Contracting in SC

7/30/2019 Contracting in SC

1/16

MANAGEMENT SCIENCEVol. 55, No. 12, December 2009, pp. 19531968issn 0025-1909 eissn 1526-5501 09 5512 1953

informs

doi 10.1287/mnsc.1090.1089 2009 INFORMS

Contracting in Supply Chains:

A Laboratory InvestigationElena Katok

Smeal College of Business, Pennsylvania State University, University Park, Pennsylvania 16802, [email protected]

Diana Yan WuSchool of Business, University of Kansas, Lawrence, Kansas 66045, [email protected]

The coordination of supply chains by means of contracting mechanisms has been extensively explored theo-retically but not tested empirically. We investigate the performance of three commonly studied supply chaincontracting mechanisms: the wholesale price contract, the buyback contract, and the revenue-sharing contract.The simplified setting we consider utilizes a two-echelon supply chain in which the retailer faces the newsven-dor problem, the supplier has no capacity constraints, and delivery occurs instantaneously. We compare the threemechanisms in a laboratory setting using a novel design that fully controls for strategic interactions between

the retailer and the supplier. Results indicate that although the buyback and revenue-sharing contracts improvesupply chain efficiency relative to the wholesale price contract, the improvement is smaller than the theorypredicts. We also find that although the buyback and revenue-sharing contracts are mathematically equivalent,they do not generally result in equivalent supply chain performance.

Key words : supply chain contracts; experimental economicsHistory : Received June 12, 2006; accepted July 28, 2009, by Aleda Roth, operations and supply chain

management. Published online in Articles in Advance October 16, 2009.

1. Introduction and MotivationPrevious studies have shown a great deal of inter-est in the analysis of contracting mechanisms that can

be used to coordinate supply chains. The alignmentof the economic incentives of supply chain partnersis important because a supply chain that consists ofindividual firms, each interested in its own welfare,yields decentralized decisions that are usually ineffi-cient. Much of the past research has focused on theanalytical design of contracting arrangements to elim-inate this inefficiency (see Cachon 2003 for a review).

In this study we undertake an experimental analysisof the simplest supply chain contracting setting thathas been analyzed theoretically, in which the retailerfaces the classic newsvendor problem and orders froma supplier, the supplier has no capacity constraint, anddelivery occurs instantaneously. This simple model isa building block for much of the contracting literaturein operations management, and as such it representsa logical starting point for our initial laboratory study.In this setting, whenever a supplier charges a whole-sale price in excess of his own production cost, dou-

ble marginalization (Spengler 1950) causes the retailerto order less than the channel-optimal amount. Thesesmaller retailer orders imply that all channel part-ners forgo potential profits; consequently, methods foravoiding this inefficiency are valuable because theycan make both parties better off. To coordinate the

supply chain, a contract must give the retailer theincentive to order the same amount that would beoptimal in a centralized setting. Cachon (2003) pro-vides a review of the analytical work that investigatesvarious contractual arrangements that facilitate thiscoordination.

One useful class of arrangements shares the de-mand risk between the retailer and the supplier

by making the suppliers profit depend on realizedsales. Cachon and Lariviere (2005) look at two suchrisk-sharing contracts, the buyback contract and therevenue-sharing contract, and show that the two aremathematically equivalent in the strongest possibleway, meaning that they generate identical outcomesfor both players for each possible realization of thestochastic demand. In the buyback contract, the sup-

plier pays the retailer a rebate on all unsold units, thusassuming some of the risk associated with overorder-ing. In the revenue-sharing contract, the supplierinduces a higher retailer order through a lower whole-sale price, but in return he receives a portion of thegross revenue. Buyback contracts are quite commonin industries such as publishing, computer softwareand hardware, and pharmaceuticals (Padmanabhanand Png 1995), whereas revenue-sharing contracts areobserved in the video-rental industry (Cachon andLariviere 2005).

1953

Additionalinformation,includingrights

andpermissionpolicies,isavailableathttp://journals.informs.org/.


2/16

Katok and Wu: Contracting in Supply Chains: A Laboratory Investigation1954 Management Science 55(12), pp. 19531968, 2009 INFORMS

These two types of contracts and the simple ana-lytical model upon which they are based have gen-erated significant scholarly interest and follow-upstudies. The problem has been extended to twostages (Donohue 2000) and to secondary markets(Rudi et al. 2001, Lee and Whang 1999). Other exten-

sions consider service levels and stockouts (Choiet al. 2004), flexibility (Kamrad and Siddique 2004),incomplete information (Corbett et al. 2004), procure-ment contracts (Wu and Kleindorfer 2005), optioncontracts (Burnetas and Ritchken 2005, Kleindorferand Wu 2003), warranty contracts (Balachandran andRadhakrishnan 2005), and target-rebate contracts onfalse failure returns (Ferguson et al. 2006).

These, along with most other analytical models ofsupply chain coordination, assume that contractingparties are fully rational expected-profit maximizers.But there is now a growing body of evidence based onlaboratory experiments (Schweitzer and Cachon 2000,

Bolton and Katok 2008) as well as some field stud-ies (Corbett and Fransoo 2007) showing that retail-ers have difficulty making optimal decisions even inthe simplest settings. Therefore, it remains unclearwhether the theoretical gains from more complex con-tractual arrangements are likely to be achieved inpractice.

We extend prior research on supply chain coordina-tion and contracting in three ways. First, we investi-gate retailers behavior when faced with coordinatingcontracts and find those contracts to be significantlyless effective than the theory suggests. Second, westudy the suppliers behavior in structuring coordi-

nating contracts and find that suppliers do not offercontracts that fully coordinate the supply chain, evenwhen retailers are programmed to order optimally,given contract parameters. Third, we compare twoequivalent coordinating contractsthe buyback con-tract and the revenue-sharing contractfrom both theretailers and the suppliers perspectives and find thatalthough they do not always induce equivalent out-comes, most of the differences disappear with expe-rience. Loss aversion (Kahneman and Tversky 1979),which Ho and Zhang (2008) have already shown to

be important in contracting, can explain these ini-tial differences. We use a controlled laboratory settingdesigned to conform to the assumptions of the con-tracting models we are testing. Our design is novel inthat it eliminates interpersonal interactions, and thusit allows us to attribute any deviations from theoreti-cal benchmarks to individual decision-making biases.

This paper is structured as follows. In the next sec-tion, we position our work relative to existing ana-lytical, empirical, and experimental research. In 3,we describe the details of the experimental designand the laboratory protocol. We develop our exper-imental hypotheses in 4, and in 5, we report the

results. In 6, we summarize our findings, identifythe limitations of our study and directions for futureresearch, and discuss the managerial implications ofour results.

2. Analytical Background and

Related Literature2.1. Analytical BackgroundIn the baseline model (Spengler 1950), the wholesaleprice contract, the retailer orders q units from a sup-plier at a wholesale price of w per unit. The retailerfaces an exogenous stochastic demand with cumula-tive distribution F( ) and an exogenous market price p,and he suffers losses whenever his actual order quan-tity q differs from the realized demand D. The retailermaximizes his expected profit by balancing the costof ordering too much or too little, and to do that hesets his order q to satisfy

Fq =p w

p

which is known as the critical fractile.In contrast, the supplier faced with the wholesale

price contract incurs no risk, because when the pro-duction cost is c and the wholesale price is w > c, hesimply makes a profit of w c q on the retailersentire order. The wholesale price that maximizes thesuppliers expected profit in the wholesale price con-tract depends on the demand distribution. If F( ) isuniform from A to B then the optimal wholesaleprice w is given by

w = min

p

c

2+

p

2

B

B A

(1)

If we let 0 < < 1 be the retailers share of the totalprofit, a continuum of coordinating risk-sharing con-tracts can be constructed, one for each . If the sup-plier uses a buyback contract in which he charges theretailer wBB per unit, and then refunds b per unit forall units unsold at the end of the selling season, sucha contract coordinates the supply chain (i.e., inducesthe retailer to place the channel-optimal order) whenpairs of parameters {wBB , b} satisfy

b = 1 pwBB = b + c

(2)

Cachon and Lariviere (2005) show that the revenue-sharing contract in which the retailer pays wRS perunit ordered and an additional r per unit sold isequivalent to the buyback contract {wBB , b} when thecost for units sold and unsold is the same under botharrangements:

Per-unit cost of units sold: wBB = wRS + rPer-unit cost of units unsold: wBB b = wRS

(3)




3/16

Katok and Wu: Contracting in Supply Chains: A Laboratory InvestigationManagement Science 55(12), pp. 19531968, 2009 INFORMS 1955

For a uniformly distributed demand, D UAB,expected sales given an order of q are given by

ES =

A + q

2

q A

B A

+ q

B q

B A

and correspondingly, the expected profit amounts for

the retailer, the supplier, and the total supply chainare given by

R ERetailer = p rES wq+ bq ES

S ESupplier = w cq+ rES bq ES

T ETotal = pES cq

(4)

The retailers expected share of the total profit, , isthen = R/T.

