Competition and Price Conduct by Bank Service Line. Wilko Bolt and David Humphrey 1. November 2016

Transcription

1 Competton and Prce Conduct by Bank Servce Lne by Wlko Bolt and Davd Humphrey November 06 In ths paper, we test the ablty of competton measures to dstngush between relatvely hgh and low prces n four bank servce lnes: busness and consumer loans plus savngs and small tme deposts. The HHI and H-Statstc often do not do better than random chance. A Mark-Up (an approxmate Lerner Index) or a fronter competton measure do better. The HHI s not predctve of realzed prce conduct and should not be used to assess or reject bank mergers. The challenge s how to use other measures to augment or replace the HHI n bank merger polcy. Keywords: Bank competton, Loans and deposts, fronter analyss. Introducton A Google Scholar search (Aprl, 06) found over 4,000 papers wth the words "HHI, banks", "Lerner Index, banks", or "H-Statstc, banks" n ther abstracts. Most of these studes concern U.S. or European banks and deal wth a sngle measure of competton. Few studes contrast ther chosen measure wth other well-known measures but, when they do, the contrast s wth the HHI. In ths study we go a step further and compare results of three standard competton measures (HHI, Lerner Index, and H-Statstc) along wth two others (a smple Mark-Up and a new fronter competton measure). Ths s done for four U.S. bank servce lnes coverng two loan products (busness and consumer loans) along wth two depost products Correspondng author: Davd Humphrey, Department of Fnance, Florda State Unversty, Tallahassee, FL USA; (850) ; emal: dhumphrey@cob.fsu.edu. Wlko Bolt, Economcs and Research Dvson, De Nederlandsche Bank, P.O. Box 98, 000 AB Amsterdam, The Netherlands, emal: w.bolt@dnb.nl. The vews expressed are those of the authors and do not necessarly represent the vews of De Nederlandsche Bank or the European System of Central Banks.

2 (savngs and small value tme deposts less than $00,000). We also assess the ablty of each competton measure to dstngush between banks wth relatvely hgh versus relatvely low loan rates or depost returns. Ths effort s not wthout dffculty. Revenue data are often avalable for general classes of loan products but not for savngs or tme deposts that bascally have no explct revenue. And whle some nformaton s avalable on depost costs, such as the rates pad, operatng costs are not allocated to any servce lnes. Except for the HHI, the data that are avalable by servce lne s at the bank level, not for each branch offce. Consequently, our competton measures reflect the value weghted average of all the locatons a bank operates n. Ths ranges from a sngle MSA, or a sngle state, or at tmes a regon composed of multple states for bllon dollar banks. Although the best ndcator of hgh loan rates and low depost rates would be the actual loan rate or depost return for each servce lne, ths would not control for the nfluence that nput cost and productvty may have on loan prce or depost returns. All but the HHI adjusts for cost dfferences across banks whle the fronter measure adjusts for productvty dfferences as well. As bankng n the U.S. s a relatvely easy entry/easy ext ndustry, our results could well dffer f a smlar analyss was appled to a dfferent ndustry. In what follows, our fve competton measures are presented n Secton wth estmatng equatons and other detals noted n the Appendx. In Secton we show how each competton measure s related to each quartle of bank servce lne prce or return. If a measure s good at dstngushng between hgh and low prces or returns, we would expect that loan rates would rse (or depost returns would fall) as a competton measure ndcates less compettve prce conduct. In some cases ths gves results that are the opposte of what we would expect. In The Communty Renvestment Act, however, can lmt the ext of branch offces n certan geographc areas.

3 Secton 4, we test statstcally the correspondence between what each measure s predctng for the prce conduct and the actual level of loan rates or depost returns. The HHI and H-Statstc do not fare well n these comparsons whle the Mark-Up (an approxmate Lerner Index) and a competton fronter measure often do better. We conclude n Secton 5 and offer some suggestons how ths nformaton may be helpful n dealng wth mportant bank merger decsons and n polcy revews before adoptng changes n the U.S. horzontal merger gudelnes (the most recent of whch was n 00).. Measures of Servce Lne Competton The 54 banks n our sample all have more than $ bllon n assets. Overall, bllon dollar banks account for almost 90 percent n total assets of the U.S. bankng ndustry. Except for all deposts as a total, data at the bank branch level does not exst. Our competton measures thus reflect the average across all branches of the varous markets each bank has at least one branch n. Ths can be local, or at most regonal, rather than natonal snce the medan bllon dollar bank has branches n only four MSAs and operates n only state out of 50. Even at the 99th percentle, the average bllon dollar bank has offces n only 6 states. Thus a bank's potental compettors are located only n the MSAs that a gven bank has a branch n.. HHI: Market Concentraton Measure The Herfndahl-Hrschman Index (HHI) s beleved to be predctve of reduced competton and possble hgher prces f mergers or acqustons occur where market concentraton exceeds a Varous screens were appled to elmnate shell banks, specal purpose banks, banks wth no loans, or no deposts, or no full tme employees, etc., or that contaned varables beyond fve standard devatons from the mean and are clearly unrepresentatve of the bankng ndustry we wsh to assess.

