Do Consumers Swtch to the Best Suppler? by Chrs M. Wlson Department of Economcs, Unversty of Oxford & Catherne Waddams Prce ESRC Centre for Competton Polcy and Norwch Busness School, Unversty of East Angla CCP Workng Paper 07-6 Abstract: Ths paper suggests that the ablty of consumers to choose accurately between alternatve supplers s substantally lmted even n a relatvely smple and transparent market. Across two ndependent datasets from the UK electrcty market we fnd, on aggregate, that those consumers swtchng exclusvely for prce reasons approprated between a quarter and a half of the maxmum gans avalable. Whle such outcomes can be explaned by hgh search costs, the observaton that at least a ffth of the consumers actually reduced ther surplus as a result of swtchng cannot. We consder and reject several alternatve explanatons to pure decson error. July 2007 (Orgnal Verson: May 2006; second verson: Aprl 2007) JEL Classfcaton Codes: L00, D83 Keywords: Search Costs, Swtchng Costs, Decson Errors, Mssellng. 1
Acknowledgements: We are very grateful to Mchael Waterson for extensve and valuable dscussons, and to partcpants at the CCRP workshop n Brmngham 2007, the workshop for Consumer Behavour and Bounded Ratonalty at UCL 2006 and EARIE 2005. Thanks also go to Tna Chang, Steve Daves, Luke Garrod, James Harvey, Morten Hvd, Khac Pham, Laurence Matheu, Peter Moffatt, Alstar Munro, Jeremy Tobacman, Matthew Olczak, John Vckers and Matthjs Wldenbeest for ther help and comments. A prevous verson of the paper was ttled Irratonalty n Consumers Swtchng Decsons: When More Frms May Mean Less Beneft. The support of the Economc and Socal Research Councl (UK) s gratefully acknowledged. The usual dsclamer apples. Contact detals: Chrstopher Wlson, Department of Economcs, Unversty of Oxford, Manor Road Buldng, Manor Road, Oxford OX1 3UQ, UK. chrstopher.wlson@economcs.ox.ac.uk 2
1. Introducton Competton polcy and other polcy ntatves n markets as dverse as health and educaton are ncreasngly based on the presumpton that consumers can play a postve role n generatng market competton by choosng to trade wth the suppler that best suts ther needs. However, consumers may be unable to perform ths role and compettve forces may be consequently weakened for several reasons. Consumers may be unwllng to change supplers because of swtchng costs, unaware of alternatve supplers because of search costs or may face dffcultes n evaluatng and comparng dfferent supplers offers because of cogntve decson-makng costs 1. Whle prevous emprcal research has largely focussed on dentfyng the effects of swtchng costs, ths paper nvestgates the mportance of the last two possbltes by analysng emprcally the accuracy wth whch swtchng consumers choose ther best avalable alternatve suppler. We explot two ndependent datasets from the UK electrcty market where consumers have been free to swtch away from ther regonal ncumbent to one of several entrants snce the market s lberalsaton n 1999. In such a market, we would expect consumers swtchng decsons to be relatvely accurate for several reasons. Frst, almost all households consume electrcty and for many, t forms a sgnfcant part of ther household budget. Second, the market s relatvely smple as frms supply a near-homogenous good and at the tme of our surveys each suppler effectvely offered only a sngle tarff opton. Thrd, the market s transparent wth the ndustry regulator and several onlne prce comparson servces provdng many forms of advce and tarff nformaton. Yet, despte such market condtons, ths paper suggests that the naccuracy of consumers swtchng decsons remans substantal. Even when focussng only on the consumers who, when asked, ndcated that they had swtched supplers exclusvely for prce reasons, we fnd that across the two datasets and under a range of assumptons, only 8-19% of consumers swtched to the frm offerng the hghest surplus and, n aggregate, swtchng 1 See Farrell and Klemperer (2006) for a revew of the market power effects of swtchng costs, Baye et al (forthcomng) for search costs, and Gabax et al (2005) for cogntve costs. 3
consumers approprated only between 28% and 51% of the maxmum gans avalable to them. Whle such behavour s wholly consstent wth the behavour of ratonal consumers facng hgh search costs, the addtonal fndng that 20-32% of swtchng consumers appear to have lost surplus through ther choce of suppler s not. These consumers lost an average 14-35 per year n ncreased blls, apart from any other swtchng costs they may have ncurred. Very lttle prevous research has examned emprcally the swtchng accuracy of consumers. As part of a much wder nvestgaton nto the effects of entry n the New York State telephone market, Economdes et al (2006) suggest that 42% of consumers swtched to a more expensve suppler, resultng n an average loss of $4.33 per month. Gulett et al (2005) suggest there may be consumer naccuracy n the UK gas market by showng that consumers (bnary) swtchng decsons appear unrelated to the monetary gans avalable from dong so, especally for consumers who expect prce dfferences to be transtory. A larger lterature however, has analysed the wdespread potental for consumers to select a non-cost mnmsng opton from a menu of tarffs offered by the same frm. Agarwal et al (2006), for example, suggest that over 40% of consumers selected the more expensve tarff when offered the opton of two credt card contracts n a market experment by a US bank, whle Lambrecht and Skera (2006) use data from a German nternet provder to estmate that around a thrd of consumers chose a more expensve fxed rate tarff, and over half of these pad more than double the cheapest alternatve. The proposed explanatons for such choces fall nto three broad categores. Frst, consumers may show a preference for certan tarff structures, such as flat-rate fees (Lambrecht and Skera 2006). We fnd no support for such an explanaton as the gans from swtchng are largely unrelated to any assocated change n tarff structure. Second, n comparng tarffs, consumers may weght napproprately the varous components of a tarff or prce, such as the ntroductory rate, shppng charge or state-tax rate (e.g. Ausubel 1999, Hossan and Morgan 2006, Ellson and Ellson 2006, respectvely). Ths explanaton s not supported by our data whch show that the gans made by consumers who swtched to supplers offerng a potentally focal dual-supply 4
