the choce of the smple. Our dt set conssts of n unblnced pnel of 2 bnks wth 420 qurterly observtons. They re the mjor commercl bnks of Nepl ccountng for bout 70% of the bnkng ctvtes of the country. Anlyss of ths smple of bnks s deemed suffcent to dscern whether reforms nd lberlzton hve ncentvzed bnkers n Nepl nd whether the bnkng sector productvty hs mproved. All dt seres hve been drectly obtned from the offce of the Governor of NRB. The relevnt nomnl vrbles re deflted by CPI s the defltor. 4 5. Emprcl Methodology Mcro-pnel dt of ths nture re wdely reported to be non-sttonry (unt root) processes (see, mong others, Luntel et l., 2008) requrng n pplcton of non-sttonry pnel dt econometrcs n estmtng the prmeters of (7). Pnel unt root nd pnel contegrton tests re shown to hve better power propertes thn the tme seres tests n smll or moderte smples. A number of pnel unt root tests re proposed n the lterture whch cn be summrzed s the frst nd second generton tests. The former ssume crosssectonl ndependence prckly ssue n mcro pnel dt whle the ltter llow for cross-sectonl dependence. The frequently ppled frst generton pnel unt root tests n the emprcl lterture nclude those of Im, Pesrn nd Sn (200; herefter IPS), Fsher-ADF (Mddl nd Wu, 999) nd Hdr (2000). The IPS test tests the null of unt root for ech cross-sectonl unt gnst the lterntve tht only frcton of cross-sectonl unts my contn unt root. Ths test does not mntn sttonrty cross ll groups under the lterntve hypothess. Further, t lso llows for the heterogenety of persstence, dynmcs nd error vrnce cross groups. 2

The Fsher-ADF test employs the p-vlues of unt root test. Under the null of unt root for ll M (cross-sectonl) unts, the quntty: M log( ) s symptotclly 2 2M ; where s the p-vlue of the unt root test on the th seres of the th pnel unt. Hdr s test tests the null of sttonrty gnst the lterntve of unt root; common persstence prmeter s ssumed cross ll cross-sectonl unts. Hdr lso derves utocorrelton nd heteroskedstcty consstent LM tests under the null of sttonrty. Hlouskov nd Wgner (2006), however, wrn tht Hdr s tests suffer from sze dstorton n the presence of utocorreltons. The second generton tests re reltvely new nd re gnng momentum n emprcl pplctons for obvous resons. Gengenbch et l.(200) show tht the cross-sectonlly ugmented IPS (CIPS) test (Pesrn, 2007) s one of the powerful second generton pnel unt root tests. Ths test ccounts for both cross-sectonl dependence nd resdul serl correlton whle testng for the null of unt root. For the ske of robustness, we employ IPS, Fsher-ADF, Hdr nd the truncted CIPS tests on ech of the dt seres of our pnel. Pedron (999) nd Ko (999), mong others, propose pnel contegrton tests to explore f non-sttonry pnel dt form lner contegrtng (long-run equlbrtng) reltonshp. They re resdul-bsed tests of contegrton extensons of the tme seres tests of Engle nd Grnger (987) on pnel settngs. Pedron (bd.) proposes seven tests of pnel contegrton four of them re wthn-dmenson tests tht ssume homogeneous contegrtng vectors cross pnel unts nd the remnng three re between-dmenson tests (referred to s Group Men Sttstcs), whch llow for heterogeneous contegrtng vectors cross pnel unts. The between-dmenson estmtors exhbt lower sze dstortons thn the wthn-dmenson estmtors nd the group t-sttstc s shown to be the most powerful one mongst the three betweendmenson pnel contegrton tests (Pedron, 2004). The Ko (999) test s smlr to Pedron s tests except tht Ko llows for heterogeneous ntercepts but ssumes homogeneous slope prmeters cross pnel unts. We report rnge of contegrton tests proposed by Pedron (999) nd Ko (999) so tht we could rech robust concluson on the contegrtng reltonshp vs-à-vs our nsttutonl producton functon.

The OLS level regressons, employed to test contegrton n the pnel, re not nformtve of the sgnfcnce or otherwse of the contegrtng vectors becuse of the well-known nference problems (cf. Engle nd Grnger, 987). Therefore, we estmte the contegrtng prmeters through Fully Modfed OLS (FMOLS; Phllps nd Hnsen, 990) nd Dynmc OLS (DOLS; Stock nd Wtson, 99; Ko et l., 999). 6. Emprcl Results Results of pnel unt root tests re reported n Tble. The frst three columns pertn to the frst generton of pnel unt root tests. The IPS nd ADF- Fsher tests do not reject the null of unt root for ny of the level seres n the pnel. Hdr s test decsvely rejects the null of level sttonrty. Both types of frst generton tests (those testng the null of unt root nd the null of sttonrty) revel tht our pnel dt s non-sttonry. Ths s further confrmed by the CIPS second generton test whch ccounts for cross-sectonl dependence. CIPS tests cnnot reject the null of unt root for ny of the dt seres n our pnel. The results n Tble re bsed on the most generl specfctons tht nclude cross-secton-specfc ntercepts nd lner trends. All ndvdul seres n the pnel re found to be frstdfference sttonry, sgnfyng tht our pnel dt seres re unt root processes. 5 Tble bout here Tble 2 reports the results of pnel contegrton tests on bnkers nsttutonl producton functon (). Both the between-dmenson nd the wthn-dmenson tests proposed by Pedron (999) re reported. These tests re performed under two determnstc settngs ncludng: () bnk-specfc constnt only, nd () bnk-specfc constnts nd lner tme trend. We lso report the pnel contegrton tests proposed by Ko (999) for the ske of robustness. We ttch more mportnce to the betweendmenson tests nd, prtculrly, the t Pedron power propertes. test Tble 2 bout here 4, whch s shown to hve better The null of non-contegrton of bnkers log lnersed nsttutonl producton functon (7) s decsvely rejected by ll the tests reported n Tble 2. The precson of these tests s very hgh nd the results re robust to dfferent test methods tht vry consderbly n ther underlyng ssumptons. Overll, there s strong