4 certan level. Importantly, and unlke other competton measures, a post-merger/acquston HHI can be determned pror to a merger and nferences drawn as to lkelhood of uncompettve behavor f the merger goes through. The other measures are solely ex post. In practce, the HHI s augmented wth addtonal market nformaton (U.S. Department of Justce, 00). The HHI focus on market shares does not account for how these shares may have been acheved through lower costs or by uncompettve behavor (Demsetz, 97). Ths led to the effcent structure controversy and the fndng that cost effcency (measured from a cost fronter) and the HHI are about equally mportant n explanng dfferences n bank proftablty (Berger, 995). Although the HHI s generally assocated wth hgher loan rates and lower depost rates (c.f., Degryse et al. 05), these results have not been adjusted for dfferences n costs or effcency across banks. If lower costs and/or greater effcency have been an mportant reason for some banks n achevng a relatvely hgh HHI, ths wll overstate the lack of competton suggested by the HHI. We follow Hrtle (007) n calculatng a depost-based HHI for each bllon dollar commercal bank for 00 n each of the 956 MSAs and non-msa countes. These MSA HHIs are then weghted by each bank's depost share across all the MSA's the bank operates n to gve a bank-level HHI. 4 Our bank-level HHIs are weghted averages of end-of-day branch-level depost data and, as averages, can understate or overstate ndvdual branch-level effects n dfferent MSAs. Due to a lack of data, the depost-based HHI s the only one avalable. Whle t s used for all servce lnes n ths study, t should be most accurate for our two depost servce lnes. 4 The Appendx has more nformaton on the calculaton process. 4

5 The 00 Department of Justce (DOJ) horzontal merger gudelne suggests that markets wth an HHI below,500 are consdered to be unconcentrated whle a moderately concentrated market exsts when the HHI les between,500 and,500. A hghly concentrated market has a HHI above,500. As the average HHI n our set of 54 bllon dollar banks s just,67 (wth a standard devaton of 570), the average bank s n an unconcentrated market. Usng the current DOJ gudelne, only percent of our sample (wth average total assets of $.6B) s operatng n a hghly concentrated market. Had we used the 98 DOJ gudelne to dentfy a hghly concentrated market (.e., HHI >,800), whch was unchanged n later revsons n 99 and 997, ths percentage would be 6 percent wth average assets of $.4B. Under ether gudelne, banks wth the hghest HHIs are concentrated among the smaller bllon dollar nsttutons not the very largest banks.. Lerner Index: the Level of the Prce-Cost Spread The Lerner Index (Lerner, 94) seeks to measure realzed (rather than potental) competton. It reflects the percentage spread of an output prce ( P o ) to estmated margnal cost ( MC ) for ndvdual banks averaged over quarters durng : ( P MC) / P. Margnal cost s o o estmated from a logarthmc cost functon lnc = f (ln Q o,ln P ) where C s total cost, Q o s output, P s composed of fundng, labor, and captal nput prces, and MC = lnc / lnq )( C / Q ). As scale economes ( SCE ) are the rato of margnal to ( o o average cost ( AC ), the Lerner Index can be reexpressed for llustratve purposes as P o AC SCE. 5 Ths reflects the level of the prce-average cost spread. 5 The varaton of ths unt spread across banks wll accord well wth ts percentage value f dvded by P o or AC. 5

6 As the separate total cost of busness or consumer loans or any of the depost categores s not reported, t s not possble to estmate a Lerner Index for any of our four servce lnes. So MC cannot be estmated nor s average cost ( AC ) avalable by servce lne. Only total cost for all servce lnes together s avalable. 6 Thus a standard Lerner Index was estmated for the whole bank and s used n place of a Lerner Index specfc to each servce lne. Our specfcaton of the bank-wde Lerner Index s shown n the Appendx and ts average value s percent wth a standard devaton of nne percentage ponts. To put ths n perspectve, f MC was cents per $ of loans the average loan rate P o would average 4.5 cents per dollar of loans. Although no Lerner Index can be computed by servce lne, a Mark-Up of loan prce over depost cost can be computed. Ths represents an approxmate Lerner Index for busness and consumer loans.. H-Statstc: Changes n the Prce-Cost Spread Lke the Lerner Index, the H-Statstc of Panzar and Rosse (987) s based on economc theory and seeks to measure realzed competton. One verson of the H-Statstc relates changes n total consumer loan revenue ( TR = P o Qo ) to changes n observed nput prces ( P ) holdng output level ( Q ) constant. An H-Statstc based on a revenue functon lntr = g(ln P, lnq,ln X ) for o busness and consumer loans s shown n the Appendx ( X represents other varables). For loans, the H-Statstc s the sum of partal dervatves or elastctes of a revenue functon: o lntr / ln ln P / ln P ) as the P and reflects the change n output prce to nput prces ( o output component of TR s beng held constant. 6 Although we can reasonably approxmate the average nterest expense ncurred n makng busness and consumer loans and determne the average depost rates pad for savngs and tme deposts, nformaton on the operatng costs (labor, physcal captal, and other non-nterest expense) that would be allocated to the varous servce lnes does not exst n the publcly avalable data. 6

7 For deposts, a prce form of the H-Statstc was used snce specfc nformaton on savngs or tme depost revenues s not separately reported and n any case s de mnmus. Whle there s depost revenue, ths s prmarly n the form of demand depost servce charges and demand and savngs depost monthly account mantenance fees (when the balance n these accounts s below a requred mnmum). 7 Importantly, servce charge revenues are often puntve n nature (e.g., account overdraft fees) and are rased prmarly from lower ncome depostors (FDIC, 008). For deposts, the prce form of the H-Statstc relates observed depost returns to changes n nput prces ln Po = f (ln P,ln X ). Consequently, the depost H-Statstc s ln Po / ln P and the specfcaton s shown n the Appendx. Banks wth a hgh rato of changes n output to nput prces ln Po / ln P wll also have a large dfference n these prces ln Po / ln P so the latter s an alternatve expresson of the former. If output and nput prces strongly rse and fall together over tme or across banks, the mplcaton s that non-cost nfluences on output prce are small. Ths would ndcate a compettve market where cost determnes prce. Competton s strong when the H-Statstc s close to.0 and non-exstent when t s close to 0.0. The closer the H-Statstc s to zero, the lower the nfluence of cost on output prce, mplyng that frms are able to set prce ndependently from cost as possbly a result of havng and usng market power to set prces or mplctly collude wth others to do so. It has been shown theoretcally that the H-Statstc s not a useful ndcator of market power, where market power s ndcated by a Lerner Index (Shaffer and Sperdjk, 05) or when market power s smulated usng a Cournot-type model (Hyde and Perloff, 995). Even so, we 7 Whle monthly demand depost account mantenance fees ranges from $7 to $5 (Moebs Servces, 0) and represent a broadly based revenue source, savng accounts have lower monthly account fees (f they have them at all). 7