2.2. Laboratory ResearchLaboratory studies that investigate the retailers con-

tracting behavior have focused almost exclusivelyon wholesale price contracts. Schweitzer and Cachon(2000) find that retailers place orders that tend to

be between the optimal orders and the averagedemand, and note that this pull-to-center effect can-not be explained by risk preferences, loss aversion,or prospect theory. They suggest that this behavioris consistent with the minimization of ex post inven-tory error and the anchoring and insufficient adjust-ment heuristic. Bolton and Katok (2008) also find thepull-to-center effect and additionally show that perfor-mance improves over time with extensive experience,although slowly, and that requiring decision makers

to place standing orders1 speeds up learning substan-tially. Lurie and Swaminathan (2009) report a similarfinding, specifically that feedback that is too frequentcan degrade performance and slow down learning.Benzion et al. (2008) vary the demand distribution andfind that while orders are affected by both the aver-age demand and the last-period demand, this bias isweakened slowly over time. This implies that partic-ipants learn over time not to chase demand. Bostianet al. (2008) find that an adaptive learning modelexplains the pull-to-center effect. Overall, the evi-dence that human players fail to place expected profit-maximizing orders when faced with the newsvendorproblem is fairly conclusive.

Several recent papers study suppliers and retail-ers behavior jointly and find that profits tend to bedistributed more equitably and the efficiency of coor-dinating contracts is lower than the standard theorypredicts. Keser and Paleologo (2004) find that in astochastic demand setting, newsvendor retailers are

1 In this setting, a standing order refers to a restriction that forcesa retailer to place one order that is used for several consecutiveperiods (in Bolton and Katok 2008, this was 10 periods).

likely to reject contracts with high wholesale prices,and therefore suppliers tend to choose wholesaleprice contracts that split profits approximately equallywhen the entire order is sold. Ho and Zhang (2008)look at a bilateral monopoly setting (deterministicdownward-sloping linear demand) and report that

two-part tariffs and quantity discount contracts fail tocoordinate the supply chain or even achieve a levelof efficiency that is significantly above the whole-sale price contract efficiency levels. They attributecoordination failure to a combination of loss aver-sion and bounded rationality. Loch and Wu (2008)also study the wholesale price contract in a bilateralmonopoly setting, but in their study the retailer andthe supplier interact repeatedly. They find that effi-ciency decreases when players are concerned aboutstatus, and increases when they are concerned abouttheir relationship. The present study is the first labora-tory investigation of the performance of risk-sharing

coordinating contracts.Standard operations management models followeconomic assumptions about human behavior, includ-ing that players are fully rational expected-profit max-imizers. This assumption implies that players want tomaximize only their own expected profit and have thecognitive ability to do so. In practice, human decisionmakers negotiate contracts, and they may violate thisstandard theoretical assumption for one of the follow-ing reasons:

(1) Bounded rationalityDecision makers want tomaximize their expected profit, but make errors indoing so, or resort to heuristics.

(2) Different utility functionsDecision makers max-imize a utility function that includes other attributesin addition to the expected profit, such as loss aver-sion or concern for fairness.

When two human players interact, it is well estab-lished that they are motivated by concerns for fairness.Cui et al. (2007) incorporate these fairness concernsinto an analytical model of contracting. De Bruyn andBolton (2008) report on a metastudy that shows that asimple model incorporating fairness explains a largevariety of data. Loch and Wu (2008) provide a reviewof the way that general social preferences affect oper-ations management models.

But to understand the effect of social preferences,it is important to recognize which deviations fromtheoretical benchmarks are due to social preferences,and which are due to other decision-making biases.We designed our study with this initial step in mind.In the present study, human retailers deal with com-puterized suppliers, and human suppliers deal withcomputerized retailers. All human decision makersin our study interact with computerized players pro-grammed to act according to theory. In Figure 1 wedepict the scope of the contracting problem and showhow our study fits into the larger picture. We call




4/16


Figure 1 Scope of the Contracting Problem and Our Study

Supplier isNot human Human

Not human

Standard theory:

Full rationality

Expected-profit maximization

The supplier game:

Bounded rationality

Individual biases

Decision: {w/b/r}

Retaileris

Human

The retailer game:

Bounded rationality

Individual biases

Decision: {q}

Practice:Bounded rationality

Individual biasesSocial preferences

Decision: {q} and {w/b/r}

the setting with human retailers interacting with com-puterized suppliers the retailer game and the settingwith human suppliers interacting with computerizedretailers the supplier game. In both games, the retailerfaces a single-period newsvendor problem.

Our study design eliminates the possibility thatplayers are motivated by social preferences such asfairness, and this is both a strength and a weakness.

It is a strength because any deviations from theorythat we discover must be attributed either to indi-vidual biases or to bounded rationality, but definitelynot to social preferences.2 Bounded rationality, andto some extent individual biases, can be eliminatedor at least lessened through education and the useof decision-support tools, but social preferences can-not be eliminated (nor is it desirable to try to do so).A thorough understanding of which deviations fromtheory can be overcome through the use of technol-ogy, and which instead should be included in modelsto make the models more valid, is valuable. A studythat looks at how two human decision makers nego-tiate contracts is an important next step, and in theconcluding section we discuss this direction for futureresearch, informed by the present study.

3. Experimental Implementation

3.1. Experimental DesignContracting arrangements in our experiments are thewholesale price contract (W, the buyback contract(BB), and the revenue-sharing contract (RS). In alltreatments we set the suppliers production cost to bec = 3 and the retail price to be p = 12 to create a settingwith potential high supply chain profits. We focus

on the high-profit condition (critical fractile > 1/2 inthis paper because it is a setting with greater possiblegains from coordination.

To investigate the effect of loss aversion on behav-ior, we use three different uniform customer demand

2 We thank an anonymous referee for pointing out that the previousstatement assumes that people are capable of consciously makingthe distinction, and that people generally change their behavior inthis task based on the presence of human beings on the other end.In Katok and Pavlov (2009) we show that this is in fact the case ina bilateral monopoly setting.

conditions. In the DLOW condition, D U0100,demand is potentially low (as low as 0), and bothretailers and suppliers can lose money under opti-mal coordinating contracts. In the DHIGH condition,D U50150, demand is potentially greater, andretailers can lose money under optimal coordinat-

ing contracts, but suppliers cannot. Under the whole-sale price contract, by contrast, suppliers cannot losemoney with either of the demand distributions, andretailers can lose with both.

We call the third demand condition DHIGH withDLOW decision frame (DHIGH/LOW). The actualdemand distribution is the same as DHIGH, but wedescribe it in a different way: as 50 guaranteed unitsand an additional number of units from 0 to 100, sothat D = 50 + X, where X U(0,100). X thus followsthe same distribution as the demand in the DLOWcondition. Participants are asked to decide on thenumber of units to order in addition to the 50 guar-

anteed units, so the order quantity is from 0 to 100,as in the DLOW condition. In this demand condition,suppliers cannot earn a negative profit, but if q ishigh and X is low, suppliers can lose money relativeto the revenue they are guaranteed from selling the50 units 50 w + r c. Thus, the DHIGH/LOWdemand condition induces loss aversion through themanipulation of framing. This allows us to exploreloss aversion as a potential explanation for any differ-ences we observe between the buyback and revenue-sharing contracts.

