Electronic tendering of pharmaceuticals and medical ...

JBR-08440; No of Pages 10 Journal of Business Research xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Journal of Business Research

Electronic tendering of pharmaceuticals and medical devices in Chile Pedro Raventós a,⁎, Sandro Zolezzi b,1 a b

INCAE Business School, Apartado 960, 4050 Alajuela, Costa Rica CINDE, Apartado 178, 1255 Escazú, Costa Rica

a r t i c l e

i n f o

Article history: Received 1 April 2015 Received in revised form 1 May 2015 Accepted 1 May 2015 Available online xxxx Keywords: Procurement Aggregation Tendering Number of bidders Corruption

a b s t r a c t The present study investigates the effect of electronic tendering on the price paid by the public sector for pharmaceuticals and medical devices in Chile. This study uses two panel regression models to analyze a data set that covers 6888 tenders for these items between 2001 and 2006, which spans 2004, the year when use of the Chilecompra electronic platform becomes obligatory. Model 1 explains the winning bid in each tender relative to the historic price, whereas Model 2 explains the winning bid relative to the concurrent price paid by drugstore chains. The regressors include variables which in the theoretical literature are indirectly associated with purchase prices (tender volume, the number of bidders and the time between tenders) and a Chilecompra dummy variable which captures the direct effect of the platform. The novel hypothesis of this paper is that e-tendering engages the market mechanism more effectively than traditional tendering, because of reduced corruption and less supplier collusion, which results in a direct platform effect. The empirical results support the volume effect. Greater aggregation of purchases leads to 2.8% lower prices. The evidence does not support the other indirect channels. More bidders result in lower prices, but the number of bidders fails to increase after Chilecompra. More frequent tendering leads to lower prices for medical devices, but tender frequency decreases after the implementation of the platform. Finally, the empirical results confirm the direct platform effect. Electronic tendering over Chilecompra leads directly to a greater than 8% reduction in prices. These results contribute to the literature on the returns to IT investments. © 2015 Elsevier Inc. All rights reserved.

1. Introduction Electronic procurement by corporations promises to save resources, accelerates cycle times and reduces errors. Aggregation of purchases, both within the organization and with other buyers and new tendering techniques promise to lower prices. Governments can achieve these advantages more readily as they do not face the “penguin problem” that corporations encounter (Farrell & Saloner, 1987 as discussed in Coles & Edelman, 2011), whereby no penguin wants to be the first to dive from an ice flow for food for fear of predators. As the largest buyer in the economy, the government can force suppliers to join its marketplace, and does not have to coordinate with other buyers to achieve buying power. Academic studies do not agree on the size of price savings. Bandiera, Prat, and Valletti (2009) show massive waste in Italian public procurement which suggests a large need for performance improvement. The least efficient decile of public buyers in Italy pays 55% more than the most efficient decile for the same goods, controlling for product quality and purchase volume. If all public buyers match the performance of the most efficient decile, they can reduce public spending by 21%. Other studies are ⁎ Corresponding author Tel.: + 506 2437 2200. E-mail addresses: [email protected] (P. Raventós), [email protected] (S. Zolezzi). 1 +506 2201 2800.

less encouraging. Pavel and Sičáková-Beblavá (2013) and Singer, Konstantinidis, Roubik, and Beffermann (2009) suggest much more modest price savings of 2.4% and 2.65%. McCue and Roman (2012) echo the more conservative sentiment. The absence of reliable spending data for the period preceding the deployment of the electronic procurement platform limits the scope of much of this research. If purchase prices for different goods and services were available for a period before the introduction of the platform — the so called “baseline period”, one could calculate saving by comparing those prices to the prices obtained after the implementation of the platform, controlling for other variables that might affect prices. The authors use a database of public purchases of pharmaceuticals (henceforth drugs) and medical devices in Chile between 2001 and 2006, which spans 2004, the year in which the Ministry of Finance makes use of its procurement platform, Chilecompra, obligatory for public agencies. The authors also have extensive data for other variables that can affect prices. Their model shows that e-Tendering over Chilecompra saves the government 8.3% in drug purchases and 9.1% in medical devices directly. The indirect effect of Chilecompra is to reduce purchase prices by 2.8% through greater aggregation and by 0.4% as a result of better rules. The authors are also able to confirm two hypotheses of great interest in health economics. First, volume discounts exist for drugs and medical devices. Second, purchase prices are lower when buyers have substitution possibilities; the greater a buyer's substitution possibilities, the lower the purchase price. The next section discusses the particular

http://dx.doi.org/10.1016/j.jbusres.2015.06.024 0148-2963/© 2015 Elsevier Inc. All rights reserved.

Please cite this article as: Raventós, P., & Zolezzi, S., Electronic tendering of pharmaceuticals and medical devices in Chile, Journal of Business Research (2015), http://dx.doi.org/10.1016/j.jbusres.2015.06.024

2

P. Raventós, S. Zolezzi / Journal of Business Research xxx (2015) xxx–xxx

context of the health care sector, followed by a discussion of the literature. The authors then explain the data, models and results, and finish with some concluding thoughts. 2. Health sector procurement Procurement in the health sector is sensitive because health care costs have risen rapidly and is challenging because there are often only a few suppliers. Several countries have attempted to implement initiatives to aggregate purchases to counteract supplier power, which coincides in some cases with the introduction of procurement platforms. In the United States, Group Purchasing Organizations (GPOs) aggregate the demand of several hundred hospitals in order to negotiate better prices. Managed care organizations, including health maintenance organizations (HMOs), provide health care to patients through a network of providers. These organizations use their purchase volume and, in the case of drugs, restrictive formularies, in order to obtain better prices. In other countries the government, as the largest provider of health care, can leverage its volume and use restrictive lists in order to gain price concessions. In the late 1970s, Chile's national system of health services creates Cenabast to purchase, manage inventories and distribute drugs and medical supplies for the public hospitals of Chile and other Ministry of Health programs. The use of Cenabast is not obligatory, but almost 100 public hospitals, responsible for 20% of public purchases, use it in 2002. In 1999 the government establishes Chilecompra, an electronic platform for all its purchases, which in practice serves mainly as an outlet to publicize transactions after they happen. In 2004, new legislation makes it compulsory for all agencies, including Cenabast and all municipalities, to use Chilecompra for procurement. On the new platform buyers from different government bodies follow standardized guidelines to carry out the entire procurement process, from posting requirements to publicizing outcomes. As of 2005, firms that want to supply the government must participate in Chilecompra. This legislation leaves public hospitals with a choice between using Chilecompra directly and going through Cenabast. By 2006, 190 public hospitals, responsible for 50% of purchases, use Cenabast. 3. Literature and hypotheses In this section the authors review four strands of literature to develop their hypotheses about the effects of e-Procurement, Aggregation, the Number of Bidders and Tender Frequency on the prices of drugs and medical devices in Chile. Geoffrion and Krishnan (2003) select the first three of these threads as important for eBusiness in their Management Science special issue. 3.1. The effect of e-Procurement De Boer, Harink, and Heijboer (2002) define Electronic Procurement as the use of the Internet in the purchasing process. Both the public and private sectors use Electronic Procurement and it takes many forms, including EDI, electronic data interchange — an inter-organizational information system that uses structured data exchange protocols, e-MRO — a mechanism for ordering indirect items (materials, repairs and operations) from an online catalog, web-based ERP, enterprise resource planning, — web-based automated procurement workflows, e-Sourcing — ways of identifying new sources of supply using Internet technologies, E-Tendering — the process of inviting offers from suppliers and receiving their responses electronically, e-Reverse auctioning or e-Auction — using Internet technologies suppliers to bid down the price of the procured item until none of them is willing to go further, and e-Informing — use of internet technologies for gathering and distributing procurement information. In their theoretical work, De Boer et al. (2002) predict that e-MRO will have a large impact on the cost of purchasing activities for inputs