dscount are not sgnfcantly dfferent from the gans made by other consumers. Thrd, consumers may evaluate alternatve supplers tarffs usng an ncorrect predcton of ther own future consumpton (Mravete 2003, Della Vgna and Malmender 2004, 2006). Ths explanaton also appears unconvncng as all results are derved from consumers own (expendture) belefs and reman robust across consumpton varatons of plus and mnus ten percent. Hghlghted by the recent wdespread allegatons about such practces wthn the ndustry, one plausble explanaton of the results concerns the pressursng or msleadng nfluence of supplers sales actvtes. However we fnd that the accuracy of consumers choces are not sgnfcantly related to the self-reported nfluence of a sales agent; nor does an ncreased number of regonal compettors, whch mght result n ncreased sales actvty, consstently reduce the accuracy of decsons. Instead, the paper concludes that consumers swtchng naccuracy s consstent wth pure decson error. Ths fndng underlnes the mportance of the growng research nto the ncentves frms may face to explot or nduce consumer confuson see Ellson and Ellson (2005) or Armstrong and Spegler (2007) for a further dscusson. Secton 2 provdes a bref theoretcal foundaton for the measures of the gans from swtchng that are later calculated. Secton 3 ntroduces the market, the data and the calculaton procedures. The descrptve results are presented n secton 4. Secton 5 proposes some potental explanatons for the results and presents some further analyss to test them; secton 6 concludes. 2. Theory To analyse the accuracy of consumers swtchng decsons t s necessary to calculate both the actual gans n surplus that each consumer made through ther choce of new suppler and the maxmum possble gans that each consumer could have acheved by swtchng to ther best suppler (gven ther demand characterstcs). We now present some smple measures to form the bass of such calculatons. 5
Consder consumer s decson to swtch away from hs old suppler, o, to a new suppler, n, chosen from hs set of alternatve supplers, S. Assumng that consumer cares only about the tarff offered by each suppler, equaton (1) descrbes the approxmate annual gan n consumer surplus (excludng swtchng costs) from decdng to swtch from suppler o to suppler, n, n o n n n o o o CS = CS CS [ u ( C ) E( C ; T )] [ u ( C ) E( C ; T )] (1) where the consumer surplus receved at any frm j conssts of the utlty from consumng C homogenous unts of electrcty annually, u ( C j ), mnus the j j j assocated bll expendture, E( C ; T ), whch depends on frm j s tarff, Wth the use of a revealed preference argument to ensure that j T. o o o n n o u ( C ) E( C ; T ) u ( C ) E( C ; T ) an upper bound for the actual gans made from such a swtchng decson, sw x, s constructed by comparng the expendtures that would result from consumng the level of post-swtchng consumpton, n C, at each suppler, (2). Such an upper bound s very close to the approxmate change n surplus descrbed by (1) when demand s hghly prce nelastc, as n the electrcty market (Baker et al 1989). CS x sw ( n ; o ) ( n ; n E C T E C T ) (2) Smlarly an upper bound for the maxmum possble gans that consumer could have made by swtchng away from suppler o, x max, o, can be constructed by comparng the expendture at s old suppler wth the lowest possble expendture avalable from the set of alternatve supplers, S, (3). One fnal upper bound measures the gans consumer would have expected to make by randomly selectng an alternatvely suppler, x, mean o. (4) compares the expendture at s old suppler wth the average expendture across suppler s set of alternatve supplers. x, = E( C ; T ) mn E( C ; T ) 0 (3) max n o n k o k S 6
mean n o n k x, = E( C ; T ) mean E( C ; T ) (4) o k S Fully ratonal and nformed consumers who care only about the tarffs offered by each frm would select the alternatve suppler that offers the maxmum reducton n expendture, x sw = x. If consumers are ratonal but not fully max, o nformed, perhaps due to the exstence of search costs, they may be wllng to select a suppler that does not offer the maxmum reducton n tarff expendture, x sw < x. However, as consumers always retan the opton of max, o not swtchng, they should never make negatve gans and so we expect x sw [0, x ]. max, o 3. Calculatons Ths secton uses the measures constructed n secton 2 to analyse the swtchng accuracy of two sets of consumers n the UK electrcty market. After an ntroducton to the market n secton 3.1, secton 3.2 presents the data and llustrates how the UK electrcty market s partcularly well suted for such an analyss. Secton 3.3 explans how the fnal calculatons are made. 3.1 The Market Snce lberalsaton of the UK resdental electrcty market was completed n md 1999, electrcty supplers have been permtted to enter each of the fourteen regonal markets to compete wth the orgnal regonal ncumbent. Whle few new supplers chose to enter the ndustry, many regonal ncumbents took the opportunty to enter most, f not all, of the regons n whch they had not prevously been ncumbent, as dd the natonal gas suppler, Brtsh Gas. Consumers were free to swtch away from ther regonal ncumbent (or any subsequent suppler) wth twenty-eght days notce and no fnancal penalty. In the subsequent eght years about half of all energy consumers moved away from ther regonal ncumbent. 7
An example of the range of tarffs on offer to consumers s dsplayed n Table 1. As tarffs vary by regon and by tme, Table 1 presents a typcal snapshot of the tarffs offered wthn an example regon, the Mdlands, n June 2000. Supplers are oblged to offer tarffs for three possble consumer payment methods - standard credt, drect debt and prepayment, but n practce, only offered a sngle tarff per payment method 2. Supplers typcally offer two-part tarffs, wth some offerng three-part tarffs that contan an addtonal margnal rate for hgher levels of consumpton beyond some threshold. The majorty of electrcty supplers who are also actve n the gas market ncreasngly partcpate n mxed bundlng by offerng a dual-supply dscount to those consumers who choose to buy both forms of energy. Whle t s common for supplers to approach consumers drectly n the hope of persuadng them to swtch, t s rare for supplers to use upfront dscounts or ncentves. Snce lberalsaton, many nternet-based prce comparson stes have offered consumers advce n choosng between supplers. Despte the ndustry regulator and consumer body endorsng the use of several comparson stes, ther popularty remaned lmted n the perod of our studes, wth only 10% of surveyed consumers havng used them n 2003 (OFGEM 2004). 2 More recently supplers have offered a wder choce of tarffs, ncludng capped tarffs, but these were not avalable at the tme of the consumer decsons analysed here. 8
