Bidder Behaviour in Swedish Simultaneous Procurement Auctions

Transcription

1 Bdder Behavour n Swedsh Smultaneous Procurement Auctons Sofa Lundberg Centre for Regonal Scence, CERUM Umeå Unversty SE Umeå, Sweden e-mal: sofa.lundberg@econ.umu.se Abstract Gven no capacty constrants and the same set of compettors one would expect the same dentty of the bdder havng the lowest cost to complete all contracts smultaneously auctoned. Results based on bd level data from Swedsh procurement auctons of nternal cleanng servce contracts suggests otherwse. The same bdder s found on smultaneous auctoned contracts although ths bdder has not submtted the lowest bd on all contracts. A possble explanaton s that bdders submt aggressve bds on some contracts n order to be gven other contracts wth less aggressve bds and thereby maxmzng the total proft. JEL: D44, H57 Fnancal support from the Swedsh Competton Authorty as well as valuable comments from Peder Axensten, Johan Lundberg, and Lars Westn s gratefully acknowledged.

2 1 1. Introducton Ths paper provdes an emprcal test of the hypothess that a bdder that wns at least one contract wthn a smultaneous procurement aucton wth an aggressve bd ncreases the probablty to wn another contract wth a less aggressve bd (.e. not the lowest submtted bd). As n the rest of the European Unon (EU), publc contracts n Sweden are auctoned by the means of sngle and multunt frst-prce sealed bd auctons. It s the law governng these procurements that makes the feature of the smultaneous procurement auctons descrbed above possble. 1 Contracts can be gven accordng to two award crtera, the lowest bd or the bd motvated to be the economcally most advantageous bd wth respect to certan crtera. It s observed n Swedsh procurement auctons that bdders mx aggressve bds wth less aggressve bds on contracts auctoned wthn the same procurement. In procurements of non-complex contracts and gven that there are no capacty constrants and the same set of bdders t s reasonable to expect the same dentty on the bdder havng the lowest cost to complete all the contracts. However, gven the two award crtera, a mxng of bds could be proft maxmzng (hgher proft on contracts won at a less aggressve bd) and rsk reducng (aggressve bds on some contracts ncreases the probablty of gettng at least one contract). Ths strategy s relevant gven the assumpton that the contractng entty fnds t benefcal to contract one or a few bdders. If the probablty to get a contract at the economcally most advantageous bd s hgher f the same bdder s awarded at least one other contract auctoned smultaneously at a low bd, the competton stuaton s affected and the resultng allocaton could be neffcent. As such, ths study has clear polcy mplcatons for the bdders, as well as the agents n the local government who evaluate the bds. The data used n ths study orgnate from procurements of nternal cleanng servce contracts on the local government level n Sweden durng the perod 1992 to The data conssts of 5,303 bds placed on 663 contracts. A logstc regresson s used to estmate the determnants of the probablty of wnnng a contract. 1 See the Publc Procurement Act n Sweden, LOU (1992:1528).

3 2 From prevous emprcal studes on procurement auctons we have knowledge about among other thngs the mpact of colluson. Examples may for example be found n an applcaton on spectrum auctons n the U.S by Cramton and Schwartz (2002) and hghway constructon procurements n the U.S, by Gupta (2001). Furthermore, Bajar, McMllan, and Tadels (2003) provde an emprcal paper on the use of auctons versus negotatons n procurements of constructon works n the U.S, and n another paper Gupta (2002) studes competton and bdder nteracton usng the same hghway constructon data as n Gupta (2001). The effect of the wnner s curse and competton on bds concernng hghway work, mantenance, and road pavng n the U.S, are analyzed n Hong and Shum (2002). Bddng behavour usng electrcty procurement data from England and Wales s studed by Wolfram (1998). Ths s also the topc n Jofre-Bonet and Pesendorfer (2000) but n a repeated hghway procurement aucton n the U.S. In a paper on U.S. procurement auctons of ol and gas leases Hendrcks, Pnkse, and Porter (2003) provde evdence of bdders takng the wnner s curse nto account when placng ther bds. Data from these procurements are also used by Hendrcks and Porter n a seres of studes of asymmetrc nformaton (1994, 1993, 1988, and 1987) as well as jont bddng (1992). The current paper s, to my notce, the frst to study the effect on bdder strateges from the two award crtera appled n procurement auctons wthn the EU. Lundberg (2005) emprcally analyses dfferences n wnnng bds due to the two award crterons wth the same data as n the current paper and fnds that wnnng bds on lowest bd offered contracts are sgnfcantly lower than contracts awarded under the economcally most advantageous bd crteron. The remander of ths paper s organzed as follows. The next secton gves a presentaton of the nsttutonal settng surroundng Swedsh procurement auctons and a bref presentaton of the theory underlyng the emprcal analyss. The followng two sectons descrbe the data and the emprcal model followed by a presentaton and dscusson of the results. Fnally, a concludng secton ends the paper.