emprcl support for the bnkers nsttutonl producton functon s long-run equlbrum reltonshp. Estmtes of the contegrtng prmeters (vectors) re reported n Tble. Results show tht two covrtes of nsttutonl producton functon, nmely, the bnk stff nd the totl depost lbltes pper postvely sgned nd hghly sgnfcnt cross ll specfctons, whch s consstent wth pror expecttons. The stock of totl fxed ssets, however, shows mxed results. It ppers postve nd sttstclly sgnfcnt under pooled (wthn-dmenson) estmtors but nsgnfcnt under grouped (between- dmensons) estmtors. The nsgnfcnce of totl fxed ssets s somewht surprsng but ths my be prtly explned by the reltve constncy (the lck of suffcent vrton) of fxed ssets n these bnks. 6 Tble bout here One of the fundmentl ssumptons of our nlytcl model s tht the bnkers producton functon follows CRTS. We explctly test ths restrcton nd report the results n row. In no cse s the CRTS restrcton rejected by the dt. We re-estmte two (Grouped) specfctons by droppng the nsgnfcnt log F t vrble nd re-ssessng the CRTS ssumpton. Results show tht CRTS s mntned. On blnce, one would prefer the between-dmenson FMOLS estmtes of ndustry-wde prmeters becuse they llow shre prmeters to dffer cross ndvdul bnks. The wthn-dmenson (Pooled) estmtes tret the shre prmeters s beng the sme cross ll bnks. In vew of the sgnfcnce of ll three covrtes, we report smulton results bsed on the pooled FMOLS estmtes. However, the qulttve nture of our smulton results s robust, rrespectve of the set of prmeter used. It s mportnt to note tht lthough the CRTS s not rejected sttstclly, the sum of the pont estmtes of the wthn-dmenson estmtes under FMOLS mounts to.049 rther thn.0 but we need prmeters to sum exctly to unty for smultons. Snce the sum of these pont estmtes s 4.9% hgher thn unty, we scled down ll three prmeters by 4.9% ech nd tested whether ths restrcton s dt cceptble. Indeed, we fnd the scled down prmeters of 0.057, 2 0.027 nd 0.96, whch sum to unty, re dt cceptble test of these prmetrc restrctons s the contegrtng vector genertes p-vlue of 2 () 0.0 under the 5

null. We use these dt congruent prmeters whch pss CRTS restrctons nd sum to unty for smultons. 7 7. Bnkers Incentve nd Bnk Productvty In order to smulte the bnkers optml level of effort (0) nd the ncentvzed optml productvty (5), we need solutons for the prmeters of techncl producton functon () - 0,, 2, ; the elstcty of substtuton, ; the dsutlty prmeter, ; nd the seres of nput nd output prces w, w2, w nd p. We use the CRTS consstent estmtes of 0.057, 2 0.027 nd 0.96 s the structurl prmeters of our nsttutonl producton functon (). Gven the restrcton 0,, some terton revels tht for 0.40,0.55 the system converges, hence we use 0. 5. The prmetrc vlue of 0. 5 combned wth equtons (2), () nd (4) nd the pont estmtes of s provde solutons: 0.0, 2 0.025 nd 0.865. The remnng prmeters nd 0 re relted through the equlbrum condton 0 2 2. Snce ll denomntors re constnts, we set whch gves.586. 8 The dsutlty prmeter hs scle 0 * effect on the optml level of effort ( ) nd productvty ( B ). Smultons convenently converge for [0.09,0.4] hence we employ 0. 0. 9 The mrgnl cost of bnk stff ( w ) s proxed by the verge hourly wge rte of bnker n 2 smple bnks. Ech bnk employee s ssumed to work 40 hours per week nd there re 48 bnk workng weeks per yer gvng, on verge, 2 bnk workng weeks per qurter. 20 However, the smulted results remn robust to hourly wges bsed on 6-44 workng hours per week nd/or weeks per qurter. The unt cost (shdow prce) of totl fxed ssets ( w 2 ) for the th bnk s tken to be the depost weghted mrket (mrket for the th bnk s defned s ll the bnks n the smple except the th bnk) nterest (one yer fxed depost) rte. The unt cost of deposts ( w ) s the verge depost rte (totl nterest pyment on deposts/totl deposts) for ech bnk. The unt output prce ( p ) s computed s the rto of lon nterest ncome over totl bnk lon (.e. the verge unt prce of lon). Usng the bove prmeter vlues nd nput-output prces, we smulte, mong others, bnkers 6