Table 1: Example Set of Tarffs (Mdlands Regon, June 2000, n pence) Payment Method: Credt Drect Debt Prepayment Dual-Supply Electrcty Suppler: Fxed Rate1 Rate2 Fxed Rate1 Rate2 Fxed Rate1 Rate2 Threshold Dscount MEB (Regonal Incumbent) 2159 6.72-2094 6.52-3734 6.72 - - - Brtsh Gas 0 10.57 5.65 0 9.01 5.65 0 10.28 6.17 900 1460 Eastern TXU Energ 2848 6.38 6.28 1856 6.38 6.28 3713 6.72-2392 - East Mdland 3541 5.99-2491 5.99-5116 5.99 - - 250 Independent 4982 5.46-4026 5.46-4497 7.77 - - - London Electrcty (1) 3048 5.86-3048 5.86-9202 7.80 - - - Northern Electrc and Gas 0 9.14 5.68 0 8.19 5.68 3990 6.52-1092 - Norweb Energ 4922 5.30-4637 5.21-3734 6.72 - - - Seeboard (2) 0 11.97 5.34 0 10.82 5.34 4112 6.72-728 - Scottsh Hydro 1873 6.08-1873 6.08-3990 6.52 - - - Scottsh Power 5408 5.26-4883 5.01-3734 6.72 - - 1050 Southern 3116 6.29-3053 6.16-3990 6.52 - - - SWALEC 1966 5.67-1886 5.44-3734 6.71 - - - SWEB 3045 5.86-2954 5.68-4523 7.39 - - - Utlty Lnk 3595 7.25-2595 7.25-7388 7.68 - - - Yorkshre 4721 5.76-4091 5.76-8669 5.76 - - - Each suppler offers a tarff across three payment methods. Each tarff conssts of an (possbly zero) annual fxed fee, Fxed, wth an addtonal margnal rate, Rate1 n pence/kwh, and, n some cases, a second margnal rate, Rate2, for consumpton over and above some annual breakpont, Threshold (n kwh). Dual supply dscounts are offered only to credt or drect debt consumers (except by East Mdland/Powergen who offer them to all consumers). Addtonal dscounts are labelled wth numbers n brackets - (1) 3% off Drect Debt f bll exceeds 10.50 (2) 8.40 off credt and drect debt. 9
3.2 Data Two datasets were constructed from two ndependent, cross-sectonal, faceto-face surveys of consumers n England, Scotland and Wales. The EA survey (Cooke et al 2001) was conducted between March and August 2000 and was ntentonally based towards low-ncome consumers 3. Of the 3417 consumers surveyed, 523 had swtched electrcty supplers and, of these, 373 had a full set of responses to questons relevant for the analyss. In contrast, the CCP survey, was desgned to be representatve of the general populaton and was conducted for the ESRC Centre for Competton Polcy n June 2005 4. Of the 2027 consumers surveyed, 370 had swtched supplers n the prevous three years, and 245 furnshed useable responses. Whle the presence of a lowncome bas and mssng nformaton lmt our ablty to draw general nferences about how swtchng behavour vares wth consumer characterstcs, we vew the measurement of swtchng accuracy wthn each of these samples as nformatve. A major constrant on the ablty to measure consumers swtchng accuracy arses from the possblty that consumers swtched for reasons other than prce. Whlst non-prce gans are lkely to be small n a near-homogeneous market lke electrcty, they may arse from two sources. Frst, although the relablty of supply s ndependent of the suppler (snce t depends upon the vertcally separated dstrbuton functon), consumers may perceve that frms vary n attrbutes such as customer servce or envronmental awareness. Second, n addton to the possble monetary benefts of beng suppled electrcty and gas by the same suppler, for whch we account for, consumers may perceve some non-prce, practcal benefts from havng to deal wth only one suppler. To elmnate these possbltes, we restrct our analyss to a subset of consumers who stated that ther swtchng decson was motvated purely by prce. Specfcally, two sub samples are created that contan 318 and 154 consumers respectvely who, when asked, cted only dfferences n 3 The EA survey and ts ntal analyss were funded by the Electrcty Assocaton an early descrpton of consumers choces and errors s contaned n Waddams Prce (2003). 4 The CCP survey was desgned to analyse search and swtchng behavour across eght dfferent product markets as analysed by Chang and Waddams Prce (forthcomng). Here, only the data from the electrcty market s used. 10
prce as a reason for swtchng and dd not menton factors such as the qualty of servce, the provson of envronmental tarffs or the practcal benefts of beng dual-suppled. A full summary of the consumers (multple) reasons for swtchng supplers s presented n Tables 2a and 2b 5. Tables 2a and 2b: Reasons for Swtchng Supplers across the Two Datasets Reason for Swtchng (EA) Mean Reason for Swtchng (CCP) Mean Cheaper 0.77 Better Prces/Rates 0.86 Dual Supply Dscounts 0.10 Better Servce/Qualty 0.19 Influence of Sales Agent 0.10 Not Satsfed wth Old Suppler 0.11 'Conned'/Unaware of swtchng 0.03 Dual Supply 0.06 Poor Servce from Old Suppler 0.03 Envronmental Tarffs 0.03 Better Servce 0.02 Other 0.10 No Standng Charge 0.01 n 245 Other 0.05 n 373 3.3 Calculatng the Gans from Swtchng Ths secton provdes further detals of how the bound measures constructed n secton 2 are used wth the selected data samples to calculate consumers swtchng accuracy. To focus only on the accuracy of consumers choce of suppler and not on the choce of payment method or gas suppler, all calculatons are made by comparng supplers relevant tarffs whlst treatng each consumer s known choce of payment method(s) and gas suppler as gven. Specfcally, the calculatons are made usng equatons (5)-(7), where the tarff of each suppler, Ttr ( m, g ), vares accordng to the consumer s date of swtchng, t, electrcty supply regon, r, choce of gas suppler, g, and choce of payment method, m, (both before and after swtchng). sw [ ˆ n o o ; (, )] [ ˆ n n n x = E C T m g E C ; T ( m, g)] (5) tr tr 5 The EA respondents were asked to provde an unstructured explanaton for why they had swtched, whch was later coded nto an exclusve lst of reasons, whereas the CCP respondents were asked to ndcate up to three reasons from a lst of possble optons. No dstncton was made between prce and non-prce benefts of dual-supply and so all consumers who cted dual-supply as a reason for changng supplers are elmnated from the sample. 11