The final factor we manipulate in this study is expe-rience. To test for the effect of experience, we conduct

each treatment twice, first with inexperienced partici-pants, and then again with participants who had priorexperience in the same role (either as retailer or sup-plier) with a different contract. Each session included100 rounds, so each participant played for 200 roundsin total: 100 rounds in the inexperienced session, fol-lowed by 100 rounds in the experienced session.

In summary, our study manipulates the decisionmakers role (retailer game and supplier game), thecustomer demand distribution (DLOW, DHIGH, andDHIGH/LOW), and experience (inexperienced andexperienced). We summarize all treatments and sam-ple sizes in Table 1. In total, 200 subjects participated

in our study.

3.2. Experimental ProtocolAll experimental sessions followed the same proto-col. Participants arrived at the computer lab at a pre-specified time and read experimental instructions thatdescribe the rules of the game, the use of the soft-ware, and the payment procedures (see the onlineappendix, provided in the e-companion).3 After all

3 An electronic companion to this paper is available as part of the on-line version that can be foundat http://mansci.journal.informs.org/.




5/16


Table 1 Experimental Design and Sample Sizes

Retailer game Supplier game

Inexperienced Experienced Inexperienced Experienced

Contract Demand n Demand n Demand n Demand n

Wholesale price DLOW 20 DLOW 20 DLOW 20 DLOW 20

DHIGH 20 DHIGH 20 DHIGH 20 DHIGH 20

Buyback DLOW 10 DLOW 10 DLOW 10 DLOW 10


DHIGH/LOW 10 DHIGH/LOW 10 DHIGH/LOW 10 DHIGH/LOW 10

Revenue-sharing DLOW 10 DLOW 10 DLOW 10 DLOW 10


DHIGH/LOW 10 DHIGH/LOW 10 DHIGH/LOW 10 DHIGH/LOW 10

Note. Demand distributions: DLOW, D U0 100; DHIGH, D U50 150; DHIGH/LOW, D= 50 +X, X U0 100.

participants had a chance to read the instructions,the experimenter read instructions to them aloud toensure common knowledge, used PowerPoint slides

to illustrate examples and formulas, and answeredquestions. Participants then completed 100 roundsunder the first of the two contracts. After all partic-ipants finished this initial (inexperienced) session,we handed them additional instructions describingthe contract used in the second (experienced) ses-sion, gave them a chance to read these new instruc-tions, read the instructions to them aloud, andanswered questions before the beginning of the sec-ond session. After completing the second session, par-ticipants were paid their actual earnings accumulatedfrom both sessions, privately and in cash. Partici-pants were not allowed to communicate during the

experiment.All sessions were conducted at the Laboratory for

Economic Management and Auctions at the SmealCollege of Business, Penn State University. Each ses-sion lasted approximately 75 minutes, and the aver-age earnings, including a $5 participation fee, were$19. Participants were Penn State students recruitedthrough a Web-based recruitment system, with cash

being the only incentive offered. The majority of ourparticipants were undergraduates from a variety ofmajors (77%), and the rest were graduate students.We compared the average earnings by student level,major, and gender for each session using a t-test

and found no response biases by those demographiccharacteristics.4

3.3. Experimental ImplementationIn the retailer game, the human decision maker playsthe role of the retailer, and we set the wholesale

4 The human subject approval for this study required us to store thedecision data using subject IDs, and therefore we do not have a wayof connecting these IDs to individuals. However, our recruitmentsystem gives us a way to track individual participants and theirearnings by session, and we used this information for the analysis.

price w optimally for both demand conditions usingEquation (1) as follows:

wDLOWW = min

123

2 +12

2

100

100 0

= 75

wDHIGHW = min

12

3

2+

12

2

150

150 50

= 105

(5)

For the buyback contract, we used Equation (2) andset = 1/3 so that both parties could benefit fromcoordination to obtain wBB = 9 and b = 8.

5 We thenconstructed the equivalent revenue-sharing contractusing Equation (3), with r= 9 and wRS = 1.

In the supplier game, the human decision makerplays the role of the supplier. Here, our aim was to

better understand the extent to which suppliers are

able and willing to offer contracts that coordinate. Thedesign includes an automated retailer programmedto act in accordance with theory: the retailer is pro-grammed to place expected-profit maximizing orders,given the contract offered by the human supplier. Thisfeature eliminates strategic interactions and also pro-vides consistent feedback. Because F( ) is UAB, the

best-response order quantity of the automated retaileris given by

q = A + B A

p w r

p b r

(6)

where A = 0 and B = 100 in the DLOW condition andA = 50 and B = 150 in the DHIGH and DHIGH/LOWconditions. Given these parameters, the order quantitythat maximizes the retailers expected profit under thewholesale price contract is 37.5 in the DLOW condi-tion and 62.5 in the DHIGH condition. Under the twocoordinating contracts, the optimal order quantity is75 in the DLOW condition and 125 in the DHIGH andDHIGH/LOW conditions.

5 We wanted to avoid = 1/2 so as not to confound our results bya 50/50 split.




6/16


We set b = r = 0 for the wholesale price con-tract, and participants selected w only. In the buy-

back contract we set r = 0, and participants selectedw and b simultaneously; in the revenue-sharing con-tract we set b = 0, and participants selected w and rsimultaneously.

Because the retailer in the supplier game is auto-mated, the system provides feedback to suppliersabout the order quantity that will follow a proposedcontract, specifically, q as defined by Equation (6).Each participant was allowed to try different con-tract parameters and observe the expected (but notthe actual) outcome as many times as desired beforemaking the final decision for the round. We repeatedthis procedure for all treatments in the supplier game(W, BB, and RS) to make certain our participantshad access to the relevant information that the theoryimplicitly assumes they have, thus giving the theoryits best shot.

4. Research HypothesesRecall that the human decision makers make differentdecisions in the two games: in the retailer game thedecision is the order quantity q, whereas in the sup-plier game the decision is a set of contract parametersw b r. However, the retailers order quantity q can

be a unifying metric for both games, because the totalchannel profit is always proportional to q. Therefore,we formulate our first three research hypotheses andconduct the corresponding data analysis in terms of q.

All four hypotheses apply to both the retailer gameand the supplier game. (When we talk about the

retailers average order in the context of the suppliergame, we mean the average order of the automatedretailer that the human supplier induces.) The firsthypothesis follows directly from the quantitative pre-dictions of the standard theory.

Hypothesis 1 (Theoretical Benchmarks). Theretailers average orders for wholesale price contracts willbe 37.5 in the DLOW condition and 62.5 in the DHIGHcondition. The average orders for the buyback and revenue-sharing contracts will be 75 in the DLOW condition and125 in the DHIGH and DHIGH/LOW conditions.

Even if the data deviate from these precise theo-

retical benchmarks, the theory can still make usefulqualitative predictions. The main point of the stan-dard theory summarized in 2.1 is that the buy-

back and revenue-sharing contracts can induce higherorders than the wholesale price contracts. Our secondhypothesis reflects this qualitative prediction.

Hypothesis 2 (Coordination). The retailers averageorders for the buyback and revenue-sharing contracts willbe higher than for the wholesale price contract.

Our third hypothesis links theoretical predictionswith known behavioral biases. Previous studies have

documented a pull-to-center effect (Schweitzer andCachon 2000, Bostian et al. 2008) in which aver-age orders are located between the optimal ordersand the average demand.6 Schweitzer and Cachon(2000) note that the data, although inconsistent withmany established behavioral models (risk preferences,

prospect theory, loss aversion) are in fact consistentwith (i) the anchoring and insufficient adjustmentheuristic, and (ii) a preference for minimizing ex postinventory error. Although these two explanationscan each account for the pull-to-center effect, theyinvolve different adjustment patterns. The anchor-ing and adjustment heuristic implies that orders startclose to the average demand and adjust over time inthe direction of optimal orders, which may be (butneed not be) away from average demand. Minimiz-ing ex post inventory error, in contrast, implies thatorders are positively correlated with past demand.Although not mutually exclusive, the two explana-

tions yield different implications about the way thatorders adjust over time, which leads us to our thirdhypothesis.

Hypothesis 3 (Causes for the Pull-to-CenterEffect).

A. Anchoring and Insufficient Adjustment:The retailers orders for all contracts will adjust, over time,toward the optimal order.

