European Journal of Operational Research 249 (2016) 1033–1049

Contents lists available at ScienceDirect

European Journal of Operational Research

journal homepage: www.elsevier.com/locate/ejor

Reference points in revenue sharing contracts—How to design optimal

supply chain contracts
Michael Becker-Peth∗, Ulrich W. Thonemann
Department of Supply Chain Management and Management Science, University of Cologne, Cologne D-50923, Germany

a r t i c l e i n f o a b s t r a c t

Article history: Coordinating supply chains is an important goal for contract designers because it enables the channel mem-
Received 27 November 2014 bers to increase their profits. Recently, many experimental studies have shown that behavioral aspects have to
Accepted 4 June 2015
be taken into account when choosing the type of contract and specifying the contract parameters. In this pa-
Available online 11 June 2015
per, we analyze behavioral aspects of revenue-sharing contracts. We extend the classical normative decision
Keywords: model by incorporating reference-dependent valuation into the decision model and show how this affects
Supply chain management inventory decisions. We conduct different lab experiments to test our model. As a result, human inventory
Contracting decisions deviate from classical normative predictions, and we find evidence for reference-dependent val-
Behavioral operations research uation of human decision makers. We also show how contract designers can use the insights we gained to
Experiments design better contracts.
Reference-dependent valuation
© 2015 Elsevier B.V. and Association of European Operational Research Societies (EURO) within the
International Federation of Operational Research Societies (IFORS). All rights reserved.

1. Introduction cerning the contract parameters. The underlying theory assumes that
the valuation of a decision outcome is determined by not only the
Numerous recent studies have analyzed the inventory decisions absolute value of the outcome but also its difference from a reference
of human decision makers in a newsvendor setting and shown
point. Contract-specific reference points are part of mental account-
that stocking quantities differ significantly from expected-profit-
ing, which is defined as a “set of cognitive operations […] to […] eval-
maximizing predictions (e.g., Schweitzer and Cachon, 2000; Bolton uate […] financial activities” (Thaler, 1999). Reference dependency
and Katok, 2008; Bostian, Holt, and Smith, 2008, and many more
has received little attention in the behavioral operations literature so
in the literature review). These studies primarily focus on wholesale
far, and we are the first to model reference-dependent contract pa-
price contracts, whereas few studies have analyzed contracts other
rameters in supply contracts.
than wholesale price contracts. Among these few studies, Katok and
The contribution of this paper is threefold. First, we incorporate
Wu (2009) study buyback and revenue-sharing contracts from a re-
reference-dependent valuation into the context of supply contract
tailer perspective, and Zhang, Donohue, and Cui (2012) analyze these
decisions and show how contract-specific reference points affect hu-
two types of contracts from a supplier perspective. Both of these man inventory decisions. Second, we analyze how reference points
studies find evidence for contract-specific behavior—i.e., different
are set in the context of revenue-sharing contracts. Third, we test our
contracts that are theoretically equivalent induce different behavior.
model analyzing actual decisions from laboratory experiments. We
Further, Becker-Peth, Katok, and Thonemann (2013) show that find evidence of reference prices, which significantly affect inventory
source-dependent valuation of money influences human inventory
decisions. The results of these analyses can serve as guidelines for
decisions under buyback contracts. In this paper, we analyze revenue-
contract designers to determine contracts and contract parameters.
sharing contracts in detail. Under such contracts we have only one The remainder of this paper is organized as follows. In Section 2,
cash-in stream, so source dependency cannot be relevant in our con-
we present the relevant analytical and behavioral background regard-
text. In contrast to wholesale price contracts, revenue-sharing con-
ing supply contracting with a focus on revenue-sharing contracts.
tracts do not allow retailers to retain all the revenue; rather, it must
In Section 3, we incorporate reference-dependent valuation into the
be shared with the supplier, which might lead to behavioral effects.
traditional newsvendor model. We then analyze different kinds of
We model these possible effects by introducing reference values con-
reference points and derive hypotheses for human inventory deci-
sions. In Section 4, we report the results of different experiments to
∗
Corresponding author. Tel.: +49 2214707942.
test these hypotheses. On the basis of the obtained results, we then
E-mail address: michael.becker-peth@uni-koeln.de (M. Becker-Peth). derive guidelines for the parameterization of supply contracts for