optml level of effort, effort drven productvty, verge nput cost nd revenue per unt of bnk output, nd the spred for the bnkng ndustry. Fgure plots the bnkng sector s verge nput cost per unt of output, verge output prce nd the bnk spred expressed s proft per unt of output. The verge nput cost per unt of output s clculted s { w } from equton (6) nd the verge output prce ( p ) s s defned bove. These verge unt cost-prce mesures re per Neplese rupee of bnk output (.e. the rentl cost of nputs to produce one rupee of bnk output nd the prce of tht output). 2 Commercl bnks verge cost nd prce per unt of output hve chnged over the yers both hve ncresed. Plots ndcte tht the bnk spred (proft per unt of output) declned durng 200() - 2004(4); t then slghtly pcked up n 2005() nd remned t hgher level untl 2006(4); subsequently t nrrowed down lttle untl 200(4) nd gn slghtly opened up therefter. Our clcultons show n verge spred of.25 percentge ponts for the whole perod (200-20);. percentge ponts for 200-2008 nd.7 percentge ponts for 2009-20. Neplese commercl bnks spred ppers to hve nrrowed down n recent yers ndctng compettve pressure. One strkng feture s tht snce 2008(4) the bnk spred hs become smller thn the nput cost per unt of output. Ths suggests tht whle bnks hve mnged to hold on to ther spred by trnsmttng costs to ther customers, the bnkng servces n generl hve become consderbly more expensve n Nepl n recent yers. The overll cost of bnkng, mesured by the verge cost nd prce per unt of bnk output, shows postve upwrd trend n recent yers. Bnkers optml effort ndex, plotted n Fgure 2, shows some fluctutons, but the overll trend s clerly n upwrd one. Durng the ntl yers (2004-2006), the optml effort ndex shows shrp rse but t decelertes qute shrply for bref perod from 2007() - 2007(4). The effort ndex then recovers nd shows rpd rse snce 2009(). Overll, the optml level of bnkers effort ppers to hve ncresed by 4% by the end of 20 compred to ts level n 2004. Plots lso revel tht the bnks ctul proft ncreses shrply, pekng t 70% hgher n 200 thn ts 2004 level. The ndex of ctul output (lons nd nvestment) peks durng 2008-2009 t 46% hgher thn ts 2004 level. The bnk output ndex decelertes durng 2009() - 200() nd then shows sgns of slow recovery therefter. The proft ndex tkes dp n 200(). These recent declnes n output nd proft ndces re due to the slump n 7

Nepl s rel estte mrket whch begn n 2008 nd hs yet to turn ts corner. Despte some dents n proftblty, the optml effort ndex shows contnuous rse. The bnkng sector s totl optml ncentvzed productvty ( B ) s plotted n Fgure. We lso plot the unscled component of B, whch s B 8 nc, just to llustrte tht sclng relly does not mtter. In fct, the sclng fctor on the RHS of equton (5) resumes vlue of.005, whch mples tht the role of techncl prmeters n our mesure of ncentve (effort) drven optml productvty s vrtully nl (these prmeters re treted s fxed). The optml productvty shows cler postve trend durng 200() - 2006(4) wth shrp blp n 2006(); productvty seems to hve flttened therefter untl 2009() nd shows some movement from 200(). The shrp rse n bnkng sector productvty s credted to the Pece Accord of 2005 between the wr wgng Mosts nd the poltcl prtes of Nepl whch effectvely ended the People s Wr. Cesston of Mosts nsurgency reduced uncertnty whch led to shrp upturn n productvty. 22 Ths lso pved the wy to the frst ever electon of the Consttuent Assembly n Nepl. Although productvty growth hs been mnml snce 2006(4), nonetheless, t ncresed by lmost 2 percentge ponts by 2007 nd beyond, compred to ts 200 level. The dervton of ncentve drven optml productvty (B) trets techncl prmeters s constnts. 2 However, technology chnges over tme. It s, therefore, mportnt to estblsh wht proporton of the bnkng sector TFP (totl fctor productvty), mesured by Solow resduls usng the CRT consstent s, s ccounted for by the bnkers ncentve-drven optml productvty. Fgure 4 plots the proporton of TFP ccounted for by the ncentve-drven productvty (B). The plot shows tht the proporton of TFP ccounted for by ncentve-drven productvty s somewht osclltory, yet t shows postve trend. The proporton of TFP ccounted for by B ncresed from below 2% to bove 26% durng 200-20 the smple verge s 24.75%. A comprson of the optml effort ndex (n Fgure 2) nd optml bnk productvty (n Fgure ) revels tht durng the ntl yers (2004-2006) both bnkers efforts nd bnk productvty went hnd-n-hnd; they ncresed. However, durng 2007() - 2007(4), the effort ndex declnes rther shrply whch concdes wth some declne n bnk productvty. However, the effort ndex shows rpd rse snce 2009() whch does correspond to some productvty gns but the ssocton