B. Minimizing Ex Post Inventory Error: Theretailers orders will be positively correlated with pastdemand.

Our fourth hypothesis deals with the mathemati-

cal equivalence of the buyback and revenue-sharingcontracts and a possible reason that this equivalencemay fail. If these contracts are equivalent, we shouldnot observe any differences in the performance of thetwo mechanisms in the supplier game, either in theretailers orders or in the contract parameters in termsof the per-unit cost of units sold and unsold. (In theretailer game, we set the parameters such that the costof sold units is 9 and the cost of unsold units is 1.)This theoretical equivalence leads to the first part ofour fourth hypothesis:

Hypothesis 4A (Mathematical Equivalence of

the Revenue-Sharing and Buyback Contracts).The retailers average orders and the per-unit cost of soldand unsold units for the buyback and revenue-sharing con-tracts will be equal.

Because the two contracts are mathematicallyequivalent as long as Equation (3) holds, any dif-ferences we observe between them in the laboratory

6 The previous literature on ordering behavior in the newsvendorproblem that we discussed in 2 deals with the wholesale pricecontract only.




7/16


must be due to framing (see Soman 2004 for a reviewof the literature on framing). In other words, themathematical equivalence of the contracts may not

be immediately apparent to participants. We proposethat loss aversion is a potential cause for these differ-ences. In a bilateral monopoly setting, Ho and Zhang

(2008) cite loss aversion as the explanation for thedifferences they observe between the quantity dis-count and the two-part tariff contract, which are alsomathematically equivalent. In our setting, loss aver-sion is a plausible driver because the wholesale price,which is low under the revenue-sharing contract andhigh under the buyback contract, is a payment thatdepends directly on the order amount, and is there-fore clearly viewed as a loss for the retailer and a gainfor the supplier. But the effect of the buyback rebateand the revenue share depends on the realization ofthe uncertain demand. Loss aversion is affected byframing because people dislike losses more than they

dislike forgone gains (Kahneman and Tversky 1979,Kahneman et al. 1990), and thus the way that weframe demand distribution affects how participantsperceive gains and losses.

In the DLOW condition, the retailer is not guaran-teed any revenue (because the demand can be as lowas 0), so the potential losses to the retailer from pay-ing the wholesale price loom large, whereas potentialgains from a rebate (b for unsold units or the forgonelosses from having to pay a revenue share (r for thesold units seem less salient. Consequently, the low up-front wholesale price of the revenue-sharing contractmay be more effective than the higher rebate of the

buyback contract in inducing high retailer order quan-tities in the retailer game. In the supplier game, how-ever, the situation is exactly reversed, because for therevenue-sharing contract to coordinate, the wholesaleprice must be below the suppliers cost. Consequently,in the supplier game, the high wholesale price of the

buyback contract may be more effective in inducinghigher retailer orders than the high revenue share ofthe revenue-sharing contract.

Hypothesis 4B.1 (Loss Aversion with DLOWDemand). In the DLOW demand condition, averageretailer orders will be higher under the revenue-sharing

contract in the retailer game and higher under the buybackcontract in the supplier game.

In the DHIGH demand condition, the retailer isguaranteed to sell at least 50 units, and thus the rev-enue share (r becomes a salient loss for the retailerin the retailer game, but a salient gain for the supplierin the supplier gameat least for those 50 units. Thismakes the revenue-sharing contract look like it penal-izes high orders in the retailer game and rewardshigh orders in the supplier game. The buyback con-tract, in contrast, appears to rewards high orders in

the retailer game and to penalize them in the suppliergame, because a substantial rebate is paid for unsoldunits.

Hypothesis 4B.2 (Loss Aversion with DHIGHDemand). In the DHIGH demand condition, averageretailer orders should be higher under the buyback contract

in the retailer game and higher under the revenue-sharingcontract in the supplier game.

When we describe the DHIGH demand as 50 guar-anteed units plus a random number of additionalunits and frame the decision in terms of the numberof additional units to order, we move the task into theDLOW frame, and just as in the DLOW condition, therevenue-sharing contract should look more appeal-ing to loss-averse retailers in the retailer game (andless appealing to loss-averse suppliers in the sup-plier game) than the buyback contract because eitherpaying or receiving the revenue share for the first

50 guaranteed units no longer seems to be part of thedecision.

Hypothesis 4B.3 (Loss Aversion with DHIGH/LOW Demand). In the DHIGH/LOW demand condition,average retailer orders will be higher under the revenue-sharing contract in the retailer game and higher under thebuyback contract in the supplier game.

Note that the comparison between the two contractsis the same in the DLOW condition (Hypothesis 4B.1)as in the DHIGH/LOW condition (Hypothesis 4B.3)and is the opposite of that in the DHIGH condition(Hypothesis 4B.2). Whether we observe this rever-

sal is the main test of the loss-aversion and framingexplanation.

5. Results

5.1. Hypothesis 1 Results: Theoretical BenchmarksIn Table 2, we provide descriptive statistics for theretailer order quantities in all treatments, includingthe average retailer order, standard deviations, andmedian orders. We used the one-sample Wilcoxon test(Siegel 1956, pp. 7583) to make the comparisons andconducted the test separately for inexperienced andexperienced sessions. The unit of analysis is the aver-

age order of an individual subject. Because the deci-sion in the supplier game cannot be fully described

by the retailer order induced by the contract, we alsoprovide, in Table 3, descriptive statistics for the con-tract parameters in the supplier game treatments.7

7 Any order quantity q can be induced in many different ways,depending on how a particular contract distributes supply chainprofits between the two parties. So the decision in the suppliergame can be fully described by the retailer order induced ( q andthe retailers profit share (), or equivalently and more directly bythe cost of the sold and unsold units.




8/16


Table 2 Average Retailer Orders for All Treatments

Retailer game Supplier gameBest-response

Contract quantity (q Inexperienced Experienced Inexperienced Experienced

DLOW: D U0 100

Wholesale 37.5 4204 4172 3744 3959

491 692 332 508

4124

4164

3768 3784Buyback 75 5555 5769 5705 5762

589 1232 1041 1548

5592 5886 5876 5337

Revenue- 75 6407 6913 4968 5060

sharing 920 1266 1181 1715

6193 7413 4993 4649

DHIGH: D U50 150

Wholesale 62.5 8123 8049 6635 6689

1019 1017 357 804

8277 8147 6469 6460

Buyback 125 11581 10610 8700 8954

1156 1729 1628 2269

11360 10445 8302 8539

Revenue- 125 10285 10456 9846 9447

sharing 1617 1965 1138 1018

10493 10423 9852 9634

DHIGH/LOW: D= 50 +X, X U0 100

Buyback 125 10271 10942 10055 9758

983 1372 2121 2195

10426 10885 10230 9577

Revenue- 125 11503 10817 10221 10120

sharing 1520 1336 1280 1932

10957 11097 10084 10807

Notes. For Hypothesis 1, Ho q= q Standard deviations are reported in

parentheses and median orders in square brackets.p < 005; p < 001.

The only retailer game treatment for which we can-not reject Hypothesis 1 is the DLOW condition ofthe revenue-sharing contract in the experienced ses-sion. Median orders in the other treatments of theretailer game are generally higher than the optimalorders under wholesale price contracts and are gen-erally lower than the optimal orders under the twocoordinating contracts.

The most convenient way to evaluate the buybackand revenue-sharing contracts in the supplier game is

by comparing the cost of sold and unsold units. In thebuyback contract, the retailer pays wBB for each unitsold and wBB b for each unit unsold. In the revenue-

sharing contract, the retailer pays wRS+ r for each unitsold and wRS for each unit unsold. The two contractsare equivalent if the cost of the sold and unsold unitsare equal. (In the wholesale price contract, the retailerpays w for each unit whether or not it is sold.) There-fore, in the supplier game, we can analyze Hypoth-esis 1 in two ways, through retailer orders induced

by the contracts (Table 2), and directly through thecontract parameters (Table 3).