http://dx.doi.org/10.1016/j.ejor.2015.06.007
0377-2217/© 2015 Elsevier B.V. and Association of European Operational Research Societies (EURO) within the International Federation of Operational Research Societies (IFORS).
All rights reserved.
1034 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049
different supply chain settings (Section 5). We summarize our find-
ings in Section 6.
2. Analytical and behavioral background—literature review
In this section, we first explain the supply contract used in this
study in Section 2.1, and we then summarize the results of ear-
lier studies on behavioral operations concerning supply contracts in
Section 2.2.
2.1. Analytical model of supply contracts in operations research
The typical setting analyzed in the literature considers a supply
chain with a single manufacturer and a single retailer. In this setting,
the retailer faces random demand with a cumulative distribution
function F(·) and an exogenous retail price r, and a supply contract be-
tween the manufacturer and the retailer determines the transfer pay-
ments, e.g., the wholesale price for purchased products. Further, the
retailer decides on the stocking quantity q and places the order with
the manufacturer, and the manufacturer then produces the quantity
and delivers the products to the retailer. This setting is also known
as the newsvendor problem (Arrow, Harris, & Marschak, 1951; Edge-
worth, 1888).
Wholesale price contracts lead to double marginalization
(Spengler, 1950). To overcome this problem, different contracts can be
used to coordinate the channel, i.e., to incentivize the retailer to stock
the channel-profit-maximizing stocking quantity (for an overview
see Cachon, 2003, chap. 6).
We focus on a commonly analyzed type of supply contract:
the revenue-sharing contract (Cachon & Lariviere, 2005). Revenue-
sharing contracts have received considerable attention in operations
research but primarily from the normative perspective. The original
single-period model (Giannoccaro & Pontrandolfo, 2004; Lariviere,
1998) has been extended in various directions, e.g., to cost sharing
(Kunter, 2012), multiple periods (Li & Hua, 2008), asymmetric power
distribution (Pan, Lai, Leung, & Xiao, 2010), and retail competition
(Yao, Leung, & Lai, 2008). Chauhan and Proth (2005) analyze supply
chain partnerships based on revenue-sharing, and Van der Veen and
Venugopal (2005) show that implementation in practice yield in a
win–win situation.
The operations research literature has focused on examining ra-
tional profit-maximizing decision makers and designing contracts
for such decision makers in different settings. In a recent study,
Hämäläinen, Luoma, and Saarinen (2013) note the importance of in-
corporating behavioral aspects in operations research models, and
we follow this line of research. Specifically, in this paper, we focus
on behavioral decision making under a revenue-sharing contract. By
gaining insight into actual decision making, we can extend existing
models to design optimal contracts for actual (non-expected-profit-
maximizing) decision makers.
2.2. Behavioral aspects of supply contracts
Decision making in the newsvendor setting that we analyze has
been the topic of various recent studies in the field of behavioral op-
erations (e.g., Schweitzer and Cachon, 2000; Benzion, Cohen, Peled,
and Shavit, 2008; Katok and Wu, 2009; Bolton, Ockenfels, and Thone-
mann, 2012; Ren and Croson, 2013).1 These studies make one com-
mon observation, the so-called pull-to-center effect, which refers to
the observation that subjects tend to stock below profit-maximizing
quantities for high critical ratios and above for low critical ratios. Sev-
eral theories have been ruled out as explanations for this effect, such
1 dency engenders the pull-to-center effect if psychological costs of
Studies with deterministic demand settings include Loch and Wu (2008), Cui, Raju,
and Zhang (2007), Ho and Zhang (2008), Lim and Ho (2007), and Katok and Pavlov
(2013).
as risk-averse or risk-seeking behavior, loss avoidance or underesti-
mation of the opportunity costs (Schweitzer & Cachon, 2000).
Su (2008) uses random optimization errors to explain the pull-to-
center effect; however, Kremer, Minner, and Van Wassenhove (2010)
show that random errors cannot explain all effects, but that the de-
cision bias is context dependent. A second explanation for the pull-
to-center bias is the anchoring and adjustment heuristic (Kahneman
& Tversky, 1979). By anchoring on mean demand subjects adjust in-
sufficiently toward profit-maximizing stocking quantities. This expla-
nation is supported by numerous studies, and it can be modeled by
using an anchoring parameter on the mean demand (Bostian et al.,
2008). A third explanation for the pull-to-center bias is demand chas-
ing. Bolton and Katok (2008) show that subjects overreact to recent
demand realizations and adjust their stocking quantity in the direc-
tion of previous demands. Further, Ho, Lim, and Cui (2010) show that
the pull-to-center effect can also result from psychological costs of
stock-outs and left-over inventory.
Behavioral supply contracting studies have primarily focused on
wholesale price contracts, whereas few studies have analyzed other
contract types. A notable exception is Katok and Wu (2009), who
study inventory decisions under revenue-sharing and buyback con-
tracts. They find evidence that human inventory decisions differ be-
tween contracts that theoretically lead to identical inventory deci-