x = E[ Cˆ ; T ( m, g)] mn E[ Cˆ ; T ( m, g)] (6) max n o o n k n, s tr tr k Sr x = E[ Cˆ ; T ( m, g)] mean E[ Cˆ ; T ( m, g)] (7) mean n o o n k n, s tr tr k Sr Usng a tme seres of the unque tarff offered by each suppler per payment method 6, an estmate of consumpton, C ˆ n, was calculated from each consumer s own estmate of ther average electrcty expendture 7. Such an approach offers two advantages. Frst, t s probably more accurate as consumers are more lkely to recall ther expendture than ther consumpton. Second, and more mportantly, all gans are calculated n a way that s consstent wth consumers own consumpton belefs, so that any naccurate consumer choces cannot be attrbuted to consumers ncorrect consumpton estmates. A potental drawback, however, comes from the possblty that each consumer s expendture belefs may have changed n the ntervenng perod between the tme of the swtchng decson and the tme of the survey. We take two approaches to allow for ths possblty and to add further robustness to the fndngs. Frst, we dentfy a subgroup of the EA consumers whose survey responses ndcated that ther consumpton was hghly prce nelastc, and stable over tme, and demonstrate that these do not dffer sgnfcantly from the rest of the sample 8. The nsgnfcant dfference supports the clams that ) the constructed upper bounds form close approxmatons to the true gans from swtchng and ) consumpton s lkely to be stable between the tme of swtchng and the tme of the survey. Second, we repeat the three measurements for all consumers usng consumpton levels whch are plus and mnus ten percent of our orgnal estmate. 6 The tarff dataset bulds on that used by Gulett et al (2005) and was obtaned by ether contactng supplers drectly or downloadng bmonthly tarffs from a consumer advce webste, www.whch.co.uk or the energy consumer body, www.energywatch.org.uk. 7 Consumers were asked to provde an estmate of ther expendture on a weekly, fortnghtly, monthly or quarterly bass as they preferred. 8 The subgroup of consumers ndcated hgh prce nelastcty by replyng the same to the followng questons: Q. If the cost of electrcty went down would you use more electrcty or use the same electrcty and use the savngs for somethng else?, and Q. If the cost of electrcty went up would you use less electrcty or use the same electrcty?, and further ndcated a stable consumpton pattern by replyng No to the followng questons, Q. Has there been any change n your household s crcumstance n the last 2-3 years that affected your fuel consumpton?, and Q. Has your household s electrcty ever been dsconnected because of unpad electrcty blls? 12
Whlst the CCP dataset s suffcently rch to provde all the requred nformaton, the EA dataset does not provde all the necessary varables drectly from the survey because of uncertanty about the exact date of swtchng and of any change n payment method. To proceed we derve the EA calculatons under the four most lkely scenaros and compare the results for robustness. Ths leads to the specfcatons, Oct99nochange, Oct99change, Jun00nochange and Jun00change, whch are detaled fully n the appendx. 4. Descrptve Results Fgure 1 plots the estmated actual gans from swtchng aganst the maxmum gans avalable for all consumers (averagng across the EA specfcatons outlned above). Two mmedate observatons can be made. Frst, many of the consumers have not approprated the maxmum gans avalable, as ndcated by the ponts located below the 45 lne. Ths s c onsstent wth the behavour of ratonal consumers facng search costs and wth expermental evdence that suggests consumers often search too lttle (Sonnemans 1998 and Tenoro and Cason 2002). Second, however, a sgnfcant fracton of swtchers appear to have actually lost surplus by swtchng to a more expensve suppler, as ndcated by the ponts below the x-axs, a fndng whch s nconsstent wth the behavour of ratonal consumers motvated to swtch only by prce. To explore the fndngs n more detal, Table 3 dsplays the man results derved from the orgnal estmates of consumpton and Table 4 ncludes the results wth the alternatve consumpton levels. 13
Fgure 1: The Actual Gans Made from Swtchng relatve to the Maxmum Gans Avalable, CCP and EA (pooled specfcaton) Datasets 200 150 Actual Gans Made (Annual, ) 100 50 0-50 0 50 100 150 200-100 -150 Maxmum Gans Avalable (Anuual, ) The results shown n tables 3 and 4 are remarkably robust across datasets, across specfcatons and across consumpton levels, provdng support for the chosen measurement methodology. Despte ncludng only decsons based exclusvely on prce, many consumers faled to swtch to the cheapest suppler. Across datasets, specfcatons and consumpton levels, the reported percentage of consumers selectng ther cheapest suppler ranges between only 8 and 19%. Although consumers as a whole made postve average gans of between 16 and 22 per annum, n aggregate, consumers approprated only between 28 and 51% of the maxmum benefts avalable to them. 14
Table 3: Descrptve Statstcs of the Gan Measures across a Range of Datasets and Specfcatons Data CCP EA EA EA EA EA Specfcaton Pooled Oct 99 no change Oct 99 change Jun 00 no change Jun 00 change Average (StDev) Average (StDev) Average (StDev) Average (StDev) Average (StDev) Average (StDev) Number of Swtchers 154 318 318 318 318 318 Average Maxmum Gans Avalable (annual, ) 49.04 (39.20) 44.22 (42.65) 43.02 (42.84) 41.42 (39.91) 47.08 (42.85) 45.35 (45.00) Average Mean Gans Avalable (annual, ) 11.43 (31.16) 8.80 (27.14) 8.62 (28.02) 7.01 (27.62) 10.64 (29.25) 8.92 (33.06) Average Actual Gans Made (annual, ) 17.92 (43.18) 19.41 (38.56) 21.36 (41.57) 19.75 (38.99) 19.13 (35.61) 17.40 (38.09) Average Mean Gans/Average Maxmum Gans 0.23 0.20 0.20 0.17 0.23 0.20 Average Actual Gans/Average Maxmum Gans 0.37 0.44 0.50 0.48 0.41 0.38 Proporton of Swtchers wth Perfect Gans 0.18 0.14 0.18 0.18 0.10 0.10 Expected Proporton f Random Alternatve Selected 0.14 0.07 0.07 0.07 0.07 0.07 Proporton of Swtchers wth Negatve Gan 0.31 (0.46) 0.25 (0.43) 0.24 (0.43) 0.26 (0.44) 0.22 (0.41) 0.29 (0.45) Average Gan gven Negatve Gan -26.96 (32.99) -17.56 (19.16) -16.78 (20.77) -19.23 (19.80) -15.76 (16.93) -18.47 (19.14) Proporton of Swtchers wth Non-Negatve Gan 0.69 (0.46) 0.75 (0.43) 0.76 (0.43) 0.74 (0.44) 0.78 (0.41) 0.71 (0.45) Average Gan gven Non-Negatve Gan 37.64 (30.55) 31.85 (35.29) 33.13 (39.27) 33.52 (34.53) 28.98 (33.24) 31.78 (34.10) Proporton of Swtchers wth Domnated Choce 0.01 0.06 0.07 0.08 0.03 0.04 Maxmum Gans Avalable refers to the change n surplus that would have been realsed by a swtcher had they swtched to ther cheapest alternatve suppler. Mean Gans Avalable refers to the change n surplus that a swtcher would expect to gan by selectng a suppler randomly. The Proporton of Swtchers wth Perfect Gans refers to the proporton of consumers who approprated all of the maxmum gans avalable. Ths s compared to the expected probablty of dong so had the consumer randomly selected an alternatve suppler. The Proporton of Swtchers wth Domnated Choce refers to the proporton of consumers that swtched to a tarff that could not be cheaper than ther prevous tarff for any level of consumpton. 15