Under the wholesale price contract in the DLOWcondition, median retailer orders do not differ from

Table 3 Average Cost of Sold and Unsold Units in Supplier Game

Treatments

Cost of units unsold

Cost of units soldOptimal

Contract Inexperienced Experienced (given sold) Inexperienced Experienced

DLOW: D U0 100

Wholesale 731 744 750 731 744081 088 081 088

750 750 750 750

Buyback 880 877 100 255 287

162 119 165 144

900 900 200 300

Revenue- 879 880 105 441 348

sharing 164 111 233 196

900 872 500 300

DHIGH: D= U50 150

Wholesale 985 1016 1050 985 1016

132 094 132 094

1050 1050 1050 1050

Buyback 1020 1025 058 390 404

107 091 224 330

1000 1050 350 300Revenue- 950 1026 068 289 239

sharing 126 081 164 162

990 1000 254 200

DHIGH/LOW: D 50 +X, X U0 100

Buyback 962 988 067 332 344

167 141 245 258

1000 1000 300 300

Revenue- 975 1001 067 238 259

sharing 133 103 155 234

1000 1000 204 200

Note. Standard deviations are reported in parentheses and median levels in

square brackets.

optimal orders (37.5), and median wholesale pricesalso are not different from the optimal prices (7.5),consistent with Hypothesis 1. In the DHIGH condi-tion, however, even though median wholesale pricesdo not differ from the optimal level (10.5), medianretailer orders are slightly above optimal, so hereHypothesis 1 is only partially supported.

We can see from Table 2 that retailer orders underboth coordinating contracts in the supplier game aresignificantly below optimal, and Table 3 reveals why. Ifwe look at the cost of unsold units, the column labeledOptimal (given sold) tells us what the cost of unsoldunits should be, given the median cost of sold units.

(Because median costs of sold units are very similarin the inexperienced and experienced sessions, we usethe average of the two medians for the optimal calcula-tion.) Note that the actual cost of unsold units that oursuppliers charge is always substantially higher than itneeds to be to coordinate the channel. This is evidencethat our suppliers are unwilling to assume enough riskto coordinate the channel.

5.2. Hypothesis 2 Results: CoordinationIn Table 4 we summarize the results of our testsof Hypothesis 2. The table shows the differences in




9/16


Table 4 Differences in Median Orders Under Wholesale Price and

Coordinating Contracts


Contract Inexperienced Experienced Inexperienced Experienced

DLOW: D= U0 100

Buyback 1468

1722

2108

1553

Revenue-sharing 2069 3249 1225 865

DHIGH: D= U50 150

Buyback 3083 2298 1833 2076


DHIGH/LOW: D= 50 +X, X U0 100

Buyback 4176 4692 3980 3327


Note. For Hypothesis 2, Ho qW = qBBqW = qRS.p < 005; p < 001.

median order under the wholesale price contract ver-sus under the coordinating contracts. We use the

MannWhitney U test (Siegel 1956, pp. 116127) tomake the comparisons, and we conduct the test sepa-rately for inexperienced and experienced sessions. Theunit of analysis for this comparison again is the aver-age order of an individual subject. We find supportfor Hypothesis 2 in all treatments. Both the buybackand revenue-sharing contracts induce higher retailerorders than the wholesale price contract.

5.3. Hypotheses 3A and 3B Results: Causes forthe Pull-to-Center Effect

Because Hypotheses 3A and 3B deal with behaviorover time, we test it using a regression model that

we fit for each demand condition and contract type.We fit the inexperienced and experienced sessionsseparately.

Qi t = Intercept + t t 2 + D1

Di t1 + i + i t (7)

In this model, the dependent variable Qi t is partic-ipant is order in period t. (In the supplier game itis the automated retailers order induced by partici-pant is choice of contract parameters in period t.) Thevariable Di t1 is the difference between the demandthat participant i observed in period t 1 and theaverage demand under the given demand condition(i.e., 50 in the DLOW and 100 in the DHIGH andDHIGH/LOW conditions). The variable t 2 capturesthe time trend. Note that there are two error compo-nents in the model: one that is independent acrossall observations, i t, and one that is participant spe-cific, i. Each error term has a mean of zero andsome positive standard deviation. This treatment ofthe individual effect is known as the random-effectsmodel, and it is used to control for individual het-erogeneity. We present regression estimates for Equa-tion (7) in Table A.1 in the appendix. In Table 5, to

make the exposition easier, we present the analysisrelated to Hypotheses 3A and 3B that includes theestimates of the Intercept term and the signs of thetwo coefficients.

Because the variable Di t1 captures the differencebetween the demand observed in the previous period

and the average demand, the intercept can be inter-preted as the average order at the beginning of thesession (at t = 2) measured at average values of thedemand in period 1. At that point, t 2 = 0.

Hypothesis 3A implies that, over time, ordersshould move toward optimal levels, meaning that ifthe intercept is above q, the coefficients on t 2should be negative; if the intercept is below q, thecoefficients on t 2 should be positive; and if theintercept does not differ from q, the coefficients ont 2 should be 0. Under the wholesale price con-tract, the intercept is always above q, and the coef-ficient on t 2 is always negative, consistent with

Hypothesis 3A. Under the buyback and revenue-sharing contracts, when the intercept is below q,some of the coefficients on t 2 are negative ornot significant, contrary to Hypothesis 3A. In thetwo cases in which the intercept is not significantly

below q, the coefficients on t 2 are negative, againcontrary to Hypothesis 3A. So we find support forHypothesis 3A under the wholesale price contract,

but not under the two coordinating contracts.Hypothesis 3B implies that the coefficients on Dt1

should be positive. We find that they are in fact posi-tive and significant in each of the retailer game treat-ments and not significant in any of the supplier game

treatments. So we find support for Hypothesis 3B inthe retailer game, but not in the supplier game.

The bottom line is that whereas anchoring andadjustment appears to be a reasonably accuratedescription of behavior under the wholesale pricecontract, we find no consistent evidence of this behav-ior under coordinating contracts. In some treatments,average orders move toward the optimum, and inother treatments they move away from it. But in alltreatments of the retailer game there is a positive cor-relation between orders and last-period demand, indi-cating that the desire to minimize ex post inventoryerror (the regret from having a mismatch between theorder and the demand, as Kremer et al. (2007) pointout) plays a role.

5.4. Hypotheses 4A4B.3 Results: EquivalenceHypotheses 4A4B.3 deals with the equivalence of the

buyback and revenue-sharing contracts. Note that inthe retailer game the two contracts are literally equiv-alent because we set contract parameters to makethem so, per Equation (3). In the supplier game,the two contracts can be equivalent if suppliers setcontract parameters appropriately, again according to




10/16


Table 5 Results of Analysis of Hypotheses 3A and 3B


DHIGH/LOW: DHIGH/LOW:

DLOW: DHIGH: D= 50 +X, DLOW: DHIGH: D= 50 +X,

D= U0 100 D= U50 150 X U50 150 D= U0 100 D= U50 150 X U50 150

Variable Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced

Wholesale price contract

Intercept 4342 4301 8459 8361 3833 4094 7102 7101

123 164 235 232 079 114 081 180

t 2

Dt1 + + + + 0 0 0 0

Buyback contract

Intercept 5802 5866 12018 10255 9382 10374 6095 5772 8787 8500 9596 9210

213 396 390 554 329 442 348 482 522 686 679 706

t 2 0 + + + 0 0 + + +

Dt1 + + + + + + 0 0 0 0 0 0

Revenue-sharing contract

Intercept 6587 6699 10013 10217 10619 11915 4844 4830 9454 8867 10128 9900

303 405 522 626 432 489 382 546 318 412 616 421

t 2 + + + + + + + + 0 +

Dt1

+ + + + + + 0 0 0 0 0 0

Note. Interceptstandard deviations are reported in parentheses.Intercept= q, p < 005; , < 0, p < 005; +, > 0, p < 005; 0, p > 005.