4 3 2. Stylzed facts regardng Swedsh procurement auctons and bdder behavour Swedsh local governments use sealed bd auctons n lne wth the frst-prce sealed bd aucton to allocate nternal cleanng servce contracts. The bdders submt sealed bds that are opened and evaluated at a predetermned date. The wnner s pad n accordance wth her bd. The phrase n lne wth s used snce there are two award crtera that may be appled. The contract can go to the lowest bdder or to some other but the lowest bdder motvated to be the economcally most advantageous bd wth respect to prce, and for example, qualty, servce, and mantenance. Although the evaluatons crtera are known to the bdders pror to the bds are placed, the weght attached to each crtera s unknown. 2 The bds are pure prce bds and t s not possble to separate the bd nto a prce and qualty share. The contracts are well defned and the bdder has nformaton about the nternal cleanng frequency, square meter to be cleaned, and the cleanng products to use before the bds are placed. Therefore t s reasonable to assume that the bdders know ther cost to complete the contract wth certanty. However, the uncertanty about the award rule could affect the bdders behavour. Results provded n Lundberg (2005) show that the award rule affects the wnnng bds. Under the economcally most advantageous award rule wnnng bds are sgnfcantly hgher than wnnng bds on contracts allocated accordng to the lowest bd rule. Ths becomes even more nterestng f the aucton format s taken nto consderaton. The contracts can namely be auctoned n sngle or multunt (smultaneous) auctons and the latter format adds another dmenson to the uncertanty regardng the award rule. Under the smultaneous format there s no combnatoral bddng and one bd cannot be condtonal on other bds n the same procurement. The bds are therefore supposed to be smultaneously submtted ndependently of each other. In the case of nternal cleanng servce contracts procured by local governments on the muncpalty level, t s observed that bdders mx aggressve bds on some contracts wth hgher bds on the other contracts smultaneously auctoned n the same procurement. Ths mght seem a bt strange. If a bddng frm has the lowest cost to complete one contract t s lkely to have the lowest cost to complete the other 2 Ths stuaton s vald for the procurements n the present paper.

5 4 contracts as well. 3 On the other hand, there could be a restrcton on how many contracts the frm can handle gven that t has a capacty constrant leadng to less aggressve bddng on some contracts accordng to some nternal predetermned rankng. If the bdder s awarded such a contract t s worth the extra effort to accomplsh t, due to the hgher proft. However, the capacty constrant for an nternal cleanng servce contract would be an access to cleaners whch, due to a labour market characterzed of unemployment, seem unlkely. Therefore, t s argued here that f the contractng entty fnds t benefcal to contract one or a few frms on a number of contracts, B, wthn the procurement, then aggressve bds on at least one contract could be the an opportunty for a bdder to be gven all the contracts (or at least more than one). Consequently, placng hgher bds on the other contracts can thereby ncrease the total proft, compared to a scenaro wth aggressve bds on all contracts. Gven ths argument a bdder, where = 1,..., n, strategcally submts aggressve bds n accordance wth the standard frst prce sealed bd aucton ( b ), on at least one contract and hgher bds than that, ( b > b ) on other contracts auctoned smultaneously n order to maxmze the expected proft. 4 Both bds are assumed to correspond to the same qualty and cost to complete the contract ( c = c ). The cost s assumed to be prvate. If there s a mx of award rules wthn the procurement, a mx of bds based on ths assumpton could ncrease the probablty to wn at least one of the contracts and thus gve hgher proft on contracts won wth a b -bd. Assume that bdder can submt b -bds on a subset K ( K B ) of the total set of contracts B n combnaton wth b -bds on another subset F ( F B, F ) of the total set of contracts B. Bdder may as an alternatve choose to submt b -bds on the total set of contracts ( F K =, F K = B ). Gven that bdder s assgned all the contracts that she has placed a bd on, the expected proft from mxng aggressve and less aggressve bds s; 3 Ths presumes that the bdders face the same set of compettors on all contracts wthn the same procurement. 4 See, for example, Vckrey (1961), Laffont, Ossard, and Voung (1995), Laffont (1997), Krshna (2002), and Mlgrom (2004) for theoretcal presentatons of the frst prce sealed bd aucton.