ppers much weker thn those wtnessed durng 2004-2006. It s hrd to elucdte why ths pprently close ssocton between bnkers efforts nd bnk productvty wned post 2009(). One possble explnton, however, s tht fter some ntl yers of productvty push, bnkers effort my hve focussed on quntty (volume). Ths s borne out by the shrp rse n the volumes of deposts nd credts post 200(2) whch re plotted n Fgure 5. These shrp rses n deposts nd credts re t the bckdrop of fll n ther volumes durng 2008() - 2009(4). Fnlly, we formlly test f the bnkers effort sttstclly explns the ncentvzed productvty by regressng the ndex of optml totl productvty on the ndex of bnkers effort. Both OLS nd Instrumentl Vrbles estmtes revel tht bnkers effort sgnfcntly explns the optml bnk productvty. A 0% ncrese n bnkers effort ncreses the bnkng sectors optml totl productvty by roughly 0.6%. 24 8. Concluson A lrge body of lterture exmnes whether bnkng sector effcency nd productvty mproves followng fnncl deregulton nd reforms. The lterture mnly employs dt envelopment nlyss (DEA). However, DEA only mesures reltve effcency nd t s not cler whether bnks on the effcent fronter re ndeed opertng optmlly. Ths pper develops n nlytcl frmework tht tkes the study of deregulton (reforms) nd bnkng sector productvty beyond the relm of DEA. We model commercl bnks s proft-cum-utlty mxmzng frms nd nlytclly derve, nter l, bnkers optml level of effort nd optml productvty whch re of polcy relevnce followng lberlzton nd reforms. We pply our model to scrutnze pnel of Neplese commercl bnks nd evlute f deregultons nd reforms hve worked n Nepl. Nepl concluded her deep fnncl reforms n 994. However, due to Most nsurgency nd poltcl nstblty, these polcy reforms strted to tke effect snce 2002 only whch quckly nd deeply trnsformed the country s bnkng nd fnncl system. Our nlyss provdes mportnt nsghts on deregultons, reforms nd bnkng sector productvty n Nepl. The mn fndngs cn be summrzed s follows. Frst, we fnd tht fnncl lberlzton hs mde Neplese bnkers more ncentvzed (effort orented) evdence shows cler rse n the level of bnkers efforts followng lberlzton. Neplese bnkers optml level of effort hs ncresed consderbly (by 4% durng 2004-20) nd t s on the rse. Second, the 9

bnkng sector s ncentve-drven productvty rose mrkedly untl 2007 (by 2% durng 200-2007) but the productvty growth hs snce slowed down sgnfcntly. The ssocton between the optml levels of effort nd optml productvty seemed very close n erly yers but ppers to hve wekened snce 2008; t s hrd to expln why ths s so. However, prm fce evdence suggests tht fter the ntl yers of productvty push, bnkers pper to hve focussed on quntty (volume) rther thn qulty (productvty). Thrd, despte the weker ssocton n the ltter perod of smple, forml tests show tht the bnkers optml effort sgnfcntly explns the bnkng sector s optml productvty n Nepl. Our clcultons show tht ncentve-drven productvty ccounts for bout qurter (24.75%) of bnkng sector TFP (mesured by Solow resduls) n Nepl. The proporton of nonperformng lons to totl lons hs declned from 24.0% n 200 to less thn 4.0% n 202. Fnlly, we fnd tht Neplese bnks erned n verge bnk spred (proft per unt of bnk output) of.25 percentge ponts durng the smple perod but ths hs slghtly declned n recent yers (.7 percentge ponts) perhps reflectng the compettve pressure. However, downsde s tht the bnkng servces n Nepl hve become costly n recent yers the verge cost nd prce per unt of bnk output hs ncresed notbly. The ltter s, however, not unexpected, gven tht deregulton bolshes uthortes control on nterest rtes nd vrous concessonry lendng progrmmes. Overll, fnncl lberlzton nd reforms hve been good experence for Nepl, especlly from the perspectves of more ncentvzed (effortorented) bnkers, ncresed optml productvty nd hgher volume of deposts, credt nd bnk proftblty. In the lght of these fndngs, ny ttempt by the uthortes to retrct from lberlst polces my do more hrm thn good. 20

Tble : Results of Pnel Unt Root Tests IPS [W-Stt] ADF-Fsher [Ch-Squre] Hdr [Consstent Z- stt] CIPS lnq 0.898 0.74 0.000-0.246 lnn 0.94 0.90 0.000-0.64 lnf 0.26 0.40 0.000-0.60 lnd 0.542 0.69 0.000-0.7 For ll tests, except for the CIPS, P-vlues under the null re reported. For CPIS t- rtos re reported. The 5% crtcl vlue for the CIPS test for T/M (200/5) s - 2.25; where T s the smple sze nd M s the cross-secton unts. The null under ll test sttstcs, except tht of Hdr, s unt root; the ltter tests the null of sttonrty. The smple conssts of 420 unblnced pnel observtons. In ll tests, constnt nd lner tme trends re ncluded s exogenous vrbles. Gven the qurterly dt, the lg length s set to 4 for ll tests except the CIPS where second order lg s used. CIPS requres led nd lg ugmenttons. W-Stt s the stndrdzed t NT test of IPS. ADF Fsher tests re 2 (24)-dstrbuted. The Hdr test s computed usng Newey-West bndwdth selecton nd Brtlett kernel; heteroskedstcty-consstent Z-sttstcs re reported. CIPS s the cross-sectonlly ugmented IPS tests ( second generton test). The vrble mnemoncs re: lnq = log of rel totl lon nd nvestment (output mesure), lnn = log of number of bnk stff, lnf = log of totl fxed ssets n rel terms; lnd = log of totl deposts n rel terms. 2