Equation (3). We compare the two coordinating con-tracts in three ways. First, in both the retailer andthe supplier games we compare the average retailerorders (Q. Second, because the average retailer orderdoes not fully describe the supplier game contract,we also compare the average cost of sold and unsoldunits in that game. Together, Q and the cost of soldand unsold units fully describe the supplier gamecontracts. To test Hypotheses 4A4B.3, we then fitthree models for the buyback and revenue-sharingtreatments in the retailer and supplier games. The firstmodel looks at the retailers order in both games:

Qit = Intercept +RSRS+t t 2+RSt

RSt 2+D1 Dit1 +i +it (8)

The next two models look at the cost of sold andunsold units in the supplier game only:

Soldit = Intercept+RSRS+t t 2+RSt

RSt 2+D1 Dit1 +i +it

Unsoldit = Intercept +RSRS+t t 2

+RSt RSt 2+D1

Dit1 +i +it

(9)

where

Soldi t =wBB for buyback

wRS + r for revenue-sharing,

Unsoldi t =wBB b for buyback

wRS for revenue-sharing.

We estimate these models for each demand conditionand for inexperienced and experienced sessions sep-arately. Note that the independent variables we usein Equations (8) and (9) are the same as in Equa-tion (7), but we include an additional variable, RS,to measure the differences between the buyback andrevenue-sharing contracts. This variable is the focusof our analysis. The variable RS takes a value of 1for revenue-sharing treatments and 0 for the buybacktreatment, so RS measures the differences betweenthe two contracts (in terms of the dependent vari-ables). Because we know from the estimates of Equa-tion (7) that time trends under the buyback andrevenue-sharing contracts are not always the same,we added the interaction variable RS t 2 tocontrol for the differences in the way that partici-pants adjust their decisions over time under the twocontracts. Because the coefficients on Di t1 are verysimilar for the buyback and the revenue-sharing con-tracts in estimating Equation (7), we do not need asimilar interaction term between RS and Di t1.

8 Wereport

RS

estimates in Table 6 and the full results ofestimates of Equations (8) and (9) in Tables A.2A.5in the appendix.

Hypothesis 4A predicts that none of the RS coeffi-cients should be significant. We do, however, observesome significant RS coefficients when the dependentvariable is Qi t . Also, one of the RS coefficients is sig-nificant when the dependent variable is Soldi t, and

8 However, when we add such an interaction term to the model,the interaction variables are not significantly different from 0, andthe rest of the estimates remain virtually unchanged.




11/16


Table 6 Estimates of RS


Qi t Qi t Sold Unsold

Contract Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced

DLOW: D U0 100 740 818 963 919 002 014 234 113

369 567 475 729 059 045 078 070

DHIGH: D U50 150 1989 079 1411 388 057 003 119 166

652 836 568 821 028 032 072 104

DHIGH/LOW: 2532 243 915 301 016 038 129 180

D= 50 +X, X U0 100 589 617 935 798 058 044 074 097

Notes. For Hypotheses 4A4B.3, Ho RS = 0. Standard deviations are reported in parentheses.p < 005; p < 001.

another is significant when the dependent variableis Unsoldi t . We conclude that the data are gener-ally not consistent with Hypothesis 4A, because weobserve some differences between the two coordinat-

ing contracts.The next question is whether the differences weobserve between the coordinating contracts are con-sistent with loss aversion, and whether the differ-ences are affected by the framing. Hypotheses 4B.1and 4B.3 predict that in the retailer game when thedependent variable is Qi t, the RS coefficient will bepositive in the DLOW and DHIGH/LOW conditions;Hypothesis 4B.2 predicts that it will be negative inthe DHIGH condition. The sign of the RS coefficientin the supplier game thus should be the opposite ofits sign in the retailer game. Because in every casein which the RS coefficient is significantly different

from 0 its sign is consistent with the predictions ofHypotheses 4B.14B.3, we find that the data offersome support for these hypotheses. The strongest evi-dence we have to offer in support of the loss-aversionhypothesis is that in the retailer game, the sign ofthe RS coefficient switches from being strongly nega-tive in the inexperienced session of the DHIGH condi-tion to strongly positive in the inexperienced sessionof the DHIGH/LOW condition. The only difference

between the inexperienced sessions of those two treat-ments is how we framed the demand distribution,and we therefore conclude that framing (in particu-lar the effect of framing on loss aversion) is the only

possible explanation.Another regularity we observe is that none of the

RS coefficients is significant in the experienced ses-sions. This implies that to the extent that loss aversionaffects behavior, it tends to lessen, and usually disap-pear, with experience in our setting. In other words,as participants gain experience, they are less affected

by framing. So we find some evidence consistent withloss aversion (Hypotheses 4B.14B.3) but also findthat the effect of loss aversion appears to wear offover time, to the point that when participants play for

the second time, there is generally no detectable dif-ference between the two coordinating contracts. Thisis consistent with Hypothesis 4A.

6. Summary, Limitations, andManagerial Implications

6.1. Summary of ResultsIn this laboratory study we compare the perfor-mance of the wholesale price contract and two typesof coordinating risk-sharing contracts, the buybackand revenue-sharing contracts. We first look at howretailers respond to different mechanisms; we thenexamine suppliers willingness and ability to takeadvantage of coordinating contracts.

We find that, consistent with earlier studies, retail-ers on average place orders that are between the

profit-maximizing order and the average demand.In the context of wholesale price contracts, aver-age retailer orders are higher than the expected-profit maximizing benchmark, and they adjust inthe direction of the optimal order over time. Thisinitial overordering behavior causes wholesale pricecontracts to perform better in our laboratory settingthan in theory.9 Coordinating contracts induce higherretailer orders than do wholesale price contracts, butthey fall short of the channel-optimal solution, so theyperform worse in the laboratory than in theory.

We also find that suppliers, when interactingwith computerized retailers that are programmed to

respond optimally, quickly find the wholesale pricethat maximizes their expected profit. We do not, how-ever, observe this under either of the coordinatingcontracts, where suppliers induce retailer orders that

9 The fact that wholesale price contracts perform better than theyshould in theory is because the optimal order in our experiments is

below average demand. When the optimal order is above averagedemand, actual orders are below the optimal level (see Schweitzerand Cachon 2000, Bolton and Katok 2008). So we cannot concludethat wholesale contracts generally perform better in the laboratorythan in theory.




12/16


are significantly below the channel-optimal level. Weoffer two explanations, the first related to boundedrationality, and the second related to preferences. Oneexplanation is that suppliers do not face any demandrisk under the wholesale price contract, but they dounder coordinating contracts, so they perform better

under the wholesale price contracts because deter-ministic problems are generally easier to solve thanare stochastic problems. A related issue is that sup-pliers have to select two different contract parame-ters under coordinating contracts, which is a morecomplex task than selecting a single parameter underthe wholesale price contract. A review of the con-tract parameters that our suppliers chose points to asecond explanation. Suppliers systematically set thecost of unsold units significantly above the channel-optimal level, given the cost they set for sold units.Thus, suppliers do not take enough risk to coordinatethe channel. With human retailers, who are likely to

place orders that are even lower than our computer-ized retailers, the effectiveness of coordinating con-tracts may well be even worse.

Our study is the first to examine coordinatingcontracts in the laboratory in an environment withstochastic customer demand. We find that particu-larly for coordinating contracts, the changes in retailer

behavior over time are more consistent with the pref-erence for minimizing ex post inventory error thanwith the anchoring and adjustment heuristic. This isevidenced by the fact that in every treatment of theretailer game, retailers placed orders that were posi-tively correlated with last-period demand. Suppliers,

though, did not choose contract parameters in a waythat caused the retailer orders they induced to be pos-itively correlated with past demand.

We find that the two mathematically equivalentrisk-sharing contracts do not initially induce identicalretailer or supplier behavior in the laboratory; how-ever, the observed differences tend to decrease anddisappear with experience. We offer a framing expla-nation for this initial lack of equivalence, specifically,that participants perception of the demand distri-

bution is an influential factor. The buyback contractemphasizes the benefit of placing higher orders, soit is more effective when the demand distribution isframed in terms of a minimum amount plus the pos-sibility of an upside. The revenue-sharing contract,which emphasizes low upfront cost, is more effec-tive when the demand distribution frames the deci-sion as one with less of an upside and a seriouspotential downside. The results from our study of theDHIGH demand distribution with the DLOW deci-sion frame demonstrate that although the framingconcept is useful for explaining initial differences in

behavior, differences due to framing tend to disappearwith experience.