8 nclude the H-Statstc here snce many studes on bankng competton contnue to use t n emprcal studes. We determne the emprcal relatonshp between the H-Statstc and other competton measures as well as how they all relate to an ndcator of realzed competton the dfference n servce lne prces across banks. Emprcally, we fnd that the H-Statstc s less nformatve than the other competton measures, whch underscores these theoretcal analyses. The average H-Statstcs for busness and consumer loans are.8 and.85, respectvely, and suggest these markets are reasonably compettve. In contrast, the H-Statstcs for savngs and small tme deposts are, respectvely,.9 and Ths suggests low to no competton for deposts snce there s a low correlaton between the depost rate pad and the varaton n factor nput prces. As Bkker et al. (0) pont out, assessng compettve conduct usng H-statstcs generally requres addtonal nformaton about costs and market equlbrum to nfer the degree of competton. In the Appendx a weak test for market equlbrum of the H-statstc s further dscussed. Assumng that average cost reflects the weghted average of nput prces ( P ), the two competton measures favored by academcs for comparson purposes can be expressed as: - Lerner Index: AC SCE (Level of the prce-cost spread), P o - H-Statstc: AC (Change n the prce-cost spread). P o In effect, the Lerner Index looks at the average level of the prce-cost spread over a sample perod whle the H-Statstc looks at changes n that spread: they measure dfferent aspects of competton. 8

9 .4 Loan Mark-Up Over Depost Costs Whle our Lerner Index (due to a lack of detaled cost data) reflects the mark-up for the entre bank, a more lmted measure s possble for busness and consumer loans. Ths would relate the average prce of these loans ( P LOAN ) to the cost of deposts that are the prmary source of fundng for these and other loans ( P DEP ). As P DEP reflects short-term fundng costs, t s close to reflectng the unknown margnal cost of busness and consumer loan fundng expenses. The Mark-Up s expressed as ( P P ) / P and represents an approxmate servce lne Lerner LOAN DEP LOAN Index even though t excludes factor costs. The average loan Mark-Up for busness loans s 7 percent and 77 percent for consumer loans, both wth a standard devaton of 8 percentage ponts. The mark-up appears hgh for two reasons: nterest rates are at hstorcal lows durng our sample perod and the numerator of the Mark-Up excludes operatng costs (captal and labor factor costs whch prmarly affects the level of PDEP, lowerng ts value). An equvalent Mark-Up measure for savngs deposts does not really exst as there s very lttle actual revenue to mark-up. Depostors do pay a "prce" but ths s the opportunty cost of holdng a savngs balance rather than puttng ther funds n a hgher earnng money market account or other short-term nvestment. Demand deposts, however, do generate drect revenues. Ths comes almost entrely from account overdraft fees, monthly account mantenance fees, charges for postng deposted checks (prmarly for busness, not consumers), ATM fees, stop payment orders, etc. Some banks do have an account mantenance fee for savngs deposts when the balance held s below some mnmum level but, typcally, the account s closed (or abandoned) before much revenue s generated. Tme deposts also have a charge f funds are removed before maturty but here agan the revenue rased s de mnmus and n any case s not 9

10 separately reported for demand, savngs, or tme accounts..5 Adjustng Revenues for Costs: A Competton Effcency (CE) Fronter Usng a procedure based on effcent fronter analyss, t s possble to derve nferences on bank competton from estmatng a revenue-cost competton fronter for each of our servce lnes. Here operatng cost and productvty dfferences for deposts plus fundng cost for loans are accounted for. Our fronter approach to measurng competton s smlar to that developed ndependently by Boone (008). 8 The goal s to determne competton based on a frm's profts but profts by servce lne are not reported. The degree of competton s obtaned by subtractng a frm's varable costs from ts revenues. Ths gves an mpled return to fxed nputs plus extra revenues assocated wth the degree of relatve competton. We apply an emprcal specfcaton of a Boone-type model whch was shown theoretcally (at least) to be more robust than the prce-cost margn of a Lerner Index. 9 In smple terms, profts = f (competton, costs). As profts are smply revenues mnus costs, proft dfferences across banks can be alternatvely measured as the "mark-up" rato of revenue to costs and, f all explct costs are ncluded, an estmate of relatve competton can be obtaned from: revenue/costs - f (costs) = g (competton). Here g (competton) represents the unexplaned mark-up over cost and ncludes a normal return on equty whch s not explctly 8 Wthn a general model of Cournot competton, Boone shows that n a more compettve ndustry, the most effcent frm gans more relatve to less effcent frms, so ts relatve profts and market share ncrease relatve to those laggng frms. A more detaled theoretcal dscusson s contaned n Bolt and Humphrey (05a). 9 Even so, our measure s smlar n concept to the Lerner Index n that we focus on a revenue-cost or prce-cost spread. One advantage s that t can be appled when only revenue data exst, but not output prce f output s not measured (Bolt and Humphrey, 04). 0