Tble 2: Results of Pnel Contegrton Tests Pedron Between Dmenson (Group Men) & t Wthn Dmenson (Pooled) & test -4.85(0.000) -2.995(0.00) -6.46(0.000) -4.87(0.000) tpedron tkao test -.922(0.000) -2.72(0.000) -9.755(0.000) 2.89(0.000) test NA NA -9.8(0.000) NA Pedron test nd tpedron tkao test re pnel rho- nd ADF-sttstcs of Pedron (999); test s the ADFt -sttstc of Ko (999). Columns nd & ndcte the determnstc component n the model, nmely, constnts (fxed effects) only, nd constnts nd lner tme trends. Lg lengths re bsed on SIC nformton crter. Superscrpts denote sgnfcnce t % or better. Ko test sttstcs re wthndmenson test whch llows fxed effects only, hence NAs n three columns. t t 22

Tble : Estmted Contegrtng Prmeters log Q log N log F log D e () 0.46 (0.000) -0.0 2 t t t 2 t t t (0.67) - 0.888 (0.000) FMOLS DOLS Grouped Pooled Grouped Pooled 0.5 (0.005) 0.89 (0.000) 0.060 b (0.050) 0.028 b (0.09) 0.96 (0.000) 0.24 (0.002) -0.05 (0.240) - 0.800 (0.000) 0.6 (0.005) 0.848 (0.000) 0.060 b (0.04) 0.0 (0.00) 0.97 (0.000) 0.686 0.900 0.26 0.625 0.758 0.44 FMOLS s the fully Modfed OLS of Phllps nd Hnsen (990) s shown n Pnel settng by Pedron (200). DOLS s the Dynmc OLS s descrbed n Ko et l. (999). Gven ther sgnfcnce, both determnstc terms (bnk specfc constnts nd lner tme trends) re retned n the estmton. Fgures wthn prentheses (.) re p- vlues under the null. reports the P-vlues of the test of the null of CRTS,.e., 2 ; the test sttstc s 2 () dstrbuted. Under the between-dmenson specfctons log F t ppers nsgnfcnt; we lso report results excludng ths nsgnfcnt covrte. Superscrpts nd b denote sgnfcnce t % nd 5% or better. 2

Fgure : Averge nput cost, output prce nd spred (ndustry); Ps per Rupee of output. 0. 0.09 0.08 0.07 0.06 0.05 0.04 Output prce Averge nput prce Spred 0.0 0.02 0.0 0 200 200 2004 2004 2005 2005 2006 2006 2007 2007 2008 2008 2009 2009 200 0 200 20 20 Industry mples the bnkng sector, comprsed of 2 smple bnks. The verge nput cost per unt of output s clculted s { w } from equton (6). The verge output prce s the verge lon rte computed s the rto of lon nterest ncome over totl bnk lon. The verge nput cost, output prce nd spred (proft) re per unt of bnk output. In the vertcl xs 0. mens ten ps. One Neplese rupee conssts of 00 ps. 24

Fgure 2: Industry optml effort ndex (2004=00), ctul output ndex nd ctul proft Index. 80 70 60 50 40 0 20 Ind output Index Ind optml effort Index Ind Proft Index 0 00 90 80 2004 2004 2005 2006 2007 2007 2008 2009 200 200 20 Industry mples the bnkng sector comprsed of 2 smple bnks. Output nd proft ndex re ctul fgures for the bnkng sector. The optml effort ndex s the smulted seres, equton (0). 25

Fgure : Bnkng Sector s Optml Productvty: ( B ) nd ( B ) nc 0.99 0.97 0.95 0.9 0.9 Industry Bnc Industry B 0.89 0.87 0.85 200 2004 2005 2006 2007 2008 2009 200 20 202 The bnkng ndustry mples 2 smple bnks. The bnkng sector s optml productvty ( B ) s s n equton (5) nd unscled, B nc, s n equton (6). Plots re bsolute fgures not ndces. 26

Fgure 4: Proporton of TFP (mesured by Solow Resdul) explned by the ncentve-drven productvty (B). 0.27 0.265 0.26 0.255 0.25 0.245 0.24 0.25 0.2 0.225 0.22 200 2004 2005 2006 2007 2008 2009 200 20 The ndustry mples the bnkng sector comprsed of 2 smple bnks. The bnkng sector s Solow Resdul-bsed TFP s computed usng the prmeters of nsttutonl producton functon (equton ()). The ncentve drven optml totl productvty, B, s defned n equton (5). 27