that are not incorporated into the firm's product and that are “clickable” by internal clients out of a catalog, especially when current activities are inefficient. They also hypothesize a large impact on the purchase price of these items when maverick buying is a problem. Consistently, Croom (2000) considers e-MRO the “killer application” of Electronic Procurement. Kaplan and Sawhney (2000) envision an even greater benefit if several firms share a hub, where different suppliers post their catalogs. The main emphasis in e-MRO is to reduce complexity, not price making. De Boer et al. (2002) hypothesize that both e-Sourcing and e-Tendering help firms reduce the cost of establishing specifications, choosing suppliers, negotiating conditions and contracting. These authors expect e-Auctions to have a direct effect on the cost of both operational and strategic inputs by allowing firms to “obtain lower prices by using the market mechanism”. In contrast, De Boer et al. (2002) expect that e-Tendering will have an impact on purchasing cost only indirectly, as firms are able to consider more alternatives over time. The benefit of expanding the supplier base also applies to e-Auctions. In the public procurement literature, in contrast, a “tender”, whether electronic or not, includes not only the interaction with suppliers, but the actual selection of a winner, and therefore entails the use of the “market mechanism”. Auction Theory refers to tenders as “competitive tenders”. McAfee and McMillan (1987) define an auction as “a market institution with an explicit set of rules determining resource allocation and prices on the basis of bids from the market participants.” In auction theory “competitive tenders” are sealed bid auctions. Their distinctive characteristic is that suppliers compete to win a project by submitting bids without knowing the bids of other suppliers. In the more familiar English auction, potential buyers place increasing bids for an item for sale in a dynamic fashion until none of them is willing to bid higher. The distinctive characteristic of English auctions, which Sotheby's uses to sell impressionist art and the Federal Communications Commission uses to sell telecommunications spectrum, is that bidders can react to the bids of their rivals. A reverse auction is an English auction used for procurement, in which potential suppliers bid down the price of an item requested by buyers. When such an auction takes place electronically it is a reverse e-Auction or sometimes simply an e-Auction, consistent with De Boer et al. (2002). When the valuation or cost of one bidder bears no relation to that of its rivals (i.e. private values setting), several auction formats lead to the same outcome, supporting the idea that both e-Auctions and e-Tenders use the “market mechanism”. See Klemperer (1999) for a discussion of several equivalence results. E-Tenders in the corporate setting also use “the market mechanism”. Snir and Hitt (2003), in their study of competitive electronic tenders for IT, model these tenders as first price auctions. Elmaghraby (2007) indicates that supplier discomfort with e-Auctions leads major e-Procurement vendors to increasingly use e-Tenders instead. Hannon (2006) reports 24% of buyers using e-Tenders and 31% using e-Auctions. De Boer et al. (2002), based on the technology available at the time, claim that e-Auctions are suitable for commodities, or items that can be clearly specified, but correctly foresee the ability to run more complex auctions. Elmaghraby (2004) argues that e-Auctions for non commodities can take place by giving quality differences a monetary value, to be added to or subtracted from monetary bids. Such adjustments can help evaluate different delivery conditions and financial terms that affect the firm's total cost. Snir and Hitt (2003) discuss how to do this in e-Tendering for informational technology contracts. Dimitri (2013) discusses scoring in government e-Tendering. Scholars and practitioners still disagree on the extent to which the market mechanism applies in procurement (Schoenherr & Mabert, 2007). On the one hand, suppliers that invest resources developing components that are specific to the buying firm, and which are vital to that firm's strategy, may no longer make these investments when they have to participate in e-Auctions (Jap, 2003). On the other hand, academic research helps resolve some of the challenges of designing



auctions that can take into account the ability of suppliers to deliver different qualities. These mechanisms still need more testing and practical application. See Pham, Teich, Wallenius, and Wallenius (2014) for a recent survey of these issues. The present paper considers competitive tenders for generic drugs and undifferentiated medical devices. Commoditization is not a requirement because they are already commodities. Although the tenders consider non-price attributes, these attributes are limited to measures of supplier experience with the government and order fulfillment. The effect of these non-price attributes is to disqualify unsuitable suppliers. Corporations and governments use tendering. In government, however, greater agency problems between the procurement officer (PO) and the ultimate user might result in additional benefits from moving to e-Tendering. Government tenders are vulnerable to bid rigging where a corrupt procurement officer (PO) favors a particular supplier in exchange for a bribe, and this practice raises purchase prices. Ingraham (2005) provides evidence of corruption in competitive tenders for school repair and construction in New York. In the theoretical model of bid rigging of Arozamena and Weinschelbaum (2009), the PO has an existing relation with one of the suppliers, referred to as the dishonest supplier (DS). All suppliers submit their bids before the tender closes. The PO examines the bids before announcing the winner and allows the DS to change its bid in the most favorable way. If the DS bid is the lowest, it can increase its bid to a little under the second lowest bid. If the DS bid is not the lowest, the PO allows the DS to reduce its bid to a little under the lowest bid. In the first case the cost to the government goes up, whereas in the second the cost remains unchanged. Clearly, bid rigging raises the expected purchasing price for the government. Arozamena and Weinschelbaum (2009) call this behavior the direct effect of corruption, which assumes that the honest suppliers are unaware of the corruption scheme. When the honest suppliers know about the bid rigging they may bid more or less aggressively, depending on the statistical distribution of their costs. Arozamena and Weinschelbaum (2009) show that this second effect, which they call the perception effect, does not outweigh the direct effect in regular cases so the expected purchasing price for the government buyer goes up. e-Tendering is likely to reduce supplier–PO corruption because it minimizes the direct human interactions between bidders and public buyers (UNODC, 2013), transforming a documentary, interpersonal interaction into an electronic interaction which is not face to face (Ho, 2002). Further, with the deployment of a secure electronic procurement platform, the PO will no longer be able to tinker with bids, as Du, Foo, Gonzalez-Nieto, and Boyd (2005) explain. When suppliers register on the platform a certification authority (CA) provides them with identities and cryptographic keys. Suppliers submit their digitally signed bids to an electronic tender box, and only after the closing time of the tender will the PO receive a key to decrypt the bids. The World Bank requires that all supplier bids use an electronic system that “maintains the integrity, confidentiality, and authenticity of bids submitted, and uses an electronic signature system or equivalent to keep bidders bound to their bids” (World Bank, 2014). In addition, the winning bid in government tenders can increase if suppliers agree on bidding high prices. In the theoretical case, in which a buyer procures an item a single time, a tender is less vulnerable to collusion than an English procurement auction, because each bidder knows that if it deviates from the agreement, its rivals will not be able to react, whereas in an English procurement auction they can respond by further undercutting (Hendricks, McAfee, & Williams, 2014). In real world settings, where governments repeatedly procure a great number of items, which have many common suppliers, dividing contracts, offering side payments not to bid, or making phony bids can enforce collusion. Pesendorfer (2000) shows evidence of these types of collusion in school milk contracts in Texas and Florida respectively. e-Tendering might limit the incidence of collusion between bidders relative to traditional tendering because they fear detection. Croom

3

(2000) discusses how the centralized storage of e-Procurement documents leaves a clearer trail of activities than the traditional decentralized paper-based system. As Bajari and Summers (2002) discuss, if one could estimate bid functions using variables that affect the costs of different bidders, the correlation of residuals between groups of bidders would be evidence of collusion, as would bidders reacting differently to objective cost conditions. Since the variables that affect the cost of bidders may not be available to monitors of the platform, bidder fear of detection may be unfounded, but may nevertheless exist. As Chilecompra only publicizes the bid of the winner, it does not inadvertently promote collusion by making deviations from agreements easier to detect and punish (Stigler, 1964). The effect of e-Tendering on limiting collusion and especially its effect on limiting corruption lead the authors to their first hypothesis. Hypothesis 1. Prices for drugs and medical devices will fall directly as a result of the implementation of e-Tendering by Chilecompra, controlling for other variables that affect prices. The World Bank reports savings in the cost of goods purchased of 2.5% in Ireland and of 5.0% in Canada, but the methodology the bank uses in these studies is unclear. Gardenal (2013) shows that discounts in relation to reserve prices increase slightly when moving from paper based tendering to e-Tendering. The authors do not find any econometric study that shows that the implementation of government e-Tendering leads directly to lower prices. The absence of reliable spending data for the period preceding the deployment of the procurement platform limits the scope of much of this research, as in Gardenal (2013), and as stated in the introduction. The data set in this study includes prices of the same drugs and medical devices for 2001–2006, which spans the introduction of Chilecompra. Evidence of reduced buying prices in the corporate sector is not much better. Pinker, Seidmann, and Vakrat (2003) complain that lack of data has made academic research difficult, and this continues to be true. Bichler, Davenport, Hohner, and Kalagnanam (2006) report average savings of 13%, based on the information obtained from a sample of software vendors. Metty et al. (2005) find that Motorola only saved 3.75%.