Table 4: Comparng the Calculated Gan Measures wth the Perturbed Consumpton Levels Data CCP EA EA EA EA EA Specfcaton Pooled Oct 99 no change Oct 99 change Jun 00 no change Jun 00 change Usng Estmated Consumpton Average (StDev) Average (StDev) Average (StDev) Average (StDev) Average (StDev) Average (StDev) Average Maxmum Gans Avalable (annual, ) 49.04 (39.20) 44.22 (42.65) 43.02 (42.84) 41.42 (39.91) 47.08 (42.85) 45.35 (45.00) Average Actual Gans Made (annual, ) 17.92 (43.18) 19.41 (38.56) 21.36 (41.57) 19.75 (38.99) 19.13 (35.61) 17.40 (38.09) Average Actual Gans/Average Maxmum Gans 0.37 0.44 0.50 0.48 0.41 0.38 Proporton of Swtchers wth Perfect Gans 0.18 0.14 0.18 0.18 0.10 0.10 Proporton of Swtchers wth Negatve Gan 0.31 (0.46) 0.25 (0.43) 0.24 (0.43) 0.26 (0.44) 0.22 (0.41) 0.29 (0.45) Usng Estmated Consumpton -10% Average Maxmum Gans Avalable (annual, ) 47.47 (37.56) 42.04 (38.00) 41.17 (41.66) 40.97 (36.27) 42.44 (38.21) 43.57 (35.85) Average Actual Gans Made (annual, ) 20.76 (41.19) 18.51 (34.89) 20.72 (40.53) 19.27 (37.05) 17.42 (31.99) 16.64 (29.99) Average Actual Gans/Average Maxmum Gans 0.44 0.44 0.50 0.47 0.41 0.38 Proporton of Swtchers wth Perfect Gans 0.16 0.13 0.19 0.14 0.10 0.08 Proporton of Swtchers wth Negatve Gan 0.25 (0.43) 0.23 (0.42) 0.24 (0.43) 0.24 (0.43) 0.22 (0.41) 0.23 (0.42) Usng Estmated Consumpton +10% Average Maxmum Gans Avalable (annual, ) 53.30 (49.22) 53.23 (59.92) 44.12 (44.46) 43.88 (38.75) 51.81 (47.86) 73.09 (108.62) Average Actual Gans Made (annual, ) 17.98 (52.50) 21.36 (39.39) 22.42 (42.48) 20.82 (39.27) 21.64 (39.19) 20.56 (36.63) Average Actual Gans/Average Maxmum Gans 0.34 0.42 0.51 0.47 0.42 0.28 Proporton of Swtchers wth Perfect Gans 0.14 0.13 0.19 0.15 0.10 0.08 Proporton of Swtchers wth Negatve Gan 0.32 (0.47) 0.22 (0.42) 0.24 (0.43) 0.24 (0.43) 0.20 (0.40) 0.21 (0.41) Maxmum Gans Avalable refers to the change n surplus that would have been realsed by a swtcher had they swtched to ther cheapest alternatve suppler. The Proporton of Swtchers wth Perfect Gans refers to the proporton of consumers who approprated all of the maxmum gans avalable. 16
They have only acheved a lttle more than would have been expected by swtchng to a randomly selected suppler; ths would have offered consumers a 7 to 14% chance of pckng the cheapest suppler 9 and approprated 17-23% of the maxmum gans avalable. More startlngly, even wthout takng nto account the (fnancal or non fnancal) costs of makng the swtch, between 20 and 32% of consumers swtched to a more expensve suppler, losng, on average, approxmately 14-35 per year. Further, between 3 and 31% of these loss-makng consumers actually swtched to a domnated tarff that could not have offered them a reducton n expendture at any level of consumpton. Fnally, although t s dffcult to make robust comparsons gven the bases wthn each of the samples, our data provde no evdence that swtchng accuracy mproved over the fve years whch elapsed between the two surveys. 5. Potental Explanatons The exstence of search costs can explan why consumers dd not select the best possble suppler, but the choce of a more expensve suppler remans puzzlng. In ths secton we explore the valdty of four possble explanatons: ) consumers exhbted some bas or preference for partcular tarff structures; ) consumers were overly-attracted to supplers offerng dual-supply dscounts; ) consumers were nfluenced by msleadng sales actvty; and v) consumers made genune decson errors. Frst, we consder the possblty that consumers choces could be explaned by a bas or preference for dfferent tarff structures, as proposed n the lterature documentng consumers naccurate tarff choces (e.g. Lambrecht and Skera 2006). Whle the potental for such bases s lmted n our market due to the narrow range of avalable tarff structures, we nvestgate the potental for consumers to have dsplayed a preference for tarff structures n 9 Ths fgure was calculated by fndng the recprocal of the number of alternatve supplers, averaged across consumers, gven ther respectve regons. The probablty doubles to 0.14 for the later CCP dataset due to the heavy market consoldaton n recent years. 17
two respects - the number of parts n the tarff (two or three) and whether or not there s a postve fxed fee. The evdence for such bases seems lmted. Table A1 n the appendx ndcates that the estmated swtchng gans are largely unrelated to the choce of a two- or three-part tarff; the only weak evdence of such a bas occurs n the EA June specfcaton where the 40 consumers who swtched from a three- to a two-part tarff made sgnfcantly less accurate decsons than other swtchers. Table 2a shows that only 1% of consumers cted the exstence of a zero fxed fee as a reason for swtchng and Table A2 n the Appendx shows that the estmated swtchng gans are, for the most part, unrelated to the magntude of the chosen fxed fee. The only possblty of a bas occurs wthn the EA dataset where the 18 consumers who swtched to a postve fxed fee made sgnfcantly worse decsons. Second, we examne the possblty that consumers could have overestmated or have been overly senstve to the dual-supply dscount, as emphassed as an explanaton n other contexts by Ausubel (1999) and Hossan and Morgan (2006). Despte excludng any consumer who cted the exstence of a dualsupply dscount as a reason for swtchng, ths explanaton may seem persuasve snce 74% of the consumers n the sample who changed suppler swtched to ther gas provder. However, Table A2 n the appendx ndcates that the dual-suppled swtchers made, f anythng, hgher gans than the nondual suppled consumers, contradctng such an explanaton. Ths evdence also elmnates the potental explanaton that consumers may have swtched to ther gas suppler to receve some unmeasured non-prce beneft. Thrd, could consumers have been nfluenced by supplers ms-sellng actvty? Such an explanaton s partcularly plausble n the UK electrcty market where there have been many allegatons of ms-sellng. Whle some complants have been targeted at nternet prce comparson stes for msleadng consumers by favourng certan supplers 10, most allegatons have been amed drectly at the use of more drect ms-sellng tactcs by supplers themselves. Indeed, the problem of aggressve or msleadng cold-callng or doorstep sellng was consdered so serous that several bodes conducted 10 See http://busness.guardan.co.uk/story/0,,1975484,00.html. December 19 th 2006. 18