Fgure 5: Totl Credts nd Deposts of Commercl Bnks, Mllons of Rupees (2004-202). 8000 7500 7000 6500 6000 Totl Credt Totl Deposts 5500 5000 4500 4000 2004 2005 2006 2007 2008 2009 200 20 The ndustry mples bnkng sector comprsed of 2 smple bnks. Plots re ctul totl deposts nd credts of 2 smple bnks. 28

Appendx: lons, (0, ] We cpture the ex post rskness of bnks by the rto of performng to totl ; where ( ) s the proporton of non-performng lons. Snce the producton functon () s homogeneous of degree one, t tkes the followng form: 2 0 ( ) Q N F D A Where, s before, 0, 0 ; [0,],,2, nd. A ht denotes tht the reltonshp ccounts for. A hgh proporton of non-performng lons, 0, scles down the totl fctor productvty, wheres hgh proporton of performng lons,, scles the totl fctor productvty up. When non-performng lons re zero, ; nd (A) collpses to equton (). The producton functon, (A), per bnker s: 2 0 (2 ) q f d A The cost mnmzton problem for the bnk remns unchnged. From (5), we hve: * TC w Q A where tc * 2 0 w 2 q (5 ). The optmum cost of producton per bnker s: (6A) The totl revenue nd proft per bnker re sme s n (7) nd (8), except tht now q q, nd. Bnkers utlty functon remns unchnged, nd the soluton to the optml level of effort s: * p 2 f d (0A) 0 Substtutng (0A) n (A), the nsttutonl producton functon of bnk s: 2 Q BN F D ( A ) The bnkng sector s optml productvty coeffcent, B, s gven by: B p (5 ) 0 A 29

It s evdent tht the ncluson of n ex post mesure of rsk ( ) does not chnge our nlytcl results, except the pure sclng effects. In Fgure A we plot the ndustry whch mproved qute rpdly n ntl yers nd hs stblsed snce 2007. The rto of performng lons hs sgnfcntly mproved n Nepl from 76% n 200 to lmost 97% n 202. Put dfferently, the rto of non-performng lons hs gone down from 24% to less thn 4%, whch s substntl mprovement n the helth of the bnkng sector. When the bnk rsk, mesured by the non-performng lon rto, goes down, both bnkers optml effort nd bnk productvty goes up nd t lso mproves ther net proft mrgn, [ p ]. Fgure A2 plots the productvty ndces computed from equtons (5) nd (5A). The lter plot shows shrp rse reflectng the mprovng proporton of performng lons. In Fgure A, bnks optml productvty (level) wth nd wthout s plotted. The ndex of optml productvty wth shows reltvely shrper trend gn reflectng the mprovements n the proporton of performng lons. 0

Fgure A: Rto of Performng to Totl Lons ( Industry) 0.95 0.9 0.85 0.8 0.75 0.7 200 200 2004 2005 2006 2006 2007 2008 2009 2009 200 20 202 The ndustry mples the bnkng sector s comprsed of 2 smple bnks.

Fgure A2: Optml Effort ( * ) wth nd wthout rsk ( ) 65 55 45 5 25 5 Effort Index (wth rsk) Effort Index (no rsk) 05 95 85 75 2004 2004 2005 2006 2007 2007 2008 2009 200 200 20 Rsk ( ) s mesured by the proporton of performng to totl bnk lons. The proporton of non-performng lons s ( ). We pply tme vryng n computng effort ndex wth rsk. 2

Fgure A: Optml Productvty (B) wth nd wthout rsk ( ).05 0.95 0.85 0.75 B (wth rsk) B (No rsk) 0.65 0.55 0.45 200 200 2004 2004 2005 2005 2006 2006 2007 2007 2008 2008 2009 2009 20 0 200 20 20 202 Plese refer to notes for Fgure A2.

References Aly, H. Y., R. Grbowsk, C. Psurk, nd N. Rngn (990) Techncl, scle nd lloctve effcences n US bnkng: An emprcl nvestgton, The Revew of Economcs nd Sttstcs, 72, 2-28. Berger, A.N., nd D.B. Humphrey (992) Mesurement nd Effcency Issues n Commercl Bnkng, n Z. Grlches (ed.), Output Mesurement n the Servce Sectors, Ntonl Bureu of Economc Reserch, Studes n Income nd Welth, Vol. 56, Unversty of Chcgo Press (Chcgo, IL), 245-279. Burgess, R. nd R. Pnde (2005) Do Rurl Bnks Mtter? Evdence from Indn Socl Bnkng Experment, Amercn Economc Revew, 95(), 780-795. Clrk, J. (984) Estmton of Economes of Scle n Bnkng Usng Generlzed Functonl Form, Journl of Money, Credt, nd Bnkng, 6, 5-68. Clrk, J. (988) Economes of Scle nd Scope t Depostory Fnncl Insttutons: A Revew of the Lterture, Federl Reserve Bnk of Knss Cty Economc Revew, 7, 6-. Dels, M. D., P. Molyneux nd F. Psours (20) Regultons nd Productvty Growth n Bnkng: Evdence from Trnston Economes, Journl of Money, Credt nd Bnkng, 4(4), 75-764. Engle, R. F nd C. J. Grnger (987) Co-ntegrton nd Error Correcton: Representton, Estmton, nd Testng, Econometrc, 55(2), 25-276. Fre, R., S. Grosskopf, M. Norrs nd Z. Zhng (994) Productvty Growth, Techncl Progress, nd Effcency Chnge n Industrlzed Countres, The Amercn Economc Revew, 84(), 66-8. Gengenbch, C. F. Plm nd J. Urbn (200) Pnel Unt Root Tests n the Presence of Cross-Sectonl Dependences: Comprson nd Implctons for Modellng, Econometrc Revews, 29(2), -45. Hdr, K. (2000) Testng for sttonrty n heterogeneous pnel dt, Econometrcs Journl (2), 48-6. Hlouskov, J. nd M. Wgner (2006) The performnce of pnel unt root nd sttonrty tests: results from lrge-scle smulton study. Econometrc Revews 25(), 85-6. Humphrey, D. B. (99) Productvty n Bnkng nd Effects from Deregulton Economc Revew, Federl Reserve Bnk of Rchmond, Mrch/Aprl: 6-28. Humphrey, D. B. nd L. B. Pulley (997) Bnk s responses to deregulton: profts, technology, nd effcency, Journl of Money, Credt nd Bnkng, 29, 7-9. 4