3.2. The effect of aggregation Galbraith (1952) develops the theory of countervailing power, whereby a large buyer can compensate the power of a large supplier. Since large buyers can achieve lower prices than small buyers, this suggests that small buyers should get together to obtain price discounts. Anand and Aron (2003) apply this theory to collective buying over the internet. Scholars have proposed different theories for the existence of countervailing market power. Katz (1987) suggests that large buyers get discounts because they can threaten to integrate backwards into the market of the supplier. Smith and Thanassoulis (2014) contend that the price at which a supplier is willing to sell depends on its expectation of incremental cost. Since large buyers affect this expectation, suppliers with decreasing incremental costs will charge lower prices. Snyder (1996) considers a large supplier which is not a monopolist that colludes with the other suppliers to keep prices high in a standard super game. If the large buyer has the possibility of substituting demand over time using inventories, then this buyer can create a “boom” in demand that requires lower prices to sustain collusion, à la Rotemberg and Saloner (1986). Snyder (1996) shows that the mere threat of doing this lowers the price for the large buyer. These theories suggest that price concessions depend on the existence of a large buyer, the substitution possibilities of the buyers and some competition on the supply side. For a further discussion of quantity discounts see Anand and Aron (2003).


4


The government is the quintessential large buyer, especially if it aggregates purchases over many government departments. The introduction of Chilecompra stimulates aggregation, as discussed in Section 2. Hypothesis 2. (a) The price of drugs and medical devices falls with the volume tendered. (b) Since the establishment of the Chilecompra platform leads to an increase in the volume of drugs and medical devices tendered, this platform leads to indirect savings through an aggregation effect. Smith and Thanassoulis (2014) find volume effects that are consistent with their theoretical framework in milk and carbonated soft drink supplies to supermarkets in the United Kingdom. Of particular interest for this paper is the evidence of the benefits of aggregation in health procurement. According to the theoretical framework of Snyder (1996), this aggregation is particularly effective when the buyer can substitute between pharmaceuticals. To test this hypothesis, Ellison and Snyder (2010) compare the prices hospitals, HMOs and drugstores pay in the United States. Hospitals pay 35% less than drugstores for branded off patent drugs where they have substitution possibilities, but only 10% less for on patent drugs, where they do not. Moore and Newman (1993) find that States that use restricted formularies spend 14.8% less per person on pharmaceuticals in their Medicaid programs. To get a sense of the importance of volume, Ellison and Snyder (2010) compare the prices paid by drugstore chains with independent drugstores, but they are not able to isolate the volume effect econometrically because they lack volume data. Wu (2009) establishes that both the volume and the substitution possibilities of buyers affect drug prices at hospitals. Yakovlev, Bashina, and Demidova (2014) investigate the impact of contract volume on the price paid for granulated sugar in Russia. 3.3. The effect of more bidders in auctions Auction theory predicts that in procurement auctions in which suppliers have private values, the expected price falls with the number of bidders (Laffont, 1997). As the number of bidders increases, each one will bid closer to its cost. The winning bid will be lower because of this aggressiveness and because the expected cost of the winning bidder is lower. Many scholars have tested this hypothesis in many settings and in empirical auction research (Brannman, Klein, & Weiss, 1987; Gupta, 2002; Lundberg, 2006; MacDonald, Handy, & Plato, 2002). In the government procurement context, Pavel and SičákováBeblavá (2013) find that one additional seller reduces the prices that Slovak municipalities pay by 3.4% and Yakovlev et al. (2014) find that one more supplier reduces the price the Russian procurement authority pays for sugar by 5.2%. The indirect effect of e-Procurement, via the attraction of more bidders, as De Boer et al. (2002) propose, results from multiplying savings per additional bidder by the number of bidders attracted. Pavel and Sičáková-Beblavá (2013) find that 0.7 additional buyers participating in the Slovak municipalities' tenders lead to savings of 2.4%. Singer et al. (2009) calibrate a first price auction model with the data for 87 public tenders, classified by size and number of bidders, to find the impact of one additional bidder, which they then multiply by estimates of the net users attracted to the platform based on a survey of suppliers and purchasing officials. They estimate that Chilecompra results in price savings of 2.65%, a figure that they consider reasonable as it is not that different to the 3.75% that Metty et al. (2005) find for Motorola. Hypothesis 3. (a) The price of drugs and medical devices falls with the number of bidders. (b) Since the Chilecompra platform attracts more suppliers to tender, this platform leads to indirect savings through more bidders.

3.4. The effect of higher tender frequency Industrial Organization Theory sustains that more frequent transactions increase prices because they facilitate collusion: firms are less likely to undercut an agreed price when they expect a rapid punishment (Hendricks et al., 2014; Tirole, 1988). In a tendering context, less time between tenders will make it easier for suppliers to divide contracts or arrange for side payments. The winning bid should decrease with the time between tenders. De Boer et al. (2002), in contrast, sustain that “testing the market more often” allows the buyer to benefit “from better deals”, which is consistent with the framework of Goeree and Offerman (2003) where a reduction in uncertainty leads to more aggressive bidding in tenders. De Boer et al. (2002) further hypothesize that the reduction in the cost of purchasing activities achieved through e-Tendering encourages more frequent tendering. This leads to the fourth hypothesis of this study. Hypothesis 4. (a) The price of drugs and medical devices increases with the time between tenders. (b) Since implementation of the Chilecompra platform leads to more frequent tenders (lower time between tenders), this platform leads to indirect savings through more frequent tenders.

4. Data & models Cenabast provides the authors with detailed information on the 6888 tenders for drugs and medical devices conducted between 2001 and 2006 on behalf of the public hospitals of Chile. For each tender, the authors have the name of the drug or medical device, the presentation, the identification number of each bidder participating, the bid of each bidder, the number of days between the announcement of the tender and the tender close, the winning bid, and whether the item is imported or produced domestically. Through the end of 2004 Cenabast implements tenders in the traditional way, and thereafter they e-Tender through Chilecompra. During this period Cenabast procures a total of $350 million; most drugs (97.5%) are off patent and most medical devices are imported. Cenabast classifies drugs into 556 codes according to their active pharmaceutical ingredient or generic name, dosage and presentation. The authors further classify these drug codes into 19 groups according to therapeutic use. They also classify a total of 821 codes for medical devices into 7 groups according to use (see Table 1). Both drugs and medical devices come in a number of presentations. The authors have a total of 1377 codes, 556 for drugs and 821 for medical devices. The authors eliminate tenders corresponding to codes not procured before 2005 or those only tendered once. This step leaves 5244 tenders;

Table 1 Classes of drugs and medical devices. Drugs 1 Anti-infective agents 2 Antineoplastic agents 3 Autonomic drugs 4 Blood formation, coagulation and thrombosis 5 Cardiovascular drugs 6 Central nervous system agents 7 Electrolytic, caloric and water balance 8 Respiratory tract agents 9 Eye, ear, nose and throat 10 Gastrointestinal drugs 11 Hormones and synthetic substitutes 12 Local anesthetics 13 Oxytocics 14 Radioactive agents

15 16 17 18

Serums, toxoids and vaccines Skin and mucous membrane agents Smooth muscle relaxants Vitamins

19 Miscellaneous therapeutic agents

1 2 3 4 5 6 7

Medical devices Intravenous Sets Surgical supplies Syringes Diagnostic equipment and supplies Dental devices and supplies Hollowware Miscellaneous


P. Raventós, S. Zolezzi / Journal of Business Research xxx (2015) xxx–xxx Table 2 Description of variables. Name

Description

Expected sign

ABit

If a bidder makes more than one bid in a tender these bids are considered additional bids Number of days between the announcement of the tender and the tender close A dummy that takes on a value of 1 in 2005 or 2006 when the new Chilecompra platform was in operation Extent to which the bidders participating in a tender encounter new competitors in the other tenders in which they participate Exchange rate at the time the tender was conducted relative to the exchange rate at the time the first tender for the same code was conducted Volume tendered relative to the volume purchased by drugstore chains Volume tendered relative to the volume in the first tender for the same code Winning bid relative to the winning bid in the first tender for the same code Winning bid relative to the purchase price of the drugstore chains Dummy variable that equals one if there is a single bidder or if all bidders have bid before for the same code Dummy variable that equals one if the drug is under patent Dummy variable that is equal to one if drugstores sell the drug under prescription Square of the number of bidders participating in a tender Number of years between two consecutive tenders for the same code Number of bidders participating in an auction

?