sions. However, their paper focuses on social preferences and does
not analyze the rationales underlying the different stocking quanti-
ties under different contract types.
Regarding buyback contracts, Becker-Peth et al. (2013) show that
mental accounting (Thaler, 1985) and source-dependent valuation
(Loewenstein & Issacharoff, 1994) can explain human inventory de-
cisions that are not in line with standard theory. Zhang et al. (2012)
also use mental accounting to explain human decision making but
consider the problem from the manufacturer’s perspective. They an-
alyze revenue-sharing and buyback contracts in a setting in which the
manufacturer is the decision maker and is able to determine the con-
tract parameters; the computerized retailer places expected-profit-
maximizing stocking quantities. Zhang et al. (2012) find that the man-
ufacturer uses different mental accounts for the different payment
times. These payments differ between the two types of contracts, i.e.,
high upfront income but subsequent payback under buyback con-
tracts and low upfront income but additional subsequent income un-
der revenue-sharing contracts. Whereas Zhang et al. (2012) describe
the contract choice and parameterization by the manufacturer using
mental accounting, we analyze the actual ordering decisions of the
retailer and will use reference prices to model actual decision making.
In many situations, reference-dependent valuation can explain
decision makers’ actual behavior that diverges from expected-profit-
maximizing behavior. Kahneman (1992) uses negotiations on salary
increases as an illustrative example. A salary increase offer can be
evaluated as a gain relative to the status quo or a loss relative to
certain reference points, e.g., the previous year’s increase. Winer
(1986) and Kalyanaram and Winer (1995) model the effect of refer-
ence prices on customer’s brand choice. Further, Hardie, Johnson, and
Fader (1993) show that reference points affect many decisions in the
buyer behavior context. In a setting where two persons with identi-
cal tastes visit a high-quality restaurant, one person might be disap-
pointed because he was previously accustomed to better quality in
the restaurant, whereas the other one might be pleasantly surprised
by the high-quality meal if he is accustomed to lower quality restau-
rants (Hardie et al., 1993).
To the best of our knowledge, reference-dependent valuation has
not yet been analyzed in the supply contracting literature. One no-
table exception is Ho et al. (2010), who model a reference point on
selling all purchased products and analyze how reference depen-
leftovers and stock-outs are present. In this paper, we address another
aspect of mental accounting and reference-dependent valuation.
 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049
Ho et al. (2010) use an outcome-based reference point in their model
where subjects have a dis-utility if the actual demand realization de-
viates from their order quantity. Such models can explain the pull-to-
center effect, but the predictions are similar across contract types;
i.e., different contracts with identical overage and underage costs
lead to the same inventory decision. In our decision model, deci-
sion makers have reference points for the contract parameters and
have dis-utility if the actual parameters deviate from their expecta-
tions/reference points.
In this paper, we extend the existing contracting literature and in-
corporate reference values and prices regarding contract parameters
into the decision model. We argue that people have reference prices
for contract parameters (revenue shares and wholesale prices), and
we provide experimental evidence that supports our arguments. Our
experimental results cannot be explained by existing rational or be-
havioral operations models but can be explained by our model. Our
results can thus help contract designers to better understand how
people respond to different contract parameters and provide valuable
insights for contract designers.
3. Reference-dependent utility—developing a behavioral model
for revenue-sharing contracts
Under a revenue-sharing contract, human subjects make in-
ventory decisions that generally deviate from the expected-profit-
maximizing quantities (Katok & Wu, 2009). A number of theories can
generally explain human subjects’ deviation from expected-profit-
maximizing behavior in newsvendor-type contexts (see the previous
section for an overview). These theories interfere with each other,
which makes it is difficult to decide which, if any, provides a good
model for actual behavior under revenue-sharing contracts. We start
our analyses with a simple but insightful experiment that will rule
out the majority of these theories and provide an indication that
reference-dependent utility is relevant (Section 3.1).
We then introduce the general concept of reference-dependent
valuation and discuss the behavioral literature (outside the opera-
tions context) that addresses this problem (Section 3.2). Finally, we
adapt the concept to the supply-contracting context and derive a new
behavioral decision model (Section 3.3).
3.1. Initial experimental analysis of revenue-sharing contracts
Under a revenue-sharing contract, the retailer purchases q units
of a product at a wholesale price wRS per unit from the manufacturer
to sell it for a retail price r. Additionally, the manufacturer receives a
share of 1 − u of the revenue, and the retailer keeps a share of u of