Ihsn, I. nd M. Hssn (200) Fnncl deregulton nd totl fctor productvty chnge: An emprcl study of Turksh commercl bnks, Journl of Bnkng & Fnnce, 27(8), 455-485. Im, K. S., Pesrn, H. nd Shn, Y. (200). Testng for unt roots n heterogeneous pnels, Journl of Econometrcs 5(), 5-74. Ko, C. (999) Spurous regresson nd resdul-bsed tests for contegrton n pnel dt, Journl of Econometrcs, 90(), -44. Ko, C., M-H. Chng nd B. Chen (999) Interntonl R&D Spllovers: An Applcton of Estmton nd Inference n Pnel Contegrton, Oxford Bulletn of Economcs nd Sttstcs (6), 69-7. Luntel, K. B., M. Khn, P. Arests nd K. Theodords (2008) Fnncl Structure nd Economc Growth, Journl of Development Economcs, 86(), 8-200. Mddl, G. nd S. Wu (999) A comprtve study of unt root tests wth pnel dt nd new smple test, Oxford Bulletn of Economcs nd Sttstcs 6(S), 6-652. McMlln, J., J. Whlley nd L. Zhu (989) The Impct of Chn s Economc Reforms on Agrculturl Productvty Growth, Journl of Poltcl Economy, 97 (August), 78-807. Mukherjee, K., S. C. Ry nd S. Mller (200) Productvty growth n lrge US commercl bnks: The ntl post-deregulton experence, Journl of Bnkng & Fnnce, 25(5), 9-99. Nepl Rstr Bnk, 40 Yers of Nepl Rstr Bnk, Kthmndu, Aprl 996. Psours, F. (2008) Interntonl evdence on the mpct of regultons nd supervson on bnks techncl effcency: n pplcton of two-stge dt envelopment nlyss, Revew of Quntttve Fnnce nd Accountng, 0(2), 87-22. Pedron, P. (999) Crtcl vlues for contegrton tests n heterogeneous pnels wth multple regressors, Oxford Bulletn of Economcs nd Sttstcs, 6(Specl Issue), 65-670. Pedron, P. (200) Purchsng power prty tests n contegrted pnels, The Revew of Economcs nd Sttstcs, 8(4), 727-7. Pedron, P. (2004) Pnel contegrton: symptotc nd fnte smple propertes of pooled tme seres tests wth n pplcton to the PPP hypothess, Econometrc Theory, 20(), 597-625. Pesrn, M. H. (2007) A smple pnel unt root test n the presence of cross-secton dependence, Journl of Appled Econometrcs, 22(2), 265-2. 5

Phllps, P. C. nd B. E. Hnsen (990) Sttstcl Inference n Instrumentl Vrbles Regresson wth I() Processes, Revew of Economc Studes, 57(), 99-25. Seley, C. W nd J. T. Lndley (997) Inputs, outputs nd theory of producton nd cost t depostory fnncl nsttutons, Journl of Fnnce, 2, 25-266. Stock, J. nd M. Wtson (99) A Smple Estmtor of Contegrtng Vectors n Hgher Order Integrted Systems, Econometrc, 55, 05-056. Trtroglu, K., K. Dnels nd E. Trtroglu (2005) Deregulton, Intensty of Competton, Industry Evoluton nd the Productvty Growth of US Commercl Bnks, Journl of Money, Credt nd Bnkng, 7(2), 9-60. Wheelock, D. C. nd P. W. Wlson, (999) Techncl Progress, Ineffcency, nd Productvty Chnge n U.S. Bnkng, 984-99, Journl of Money, Credt nd Bnkng, (2), 22-24. 6