6.2. Limitations and Directions forFuture Research

Our study is subject to two main limitations that pointtoward fruitful directions for future research. The firsthas to do with the subject pool. We used a sub-

ject pool that is common in experimental economics

(see Holt 1995, Kagel and Roth 1995) comprisingstudents, mostly undergraduates, recruited throughadvertisements offering an opportunity to earn cash.Our subject pool is representative of the larger stu-dent population at Penn State in terms of gender,majors, and, to the extent we are able to determine,ethnicity. A number of studies have found no differ-ence between the performance of students and profes-sionals in laboratory experiments (see, e.g., Plott 1987,Ball and Cech 1996, Katok et al. 2008). However, inthe context of the newsvendor problem, Bolton et al.(2008) investigated the effect of learning and foundthat instruction regarding how to solve the newsven-

dor problem was highly effective with students buthad almost no effect on managers. Clearly, the effectof the subject pool on the outcome of operations man-agement experiments is a complicated and importantquestion, deserving of future study.

The second limitation is that we study the behav-ior of retailers and suppliers separatelyour retail-ers and suppliers do not interact. This design wasintentional, because we wanted to understand indi-vidual decision making unaffected by social utilityconsiderations such as preferences for fairness (Fehrand Schmidt 1999, Bolton and Ockenfels 2000, Cuiet al. 2007). In that respect, our study represents an

intermediate step, because it offers a standard forcomparison that can be used to separate the effectof social preferences from other individual decision-making biases. Whereas Keser and Paleologo (2004)reported that in their study, which included twohuman players, wholesale prices were consistently

below optimal, we found that in our study suppliersquickly identified profit-maximizing wholesale pricecontracts. Based on our findings, we can say with afair degree of confidence that Keser and Paleologos(2004) result is likely a consequence of social prefer-ences, which was not their original conclusion.

Real contractual arrangements, however, are nego-tiated by human participants on both sides, so a bet-ter understanding of how these contracts compare inreality would be gained through a study that includesthe interaction of human retailers and suppliers. Wu(2009) follows this direction, extending our exper-imental design to investigate the strategic interac-tions between supply chain partners under buybackand revenue-sharing contracts. In Wus (2009) study,retailers are constrained to place the optimal order orto punish the supplier by either ordering a quantity of0 or ordering the minimum possible demand amount.




13/16


Wus (2009) main finding is that in that environmentparticipants learn, over time, to negotiate contractsthat are more efficient than the contracts in our sup-plier game, while dividing expected profits in a moreequitable manner.

6.3. Managerial ImplicationsOur results indicate that decision-support tools forcontract design could increase the effectiveness of con-tractual arrangements for both suppliers and retail-ers. From the retailers perspective, tools are neededto counteract the demand-following behavior thatresults from participants trying to minimize ex postinventory error. It is this behavior that decreasesthe effectiveness of coordinating contracts by makingretailers less responsive to economic incentives (seeKremer et al. 2007). From the suppliers perspective,decision-support tools that help set contract parame-ters properly may well go a long way to increase not

only the total supply chain efficiency, but also the sup-pliers profit. When buyback contracts are prevalent inan industry, the natural tendency of contract design-ers may be to set both the wholesale price and therebate too low. Alternatively, a lack of understandingof contracting mechanisms may lead to such obviouslysuboptimal contracts as the offering of full rebates,

Appendix

Table A.1 Tests of Hypotheses 3A and 3B: Estimates of Equation (7)


DHIGH/LOW: DHIGH/LOW:

DLOW: DHIGH: D= 50 +X, DLOW: DHIGH: D= 50 +X,

D= U0 100 D= U50 150 X U50 150 D= U0 100 D= U50 150 X U50 150

Variable Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced

Wholesale price contract

Intercept 4342 4301 8459 8361 3833 4094 7102 7101

123 164 235 232 079 114 081 180

t 2 0028 0025 0069 0068 0013 0023 0097 0072

00109 00102 00109 00090 00045 00041 00053 00054

Dt1 00948 00919 01079 00932 00006 00024 00024 00047

00108 00101 00108 00090 00044 00040 00053 00054

Buyback contract

Intercept 5802 5866 12018 10255 9382 10374 6095 5772 8787 8500 9596 9210

213 396 390 554 329 442 348 482 522 686 679 706

t 2 00510 00190 0069 00751 01829 01144 00915 00020 00180 00972 00881 01054

00211 00154 00193 00170 00205 00144 00190 00125 00139 00114 00157 00194

Dt1 01163 01075 00648 00717 00529 00263 00065 00153 00024 00017 00073 00008

00217 00158 00198 00174 00213 00149 00195 00128 00143 00117 00163 00201

Revenue-sharing contract

Intercept 6587 6699 10013 10217 10619 11915 4844 4830 9454 8867 10128 9900

303 405 522 626 432 489 382 546 318 412 616 421

t 2 00364 00441 00578 00506 00416 00823 00280 00487 00590 01734 00081 00624

00169 00121 00177 00131 001837 00174 00142 00112 00151 00186 00153 00187

Dt1 00651 00912 00828 00245 00351 0057 00144 00102 00051 00221 00191 00265

00174 00125 00181 00135 00188 00179 00146 00115 00155 00191 00157 00193

Note. Coefficient estimates and standard errors are reported in parentheses.p < 010 p < 005.

as is the standard in the pharmaceutical industry, forexample. Although such contracts may be rational inthe face of retailer underordering, they are sure tolead to over-ordering. Thus, effective decision-supporttools for both retailers and suppliers offer promise fordecreasing waste and increasing profitability.

7. Electronic CompanionAn electronic companion to this paper is available aspart of the online version that can be found at http://mansci.journal.informs.org/.

AcknowledgmentsThis research was funded in part by a National ScienceFoundation Doctoral Dissertation Grant (SES-0412900) and

by the Smeal College of Business Doctoral DissertationAward, Pennsylvania State University. The authors thankAxel Ockenfels and the Deutsche Forschungsgemeinschaftfor financial support through the Leibniz Program. Katok

gratefully acknowledges support from the Smeal Collegeof Business and the Center for Supply Chain Research atPennsylvania States Smeal College of Business through theSmeal Summer Grants Program. The authors thank Kay-YutChen, Karen Donohue, Mirko Kremer, and Enno Siemsenfor helpful suggestions and feedback, and Lynn Hand forcopyediting. All remaining errors are their own.




14/16


Table A.2 Tests of Hypotheses 4A4B.3: Estimates of Equation (8) for Retailer Game

DHIGH/LOW:

DLOW: D U0 100 DHIGH: D U50 150 D= 50 +X, X U0 100

Variable Inexperi enced Experienced Inexperienced Experienced Inexperienced Experienced

Intercept 5825 5873 12010 10275 9385 103.79

261 401 461 591 416 437RS 740 818 1989 078 2532 243

369 567 651 001 589 618

t 2 0054 00203 00676 00710 01824 01133

00190 00127 00183 00151 00189 00164

RS t 2) 00234 00658 01239 00164 02651 00724

00266 00193 00258 00211 00269 02339

Dt1 00908 00994 00738 00481 00548 00308

00139 00101 00134 00110 00139 00120

Note. Coefficient estimates and standard errors are reported in parentheses.p < 010; p < 005.

Table A.3 Tests of Hypotheses 4A4B.3: Estimates of Equation (8) for Supplier Game

DHIGH/LOW:

DLOW: D U0 100) DHIGH: D U50 150 D= 50 +X, X U0 100

Variable Inexperienced Experienced Inexperienced Experienced Inexperienced Experienced

Intercept 5949 5761 8414 8490 9637 9244

354 515 424 567 564 661

RS 964 919 1411 388 915 301

475 729 569 821 935 798

t 2 01167 00463 00792 00721 01108 00235

00234 00165 00202 00211 00247 00245

RS t 2 00898 00002 00187 00991 01021 00836

00166 00118 00144 00147 00173 00172Dt1 00103 00026 00023 00096 00095 00103

00122 00086 00105 00110 00128 00127

Note. Coefficient estimates and standard errors are reported in parentheses.p < 010; p < 005.