the revenue, with 0 ≤ u ≤ 1 and 0 ≤ wRS ≤ ur. The expected-profit-
maximizing stocking quantity is:
u r − w 
RS
q∗ = F −1 , (1)
ur
ur−w
where CR = ur RS is referred to as critical ratio. Solving the critical
ratio for u, we obtain
wRS
uCR (wRS ) = . (2)
r (1 − CR )
Each combination of wRS and uCR (wRS ) satisfying Eq. (2) has
the same critical ratio and therefore the same expected-profit-
maximizing stocking quantity q∗ .
The classical wholesale price contract can be considered as a spe-
cial case of the revenue-sharing contract. Under a wholesale price
contract, the retailer pays a wholesale price of wWP for each unit and
keeps all the revenue. For u = 1 and wRS = wWP , a revenue-sharing
contract and a wholesale price contract have the same optimal stock-
ing quantity and the same costs, revenues, and profits. We refer to
revenue-sharing contracts with u = 1 and wRS = wWP as reduced con-
tracts because setting u = 1 leaves the contract with a single param-
eter and reduces it to the wholesale price contract.
1035
Most of the supply contracting literature assumes that peo-
ple strive to optimize expected profit. Under such assumption, the
expected-profit-maximizing stocking quantity depends only on the
critical ratio and the demand distribution. Wholesale price contracts
and reduced revenue-sharing contracts (revenue-sharing contracts
ur−w r−wWP
with u = 1) have the same critical ratio CR = ur RS = r and
for a given demand distribution, F(y), the contracts have the same
expected-profit-maximizing stocking quantity of q∗ = F −1 (CR ). We
take the expected-profit-maximizing stocking quantity as a baseline
(null) hypothesis:
Hypothesis 1. The expected stocking quantity under a wholesale
price contract and a reduced revenue-sharing contract is q∗ =
F −1 (CR ).
The behavioral operations literature shows that actual stock-
ing quantities generally differ from the expected-profit-maximizing
quantity and we do not expect Hypothesis 1 to be fully borne out.
Although the actual quantity might differ from the expected profit-
maximizing quantity, all outcome-based behavioral theories that are
based on cost, revenue, or profit (including most of the theories
described above, e.g., anchoring, prospect theory and psychological
overage and underage costs), predict identical inventory decisions
under both types of contracts. Hypothesis 2 formally addresses the
similarities in inventory decisions under a wholesale price contract
and a reduced revenue-sharing contract:
Hypothesis 2. The expected stocking quantity is the same under
wholesale price contracts and reduced revenue-sharing contracts.
3.1.1. Experimental design of Experiment 1
We test our hypotheses by conducting a laboratory experiment
with two treatments: one with a wholesale price contract and one
with a reduced revenue-sharing contract. In Experiment 1, we set
the wholesale price to 50 for both contracts (wWP = wRS = 50), unit
revenues to r = 100 and we use demand that is discrete uniformly
distributed between 1 and 100. Most previous experimental stud-
ies also use this demand distribution because it is easier to under-
stand for the subject than more complex distribution functions. For
the chosen parameters, the inventory decision that maximizes the
expected profit is q∗ = F −1 ((100 − 50 )/100 ) = 50.2 The optimal so-
lution equals mean demand for this experiment. We use this param-
eterization to account for the pull-to-center effect. In our setting, de-
viations from the expected-profit-maximizing quantities cannot be
described by the pull-to-center effect for any of the contracts because
CR = 0.5. This feature is important because the underlying explana-
tions (anchoring, bounded rationality) are unclear and might differ
between contract types.
Experiment 1. Demand U(1, 100), CR = 0.50, q∗ = 50
Revenue-sharing contracts: wRS = 50, u = 1, NRS = 26.
Wholesale price contract: wW P = 50, NW P = 23.
In the laboratory experiment, we asked N = 49 students of the
University of Cologne in the Cologne Laboratory for Experimental Re-
search (CLER) to make inventory decisions under a wholesale price
contract and under a reduced revenue-sharing contract. We briefed
the subjects on the general setting and the contract that we used in
the experiment (screen shots can be found in Appendix C and the in-
structions can be found in the supplementary material). To test the
subject’s understanding of the setting, we had them calculate the
profit for one setting. The experiment lasted ten rounds, and in each
round, the subjects were reminded of the contract parameters, their
previous decisions and the profit in the previous rounds. Each sub-
ject was exposed to exactly one type of contract and there was no
2
And 51, which results in the same expected profit as q = 50.
1036 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049
100%
90%
80%
70%
Cumulative density
60%
50%
40%
30%
20%
10%
0%
0 5 10 15 20 25 30 35 40 45 50 55 60
Order quantities
Fig. 1. Cumulative density of order quantities between revenue-sharing and wholesale
price contracts.
communication or interaction between the subjects in any of the ex-
periments. All experiments in this paper were programmed using the
z-Tree system (Fischbacher, 2007).
3.1.2. Results of Experiment 1
To test our hypotheses, we focus our analyses on the decisions in
the first period of the experiment. We find significant differences be-
tween the reduced revenue-sharing and the wholesale price contract.
The mean stocking quantity in the first round is q̄WP = 53.3 for the