Daysit Chilecomprait

CIit

ln (ERit/ERio)

ln (Qit/QPhit) ln (Qit/Qio) ln (WBit/WBio) ln (WBit/PPhit) OneBidderit

Patentit Prescriptionit Bidders2it TimeBetweenTendersit Biddersit

− −(H1)

−

+

−(H3) −(H3) DV1 DV2 +

? − + +(H4) −(H2)

DV1 and DV2 are the dependent variables for Model 1 and 2, respectively.

2122 for drugs and 3122 for medical devices. On average they have 112 tenders per drug group and 446 tenders per medical device group. The authors also use volume and price data on the sale of drugs to the three pharmacy chains that control 95% of the private market in Chile. They match this data, which they purchased from IMS Health Chile, with the Cenabast data by active pharmaceutical ingredient. Then, the authors transformed prices and volumes to match the units of Cenabast codes. To evaluate their four hypotheses, the authors specify two regression models. Table 2 explains the variables used. All monetary variables are in Chilean Pesos. The winning bid, tendered volume and exchange rate variables are normalized in order to make the units comparable and to avoid heteroskedasticity, using the same approach as Zhong and Wu (2006) and Mithas and Jones (2007). Model 1 explains the winning bid in each tender relative to the historic price for the corresponding code. The independent variables include time between tenders, the number of days between the announcement of the tender and its close, the number of bidders, CI, OneBidder, volume, the exchange rate, additional bids, and a dummy variable which equals one if the tender takes place over the Chilecompra platform. The regression model for Model 1 is:

ln ðWBit =WBio Þ ð1Þ " # Daysit ; Chilecomprait ; ABit ; CIit ; ln ðQ it =Q io Þ; ln ðERit =ERio Þ; ¼f ; 2 OneBidderit ; Biddersit TimeBetweenTendersit ; Bidders it

where i are the codes and t the years 2001 through 2006.

5

Table 2 shows the expected sign for each explanatory variable. Table 3 shows descriptive statistics for the variables. Since the authors use first observation for each code to calculate the ratios indicated above, they end up with 1566 observations for drugs and 2301 observations for medical devices. Model 2, run only for drugs, explains the winning bid in each tender relative to the concurrent price that drugstore chains paid for the corresponding code. Several of the independent variables are the same as in Model 1. In addition, Model 2 includes patent status and whether the drug is sold under prescription. The authors measure volume relative to concurrent drugstore chain volume. ln WBit =PPh it # " Daysit; Chilecomprait ; ABit ; ln Q it =Q Ph it ; ¼g 2 Patentit ; Prescriptionit : Biddersit ; Bidders it

ð2Þ

where i are the codes and t the years 2001 through 2006. Table 4 shows descriptive statistics for the variables. Since concurrent private sector transactions are not available for 895 tenders, the researchers end up with 1227 observations. Hypothesis 1 proposes that the coefficient of the Chilecompra dummy is negative. Hypothesis 2a proposes that the coefficient of the volume variable is negative. Hypothesis 3a proposes that the coefficient of the number of bidders is negative. To allow for the possible non-linear effect of the number of bidders the authors also include the square of the number of bidders, consistent with Lundberg (2006) and Ingraham (2005). Hypothesis 4a proposes that the coefficient of the time between tenders variable is positive. A discussion on the expected sign for the coefficients of the other variables follows, based on the predictions of auction theory for private value first price auctions, that is, competitive tenders, and on the insights from industrial organization theory, I.O., which Klemperer (2005) argues are valid in a repeated auction setting. The authors start with tender characteristic variables. As the number of days between the announcement of a tender and its close increases, the relative price should be lower as bidders have more time to prepare. A corrupt arrangement with the purchasing officer is also less likely to exclude a competent bidder because there is a larger window for pre tender appeals. For this reason international guidelines like those of the World Bank (2004) try to extend this period. To pick up rivalry in Model 1, the authors use variables CI and OneBidder. The CI variable tries to capture the effect on bidders of contact across multiple markets. The “multi-market contact” story in I.O. says that when bidders meet in more than one market, they are more likely to collaborate, as they will hold their punches in a market where they are strong to avoid punishment in markets where they are weak (Bernheim & Whinston, 1990). In non-auction markets, collaboration means splitting the market in ways that maximize cartel stability, which may mean not participating in one of the markets when costs are asymmetric. In auction markets, when side payments are not possible, dividing contracts may sustain collaboration. Feinstein, Block, and Nold (1985) develop CI. When the bidders that participate in the tender are bidders that in other tenders encounter different rivals, CI will be large. CI, in a loose sense, picks up the absence of multimarket contact. More precisely, CI for bidder i is defined as the ratio of the number of distinct bidders, other than i, that bid in the same tenders as bidder i and the total bids made by bidders, other than i, in these same tenders. This index will be large when the bidder participates in tenders where the bids come from different bidders and small when it participates in tenders with the same bidders. The CI index for the tender is large when the participating bidders have high individual CIs. The authors expect the coefficient of this variable to be negative. OneBidder is a dummy variable which is equal to one if there is only one bidder or if all bidders participating in the tender coincided in a


6


Table 3 Descriptive statistics for Model 1. Variable

Full sample (n = 3867)

Daysit Chilecomprait ABit CIit ln (Qit/Qi0) ln (ERit/ERi0) OneBidderit Bidders2it TimeBetweenTendersit Biddersit ln (WBit/WBi0) Variable

Mean

Std. dev.

9.57 0.40 0.52 0.11 0.39 −0.05 0.69 8.34 0.88 2.51 −0.03

1.57 0.49 1.08 0.12 1.59 0.10 0.46 9.64 1.00 1.42 0.37

Before Chilecompra (n = 2339) Min

Max

Mean

3 0 0 0 −7.35 −0.32 0 1 0 1 −3.19

10 1 12 1 8.44 0.23 1 81 5.45 9 2.92

9.34

1.89

3

0.65 0.10 0.11 0.02 0.74 9.16 0.77 2.64 0.03

1.24 0.12 1.48 0.08 0.44 10.47 1.02 1.48 0.30

0 0 −7.35 −0.24 0 1 0 1 −3.19

Drugs (n = 1566)

Std. dev.

Std. dev.

Min

Max

Mean

Daysit Chilecomprait ABit CIit ln (Qit/Qi0) ln (ERit/ERi0) OneBidderit Bidders2it TimeBetweenTendersit Biddersit ln (WBit/WBi0)

9.66 0.40 0.27 0.13 0.47 −0.06 0.70 6.14 0.96 2.19 −0.06

1.45 0.49 0.83 0.12 1.75 0.10 0.46 6.83 1.01 1.17 0.34

3 0 0 0 −6.91 −0.32 0 1 0 1 −3.19

10 1 12 1 8.44 0.22 1 49 5.45 7 1.23

9.48

1.77

3

0.25 0.13 0.15 0.00 0.75 6.13 0.87 2.19 0.01

0.95 0.13 1.66 0.07 0.43 6.65 1.03 1.15 0.30

0 0 −6.91 −0.24 0 1 0 1 −3.19

Variable

Medical devices (n = 2301) Mean

Std. dev.