11 specfed n the model. Our revenue/cost rato s the nverse of the popular Cost Income Rato used n bankng: revenue/cost = (nterest revenue - nterest expense + fee ncome)/(labor + captal + other non-nterest expense). Banks wth a lower operatng nput cost per unt of output net of nterest revenue rased are (by defnton) more proftable. If the productvty of factor nputs dffers across banks then nput prces (reflected n operatng cost) wll not reflect ther true cost to the bank. Observed nput cost wll be hgher for banks wth greater productvty makng them appear more compettve than they are as the observed revenue-cost spread wll be lower. The productvty varables we specfy have been mportant n reducng cost neffcency to low levels n both stochastc and lnear programmng fronter models (Carbo et al. 007). 0 Specfcally, labor ( L ) s more productve and real labor costs are lower when there are fewer employees (lne, back offce, and management) per branch offce ( BR). As well, captal productvty s mproved when a bank produces more deposts ( DEP ) per branch offce. In sum, a hgh labor/branch rato ( L / BR ) and a low depost/branch rato ( DEP / BR ) suggests cost neffcences that rolls over nto proft neffcences relatve to other banks. In estmatng the competton fronter, we use the composed error Dstrbuton Free Approach (DFA) n Berger (99). In a composed error framework, the DFA model llustrated n () relates the rato of busness or consumer loan revenues, net of approxmate fundng expenses, to overall operatng costs. Specfcally, t attempts to explan the varaton across banks n the rato of unt loan revenues ( PLOAN ) mnus unt depost cost ( PDEP) dvded by overall 0 Berger and Mester (997) and Fre et al. (000) have also shown productvty nfluences to be a prmary determnant of prevously unexplaned bank cost neffcency. Some banks operate n-store branches where the staffng level s about half that of a stand-alone offce. As well, the captal cost of an n-store branch s only about one-ffth of a conventonal branch (Radeck et al. 996).

12 operatng cost (OC ), comprsed of labor, physcal captal, and materals expense: ln(( PLOAN PDEP) / OC) = R(ln Q,ln P,ln X,ln Z) + ln e + lnu. () As noted above, there s no output prce or explct revenue for savngs or small tme deposts. The depost rate pad s a cost to the bank. However, relatve to an alternatve nterbank or longer term fundng source, we consder the dfference between the savngs depost rate pad ( RS ) and some (typcally) hgher cost alternatve source of funds to be mplct revenue. For savngs accounts, the alternatve could be a weghted average or separate fundng costs of borrowng federal funds, ssung commercal paper, generatng more tme deposts, or ssung more subordnated notes and debentures or even more equty snce all of these actvtes generate fundng. We use two alternatve revenue benchmarks for savngs deposts: the natonal quarterly average federal funds rate ( FFRATE) and each bank's small tme depost rate ( RTDS ). The pont s that havng fewer savngs accounts, f assets are not also reduced, has an opportunty cost for a bank and we use ths as a benchmark to generate an estmate of mplct revenues (saved fnancng cost). Whle the RHS of () remans the same for savngs deposts, the dependent varable s ln(( FFRATE RS ) / OC) or ln(( RTDS RS ) / OC). 4 A smlar procedure s adopted for small tme deposts. Snce the average rate pad on large tme deposts (> $00,000) s Although operatng cost s not allocated to the varous bankng servces n the reported data, we do specfy varables to explan the overall level and varaton of operatng cost among banks whch, n turn, nfluence the effect of operatng cost on consumer loans and other bankng servces. The purpose s to llustrate how the results may vary dependng on the benchmark alternatve fundng source chosen. 4 As none of the nterest rates n these expressons ncludes the assocated operatng costs (tradng facltes, tellers, branch space, etc., whch are not allocated n the avalable data) the numerator values wll not fully reflect all ther expenses.

13 actually slghtly less than the average rate pad on small tme deposts, we use each bank's prce of consumer loans as the benchmark gvng ln(( PCLOAN RTDS) / OC) as the dependent varable for small tme deposts. Strctly speakng, we could have used the rate pad on subordnated notes and debentures or even the prce of busness loans snce what we want here s a spread that reflects a dfference between mplct revenue (saved cost) and actual fundng cost. Ths approach to specfyng a dependent varable for obtanng a CE depost measure llustrates an mportant pont. Whle t s obvous how one would construct a Lerner Index or Mark-Up measure for loan products (.e., ( P MC) / P ), smlar data are not avalable to apply o o the same approach for deposts. Deposts bascally have no or very lttle revenue and hence no or a very low explct output prce ( P o ). Also, cost data needed to estmate MC are not publcly avalable by servce lne. But we do have nformaton on a number of dfferent alternatve fundng sources ( RTDS ) and an estmate of actual fundng cost ( RS ) whch can represent an mplct spread ( RTDS RS ) smlar n concept to ( P o MC ). Indeed, a Lerner Index of the form ( P P ) / P could be used to assess the LOAN DEP LOAN compettve nature of loans as well as deposts. For example, hgh loan rates at a bank could be due to havng a relatvely larger prce-cost spread wth normal depost costs whle other banks could have much the same spread but, due to payng a low depost rate, the output prce s not much hgher than the average. The frst bank could be consdered less compettve on the loan sde whle the second bank could be less compettve on the depost sde. It s n ths sense that the same spread measure or Lerner Index could be used to assess loan (output) as well as depost (nput) competton. Effectvely, ths s what we are proposng to do for our CE measure. The total resdual ( ln e + ln u ) n () reflects the unexplaned porton of the net revenueoperatng cost dependent varable remanng after output ( Q ), nput cost ( P ), productvty ( X ),