wholesale price contract and q̄RS = 66.0 for the revenue-sharing con-
tract. The mean stocking quantity for the revenue-sharing contract
q̄RS is significantly different from 50 (p = 0.001, two-sided t-test) and
significantly different from q̄WP (p = 0.039, two-sided t-test). Fig. 1
shows the cumulative density of the stocking quantities between the
contract types. As shown, the quantities for the wholesale price con-
tract are stochastically below the quantities for the reduced revenue-
sharing contract. Based on the results, we reject Hypotheses 1
and 2.3
The results of this simple experiment show that subjects make
inventory decisions that are above the mean demand for reduced
revenue-sharing contracts. This effect cannot be explained by the
behavioral operations theories used in previous studies. In partic-
ular, mean anchoring and bounded rationality are often used to
explain observed deviations from optimal ordering decisions (e.g.,
Schweitzer and Cachon, 2000; Bolton and Katok, 2008; Katok and
Wu, 2009); however, these theories lead to order quantities that
are pulled away from the optimal order quantity toward the mean
demand. In our experiment, the mean demand equals the optimal
order quantity for both types of contracts. Further, we observe signif-
icant deviations away from mean demand for the reduced revenue-
sharing contract. The results of our experiment thus cannot be ex-
plained by mean anchoring or bounded rationality. Note that we
do not claim that mean anchoring is invalid; rather, we eliminate
it as an explanation for suboptimal behavior by our experimental
design.
Generally, no transformation function of the outcome—on final
outcome and on separated cash streams (Thaler, 1985; Zhang et al.,
2012)—can explain the differences between reduced revenue-sharing
contracts and corresponding wholesale price contracts because the
contracts are identical in all monetary streams. Furthermore, differ-
ent framing and valuations of different streams cannot explain the
observed stocking pattern, as might be the case for a second income
3
Analyzing the inventory decisions over time, we actually find demand chasing pat-
terns in our data for 7 of our 49 subjects: there is a significant correlation between the
inventory decision and last period’s demand (Kendall’s-tau, p < 0.05). As a result, the
differences between contracts decrease over time. Learning and other theories could
also explain this effect (for a discussion on this issue, see Section 6).
stream for returns or a second cash stream toward the supplier (the
revenue share the supplier receives). For our special case (reduced
contracts), we have only a single simple revenue stream and a single
purchase stream; hence, there is no reason to assume a difference in
our setting.
A possible explanation of these differences is reference-
RS Contract dependent valuation, where different contracts induce different ref-
WP Contract
erence points. Placing the first stocking quantity, an intrinsic expecta-
tion of the revenue share uRef < 1 might increase the stocking quan-
tity facing a contract with u = 1. To analyze the intrinsic expectations
regarding the revenue share, we asked the subjects in Experiment 1 to
set the contract parameters in the profit calculation task in the brief-
ing part of the experiment. Note that to avoid an anchoring effect,
65 70 75 80 85 90 95 100
we did not provide any value for wholesale prices or revenue shares
beforehand. We provided only the selling revenue of r = 100. On av-
erage, the subjects set a revenue share of ū = 61.2 percent, offering
an initial indication that the subjects might have reference prices
uRef < 1, which might affect their inventory decisions. The use of such
a reference point for the revenue share might explain subjects’ inven-
tory decisions above the profit-maximizing prediction under the re-
duced revenue-sharing contract. We analyze the effect of reference
points in detail in the next section.
3.2. Reference-dependent valuation
The valuation of a decision outcome is often determined by not
only the absolute value of the outcome but also by its difference from
a reference point. The reference point is not necessarily the status quo
of a decision maker; rather, it can be affected by aspirations, expecta-
tions, or social comparisons (Tversky & Kahneman, 1991).
Following (Thaler, 1985), we can divide the utility of a decision
into two parts: the acquisition value and the transaction value. The
first part is the context-independent value of the decision, e.g., the
profit of a decision, or the value of a purchased product. The second
part (the transaction value) compares the outcome of the decision to
a certain reference value.
In the context of consumer behavior, the acquisition value de-
scribes the utility of buying a product for a given (actual) price, and
the transaction value describes the utility of making a good or bad
deal, as the actual price might be below or above the expected (ref-
erence) price, e.g., in price promotions. Therefore, the total utility of
a purchased product can vary for a given actual price, depending on
the reference price (Thaler, 1985).
In our context, the acquisition value can be interpreted as the
profit generated by buying q units and selling min(q, y) units, where y
is the realized demand. The transaction value compares the expected
profit for the actual prices with the expected profit with certain ref-
erence prices.
The overall expected utility EU of a stocking quantity q is