6 5 (1) E [ ] 1 π = ( bj cj ) + ( b j c j ) j F j K for j B, F B,and K B Under the same assumpton the expected proft from submttng bds accordng to the bddng strategy gven by the frst-prce sealed bd aucton the expected proft to bdder s; (2) E [ ] 2 π = ( b j B for j B j c ) j Snce b and c c b > = the expected proft accordng to expresson (1) s hgher than the expected proft accordng to expresson (2) (3) E [ ] E [ π ] 1 π > 2 The bdder also has to decde whch contract to bd aggressvely and less aggressvely on based on, her belef about the award rule. However, ths s not modelled here because t s not necessary for testng the hypothess. 3. Data from Swedsh procurement auctons In order to test the hypothess about bdder behavour, feld data from smultaneous Swedsh procurement auctons of nternal cleanng servce contracts are used. The contracts are fxed prced, well defned, and not complex. The data orgnate from the perod and are rch on nformaton about the bdders, contract specfcs, procurement procedure, and muncpalty characterstcs. All the 289 Swedsh muncpaltes receved a request askng them for documents concernng nternal cleanng servce procurements. 5 The response rate was 79.5 percent and about 22 percent of the respondents had actually procured nternal cleanng servces durng the tme frame. The rest had n-house producton wthout procurement. It s a poltcal decson n Sweden whch type of servces the local government should produce by 5 The contract notce, techncal specfcaton, lst of tenders, and decson protocol.

7 6 them selves and whch to put up for competton. If an n-house bdder competes for the contract, ths bdder should not be favoured n any way. The data s organzed n three levels; there are 50 procurements and 664 contracts on whch 5,303 bds were placed. The number of smultaneously auctoned contracts vares between 2 and 74 and the bd level data wll be used n the present paper. There are dfferent contract types n the data for dfferent premses, such as for example a school, a day care centre, a medcal health care centre, or an offce. Each of the premses (buldng type) s contracted separately. The contract perod ranges from a couple of months up to four years. The varables used n the emprcal analyss are categorsed nto procurement characterstcs, the bdders, and muncpalty characterstcs. The causal varable s a dummy varable (Advantage) capturng the effect of a bdder wnnng at least one contract wth a lowest bd on the probablty that the same bdder wll wn another contract auctoned n the same procurement wth an economcally most advantageous bd. Ths dummy varable takes the value one f ths s the case (that s, wns wth a b -bd gven that has won another contract n the same procurement wth a b -bd) and zero otherwse. Descrptve statstcs show that 252 of the 664 contracts were assgned to a bdder accordng to the economcally most advantageous award crteron, and who also was assgned a contract n accordance wth the lowest bd crteron. The procurement characterstcs are the contract perod and dummy varables for the contract type (the premses to be cleaned) where a school s the reference category. It s very common that the contracts specfy a contract perod and a prolongaton perod (usually one or two years). If everythng runs smoothly durng the contract perod, the contract s automatcally prolonged n accordance wth the stpulated prolongaton perod. Snce ths s the expected outcome, the contract perod varable s the sum of the contract and prolongaton perod. Varaton n the number of contracts auctoned smultaneously s also consdered. On average, 26 contracts were smultaneously auctoned. See Table 1 for descrptve statstcs on the contract and prolongaton perod and number of contracts.