14 and rsk ( Z ) have been accounted for. Here mantaned hypothess s that returns for deposts. The DFA concept assumes that whle the average of the lowest average resdual ( lnu ln e represents the value of random error whle our ln u represents the effect of competton on revenues for loans and ln e wll average to a value close to zero ln u wll reflect the average effect of competton ( ln u ). 5 The bank wth mn ) s also the bank where the varaton n underlyng cost and productvty explans the greatest amount of the varaton n loan revenues or depost returns relatve to operatng costs, reflectng the strongest effect of cost-based prcng on the revenue/cost rato through competton. Ths mnmum value defnes the competton fronter and the relatve competton effcency ( CE ) of all the other banks n the sample s determned n () by ther dsperson from ths fronter: CE = exp(lnu lnumn ) = ( u / umn ). () The smaller (larger) s CE, the stronger (weaker) s the cost-based nfluence on loan revenues or depost mplct returns, so CE s smlar to neffcency n a cost fronter. Ths s a relatve measure so f the mnmum resdual s small, the CE value can stll appear to be large as an ndex. The rankng of ths ndex, lke the rankng of the other competton measures, s what s nformatve wth respect to prce conduct. It s possble that we have excluded some mportant cost nfluence from the estmated fronter model shown n the Appendx. Even so, our statstcal ft across sx two-quarter panel 5 DeYoung (997) has suggested that 6 separate cross-secton estmatons may be needed for random error n the composed error term to acheve an average value close to zero. Our composed error terms are averaged over 6 cross-secton estmatons of two quarters each. 4

15 estmatons averaged.96 (.86) for busness (consumer) loans and.9 (.9) for savngs deposts (small tme deposts). Ths suggests that there s not much left to be explaned by unreported and/or excluded dfferences n cost. 6 Indeed, f all costs have been ncluded, nfluences other than cost or productvty on the loan dependent varable are only 4 to 4 percent of the total varaton. The unexplaned varaton for deposts s just 7 percent for both depost types.. Servce Lne Prces and ther Relatonshp to Measures of Competton One way to gauge the usefulness of the dfferent competton measures outlned above would be to determne ther ablty to dentfy ndvdual banks that appear to have relatvely hgh loan prces and/or pay relatvely low depost rates due to a lack of effectve market competton rather than beng a result of ther underlyng cost structure. The goal s to have prce predomnately determned by cost for effcent resource allocaton. Ths, certanly, s the purpose of havng the HHI serve as way to judge the possble effects of bank mergers and acqustons on market competton.. Servce Lne Prces Fgure shows the average prce for busness and consumer loans (Y-axs) for each of our 54 bllon dollar banks over arrayed aganst the log of each bank's average value of total assets (X-axs). Fgure shows the same thng for savngs and small tme deposts (those < $00,000). The ftted cubc splnes (sold lnes) n both fgures are very flat wth almost a zero 6 Whle a fxed effects approach would lkely mprove the adjusted R, the tme and bank-specfc dummy varables cannot dscrmnate between excluded cost effects (whch we would lke to nclude) and the excluded effect of competton on revenues (whch we wsh to exclude and keep n the resdual). The Lerner Index has a smlar problem when appled to the entre bank as t excludes around 5% of revenue generated by bankng output when estmatng margnal cost n a cost functon snce output assocated wth non-nterest ncome (such as depost and payment processng fees) s not reported. 5

16 slope. The major dfference n Fgure s that the average consumer loan rate (4.9 percent) s 40 percent hgher than that for busness loans (.5 percent). The dfference n deposts rates n Fgure s smlar: the average small tme depost rate (.9 percent) s 54 percent hgher than the savngs rate (0.8 percent) INSERT FIGURE AND FIGURE ABOUT HERE --- Table shows how much hgher average loan rates are at banks n the quartle where loan rates are hghest compared to the quartle where loan rates are lowest. Busness loan rates are 9 percent hgher at the hghest quartle relatve to the lowest quartle but 06 percent hgher for consumer loans. At these same banks, however, the maxmum dfference n savngs and small tme depost rates s only 9 percent. In farness, we have not controlled for dfferences n factor prces, scale economes, credt rsk, ndustry concentraton or borrower type. As well, we do not know the all-n cost of provdng transacton and safekeepng servces. Ths s the qud pro quo for payng a zero rate on demand deposts whch s the major source of bank funds. Even so, the over 00 percent dfference n consumer loan rates n Table seems surprsngly large INSERT TABLE ABOUT HERE --- Lookng at the last column n Table, t seems that smaller bllon dollar banks, on average, charge the hghest loan rates whle the lowest rates are charged by much larger banks. 7 Ths dfference s partly due to the fact that market rates fell rapdly over our tme perod and a large porton of tme deposts were pad a locked-n rate from an earler perod. 8 Ths dfference would be somewhat larger f t was possble to nclude the effect of compensatng balance requrements often assocated wth busness loans, as ths rases the effectve loan cost to the borrower. 6

17 Ths pattern s reversed on the depost sde. Table shows that smaller banks pay the hghest depost rates but larger nsttutons pay the lowest. It s temptng to generalze ths result and conclude that smaller banks charge the most for loans but pay the hghest for deposts, and vce versa for larger banks. However, Table shows that smaller banks n the hghest loan rate quartle (top two rows) pay ther depostors about the same depost rates as larger banks n the lowest loan rate quartle (bottom two rows). Snce there s not much dfference n depost rates pad, t would be best to borrow from a bank n the lowest loan rate quartle. Table shows a smlar stuaton but for the hghest and lowest depost rate quartles. It s better to have deposts at a bank n the hghest depost rate quartle than n a bank n the lowest quartle snce the assocated loan rates are hardly any dfferent. In practce, of course, consumers and small busnesses are restrcted to banks that happen to be n ther local area. Large frms can seek out banks offerng lower loan rates n the regon or natonwde and have lttle use for savngs or tme deposts anyway. They typcally purchase hgher earnng short-term nvestments and keep ther depost balances close to zero overnght (except for dle compensatng balances requred n ther bank loan agreements or to pay for bank servces used). In sum, banks seem to prce loans and deposts as f they were segmented markets, whch s probably the case for most depostors and borrowers snce busnesses are the prmary bank borrowers wth low (end-of-day) depost balances whle consumers mostly use depost servces along wth mnmal borrowng. --- INSERT TABLE ABOUT HERE --- 7