EU(q ) = (v((q, y )) + λv((q, y ) − Ref (RP )) f (y ))dy, (3)
y
where λ is a weighting factor of the transaction value and v(x) is the
value function for the outcomes. We use the standard value function
v(x) from Prospect Theory, which is convex for losses and concave for
gains (Kahneman & Tversky, 1979). To keep the analyses tractable, we
assume that expected utility is linear in the probabilities (Kahneman
& Tversky, 1979; Köszegi & Rabin, 2006) and use the linear form of
the value function:

x x≥0
v (x ) = , (4)
βx x<0
where β is the loss-aversion factor. Generally, people can be loss
averse (losses count more than gains: β > 1), loss neutral (losses are
equal to gains: β = 1) or loss seeking (losses count less than gains:
β < 1). Numerous studies show that most people are loss averse and
 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049 1037

that losses count about twice as much as gains (β ≈ 2) (Ockenfels &
Selten, 2014).
(q, y) is the profit for a stocking quantity q and demand y, and
Ref (RP ) is the reference profit regarding some reference point RP.
An open issue is the structure of the reference profit Ref (RP ). We
discuss this issue in the next section.
3.3. Reference points and the effect on supply contracts
Köszegi and Rabin (2006) argue that the reference profit can be
an external aspiration profit or an endogenous profit. Note that these
(contract-independent) outcome-based explanations cannot explain
the effect that we observe because the resulting stocking quantities
for these explanations are similar between wholesale price and re-
duced revenue-sharing contracts. The same holds for other outcome-
focused reference points, i.e., a reference point on sales quantities as
in Ho et al. (2010). A reference point that can generally explain the
results of the experiment is a reference value of contract parameters.
We discuss this issue next.
3.3.1. Reference contract parameters
Rather than defining an external reference profit or a reference
demand realization, subjects might have expectations regarding con-
tract parameters. A revenue-sharing contracts might induce a refer-
ence point for the revenue share the retailer is keeping because cus-
tomers pay him for selling his products.
In general, subjects can have reference points on any contract pa-
rameter. For the revenue-sharing contract with wholesale price w,
revenue share u and reference prices wRef and uRef , the expected util-
ity for a given stocking quantity q is:
 