9.50 0.39 0.69 0.09 0.34 −0.04 0.68 9.84 0.83 2.74 −0.01

Max

1.64 0.49 1.19 0.12 1.46 0.11 0.47 10.90 0.99 1.53 0.38

Std. dev.

12 1 5.73 0.23 1 81 3.68 9 1.75

Max

Max

Mean

10 1 6 1 5.67 0.23 1 81 4.96 9 2.92

9.24

1.96

3

0.91 0.08 0.09 0.03 0.73 11.19 0.70 2.94 0.05

1.33 0.11 1.35 0.08 0.44 11.97 1.00 1.60 0.30

0 0 −7.35 −0.21 0 1 0 1 −1.29

previous tender for the same product. A single bidder could bid the buyer's reservation price, as could a group of bidders that have so agreed, after competing in previous tenders. When this condition exists, the price tends to be higher. Pesendorfer (2000) and Lee (1999) use variables similar to OneBidder. Model 2, which is run only for drugs, includes two additional variables: prescription and patent. Pharmacies filling prescriptions in Chile are not able to substitute the branded drug for a generic if the prescription specifies the brand name, similar to the situation in the U.S. till 1989 (Hellerstein, 1998). For this reason drugstore chains are unable to negotiate price concessions in exchange for higher volumes. In contrast, the government awards its full volume for a code to the bidder with the

Std. dev.

Min

0.74

3

0.33 0.12 0.83 −0.14 0.61 7.08 1.05 2.32 −0.12

0.74 0.13 1.64 0.06 0.49 8.04 0.94 1.30 0.43

0 0 −5.21 −0.32 0 1 0 1 −2.72

Mean

10 12 1 5.73 0.22 1 49 3.52 7 1.23

Before Chilecompra (n = 1403)

3 0 0 0 −7.35 −0.32 0 1 0 1 −2.72

Std. dev.

9.92

Max 10 5 1 8.44 0.04 1 64 5.45 8 2.92

After Chilecompra (n = 630)

Min

Min

Mean

10


Mean

Daysit Chilecomprait ABit CIit ln (Qit/Qi0) ln (ERit/ERi0) OneBidderit Bidders2it TimeBetweenTendersit Biddersit ln (WBit/WBi0)


Min

Std. dev.

Min

9.93

0.68

3

0.30 0.12 0.95 −0.15 0.63 6.15 1.09 2.17 −0.15

0.61 0.11 1.77 0.07 0.48 7.10 0.97 1.20 0.38

0 0 −5.13 −0.32 0 1 0 1 −2.04

Max 10 5 1 8.44 0.04 1 49 5.45 7 1.21


Min

Max

Mean

10 6 1 5.16 0.23 1 81 3.68 9 1.75

Std. dev.

Min

9.91

0.79

3

0.36 0.11 0.75 −0.13 0.60 7.74 1.02 2.43 −0.11

0.82 0.14 1.54 0.06 0.49 8.58 0.93 1.36 0.47

0 0 −5.21 −0.32 0 1 0 1 −2.72

Max 10 4 1 5.67 0.02 1 64 4.96 8 2.92

best offer, which is most often the lowest bid because quality differences are insignificant. For this reason the relative price paid by the government should be lower for prescription drugs, and the authors expect the prescription dummy to have a negative coefficient. Model 2 also includes the patent status of the drug, a variable which Ellison and Snyder (2010) find is associated with much smaller discounts. A higher exchange rate should increase the price paid by Cenabast, because both drugs and medical devices are tradable. Most of the drugs are generics produced by local laboratories using imported inputs. As the Chilean Peso cost of these inputs increases with a depreciation of the currency, laboratories will tend to raise their bids, and the winning bid should be higher. Since medical devices are mostly imported, the

Table 4 Descriptive statistics for Model 2. Variable

Daysit Chilecomprait ABit ln (Qit/QPhit) Patentit Prescriptionit Bidders2it Biddersit ln (WBit/PPhit)

Full sample (n = 1227)


Mean

Std. dev.

Min

Max

9.65 0.25 0.41 1.97 0.03 0.86 7.11 2.36 −0.73

1.43 0.44 1.33 2.50 0.16 0.35 7.72 1.24 0.73

3 0 0 −5.93 0 0 1 1 −4.82

10 1 12 10.54 1 1 64 8 1.70

Mean

Std. dev.

Min

9.57

1.58

3

0.42 1.65 0.02 0.86 6.95 2.34 −0.69

1.49 2.45 0.15 0.35 7.61 1.22 0.76

0 −5.93 0 0 1 1 −4.82

After Chilecompra (n = 312) Max 10 12 10.00 1 1 64 8 1.70

Mean

Std. dev.

Min

9.91

0.79

3

0.38 2.89 0.03 0.87 7.60 2.44 −0.88

0.65 2.40 0.17 0.34 8.02 1.28 0.63

0 −3.20 0 0 1 1 −3.18

Max 10 4 10.54 1 1 49 7 0.63



authors expect the pass-through from the exchange rate to the local price to be higher than for drugs. The authors are surprised to find more than one bid per bidder in close to 30% of the tenders. If a bidder submits more than one bid, the difference is added to a variable called Additional Bids (ABit). The winner in these tenders is the bidder that offers the best combination of price and non-price attributes. The non-price attributes in drugs and medical devices, as discussed above, are limited to measures of supplier experience with the government and order fulfillment. The effect of these non-price attributes is to disqualify unsuitable suppliers. Additional bids may help sustain a corrupt scheme when a simple substitution of bids is not feasible, along the lines of Ingraham (2005). 5. Estimation and results The data in this study represents an unbalanced panel with observations on 1377 codes over the six years from 2001 to 2006. The authors first estimate Model 1 using ordinary least squares (OLS) with the full sample (drugs and medical devices), to get a sense of how well their model fits. They use the panel robust standard errors that Arellano (1987) proposes to compute t ratios. Table 5 shows the results incrementally adding in the CI, OneBidder and exchange rate variables. In the first column, most coefficients take on the expected sign. Price falls when bidders have more time to prepare their bids, when the tender takes place over Chilecompra, when tendered volume is greater and more bidders participate. When the two rivalry variables, CI and OneBidder, are added in columns 2 and 3, they are significant at the 10% and 5% levels respectively and have the expected sign. In the process, the size of the Chilecompra effect changes little, remaining close to 14%, which is quite large. Since this effect is measured by a dummy, it can be picking other changes that occurred in 2005–2006 that affected prices. Careful consideration of possible variables leads the authors to the exchange rate and to changes in government coverage. If any pass-through from the exchange rate to domestic prices exists, an appreciating exchange rate should lead to lower prices. The exchange rate appreciates from 691.4 Chilean Pesos per US Dollar in 2003 to 530.3 in 2006. When the relative exchange rate is included in column 4 the coefficient on Chilecompra drops by 35%. AUGE, the set of drugs that the government covers for all Chileans, expands in 2005. The increased volume for certain codes that occurs as a result of this expansion is picked up by the volume variable in this study, and should not inflate the author's estimated coefficient for Chilecompra. Furthermore, even in 2005, the AUGE program represents less than 10% of the authors' observations. In the regressions in Table 5, the authors use the full sample of drugs and medical devices, which are very heterogeneous, and a Chow test

7

(Chow, 1960) shows that the full data sample should not be pooled. For this reason for Model 1, they separate drugs and medical devices and, for each, allow for the possibility of group effects in the following manner: ln ðWBit =WBio Þ ¼ βo þ β1 Daysit þ β2 Chilecomprait þ β3 ABit þ β4 CIit þβ5 ln ðQ it =Q io Þ þ β6 ln ðERit =ERio Þ þ β7 OneBidderit þβ8 Biddersit þ β9 TimeBetweenTendersit 2