18 .. Smlarty Among Competton Measures There s no economc relatonshp between the HHI and any of the other competton measures. The adjusted R between the HHI and all the other measures, s between.0 and zero. The hghest correlatons between competton measures for each servce lne are shown n Table. In all cases, ths nvolves the Lerner Index (whch s for the entre bank). One would expect a relatonshp between the Lerner measure and the Mark-Up snce t s an approxmate Lerner Index to begn wth (but wth less nformaton on cost). The CE measure s smlar to a Lerner Index but n a dfferent way, ether by usng revenue when output prce s not avalable (Bolt and Humphrey, 04) or when nether explct revenue or output prce exst and cost nformaton s lmted (ths paper). The strongest relatonshp s between the Lerner measure and the H-Statstc wth R =.5. Although less extreme, the lack of an economc relatonshp between competton measures s the general case for the bankng ndustry n Europe as well (Carbo et al. 009). None of these relatonshps would be consdered to be strong and no measure s related to the HHI, whch s the one reled on for anttrust polcy n the U.S., although not as promnently n Europe (c.f., European Commsson, 007). Unfortunately, there seems to be lttle scope to use other measures of competton to benchmark the HHI gong forward. 9 Although there s only a weak economc relatonshp between our competton measures, we now see how these measures are related to relatve prces across banks for four servce lnes. The HHI has long been used to "predct" the possble effect on bank prces pror to a merger or acquston by smply addng the pre-merger market shares of the merger partners. In ths regard, one would expect the HHI to vary wth bank prces across banks n general. Anttrust authortes 9 Investgatng the possblty that other competton measures could be used to benchmark the HHI was one reason for ths study. 8

19 have devsed an alternatve ndcator of the possble effect of a merger/acquston on output prce of mergng frms called GUPPI Gross Upward Prcng Pressure Index (c.f., Shapro, 00). Ths reles on pre-merger estmates of cross-prce and own prce demand elastctes, prce-cost margns, and prce levels of products of the mergng frms. The end result s an estmate of the output prce change f the merger goes forward. In what follows we see how bank servce lne prces vary across banks when competton measures are beleved to reflect greater versus reduced market competton. --- INSERT TABLE ABOUT HERE ---. Competton Measures and Loan and Depost Rates Do loan prces rse or depost rates fall when a competton measure ndcates less competton? The top part of Table 4 shows how the average values of fve dfferent competton measures vary across quartles of banks. Each measure s ranked from ts lowest value (suggestng that the banks n ths quartle would be the most compettve) to ts hghest value (suggestng these banks would be n the most compettve quartle) INSERT TABLE 4 ABOUT HERE --- The values n the frst fve rows ndcate the average value of each competton measure n each quartle. The values n the second fve rows show the correspondng average busness loan prce for the banks n each quartle of the frst fve rows. For example, when the average 0 As seen n the table, the H-Statstc rankng s reversed so that hgher values suggest greater (not lessor) competton. 9

20 HHI = 794, the correspondng banks have an average busness loan rate of.5 percent. The busness loan rate remans at.5 percent as the average HHI rses to,6 and then to,409 for our 54 bllon dollar banks over When the HHI rses to,04 and s thought to dentfy the quartle of least compettve banks whch are expected to have the hghest loan prces, the busness loan rate s.6 percent. The change n prce from most to least compettve s +0. percentage pont or 0 bass ponts (all numbers are rounded). The banks that are responsble for the 0 bps hgher average loan rate account for 7 percent of the total assets of our sample of bllon dollar banks. Ths s seen from the last set of fve rows n the table. Recall that the HHI s computed from branch level data on all deposts whle the Lerner Index, due to data lmtatons, refers to all bank servce lnes. Only the H-Statstc, Mark-Up, and CE measures are specfc to busness loans n Table 4. The H-Statstc for busness loans s seen to fall, suggestng less competton, but the average loan prce falls by 0 bass ponts nstead of rsng as would be expected. As the H-Statstc reles on the correlaton between loan prce and nput unt cost (the prce labor, captal, and all lablty fundng), ts normal behavor may well have been affected by the unusual experence of where rates fell but factor prces were stable. The Lerner Index, Mark-Up, and CE measures all rse (by 0, 0 and 0 bass ponts, respectvely) when movng from most to least compettve values for these measures. The H-Statstc and Mark-Up both suggest that banks n the least compettve quartle comprse 70 percent of our sample bank assets whle the HHI and CE suggests t s lttle more than half that value (from 7 to 4 percent). The clear outler s the Lerner Index whch suggests If loan prces are hgh due to hgh costs resultng from a lack of competton (a "quet lfe" vew, see also Koetter et al. 0), controllng for dfferences n costs across banks could reduce the ablty of most of our competton measures to dentfy banks wth relatvely hgh output prces. However f loan and depost markets are largely segmented wth respect to prce conduct, the quet lfe ssue s less mportant and lkely restrcted to very small bankng markets whch can only support a few small banks (certanly not bllon dollar banks). 0

21 only 5 percent of assets are n the least compettve quartle. Clearly, dfferent sets of banks are beng dentfed as beng least compettve. Generalzng, t appears that all but the H-Statstc yelds the expected result that hgher busness loan prces are hgher at banks when the competton measure ndcates they should be least compettve. As well, the Lerner Index and CE suggest that the loan prce rse s on the order of 0 to 0 bps. Snce the asset shares for these two measures are qute dfferent (5 versus 4 percent), ths means that we cannot be sure of how mportant these prce ncreases are for the bankng ndustry. The Lerner Index suggests lttle mportance whle the CE puts t at 4 percent of the ndustry. One would have hoped for a greater correspondence of results n terms of prce changes and asset composton. We perform the same analyss for consumer loans n Table 5. The varaton of the H- Statstc, Mark-Up, and CE measures across quartles n the frst fve rows dffer from those n Table 4. These three competton ndcators can be estmated specfcally for consumer loans whle the HHI and Lerner Index are sngle measures and by default are appled to all servce lnes rather than vary across them. --- INSERT TABLE 5 ABOUT HERE --- Overall, there are not many dfferences for consumer loans from what we saw for busness loans. Both the HHI and the H-Statstc show that consumer loan prces fall (rather than rse) as the rankng of these measures go from most to least compettve. And, as before, prces rse when the Lerner Index, Mark-Up, and CE measures are appled. Table 5 also tells us that the asset shares of banks dentfed as beng least compettve are agan qute dfferent, even though