nvestgatons (energywatch 2002, OFGEM 2002 and OFT 2004) and OFGEM subsequently fned London Electrcty two mllon pounds 11,12. 5.1 Potental Ms-sellng In ths secton we estmate whether the consumers swtchng accuracy s related to two sets of test varables assocated wth potental ms-sellng. We analyse each n turn. Frst, we explore whether the accuracy of consumers swtchng decsons s adversely affected by the self-reported nfluence of supplers sales actvty, as captured by two dummy varables from the EA survey. These correspond to consumers ether reportng that they had been conned nto swtchng wthout ther consent, had been actve n ther swtchng decson, conned, or that a sales agent agent 13. Consumers could cte both nfluences. To analyse how these varable relate to swtchng accuracy, two procedures are used to estmate varatons of equaton (6), where the swg gans from swtchng, y *, are modelled as a functon of the two test varables agent and conned whle controllng for a vector of consumer demographcs, D, and each consumer s maxmum avalable gans, max x. 11 See http://news.bbc.co.uk/1/h/busness/2315115.stm. October 10th 2002. 12 We fnd no evdence that those consumers who swtched to London Electrcty made sgnfcantly dfferent gans to those who swtched to other supplers. 13 The CCP data do not nclude these varables 19
Table 5: Summary Statstcs of the Demographc and Test Varables Varable Name Varable Defnton Mean (StDev) hghsoc Household socal grade: A, B or C1 0.28 (0.45) mdsoc Household socal grade: C2 or D 0.49 (0.50) lowsoc Household socal grade: E 0.22 (0.42) hghnc Household ncome: 25000 + 0.13 (0.33) mdnc Household ncome: 12500-25000 0.25 (0.43) lownc Household ncome: Less than 12500 0.43 (0.50) ncref Income status refused 0.20 (0.40) age Age of respondent 44.86 (15.96) sngle The household respondent s sngle 0.15 (0.36) marred The household respondent s marred 0.62 (0.49) exmar The household respondent s wdowed or dvorced 0.23 (0.42) arrears The household has electrcty arrears 0.04 (0.21) gassw The household has prevously swtched gas suppler 0.51 (0.50) rent The household lves n rented accommodaton 0.43 (0.50) dsable The household has some form of dsablty beneft 0.19 (0.47) agent The household cted the nfluence of a sales agent 0.11 (0.31) conned The household swtched wthout consent 0.03 (0.18) n The number of regonal compettors 14.75 (0.85) Number of Observatons 318 A further varable, stable, s ncluded to nvestgate whether the measured swtchng accuracy of the sub group of consumers who reported hghly prce nelastc and stable consumpton dffers from the rest of the sample. Ths varable s later reported to be nsgnfcantly dfferent from zero, as dscussed prevously n Secton 3.3. All relevant varables are descrbed and summarsed n Table 5. y * = β + agent β + conned β + D ' β + x β + stable β + ε (6) swg max 1 2 2 3 4 5 We use equaton (6) to explore how consumers swtchng gans depend on a set of ndependent varables n two ways. In the frst case, y * s treated as a latent varable and we estmate the probablty of a consumer makng a postve gans usng a probt model, and n the second case, we model the gans from swtchng as a contnuous varable usng OLS wth heteroscedastcty-consstent standard errors. For robustness, the two estmatons are conducted across each of the four EA data specfcatons and the results are reported n Tables 6 and 7. swg 20
The self-reported ncdences of sales and connng actvty have no sgnfcant effect on swtchng accuracy across all specfcatons. The estmatons also ndcate, n lne wth the fndngs of Economdes et al (2006) and Mravete (2003), that very few demographc varables are useful predctors of the ablty of consumers to make accurate decsons. Consumers lvng n rented property make less accurate decsons, probably because they expect to enjoy any benefts for a shorter tme. Some of the specfcatons suggest that consumers wth hgher ncomes (and those who declned to reveal ther ncomes) approprate less of the avalable gans. Consumers are less lkely to make a loss from swtchng supplers f the maxmum gans avalable are hgher, a fndng consstent wth consumers havng a hgher ncentve to make an accurate decson when the rewards from dong so are greater. Table 6: Estmatons of the Probablty of Makng a Postve Gan 14 June June October October No Method Change Method Change No Method Change Method Change M.Effect z M.Effect z M.Effect z M.Effect z agent 0.03 0.53-0.16-1.62 0.08 1.39 0.04 0.61 conned -0.18-1.16-0.23-1.24 0.07 0.79-0.07-0.45 ganmax 0.00 4.23** 0.01 7.16** 0.01 5.52** 0.01 7.18** stable -0.03-0.55-0.02-0.46-0.05-1.04-0.06-1.31 hghsoc -0.01-0.11-0.07-0.74 0.02 0.21-0.08-0.89 mdsoc -0.02-0.39-0.07-1.00-0.05-0.79-0.14-2.12* hghnc -0.24-2.03* -0.22-1.78-0.13-1.21-0.16-1.37 lownc -0.05-0.69-0.04-0.55-0.03-0.43-0.09-1.40 ncref -0.09-1.13-0.11-1.21-0.08-1.05-0.10-1.17 age 0.00 0.63-0.01-0.83 0.00 0.30 0.00 0.21 age2 0.00-0.71 0.00 0.78 0.00-0.01 0.00 0.12 dsable -0.05-0.96-0.07-1.25 0.00-0.01-0.04-0.70 sngle -0.10-1.17-0.08-0.86-0.12-1.33-0.21-2.07 exmar 0.01 0.09 0.03 0.50 0.02 0.29 0.02 0.29 rent -0.15-2.87** -0.16-2.58** -0.10-1.93-0.14-2.55** arrears 0.03 0.27-0.01-0.05 0.09 1.29 0.08 1.02 gassw -0.12-2.77** -0.12-2.44* -0.05-1.20-0.04-0.84 n 318 318 318 318 Log-Lk -141.7-145.6-144.3-137.0 LR(17) 51.90** 89.65** 58.78** 91.07** McF R2 0.15 0.24 0.17 0.25 14 All sgnfcant tests are ndcated by * for the 5% level and by ** for the 1% level. Where applcable, all margnal effects are calculated for the average swtcher relatve to the base case of a consumer who s marred, of low socal class and wth mddle ncome. 21