þβ10 Bidders

it

ð3Þ

þ cg þ εit

where εit is the error and cg is a group effect, following the classification in Table 1. In both cases the hypothesis of random effects is rejected in favor of fixed effects using a Hausman test as explained in Greene (2007). The fixed effect model is estimated using the within estimator, where variables are taken as a difference from their group mean. An F test shows that the group effects are significant for both drugs and medical devices. Tables 6 and 7 show the results. For Model 2 the authors also allow for the existence of group effects: ln WBit =PPh it ¼ βo þ β1Daysit þ β2 Chilecomprait þ β3 ABit þ

þβ4 ln Q it =Q Ph it þ β5 Patentit þ β6 Prescriptionit ð4Þ 2

þβ7 Biddersit þ β8 Bidders

it

þ cg þ εit

where εit is the error and cg is a group effect, following the classification of Table 1. The Hausman test is unable to reject random effects compared to fixed effects. The Breusch–Pagan Lagrange multiplier test, as explained in Greene (2007), shows that random effects are significant. The authors estimate the random effect using feasible generalized least squares (FGLS). Table 8 shows the results. The authors discuss and interpret the coefficients in these regressions as follows. The coefficient for tender days is significant at the 10% only for drugs. Each additional day lowers price by 0.89%, but the standard deviation of tender days is only 1.45. Table 8 shows that each additional day lowers the price obtained by Cenabast relative to the price charged to the drugstore chains by 4%, controlling for volume and for the number of bidders. The variables related to rivalry have the expected effect on price. Higher CI (i.e. the absence of multi-market contact) leads to lower prices, but only for drugs. In contrast, the presence of a single bidder or only repeated bidders increases the price only for medical devices. These last variables are not significant in Model 2. Model 2 considers two additional variables: prescription and patent. The coefficient for prescription is negative and fairly large in Table 8. In so far as the government can threaten to substitute one prescription

Table 5 Model 1 (full sample). Pooled OLS estimation. Variable

Estimate

t Arellano

Estimate

t Arellano

Estimate

t Arellano

Estimate

t Arellano

(Intercept) Daysit Chilecomprait ABit ln (Qit/Qio) CIit OneBidderit ln (ERit/ERio) Bidders2 it TimeBetweenTendersit Biddersit R2 F

0.27649 −0.00470 −0.14079 −0.00813 −0.04300

8.302 −1.428 −9.546 −0.946 −7.177

0.30647 −0.00548 −0.13993 −0.00938 −0.04236 −0.12526

8.119 −1.677 −9.439 −1.098 −7.161 −2.018

0.19737 −0.00518 −0.13187 −0.01345 −0.04258 −0.10710 0.07014

4.414 −1.586 −8.969 −1.566 −7.210 −1.737 4.843

0.01388 0.01333 −0.12361 0.1164 72.64

5.350 2.337 −6.882

0.01446 0.01350 −0.12921 0.1180 64.52

5.415 2.364 −6.853

0.01187 0.02212 −0.10182 0.1234 60.31

4.406 3.679 −5.184

0.18868 −0.00532 −0.08582 −0.01471 −0.04042 −0.10784 0.07126 0.31686 0.01176 0.02781 −0.10126 0.1268 56.01

4.239⁎⁎⁎ −1.705⁎ −3.916⁎⁎⁎ −2.139⁎⁎ −6.831⁎⁎⁎ −1.769⁎ 4.953⁎⁎⁎ 2.610⁎⁎ 4.283⁎⁎⁎ 4.367⁎⁎⁎ −5.084⁎⁎⁎

t Arellano are t ratios calculated using Arellano robust standard errors for panel regressions. ⁎ p b 0.10. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.


8


Table 6 Model 1. Drugs. Fixed effects.

Table 8 Model 2. Drugs. Random effects.

Variable

Estimate

Std. error

Robust S. E.

Daysit Chilecomprait ABit CIit ln (Qit/Qio) ln (ERit/ERio) OneBidderit Bidders2 it TimeBetweenTendersit Biddersit R2 F

−0.00889 −0.08250 0.00311 −0.13034 −0.03479 0.26863 0.01638 0.01106 0.00915 −0.12574

0.0059 0.0236 0.0099 0.0732 0.0048 0.1199 0.0212 0.0043 0.0084 0.0265 0.1613 29.54

0.0047 −1.880* 0.0383 −2.153** 0.0043 0.718 0.0783 −1.664* 0.0050 −7.015*** 0.2357 1.140 0.0155 1.059 0.0050 2.205** 0.0093 0.978 0.0305 −4.118*** *** p b 0.01 ** p b 0.05 * p b 0.10

t Arellano

t Arellano are t ratios calculated using Arellano robust standard errors for panel regressions. F test for group effects. F = 2.6938, df1 = 18, df2 = 1537, p-value = 0.0001512. Hausman test. chisq = 37.1758, df = 10, p-value = 5.276e−05.

drug for another in its formulary, it can obtain better prices than the drugstore chains. This claim is consistent with Ellison and Snyder (2010). The patent status of drugs is not significant. The exchange rate coefficient has the expected sign, but is significant at the 10% only for medical devices. An appreciation of the exchange rate leads to a reduction in price. The coefficient for a number of additional bids is significant at the 5% for medical devices, but not for drugs. For medical devices the authors find additional bids lower the price, which is contrary to their expectation. The results of this study support Hypothesis 1 concerning the direct effect of Chilecompra on prices. For Model 1 the coefficient of Chilecompra (− 0.083) is significant at the 5% level for drugs and the coefficient for medical devices (−0.091) is significant at the 10% level. For Model 2 the Chilecompra coefficient (−0.102) is significant at 10%. The results support Hypothesis 2a. The volume variable has the expected sign and is significant at the 1% for both drugs and medical devices using Model 1, and for drugs using Model 2. Unlike Ellison and Snyder (2010), the authors are able to measure a volume effect independent of the restricted formulary effect. The evidence also supports Hypothesis 3a. The number of bidder's variable has the expected sign and is significant at the 1% for all specifications. Finally, the evidence supports Hypothesis 4a for medical devices. The time between tenders variable has the expected sign and is significant at the 1%. For drugs, the authors find evidence that restricted formularies and aggregation of purchases allow the government to obtain better prices,

Table 7 Model 1. Medical devices. Fixed effects. Variable

Estimate

Std. error

Robust S. E.

Daysit Chilecomprait ABit CIit ln (Qit/Qio) ln (ERit/ERio) OneBidderit Bidders2 it TimeBetweenTendersit Biddersit R2 F

−0.00037 −0.09141 −0.03387 −0.10184 −0.04304 0.38114 0.10706 0.00721 0.04549 −0.05421

0.0048 0.0235 0.0078 0.0676 0.0052 0.1114 0.0190 0.0026 0.0085 0.0195 0.1196 31.01

0.0078 −0.047 0.0501 −1.825* 0.0142 −2.385** 0.1342 −0.759 0.0092 −4.702*** 0.2106 1.810* 0.0435 2.463** 0.0023 3.196*** 0.0070 6.472*** 0.0111 −4.865*** *** p b 0.01 ** p b 0.05 * p b 0.10

t Arellano

t Arellano are t ratios calculated using Arellano robust standard errors for panel regressions. F test for group effects. F = 19.6341, df1 = 6, df2 = 2284, p-value b 2.2e−16. Hausman test. chisq = 832.6191, df = 10, p-value b 2.2e−16.

Variable

Estimate

Std. error

Robust S. E.