8 7 The bdders are heterogeneous. There are fve natonal frms, the n-house producton as well as a number of local frms that compete wth each other. Instead of usng frm specfc dummy varables as n Bajar and Hortacsu (2003), the effect of bdder heterogenety s captured by bdder type dummy varables. Three dummy varables are constructed, one that takes the value one f the bdder s one of the fve natonal frms and another dummy varable that takes the value one f the bd s placed by the n-house producton. Although each muncpalty has ts own n-house producton, these are all consdered to belong to the same bdder category. The local frms are the reference category. Almost 30 percent of the bds were placed by a natonal frm and about 7 percent by the n-house producton. In order to capture the effects of competton, the number of bdders s ncluded n the analyss. It s reasonable to assume that the probablty that bdder wns s negatvely affected by the number of bdders. Ths effect s assumed to be non lnear. The average number of bdder per contract s approxmately 8 (see Table 1 for more detals). One contract attracted only one bdder. Ths case s excluded from the estmatons. The most obvous determnant f a bddng frm wll become a wnner s the bd t has placed. The theoretcal predcton s a negatve relatonshp between wnnng optons and the bd level but snce we have the two award crtera ths s not self-evdent. As dsplayed n Table 1, there s a hgh degree of varaton n the bd varable. Ths s measured as the annual prce per square meter to be cleaned n the 1994 prce level. 6 Fnally there s a dummy varable (Prevous wnner) that takes the value one f the bdder s a frm that has been contracted by the local government n one or more prevous procurements. A prevous wnner placed approxmately 23 percent of the bds. The n-house producton s the most successful bdder n terms of contract won n relaton to number of contracts bd on; t wns n 55 percent of the cases. See Table A1 n the appendx for more nformaton about success rato gven bdder type. The muncpaltes are characterzed by the poltcal characterstc, populaton densty, and unemployment rate. These varables are assumed to measure dfferences n the bddng envronment. The poltcal characterstc (Red) s the share of seats n the 6 The bds are normalzed to a cleanng frequency correspondng to 260 days per year.

9 8 local councl assgned to the socal-democratc and left wng partes measured n percent. The populaton densty s computed as the populaton dvded by land area and the unemployment rate s the unemployment per capta n percent. Table 1. Descrptve statstcs Varable N Mn Max Mean Standard devaton Bd 5, Number of bdders Number of contracts Contract perod Prolongaton perod Red Populaton densty Unemployment rate Emprcal model The prevous dscusson about bdder strategy bols down to the followng hypothess to be emprcally tested; The probablty that bdder wns a contract based on the economcally most advantageous award rule s hgher, gven that bdder, s assgned at least one other contract n accordance wth the lowest bd award rule auctoned smultaneously. Ths wll be tested wth a bnary choce model, where y = 1 f the bdder s a wnner and zero otherwse. Ths probablty s estmated wth a bnary logt model (4) Pr [ y = 1 z ] = Λ( β z ) k k β exp = 1+ exp zk β zk where β s a parameter vector and Λ s the logstc cumulatve dstrbuton. Contract specfcatons, characterstcs of the bddng envronment, bdder heterogenety, and muncpalty characterstcs as specfed n the prevous secton are ncluded n the z vector. The margnal effects are computed accordng to 7 7 See Greene (2000) and Las (1994) for a descrpton of bnary choce models.

10 9 (5) Λ z ( β z ) k k = Λ ( β z ) 1 Λ( β z ) k ( )β k 5. Results Parameter estmates from the regresson of expresson (4) are presented n Table 2 below. The results suggest that the probablty that a bdder wll wn a contract wth an economcally most advantageous bd s sgnfcantly hgher f the bdder has won another contract n the same procurement to a lowest bd, and as such the hypothess can not be rejected. When t comes to the procurement characterstcs t s evdent that the contract type s not relevant for the probablty that a bdder wns a contract n comparson wth school contracts. Results from a Wald test show that t s easer to wn an offce contract than a chld-care centre contract, whch n turn s easer to wn than another contract. The test statstcs are found n Table A2 n the Appendx. The number of contracts smultaneously auctoned matters; the odds of becomng a wnner are hgher the more contracts that are auctoned smultaneously. The model further suggests that bdders are not equally lkely to become a wnner. Both natonal frms and the n-house producton are sgnfcantly more lkely to wn a contract than the local frms. Moreover, from a Wald test (the χ 2 (1) statstc s ) t s evdent that the n-house producton s more lkely to wn compared to the natonal frms. There s a clear compettve effect concernng the lkelhood of becomng a wnner. The more bdders the harder ths wll be. Based on the fact that the parameter for number of bdder squared s postve and sgnfcant one may conclude that ths effect s decreasng. The bd parameter has a sgnfcant negatve effect on the probablty of gettng a contract. The lower the bd the more lkely t s that the bdder wll wn a contract. The model gves no evdence of the assumpton that hstory matters; bdders who have been contracted by the same local government n prevous procurements are equally lkely to wn a contract as any other bdder. Bdders possblty to be awarded a contract s also sgnfcantly affected by the bddng envronment. The muncpalty characterstc parameters are negatve and sgnfcant.