There s no evdence that prevous experence mproves decson accuracy. Whle Gulett et al (2005) suggest that consumers are more lkely to swtch n a gven market f they have prevously swtched n others, we fnd that a past experence of swtchng gas supplers does nothng to mprove (and sometmes reduces) swtchng accuracy. To provde a further (less drect) test of the effects of ms-sellng, the estmatons are repeated wth the ncluson of a dfferent test varable - the number of compettors n each consumer s regonal market. Whle conventonal theores of consumer search do not predct any negatve relatonshp between consumers ablty to approprate the gans avalable Table 7: Estmatons of the Gans Made From Swtchng 15 June June October October No Method Change Method Change No Method Change Method Change Coeff t Coeff t Coeff t Coeff t agent -0.70-0.14-2.57-0.49-2.10-0.41-3.34-0.58 conned 0.22 0.05-0.03-0.01-3.64-0.45-3.75-0.49 ganmax 0.01 9.43** 0.01 11.05** 0.01 14.24** 0.01 10.55** stable 0.93 0.31 0.79 0.27-1.22-0.42-1.54-0.54 hghsoc -4.21-0.90-2.98-0.61-2.26-0.56-2.28-0.54 mdsoc -3.88-1.00-3.91-0.95-3.08-0.85-4.45-1.16 hghnc -13.90-2.21* -13.23-2.08* -1.08-0.19-0.36-0.06 lownc -5.12-1.39-5.80-1.50 1.89 0.52 1.55 0.41 ncref -13.57-3.22** -13.73-3.22** -6.87-1.57-5.63-1.41 age -0.02-0.04-0.27-0.51 0.39 0.81 0.27 0.55 age2 0.00 0.18 0.00 0.66 0.00-0.25 0.00 0.04 dsable -4.87-1.30-4.52-1.16-6.53-1.77-6.30-1.71 sngle -5.66-1.25-4.94-1.06-0.33-0.08-3.25-0.75 exmar -0.49-0.16-0.33-0.10 0.16 0.05-0.44-0.13 rent -6.08-2.17* -4.54-1.58-8.40-2.77** -7.71-2.46* arrears -8.98-1.21-8.22-1.08-4.17-0.66-4.48-0.72 gassw -3.92-1.33-3.44-1.15-4.27-1.53-3.32-1.20 constant 5.28 0.38 7.29 0.52-15.52-1.23-12.03-0.92 n 318 318 318 318 F(17,300) 10.34** 14.06** 18.37** 14.06** McF R2 0.58 0.61 0.68 0.64 15 All sgnfcant tests are ndcated by * for the 5% level and by ** for the 1% level. Where applcable, all coeffcents are estmated relatve to the base case of a consumer who s marred, of low socal class and wth mddle ncome. 22
and the number of compettors 16, t s reasonable to conjecture that ms-sellng strateges may be more attractve to frms as the profts from more standard forms of competton are reduced from ncreases n the number of supplers. In a related sense, recent work by Spegler (2005) llustrates how frms face an ncreased ncentve to obfuscate by ncreasng the varance of ther utlty offers when faced wth more compettors, whle Mravete (2007) offers evdence to suggest that frms are more lkely to employ domnated tarff optons when competton ncreases. To test for such an effect, we explot the fact that the number of regonal compettors vared between twelve and sxteen at the tme of the EA survey 17. If ms-sellng were an explanaton, consumers would make less accurate decsons n regonal markets wth a hgher number of competng supplers 18. Formally, the two estmaton procedures are repeated wth the replacement of the prevous test varables, agent and conned, wth the new test varable, n, measurng the number of regonal supplers faced by each consumer 19. As the estmated coeffcents dffer very lttle from those prevously reported, only the effects of the test varable are dsplayed n Tables 8 and 9. 16 Indeed, for any gven prce dstrbuton and cost of search, a consumer should accept any dscovered prce below the optmal reservaton prce whch s defned ndependently from the number of frms (Kohn and Shavell 1974). 17 These numbers refer to the number of large frms that were patronsed by consumers n the EA sample and do not nclude some smaller frms that also operated across all regons. Includng such frms n the estmatons ncreases the number by a constant and does not affect our qualtatve results. No such varaton n frm numbers exsted at the tme of the CCP survey due to later market consoldaton. 18 It s feasble, but unlkely gven the lmted varaton n the number of frms, that consumer naccuracy may also be prompted by a choce overload effect from the ncreased complexty of the decson (e.g. Iyengar and Lepper 2000 and Iyengar and Kamenca 2007). 19 Both the number of compettors and the maxmum gans can be ncluded as explanatory varables, snce they have a neglgble correlaton of approxmately 0.02 across specfcatons. 23
Table 8: Estmated Margnal Effects of the Number of Regonal Compettors on the Probablty of Swtchng to Make a Postve Gan 20 June June October October No Method Change Method Change No Method Change Method Change M.Effect z M.Effect z M.Effect z M.Effect z n -0.01-0.54 0.03-0.96-0.04-1.43-0.05-1.77 Table 9: Estmated Margnal Effects of the Number of Regonal Compettors on the Actual Gans Made from Swtchng June June October October No Method Change Method Change No Method Change Method Change Coeff t Coeff t Coeff t Coeff t n -3.76-2.47* -3.84-2.47* -1.87-0.99-2.66-1.36 Whle there s no evdent relatonshp between the number of regonal compettors and the probablty of makng a postve gan by swtchng, Table 9 suggests that n two out of four specfcatons, consumers approprated relatvely less of the maxmum avalable gans n regons wth a hgher number of supplers. However as much of the varaton n the number of regonal compettors arses, however, from the relatve lack of market entry n the two Scottsh electrcty regons, such a fndng s also consstent wth the presence of some unobserved characterstc of frms or consumers wthn the Scottsh markets. The results are therefore unclear and do not provde drect evdence that ms-sellng explans the naccuracy of consumers swtchng decsons. The evdence presented n ths secton does not ndcate that consumers poor swtchng choces are explaned by tarff bases or supplers ms-sellng actvty. We deduce that much of the swtchng naccuracy results from genune consumer confuson and decson error. 20 Sgnfcance s denoted at 5% by * and at 1% by **. 24
6. Concluson Usng two ndependent datasets from the UK electrcty market our results show that the capacty of consumers to choose effcently between supplers may be lmted, even when swtchng purely for prce reasons. Whle the results are not necessarly representatve of the general populaton, our estmatons show that, at best, a ffth of the consumers n our samples actually lost surplus as a result of swtchng; and that, n aggregate, swtchng consumers approprated only half of the maxmum gans avalable to them. Such a falure of consumers to compare accurately between alternatve supplers can damage ther welfare, both drectly n lost savngs, and ndrectly by delverng frms wth a source of market power. Indeed, together wth the well establshed effects of swtchng costs n reducng the wllngness of consumers to swtch supplers, such behavour may serously mpede the compettve process, even after a market has been lberalsed or made subject to standard competton polcy (as recently argued by Waterson 2003). We have examned and rejected several explanatons of consumer errors, ncludng preferences for partcular tarff structures or dual fuel supply, and msleadng sales actvtes by frms. Instead, despte the apparent smplcty and transparency of the market, consumers poor choces seem more consstent wth an explanaton of pure decson error. Ths fndng casts doubt on the ablty of consumers to generate compettve forces through accurate swtchng decsons and rases many mportant polcy concerns. Future research would be valuable n understandng how competton and consumer authortes should respond to consumer errors, f at all, and n nvestgatng the mplcatons for current polces amng to ncrease competton n less famlar markets, such as health and educaton. 25