(Intercept) Daysit Chilecomprait ABit ln (Qit/QPhit) Patentit Prescriptionit Bidders2 it Biddersit R2 F

0.21525 −0.04045 −0.10198 0.00977 −0.05016 −0.26765 −0.13682 0.01672 −0.22465

0.1753 0.0138 0.0450 0.0149 0.0082 0.1339 0.0658 0.0089 0.0561 0.1249 21.73

0.3319 0.648 0.0228 −1.774* 0.0580 −1.757* 0.0133 0.734 0.0140 −3.581*** 0.2042 −1.311 0.0780 −1.754* 0.0077 2.158** 0.0622 −3.609*** *** p b 0.01 ** p b 0.05 * p b 0.10

t Arellano

t Arellano are t ratios calculated using Arellano robust standard errors for panel regressions. Lagrange multiplier test — (Breusch–Pagan). chisq = 2883654, df = 1, p-value b 2.2e−16. Hausman test. chisq = 4.3943, df = 8, p-value = 0.8199.

a result which sits comfortably with the mounting evidence in Health Economics (Sorensen, 2003; Wu, 2009). 6. Conclusions Many governments implement procurement platforms to signal their openness to scrutiny and, therefore, their honesty, but the returns to such investments through reduced prices are hard to pin down. According to De Boer et al. (2002), e-Tendering reduces the price of purchased products only indirectly by “attracting more suppliers over time” and “testing the market more often”. Hypothesis 1, developed in Section 3.1 of this paper, sustains that government e-Tendering reduces costs directly because of less corruption and less supplier collusion. The models estimated in this paper find no evidence for the indirect channels that De Boer et al. (2002) propose. They find that more bidders lead to lower prices (Hypothesis 3a), but that the number of bidders is unchanged for drugs and actually decreases for medical devices, contrary to Hypothesis 3b. The empirical results also confirm that more frequent tenders lead to lower prices for medical devices (Hypothesis 4a), but that tender frequency actually decreases contrary to Hypothesis 4b. The evidence in this paper supports Hypothesis 1. Chilecompra results in direct price saving of 8.3% for drugs and 9.1% for medical devices. This finding challenges the theoretical proposition of De Boer et al. (2002) that e-Tendering only reduces prices indirectly. This effect is similar in size to the one found by other research that has focused on waste, rather than the number of bidders. Bandiera et al. (2009) find that corruption, or what they call active waste, can add an additional 11% to purchase prices in Italy. Auriol (2006) estimates that the cost of capture represents 4.1–9.9% of global procurement spending. Di Tella and Schargrodsky (2003) find that a crackdown on hospital overcharging in Buenos Aires has the effect of reducing prices by 10%. This paper develops and finds evidence for important indirect effects of e-Tendering not considered by De Boer et al. (2002). The empirical analysis strongly supports Hypothesis 2, which sustains that aggregation leads to lower prices. Multiplying the estimated coefficient of the volume variable (Tables 6 and 7) with the change in mean volume after the implementation of Chilecompra (Table 2), results in aggregation savings of 2.8% for both drugs and medical devices. Aggregation has been discussed in the health sector for years (Ellison & Snyder, 2010; Wu, 2009), and the authors are able to find that the effect is important and larger on its own than the effect of the number of bidders that is sometimes uncovered. Even though Chilecompra does not attract more bidders for medical devices, the platform does attract new bidders. Multiplying the coefficient of the OneBidder variable by the reduction in the mean level of that variable, results in indirect price saving of 1.4%. This finding is



consistent with the work of De Silva, Dunne, and Kosmopoulouz (2003), which shows that in repeated auctions for highway contracts, new participants bid more aggressively. Finally, the authors are able to determine the effect of better rules for drugs. After the implementation of Chilecompra, the time between the posting and the award of tenders is extended by half a day, leading to indirect price savings of 0.4%. In addition to their theoretical and empirical importance, these results are extremely encouraging for governments. e-Tendering might reduce prices, through reduced corruption between purchasing officers and suppliers, reduced collusion between suppliers, better rules associated with the implementation of the platform, greater aggregation of purchases, more bidders in some sectors (though not for drugs and medical devices in Chile). This work also contributes to the literature on the returns to IT investments. Over twenty-five years ago, Solow (1987) indicates that one can find computers everywhere but in the productivity statistics. Work over the following years resolves this enigma. Brynjolfsson and Hitt (1996) first identify the productivity of computer investments and IS labor by estimating production functions with firm level data. Aral, Brynjolfsson, and Wu (2006) take the firm level research a step further by finding the impact of ERP (Enterprise Resource Planning), SCM (Supply Chain Management) and CRM (Customer Relationship Management) systems on a series of operational metrics. In 2006, $65 million in drugs and $37 million in medical devices are e-Tendered over the Chilecompra platform. Using only the direct effect of e-Tendering on price the authors find savings of $8.7 million. On the other hand, the Chilecompra platform for the whole public sector entailed an investment of $14.9 million. Assuming indirect investments and implementation expenses of five times that amount, to be conservative, the authors get an overall investment of $89.4 million. Since Cenabast represents 6.5% of the value of tenders conducted over Chilecompra, the authors consider that its share of investment is $5.8 million. The return on this investment from direct price saving alone would be 151%. Lack of data makes empirical research on procurement difficult. By using data for a period that spans the introduction of Chilecompra and by controlling for other variables that affect prices, the authors are able to isolate the impact of an important category of IT investment. A limitation of this study is that it is based on a single product class and a single country. Further work with similar data sets might allow researchers to determine whether the results of this study extend to other industries and settings. Acknowledgments The authors thank an anonymous reviewer for the helpful comments. References Anand, K.S., & Aron, R. (2003). Group buying on the Web: A comparison of pricediscovery mechanisms. Management Science, 49(11), 1546–1562. Aral, S., Brynjolfsson, E., & Wu, D.J. (2006). Which came first, IT or productivity? The virtuous cycle of investment and use in enterprise systems. Proceedings of the Twenty Seventh International Conference on Information Systems, Milwaukee, Wisconsin. Arellano, M. (1987). Computing robust standard errors for within group estimators. Oxford Bulletin of Economics and Statistics, 49(4), 431–434. Arozamena, L., & Weinschelbaum, F. (2009). The effect of corruption on bidding behavior in first-price auctions. European Economic Review, 53, 645–657. Auriol, E. (2006). Corruption in procurement and public purchase. International Journal of International Organization, 24, 867–885. Bajari, P., & Summers, G. (2002). Detecting collusion in procurement auctions. Antitrust Journal, 70, 143–170. Bandiera, O., Prat, A., & Valletti, T. (2009). Active and passive waste in government spending: evidence from a policy experiment. American Economic Review, 99(4), 1278–1308. Bank, World (2004). Electronic Government Procurement — Roadmap. Retrieved from http://idbdocs.iadb.org/wsdocs/getdocument.aspx?docnum=645469 (accessed May 1, 2009) Bernheim, B.D., & Whinston, M.D. (1990). Multimarket contact and collusive behavior. Rand Journal of Economics, 21, 1–26.