11 10 Table 2. Results from maxmum lkelhood estmaton. N=5,302 Logt model Fxed effect model Varable β t-value β t-value Constant Advantage Procurement characterstcs School Chld care centre Medcal health care centre Offce Other Contract perod Number of contracts The bdders Local frms Natonal frms In-house producton Bd Number of bdders Number of bdders squared Prevous wnner Muncpalty characterstcs Red Populaton densty Unemployment rate LogL LogL 0 (The constant only) χ (16) / χ 2 (38) McFadden nformaton crteron The explanatory power of the model s good. The χ 2 (16) -value s well above the crtcal value and as such, the model cannot be rejected. The McFadden nformaton crteron suggests the same concluson. The estmated probablty that a bdder wll wn a contract s approxmately The percentage of correct predctons s 91 (see Table 3). Table 3. Actual and predcted values (wnner=1) Predcted Total Actual , , Total 4, ,302 8 The McFadden nformaton crteron s computed accordng to1 L / L0. See Greene (2000).

12 11 The margnal effects, computed accordng to expresson (3), show a small but sgnfcant negatve margnal effect of competton and postve (tny) margnal effect of bd level. There s clear margnal effect on the probablty of wnnng a contract from the advantage varable. See Table A3 n the appendx for all margnal effect estmates. The correlaton between the unemployment rate and the poltcal stuaton (Red) s hgh (0.77), therefore two separate models have been estmated, wth each of these varables. The results proved the model as t s specfed above to be robust for the model specfcaton. In order to allow for muncpal specfc effect not ncluded n the data, a fxed effects model s estmated. Snce estmaton of a logt model wth fxed effects requres the observatons to be adjacent, dummy varables for each of the muncpaltes are created. However, the ncluson of the fxed effects does not affect the parameter estmates. See Table 2 for presentaton of the results. 6. Conclusons Ths paper focuses attenton on bdder behavour wthn Swedsh procurement auctons. The man ssue has been to emprcally evaluate the hypothess that bdders mx aggressve bds (n accordance wth the standard sealed bd frst-prce aucton) and less aggressve bds than that on nternal cleanng servce contracts n order to maxmze profts. The man reason for applyng ths bddng strategy s a belef that an aggressve bd on at least one contract ncreases the probablty to wn smultaneously auctoned contracts to a less aggressve bd. Ths belef s motvated by an assumpton that f the contractng entty fnds t benefcal to contract one or a few frms nstead of a bundle of bdders then aggressve bds on some contracts could lead to gettng other contracts wthn the same procurement to less aggressve bds. The profts from the latter are hgher than from the former. The hypothess s by an emprcal analyss based on Swedsh procurement data regardng nternal cleanng servce contracts concluded to be accurate. Conclusvely, the award crteron rules dd affect the bddng behavour n these procurements. The data covers the 1992 to 1998 perod and conssts of 5,303 bds placed on 663 contracts.