References: Agarwal S., Chomssengphet S., Lu C. and Souleles N.S. (2006) Do Consumers Choose the Rght Credt Contracts? Federal Reserve Bank of Chcago Workng Paper 2006-11 Armstrong M. and Spegler R. (2007) Boundedly Ratonal Consumers June 2007 Ausubel L.M. (1999) Adverse Selecton n the Credt Market Workng Paper Unversty of Maryland Baye M.R., Morgan J. and Scholten P. (forthcomng) Informaton, Search and Prce Dsperson n Handbook on Economcs and Informaton Systems, Elsever Baker P., Blundell R. and Mcklewrght J. (1989) Modellng Household Energy Expendture Usng Mcro-Data Economc Journal vol.99 p.720-738 Chang Y. and Waddams Prce C. (forthcomng) Search and Swtchng Across Relatonshp Markets ESRC Centre for Competton Polcy Workng Paper Cooke D., Ferrar A., Gulett M., Sharratt D. and Waddams Prce C. (2001) Affordng Gas and Electrcty: Self Dsconnecton and Ratonng by Prepayment and Low Income Credt Consumers and Company Atttudes to Socal Acton www.ccp.uea.ac.uk/publcatons.asp#2001 Della Vgna S. and Malmender U. (2004) Contract Desgn and Self-Control: Theory and Evdence Quarterly Journal of Economcs vol. 119(2) p.353-402 Della Vgna S. and Malmender U. (2006) Payng Not to Go to the Gym Amercan Economc Revew vol. 96 (3) p. 694-719 Economdes N., Sem K. and V.B. Vard (2006) Quantfyng the Benefts of Entry nto Local Phone Servce NET Insttute Workng Paper, August 2006 Ellson G. and Ellson S.F. (2005) Search, Obfuscaton and Prce Elastctes on the Internet Workng Paper February 2005 Ellson G. and Ellson S.F. (2006) Internet Retal Demand: Taxes, Geography and Onlne-Offlne Competton Workng Paper Aprl 2006 energywatch (2002) Doorstep and Telephone Sellng: A Gude for Gas and Electrcty Consumers Informaton Sheet Farrell J. and Klemperer P. (2006) Coordnaton and Lock-In: Competton wth Swtchng Costs and Network Effects Forthcomng n 26
Handbook of Industral Organzaton Vol 3, M. Armstrong and R. Porter (eds), North-Holland Gabax X., Labson D. and L H. (2005) Extreme Value Theory and the Effects of Competton on Profts Workng Paper March 2005 Gulett M., Waddams Prce C. and Waterson M. (2005) Consumer Choce and Competton Polcy: A Study of UK Energy Markets Economc Journal vol. 115 p.949-968 Hossan T. and Morgan J. (2006a) Plus Shppng and Handlng: Revenue (Non) Equvalence n Feld Experments on ebay Advances n Economc Analyss and Polcy vol. 6(2) Iyengar S.S. and Lepper M. (2000) When Choce s Demotvatng: Can One Desre Too Much of a Good Thng? Journal of Personalty and Socal Psychology vol.79 p.995-1006 Iyengar S.S. and Kamenca E. (2007) Choce Overload and Smplcty Seekng Workng Paper February 2007 Kohn M.G. and Shavell S. (1974) Optmal Adaptve Search Journal of Economc Theory vol. 9 p.93-123 Lambrecht A. and Skera B. (2006) Payng Too Much and Beng Happy About It: Exstence, Causes and Consequences of Tarff Choce Bases Journal of Marketng Research, forthcomng, May 2006 Mravete E.J. (2003) Choosng the Wrong Callng Plan? Ignorance and Learnng Amercan Economc Revew vol. 93(1) p.297-310 Mravete E.J. (2007) The Doubtful Proftablty of Foggy Prcng Workng Paper July 2007 OFGEM (2002) OFGEM Keeps up Pressure on Supplers to Reduce Mssellng The Offce of Gas and Electrcty News, Press Release 30 January 2002 OFGEM (2004) Domestc Compettve Market Revew Aprl 2004 OFT (2004) Doorstep Sellng: A Report on the Market Study Offce of Far Tradng, May 2004 Sonnemans J. (1998) Strateges of Search Journal of Economc Behavor and Organzaton vol.35, p.309-332 Spegler R. (2005) Competton Over Agents wth Boundedly Ratonal Expectatons Theoretcal Economcs vol. 1(2) p.207-231 27
Tenero R. and Cason T.N. (2002) To Spn or Not to Spn? Natural and Laboratory Experments from The Prce s Rght Economc Journal vol.112 p.170-195 Waddams-Prce C. (2003) Spolt for Choce? The Costs and Benefts of Openng UK Resdental Energy Markets Workng Paper December 2003 Waterson M. (2003) The Role of Consumers n Competton and Competton Polcy Internatonal Journal of Industral Organzaton vol.21 p. 129-150 28
Appendx 1: Identfyng tarffs for the EA dataset Two aspects of the EA dataset make t dffcult to dentfy drectly the exact set of tarffs relevant for each consumer s swtchng decson. The frst s the exact date of the swtchng decson. (Economdes et al (2006) faced the same problem and were forced to assume that consumers had swtched at the date of nformaton collecton.) The second problem arses from the tmng of the change n payment method for the 32% of consumers who reported such a change. To calculate the gans on swtchng we need to know whether they, changed ther payment method before, after, or at the same tme as they swtched supplers. To resolve these uncertantes and to enhance the robustness of our fndngs we report the results over four dfferent specfcatons. As the EA survey was conducted n March-August 2000, very soon after lberalsaton, consumers could have swtched usng one of only four possble tarff sets, namely those commencng n June 1999, October 1999, Aprl 2000 and June 2000. Consumers are most lkely to have swtched under ether the October 1999 tarffs, as these were stable for the longest perod (October 1999 -Aprl 2000), or the June 2000 tarffs, as the proporton of consumers swtchng supplers was rsng over the perod. Usng both of these tme perods, the calculatons are then made under two further assumptons to provde a total of four specfcatons. These two assumptons concern whether the 32% of consumers who had changed ther payment method, changed ether before they swtched supplers (the consumers traded wth both ther orgnal and current suppler under ther current payment method) or, perhaps more realstcally, at the tme of swtchng (the consumers traded wth ther orgnal suppler usng ther prevous payment method but traded wth ther current suppler under ther current payment method) 21. The four specfcatons are respectvely labelled as Oct99nochange, Oct99change, Jun00nochange and Jun00change (see appendx for further detals). 21 The most commonly reported method changes are movng from credt to drect debt (41%) and credt to prepayment (38%). We do not allow for the unlkely possblty that the change was made after the process of changng supplers. 29