22 the shares for each competton measure are (except for CE) smlar for busness and consumer loans. The major dfference between consumer and busness loans concerns by how much prce rses gong from the most to least compettve quartles. The prce ncrease s 00 bps for the Mark-Up and CE measures for consumer loans but only 0 to 0 bps for busness loans. That the consumer loan market seems much less compettve than the market for busness loans s not news to bank regulators. The standard explanaton s that busness borrowers are more nformed about fnancal matters. Busnesses shop around for better rates much more than consumers do and large frms often obtan loans outsde ther local area. Also, snce busnesses borrow larger amounts and also purchase many other bankng servces, they are a more proftable customer and so can drve a better bargan. Hgher returns on these extra servces can cross-subsdze the busness loan rate charged. In Table 6, the average depost rate pad s expected to fall as the varous competton measures move from most to least compettve. For savngs deposts ths occurs for all measures except the H-Statstc. When appled to small tme deposts (deposts < $00,000) there are no exceptons: all measures generate the expected prce reducton. Overall, the HHI ndcates the rate reducton s only 5 bps for both depost categores, suggestng that the market for savngs and small tme deposts could be consdered to be compettve. The H-Statstc s nconsstent and shows that rates rse for savngs but fall (as expected) for small tme deposts when movng from most compettve to least compettve quartles. All the other measures for savngs and small tme deposts show the expected result depost rates are lower at banks dentfed as beng least compettve, wth prce reductons between -5 to -4 bps. As the savngs rates are so low, the prce changes are taken to an extra decmal place.

23 --- INSERT TABLE 6 ABOUT HERE --- Recall that the CE dependent varable was expressed two ways to llustrate how the choce of an alternatve benchmark fundng rate for savngs deposts may affect the CE results. It was run once usng the natonal quarterly average federal funds rate as the benchmark (suggestng a -46 bps depost rate change) and agan usng each bank's own small tme depost rate (gvng a - bps change). The federal funds benchmark s the same for each bank for each quarter and only vares over tme whle each bank's small tme depost rate vares both across banks and over tme. The CE dependent varable s defned as ( RTDS RS ) / OC usng the small tme depost rate as the alternatve fundng source. If there s lttle varaton n the RTDS, the numerator should more closely correspond to the varaton n the savngs rate ( RS ) by tself. And when the resultng CE estmate s ranked, t wll tend to nclude banks that have a lower savngs rate as beng least compettve. Ths s because a smaller porton of the dfference n the numerator wll lkely be explaned by the RHS cost nfluences we have specfed. The lmtng case would be f RTDS was a constant and had no varaton. Ths s lkely the reason usng the natonal federal funds rate n place of RTDS, whch reduces the varaton n the numerator of the dependent varable, results n a hgher assocated depost rate for the most compettve quartle n Table 6 (.07 percent) as well as a lower depost rate for the least compettve quartle (0.6 percent). Ths results n a larger reducton n the depost rate (-0.46) compared to the reducton (-0.) when RTDS s used as the alternatve fundng source.

24 4. Are the Competton Measure-Prce Relatonshps Sgnfcant? Across our four bank servce lnes, from Tables 4, 5 and 6, 4 out of the 8 possble competton measure-prce relatonshps shown there moved n the expected drecton. Ths means that when a competton measure suggests banks are lkely to be most compettve, the assocated loan rate s lower and the depost rate s hgher. When the same measure suggests banks are lkely to be least compettve, the loan prce s hgher and the depost prce lower. A competton measure s consdered to be nformatve here even though the sze of the prce change often dffers as seen above. But are these relatonshps sgnfcant? Out of a sample of 54 bllon dollar banks, the (rounded) number n each quartle s 89. The expected number of competton measure-prce relatonshps possble for each quartle va random chance alone s banks. 4 The 95-percent confdence nterval around ths chance matchng s 5 to 0 banks so we are lookng for 0 or more matches to determne whch competton measures are sgnfcantly assocated wth the expected prce conduct. For busness and consumer loans n Table 7, only the Lerner Index, Mark-Up, and CE measures have matches of 0 or more and are sgnfcant at the 95-percent level. If we focus on the most mportant quartle, where a hgher competton measure s assocated wth a hgher loan rate, only the Mark-Up and CE measures are sgnfcant. Smlar results are obtaned for consumer loans. If we reduce the sgnfcance to 90 percent, nothng changes snce the confdence nterval s (6, 9). The relatonshp s reversed, of course, for the H-Statstc. 4 The jont probablty of matchng two ndependent seres s 54*/4*/4 = 88.5*/4 = just by chance. 4