13 12 The bddng behavour descrbed above accords wth the law governng these auctons. Publcly auctoned contracts wthn the EU can be awarded to the lowest bddng frm or the frm that has placed a bd consdered beng the economcally most advantageous one. Ths s a central part of the procurement wthn the EU wth clear effects on the outcome, possbly resultng n neffcent contract allocatons where more research s needed. A natural extenson of the present paper s the ncluson of systematc bddng behavour regardng whch type of bd to place on whch contract. References Bajar, Patrck and Al Hortacsu Are Structural Estmates of aucton Models Reasonable? Evdence from Expermental Data. Natonal Bureau of Economc Research. Inc. NBER Workng papers Bajar, Patrck, Robert McMllan, and Steven Tadels Auctons versus Negotatons n Procurement: An Emprcal Analyss. Natonal Bureau of Economc Research, Inc, NBER Workng Papers: Cramton, Peter and Jesse A. Schwartz Collusve Bddng n the FCC Spectrum Auctons. Contrbutons to Economc Analyss 1(1): Greene, Wllam H Econometrc Analyss, fourth edton. Prentce Hall Internatonal, London. Gupta, Srabana The Effect of Bd Rggng on Prces: A Study of the Hghway Constructon. Revew of Industral Organzaton 19(4): Gupta, Srabana Competton and Colluson n Government Procurement Aucton Market. Atlantc Economc Journal 30(1): Hendrcks, Kenneth and Robert H. Porter Informaton, Returns, and Bddng Behavor n OCS Auctons: Journal of Industral Economcs 35(4): Hendrcks, Kenneth and Robert H. Porter An Emprcal Study of an Aucton wth Asymmetrc Informaton. Amercan Economc Revew 78(5): Hendrcks, Kenneth and Robert H. Porter Jont Bddng n Federal OCS Auctons. Amercan Economc Revew 82(2): Hendrcks, Kenneth and Robert H. Porter Bddng Behavour n OCS

14 13 Dranage Auctons: Theory and Evdence. European Economc Revew 37(2-3): Hendrcks, Kenneth and Robert H. Porter Auctons for Ol and Gas Leases wth an Informed Bdder and a Random Reservaton Prce. Econometrca 62(6): Hendrcks, Kenneth, Jors Pnkse, and Robert H. Porter Emprcal Implcatons of Equlbrum Bddng n Frst-Prce, Symmetrc Common Value Auctons. Revew of Economc Studes 70(1): Hong, Han and Matthew Shum Increasng Competton and the Wnner s Curse: Evdence from Procurement. Revew of Economc Studes 69(4): Jofre-Bonet, Mrea and Martn Pesendorfer Bddng Behavor n a Repeated Procurement Aucton: A Summary. European Economc Revew 44(4-6): Krshna, Vjay Aucton Theory. Academc Press, San Dego, CA. Laffont, Jean-Jacques Game Theory and Emprcal Economcs: The Case of Aucton Data. European Economc Revew 41(1): Laffont, Jean-Jacques, Herve Ossard, and Quang Vuong Econometrcs of Frst-Prce Auctons. Econometrca 63(4): Las Futng, Tm Interpretng probablty Models Logt, Probt, and Other Generalzed Lnear Models. Sage Unversty Paper seres on Quanttatve Applcatons n the Socal Scences, Thousand Oaks, CA. LOU (1992:1528) The Publc Procurement Act n Sweden. Lundberg, Sofa Aucton Formats and Award Rules n Swedsh Procurement Auctons. CERUM Workng Paper 79:2005. Mlgrom, Paul Puttng Aucton Theory to Work. Cambrdge Unversty Press, New York, NY. Wolfram, Catherne D Strategc Bddng n a Multunt Aucton: An Emprcal Analyss of Bds to Supply Electrcty n England and Wales. RAND Journal of Economcs 29(4): Vckrey, Wllam Counterspeculaton, Auctons, and Compettve Sealed Tenders. Journal of Fnance 16(1): 8-37.

15 14 Appendx Table A1. Statstcs on bddng behavor and success rate for natonal frms and n-house producton. Number of placed bds Number of wns Success rato # % # % Frm Frm Frm Frm Frm In-house N 664 Table A2. Wald test. χ 2 (1) -statstc for contract categores. Chld care Medcal Offce Other centre health centre Chld care centre Medcal health centre Offce Table A3. Margnal Effects. 9 N=5,302 Varable β t-value Constant Advantage Procurement characterstcs School - - Chld care centre Medcal health care centre Offce Other Contract perod Number of contracts The bdders Local frms - - Natonal frms In-house producton Bd Number of bdders Number of bdders squared Prevous wnner Muncpalty characterstcs Red Populaton densty Unemployment rate The margnal effects for the dummy varables are Λ β z )[ 1 Λ( β z )]0. ( k k

16 15