Table A.4 Tests of Hypotheses 4A4B.3: Estimates of Equation (9) for Units Sold

DHIGH/LOW:



Intercept 817 861 966 1015 918 937

042 032 020 022 041 031

RS 002 014 057 003 016 038

059 045 028 032 058 044

t 2 00128 00042 00112 00022 00093 00109

00012 00007 00010 00005 00009 00009

RS t 2) 00009 00023 00025 00003 00008 00054

00017 00010 00015 00007 00013 00012

Dt1 00005 00009 00005 00010 00007 00004

00009 00005 00008 00004 00007 00006

Notes. The dependent variable is units sold. Coefficient estimates and standard errors are reported in parentheses.p < 010; p < 005.




15/16


Table A.5 Tests of Hypotheses 4A4B.3: Estimates of Equation (9) for Units Unsold

DHIGH/LOW:



Intercept 254 278 470 479 397 472

055 049 051 072 053 069RS 234 113 119 166 129 180

078 070 072 104 074 097

t 2 00014 00009 00166 00155 00130 00259

00014 00010 00014 00014 00015 00014

RS t 2) 00099 00109 00039 00000 00066 00192

00019 00013 00019 00020 00021 00020

Dt1 00012 00008 00002 00001 00023 00002

00010 00007 00010 00010 00011 00010

Notes. The dependent variable is units unsold. Coefficient estimates and standard errors are reported in parentheses.p < 010; p < 005.

References

Balachandran, K. R., S. Radhakrishnan. 2005. Quality implica-tions of warranties in a supply chain. Management Sci. 51(8)12661277.

Ball, S. B., P. Cech. 1996. Subject pool choice and treatment effects ineconomic laboratory research. Experiment. Econom. 6 239292.

Benzion, U., Y. Cohen, R. Peled, T. Shavit. 2008. Decision-makingand the newsvendor problem: An experimental study. J. Oper.Res. Soc. 59(9) 12811287.

Bolton, G., E. Katok. 2008. Learning-by-doing in the newsven-dor problem: A laboratory investigation. Manufacturing ServiceOper. Management 10(3) 519538.

Bolton, G., A. Ockenfels. 2000. A theory of equity, reciprocity, andcompetition. Amer. Econom. Rev. 90(1) 166193.

Bolton, G., A. Ockenfels, U. Thonemann. 2008. Managers and stu-dents as newsvendors: How out-of-task experience matters.

Working paper, University of Cologne, Cologne, Germany.Bostian, A. J. A., C. A. Holt, A. M. Smith. 2008. Newsvendor pull-to-center effect: Adaptive learning in a laboratory experiment.

Manufacturing Service Oper. Management 10(4) 590608.

Burnetas, A., P. Ritchken. 2005. Option pricing with downward-sloping demand curves: The case of supply chain options.

Management Sci. 51(4) 566580.

Cachon, G. P. 2003. Supply chain coordination with contracts. A. G.de Kok, S. C. Graves, eds. Handbooks in Operations Research and

Management Science: Supply Chain Management, Vol. 11. ElsevierB.V., Amsterdam, 229340.

Cachon, G. P., M. A. Lariviere. 2005. Supply chain coordination withrevenue sharing contracts: Strengths and limitations. Manage-ment Sci. 51(1) 3044.

Choi, K., J. G. Dai, J. Song. 2004. On measuring supplier perfor-mance under vendor-managed inventory programs in capac-itated supply chains. Manufacturing Service Oper. Management6(1) 5372.

Corbett, C. J., J. C. Fransoo. 2007. Entrepreneurs and newsven-dors: Do small businesses follow the newsvendor logic whenmaking inventory decisions? Working paper, University ofCalifornia, Los Angeles.

Corbett, C. J., D. Zhou, C. S. Tang. 2004. Designing supply con-tracts: Contract type and information asymmetry. ManagementSci. 50(4) 550559.

Cui, T. H., J. S. Raju, Z. J. Zhang. 2007. Fairness and channel coor-dination. Management Sci. 53(8) 13031314.

De Bruyn, A., G. E. Bolton. 2008. Estimating the influence of fair-ness on bargaining behavior. Management Sci. 54(10) 17741791.

Donohue, K. 2000. Efficient supply contracts for fashion goods withforecast updating and two production modes. Management Sci.46(11) 13971411.

Fehr, E., K. M. Schmidt. 1999. A theory of fairness, competition andcooperation. Quart. J. Econom. 114(3) 817868.

Ferguson, M. E., V. D. Guide, G. C. Souza. 2006. Supply chain coor-dination for false failure returns. Manufacturing Service Oper.

Management 8(4) 376393.

Ho, T., J. Zhang. 2008. Designing pricing contracts for boundedlyrational customers: Does the framing of the fixed fee matter?

Management Sci. 54(4) 686700.

Holt, C. A. 1995. Industrial organization: A survey of laboratoryresults. J. Kagel, A. Roth, eds. Handbook of Experimental Eco-nomics. Princeton University Press, Princeton, NJ, 349443.

Kagel, J. H., A. E. Roth, eds. 1995. Handbook of Experimental Eco-nomics. Princeton University Press, Princeton, NJ.

Kahneman, D., A. Tversky. 1979. Prospect theory: An analysis ofdecision under risk. Econometrica 47 263291.

Kahneman, D., J. L. Knetsch, R. H. Thaler. 1990. Experimental testsof the endowment effect and the Coase theorem. J. PoliticalEconom. 98 13251348.

Kamrad, B., A. Siddique. 2004. Supply contracts, profit sharing,switching, and reaction options. Management Sci. 50(1) 6482.

Katok, E., V. Pavlov. 2009. Fairness and coordination failures in sup-ply chain contracts. Working paper, Pennsylvania State Uni-versity, University Park.

Katok, E., D. Thomas, A. Davis. 2008. Inventory service level agree-ments as coordination mechanisms: The effect of review peri-ods. Manufacturing Service Oper. Management 10(4) 609624.

Keser, C., G. Paleologo. 2004. Experimental investigation of retailer-

supplier contracts: The wholesale price contract. Workingpaper, IBM Research, New York.

Kleindorfer, P. R., D. J. Wu. 2003. Integrating long- and short-termcontracting via business-to-business exchanges for capital-intensive industries. Management Sci. 49(11) 15971615.

Kremer, M., S. Minner, L. Van Wassenhove. 2007. Regret and fram-ing effects in the newsvendor modelThe impact of framingand complexity. Working paper, INSEAD, France.

Lee, H., S. Whang. 1999. Decentralized multi-echelon supply chains:Incentives and information. Management Sci. 45(5) 633640.

Loch, C. H., Y. Wu. 2008. Social preferences and supply chainperformance: An experimental study. Management Sci. 54(11)18351849.




16/16


Lurie, N. H., J. M. Swaminathan. 2009. Is timely information alwaysbetter? The effect of feedback frequency on decision making.Organ. Behav. Human Decision Processes 108(2) 315329.

Padmanabhan, V., I. P. L. Png. 1995. Returns policies: Make moneyby making good. Sloan Management Rev. 37(1) 6572.

Plott, C. R. 1987. Dimensions of parallelism: Some policy applica-tions of experimental methods. A. E. Roth, ed. Laboratory Exper-imentation in Economics: Six Points of View. Cambridge Univer-sity Press, New York, 193232.

Rudi, N., S. Kapur, D. F. Pyke. 2001. A two-location inventorymodel with transshipment and local decision making. Manage-ment Sci. 47(12) 16681680.

Schweitzer, M., G. Cachon. 2000. Decision bias in the newsvendorproblem: Experimental evidence.Management Sci. 46(3) 404420.

Siegel, S. 1956. Nonparametric Statistics for Behavioral Sciences.McGraw-Hill, New York.

Soman, D. 2004. Framing, loss aversion, and mental account-ing. D. J. Hoehler, N. Harvey, eds. Blackwell Handbook of

Judgment and Decision Making. Blackwell Publishing, Boston,379398.

Spengler, J. J. 1950. Vertical integration and antitrust policy. J. Polit-ical Econom. 58(4) 347352.

Wu, D. J., P. R. Kleindorfer. 2005. Competitive options, supplycontracting, and electronic markets. Management Sci. 51(3)452466.

Wu, Y. D. 2009. The impact of long-term relationship on supplychain contracts: A laboratory study. Working paper, Universityof Kansas, Lawrence.



Documents

Contracting in SC