25 --- INSERT TABLE 7 ABOUT HERE --- For savngs and small tme deposts, Table 8 shows that the HHI and one of the CE measures are sgnfcant when these competton measures suggest a less compettve market. In contrast to the loan results, there s only one measure (CE) that s sgnfcant n both columns. Nothng s sgnfcant for small tme deposts. Unlke the other measures, the HHI does not control for how cost dfferences may affect prce. The HHI could be sgnfcant (a) because banks n the least compettve quartle have low depost rates due to lower costs or (b) because the low rates reflect uncompettve behavor the Demsetz (97) dchotomy agan. As there s almost no overlap between the least compettve banks suggested by the HHI n Table 8 wth the least compettve banks dentfed usng the sgnfcant cost-adjusted CE measure, ths would support reason (a) for the sgnfcance of HHI, although ths s not defntve. We note that these quartle results are the very much the same even f we use decles, as shown n Appendx Tables A. and A INSERT TABLE 8 ABOUT HERE Conclusons We have computed fve dfferent measures of bankng competton: HHI, H-Statstc, Lerner Index, Mark-Up, and a fronter ndcator of competton effcency (CE). These were related to the prces of four servce lnes for 54 U.S. bllon dollar banks coverng busness and consumer loans as well as savngs and small tme deposts. The purpose was to see how competton measures may dffer across bank servce lnes and how these measures are related to low and 5

26 hgh loan rates and depost returns. An mportant qualfer s that our tme perod 008 to 00 s defntely not a normal one. Ths could have based our results, especally for the H-Statstc whch reles on the correlaton between fnancal output prces and fnancal and factor nput prces. There was a complete lack of a meanngful relatonshp between the HHI and any of the other measures of bankng competton. The R here was.0 or lower. 5 Indeed, out of a possble bvarate correlatons among competton measures, only fve had an R larger than.0 (the hghest was.5). A very weak economc relatonshp between competton measures also exsts n Europe (Carbo et al. 009). It was hoped that other measures of competton could be used to benchmark the HHI. The goal would be to more accurately nform bank and Justce Department regulators on when mergers and acqustons may be n the publc nterest even whle retanng the HHI as the "publc face" of anttrust polcy. The HHI may have some value regardng prce conduct n more olgopolstc ndustres but ths seems not to be the case for bankng. The excepton may be very small bankng markets where there s only one to three banks. Here the small sze of the market serves as a barrer to new entry smply because the market s too small and not growng to support a new entrant. Ths s certanly not the case for the bllon dollar banks examned here and that, overall, account for some 90 percent of the ndustry. It seems that a mark-up type of approach to assessng bank servce lne competton a Lerner Index, Mark-Up, or CE measure appear to be the most nformatve. They have the most success n dentfyng banks that bear further nvestgaton for uncompettve prce conduct. The 5 Our vew s that havng a statstcally sgnfcant lnear or non-lnear relatonshp between two competton measures s bascally mmateral when the resultng R s de mnmus. 6

27 HHI had one success for savngs deposts, not loans but s problematc. The possblty that the HHI dentfed possble uncompettve prce conduct, rather than a lower cost structure leadng to lower depost rates, s apparently not supported. The banks the HHI dentfed as havng lower depost rates, do not overlap much wth those dentfed usng the other measures whch are structured to dentfy banks wth lower depost rates after adjustng for dfferences n cost. It turns out that the H-Statstc s also less nformatve than the other measures. Ths supports theoretcal analyses that came to the same concluson (Shaffer and Sperdjk, 05; Hyde and Perloff, 995). Three out of four prce predctons usng the H-Statstc were the opposte of what they should be (Tables 4, 5, and 6) and none were sgnfcant (Tables 7 and 8). Ths sgn reversal occurred only once for the HHI and not at all for the other measures. The mplcaton of these results for bankng research s that a smple Mark-Up (or Lerner Index) wll lkely dentfy better the dfferental nfluence of competton when tryng to hold these nfluences constant n a statstcal model. Indeed, nstead of havng to estmate margnal cost for a standard Lerner Index, one can use observed/computed average cost nstead. Scale economes exst n bankng and ths tes margnal and average cost together. 6 The CE measure s dffcult to estmate whle the HHI and H-Statstc, n ths exercse at least, are not nformatve. What are the mplcatons for competton polcy? The HHI s smple to compute and apply. But other measures would be smple to apply as well f the FDIC publcly or prvately collected approxmate bank cost accountng data by servce lne. 7 Unfortunately, the HHI wll 6 Scale economes are just the rato of margnal to average cost and changes n the former lead to changes n the latter. As they predct each other ether one can be used n a Lerner Index to "predct" prce conduct. 7 Although case law prevents the applcaton of anttrust acton to ndvdual bank servce lnes, the HHI regulators use to determne the possble compettve mplcatons of mergers and acqustons s tself based on the depost servce lne. A servce lne approach to assessng competton, because t s less nfluenced by dfferences n balance sheet composton across banks, s lkely to be more nformatve than applyng competton measures to all servce lnes as an aggregate. 7

28 lkely stll be used for merger analyss regardless of results whch suggest t performs poorly n practce and s nferor to other competton ndcators. Indeed, the HHI concept s effectvely enshrned n legslaton (the Regle-Neal Act of 994) snce 0 percent s the maxmum amount any one bank can hold of natonwde nsured deposts through a merger or acquston. 8 Realstcally, pressure for change would only come f regulatory opnons based on the HHI were challenged (n court or otherwse) usng evdence that showed decsons relyng on the HHI do not measure what regulators' assert t does. If the HHI s kept as the "publc face" of bank anttrust polcy, t would be useful to consult what other measures of competton are reportng when the DOJ merger gudelnes are beng reconsdered (as they were n 00 when the "hghly concentrated" gudelne was rased by 40 percent relatve to the one establshed n 98 and retaned n 99 and 997). Alternatve measures of competton should be among the evdence presented to support, or not, changes n the HHI gudelnes. Ths would not be dffcult or controversal f the Mark-Up wth addtonal cost nformaton was used n ths capacty, n contrast to the Lerner Index or Fronter CE measures due to ther need to be statstcally estmated and the dffculty of explanng ths process to the publc. 8 At the state level, the maxmum concentraton was set at 0% of statewde deposts, although 0 state legslatures have opted out of ths restrcton. 8