For exmple, McMlln et l. (989) n nlysng Chnese grculturl reforms tht ncentvsed (rewrded) ndvdul frmers, concluded tht rewrdng ndvdul effort yelds lrge beneft (bd., p. 78). 2 Some of these studes subsequently employ prmetrc methods to model the productvty nd effcency mesures computed through DEA. To see ths, mgne scenro where the whole of the bnkng ndustry mkes n verge effcency gn of 5% but ths gn would look neffcent f the effcent fronter moves up by 0%. 4 The non-zero proft n the long-run rses the theoretcl possblty of bnks producng n nfnte mount of output. Whle we cknowledge ths theoretcl possblty we do not regrd t s beng of much prctcl relevnce. 5 Lck of dt prevented us from usng the off-blnce-sheet tems n our mesure of bnk output. However, our mesure of bnk output s consstent wth those of Dels et l. (20), mong others. 6 Utlty from proft could be seen s tngble (bonuses) nd ntngble (fme) rewrds whch tend to be postvely ssocted wth the level of profts. Gven tht our focus s on bnk productvty, we do not model the detls of ncentve pckges. 7 It s trvl tht from (2), () nd (4) tht:, 2 2 nd. 8 Dt on GDP per cpt n current US dollrs re for 20 nd obtned from the World Bnk t http://www.worldbnk.org/en/country/nepl. Populton fgures re from the 20 census, Nepl. 7

9 Under the socl bnkng phlosophy uthortes, mong other thngs, dctted tht commercl bnks expnd ther brnch networks cross countres n tndem wth ntonlly declred bnkng densty objectves of one bnk brnch per 0,000 hed of populton. 0 The precse yer 2002 s suggested by the semnr prtcpnts (hgh rnkng offcls of the Centrl Bnk, Commercl Bnks nd the Government nd of Nepl) t the Neplese Bnkers Assocton semnr, Kthmndu, December 20. We re grteful for ths suggeston. Fgures on the growth of fnncl nsttutons re tken from Bnkng nd Fnncl Sttstcs, md-july, 202, No. 58, NRB. 2 One of our smple bnks hs mnmum dt length of 7 yers 2006() to 202() s t ws opened n 2006. We re thnkful to the Governor of NRB, Dr Y R Khtwd, for the dt. The 2 commercl bnks covered re: Bnk of Kthmndu Ltd. (995), Everest Bnk Ltd. (994), Globl Bnk Ltd. (2006), Hmlyn Bnk Ltd. (99), Kumr Bnk Ltd. (200), Lxm Bnk Ltd. (2002), Mchhpuchhre Bnk Ltd. (2000), Nbl Bnk Ltd. (984), Nepl Bnk Ltd. (97), Nepl Investment Bnk Ltd. (986), Rstry Bnjy Bnk Ltd. (966) nd Stndrd Chrtered Bnk Nepl Ltd. (987). Fgures wthn prentheses (.) ndcte the dte of bnk operton. As stted n the text, they consttute the mjor commercl bnks of Nepl. 4 Dt on qurterly GDP defltor re not vlble n Nepl. 5 Results of frst dfference sttonrty re not reported, to conserve spce, but re vlble on request. 8

6 The fxed ssets of bnks tend to chnge slowly compred to bnk output, employment nd totl depost lbltes. 7 The prmeter estmtes of the lst column of DOLS results reported n Tble sum to.0. A reducton of.0% of ech prmeter to mke them sum to unty s lso not rejected by the test. The p-vlue of the test s 2 () 0.94. The Grouped prmeters under FMOLS nd DOLS sum to.006 nd 0.984, respectvely (columns whch delete the nsgnfcnt F t ). 8 Prmeter 0 s the constnt term of the techncl producton functon (equton ()). The estmtes of the constnt term of the nsttutonl producton functon rnge between 0.024 to 2.498 under dfferent specfctons (not reported n Tble ). Our smulton results re robust to vlues of 0 0 0. 9 Iterton revels tht for wde-rngng vlues of 0.0 0 the smulted vlues * of nd B remn frly robust. 20 The verge hourly wge rte ( w ) s clculted s follows. Frst, the qurterly verge wge bll for stff s computed by dvdng the totl qurterly wge bll by the totl number of stff. Then the qurterly verge wge bll s dvded by 40x2; where 40 represents the hours worked per week nd there re 2 workng weeks n qurter. 2 Neplese currency s known s Rupees nd one Rupee conssts of 00 Ps. 22 We re thnkful to the semnr prtcpnts t the Nepl Bnkers Assocton, Kthmndu, n explnng (reconclng) ths shrp upturn n bnkng sector productvty wth the Pece Accord of 2005. 2 A tme vryng pproch of estmton would llow the techncl prmeter to be tme dependent but dt constrnts preclude us from usng t. 9

24 The estmted regresson s: log B 4.44 0.054log * 0.005sldmy ; where B s the ndex of smulted totl optml productvty, * s the ndex of smulted optml effort nd sldmy s the slope dummy for 2006()-(2) to cpture the blp n productvty (see Fgure ). The constnt term nd the slope prmeters resume p- vlues of 0.000, 0.024 nd 0.000, respectvely, hence re sttstclly hghly sgnfcnt. The reported results pertn to the frst order resdul serl correlton (AR()) correcton. The R-br squre s 0.9 nd DW sttstc s.72. Results pper lmost dentcl when the IV estmtor s used. 40