9

Bichler, M., Davenport, A., Hohner, G., & Kalagnanam, J. (2006). Industrial procurement auctions. In P. Crampton, Y. Shoham, & R. Steinberg (Eds.), Combinatorial auctions (pp. 593–612). Cambridge, MA: The MIT Press. Brannman, L., Klein, J.D., & Weiss, L.W. (1987). The price effects of increased competition in auction markets. Review of Economics and Statistics, 69(1), 24–32. Brynjolfsson, E., & Hitt, L. (1996). Paradox lost? Firm-level evidence on the returns to information systems spending. Management Science, 42(4), 541–558. Chow, G.C. (1960). Tests of equality between sets of coefficients in two linear regressions. Econometrica, 28(3), 591–605. Coles, P., & Edelman, B. (2011). Mobilizing networked businesses. HBS No. 9-911-048. Boston, MA: Harvard Business School Publishing. Croom, S.R. (2000). The impact of web-based procurement on the management of operating resources supply. The Journal of Supply Chain Management, 36(1), 4–13. De Boer, L., Harink, J., & Heijboer, G. (2002). A conceptual model for assessing the impact of electronic procurement. European Journal of Purchasing & Supply Management, 8(1), 25–33. De Silva, D., Dunne, T., & Kosmopoulouz, G. (2003). An empirical analysis of entrant and incumbent bidding in road construction auctions. The Journal of Industrial Economics, 51(3), 295–316. Di Tella, R., & Schargrodsky, E. (2003). The role of wages and auditing during a crackdown on corruption in the city of Buenos Aires. Journal of Law & Economics, 46(1), 269–292. Dimitri, N. (2013). Best value for money in procurement. Journal of Public Procurement, 13(2), 149–175. Du, R., Foo, E., Gonzalez-Nieto, J., & Boyd, C. (2005, Aug. 22–26). Designing secure e-Tendering systems in trust, privacy, and security in digital business. Proceedings in the Second International Conference, TrustBus 2005, Copenhagen, Denmark. Ellison, F.M., & Snyder, C.M. (2010). Countervailing power in wholesale pharmaceuticals. Journal of Industrial Economics, 58(1), 32–53. Elmaghraby, W. (2004). Auctions and pricing in e-Marketplaces. In D. Simchi-Levi, S. David Wu, & Z. Max Shen (Eds.), Handbook of quantitative supply chain analysis: Modeling in the business era. Norwell, MA: Kluwer Academic Publishers. Elmaghraby, W. (2007). Auctions within e-Sourcing events. Production and Operations Management, 16(4), 409–422. Farrell, J., & Saloner, G. (1987). The economics of horses, penguins and lemmings. In L. Gable (Ed.), Production standardization and competitive strategies, Amsterdam, North Holland. Feinstein, J.S., Block, M.K., & Nold, F.C. (1985). Asymmetric information and collusive behavior in auction markets. The American Economic Review, 75(3), 441–460. Galbraith, J.K. (1952). American capitalism: The concept of countervailing power. Boston: Houghton Mifflin. Gardenal, F. (2013). A model to measure e-Procurement impacts on organizational performance. Journal of Public Procurement, 13(2), 214–242. Geoffrion, A.M., & Krishnan, R. (2003). e-Business and management science: Mutual impacts (parts 1 and 2). Management Science, 49(11), 1275–1445. Goeree, J.K., & Offerman, T. (2003). Competitive bidding in auctions with private and common values. The Economic Journal, 113, 598–613. Greene, W.H. (2007). Econometric analysis (6th ed.). Prentice Hall. Gupta, S. (2002). Competition and collusion in a government procurement auction market. Atlantic Economic Journal, 30(1), 13–25. Hannon, D. (2006). e-Sourcing finds its place. Purchasing, 135(16), 51–52. Hellerstein, J.K. (1998). The importance of the physician in the generic versus trade-name prescription decision. RAND Journal of Economics, 29(1), 108–136. Hendricks, K., McAfee, R.P., & Williams, M.A. (2014). Auctions and bid rigging. In R.D. Blai, & D. Sokal (Eds.), Oxford handbook of international antitrust economics (pp. 498–522). NEWGEN Electronic Publishing. Ho, T.K. (2002). Reinventing local governments and the e-Government initiative. Public Administration Review, 62(4), 434–444. Ingraham, A.T. (2005). A test for collusion between a bidder and an auctioneer in sealed-bid auctions. Contributions to Economic Analysis & Policy, 4(1), 1–32. Jap, S.D. (2003). An exploratory study of the introduction of online reverse auctions. Journal of Marketing, 67, 96–107. Kaplan, S., & Sawhney, M. (2000, May–Jun.n). e-Hubs: The new B2B marketplaces. HBR, 97–103. Katz, M.L. (1987). The welfare effects of third degree price discrimination in intermediate goods markets. American Economic Review, 77, 154–167. Klemperer, P. (1999). Auction theory. A guide to the literature. Journal of Economic Survey, 13(3), 227–286. Klemperer, P. (2005). Bidding markets. UK Competition Commission, 6, 1–40. Laffont, J.J. (1997). Game theory and empirical economics. The case of auction data. European Economic Review, 41(1), 1–35. Lee, I.K. (1999). Non-cooperative tacit collusion, complementary bidding and incumbency premium. Review of Industrial Organization, 15(2), 115–134. Lundberg, S. (2006). Auction formats and award rules in Swedish procurement auctions. Rivista di Politica Economica, 96(1), 91–114. MacDonald, J.M., Handy, C.H., & Plato, G.E. (2002). Competition and Prices in USDA commodity procurement. Southern Economic Journal, 69(1), 128–143. McAfee, P., & McMillan, J. (1987). Auctions and bidding. Journal of Economic Literature, 15, 699–738. McCue, C., & Roman, A.V. (2012). e-Procurement: Myth or reality? Journal of Public Procurement, 12(2), 221–248. Metty, T., Harlan, R., Samelson, Q., Moore, T., Morris, T., Sorensen, R., et al. (2005). Reinventing the supplier negotiation process at Motorola. Interfaces, 35(1), 7–23. Mithas, S., & Jones, J.L. (2007). Do auction parameters affect buyer surplus in e-Auctions for procurement? Production and Operations Management, 16(4), 445–470. Moore, W.J., & Newman, R.J. (1993). Drug formulary restrictions as a cost-containment policy in medicaid programs. Journal of Law and Economics, 36(1), 71–97.


10


Pavel, J., & Sičáková-Beblavá, E. (2013). Do e-Auctions really improve the efficiency of public procurement? The case of the Slovak municipalities. Prague Economic Papers, 1, 111–124. Pesendorfer, M. (2000). A study of collusion in first-price auctions. Review of Economic Studies, 67(3), 381–411. Pham, L., Teich, J., Wallenius, H., & Wallenius, J. (2014). Multi-attribute online reverse auctions: Recent research trends. European Journal of Operational Research, 242(1), 1–9. Pinker, E.J., Seidmann, A., & Vakrat, Y. (2003). Managing online auctions: Current business and research issues. Management Science, 49(11), 1457–1484. Rotemberg, J.J., & Saloner, G. (1986). A supergame-theoretic model of price war during booms. The American Economic Review, 76(3), 390–407. Schoenherr, T., & Mabert, V.A. (2007). Online reverse auctions: Common myths versus evolving reality. Business Horizons, 50, 373–384. Singer, M., Konstantinidis, G., Roubik, E., & Beffermann, E. (2009). Does e-Procurement save the state money? Journal of Public Procurement, 9(1), 58–78. Smith, H., & Thanassoulis, J. (2014). Upstream competition and downstream buyer power. Research in Economics, 68(1), 84–93. Snir, E.M., & Hitt, L.M. (2003). Costly bidding in online markets for IT services. Management Science, 49(11), 1504–1520. Snyder, C.M. (1996). A dynamic theory of countervailing power. Rand Journal of Economics, 27(4), 747–769. Solow, R. (1987, Jul. 12). We'd better watch out. [Review of the book Manufacturing Matters: The Myth of the Post-Industrial Economy by S.S. Cohen & J. Zysman]. The New York Times July 12, 1987, p. 36.

Sorensen, A.T. (2003). Insurer–hospital bargaining: Negotiated discounts in postderegulation Connecticut. The Journal of Industrial Economics, 51(4), 469–490. Stigler, G. (1964). A theory of oligopoly. Journal of Political Economy, 72(1), 44–61. Tirole, J. (1988). The theory of industrial organization. Cambridge, MA: The MIT Press. United Nations Office on Drugs and Crime (2013). Guidebook on anti-corruption in public procurement and the management of public finances: Good practices in ensuring compliance with Article 9 of the United Nations Convention against Corruption. Retrieved from http://www.unodc.org/documents/corruption/Publications/2013/ Guidebook_on_anti-corruption_in_public_procurement_and_the_management_of_ public_finances.pdf World Bank (2014). Guidelines procurement of good, works and non-consulting services. Retrieved from http://siteresources.worldbank.org/INTPROCUREMENT/Resources/ Procurement_GLs_English_Final_Jan2011_revised_July1-2014.pdf Wu, V.Y. (2009). Managed care's price bargaining with hospitals. Journal of Health Economics, 28(2), 350–360. Yakovlev, A., Bashina, A., & Demidova, O. (2014). The effectiveness of simple homogeneous commodity procurement under rigid governmental regulation: The case of granulated sugar procurement in Russia (BRP No. 13). Moscow: Basic Research Program. Higher School of Economics. Zhong, V., & Wu, D.J. (2006). e-Sourcing: Impact of bidding behavior and non-price attributes. Proceedings of the Twenty Seventh International Conference on Information Systems, Milwaukee, Wisconsin.