EU(q ) = v((u, w, q, y )) + λv((u, w, q, y )
y to the same stocking quantity, i.e., that have the same critical ratio but
 different contract parameters (see Eq. (2)). We refer to such contracts
−(uRef , wRef , q, y )) f (y ) dy.
Using the linear value function from Eq. (4) yields the first-order
condition in Property 1. All proofs can be found in Appendix A.
Property 1. The first-order condition for reference-dependent decision
makers follows:
⎧ u r−w+λβ (u r−uRe f r+wRe f −w)
⎪
⎪ u r+ w (uβr−1) +λβ (u r−uRe f r )
if A
⎪
2
Corollary 1. The optimal stocking quantity for β = 1 is
 
u r − w + λ(u r − uRe f r + wRe f − w )
qRe f
= F −1 . (7)
RS
u r + λ(u r − uRe f r ))
For β = 1, the expected utility function is concave for sufficiently
λ
small uRef ≤ 1+ λ u. The condition is violated if the reference loss is
dominating the actual profit. Note that this condition is not restric-
tive in our setting; stocking quantities decrease in uRef and reach zero
beforehand. To reduce the complexity of the analyses (without los-
ing the relevant implications), we use β = 1 for the remainder of the
paper. Based on Corollary 1, we can derive the following effect of ref-
erence values on stocking quantities.
Corollary 2. For wRe f = w, the stocking quantities under reduced
revenue-sharing contracts are greater than or equal to those under
Re f
wholesale price contracts because uRe f ≤ u = 1: qRS (wRS = wW P , u =
1, u ) ≥ qW P .
Re f
The effect for wRef is unclear because wRef might be greater or
smaller than w; however, we assume the effect is equivalent for the
contracts in Experiment 1. Note that the effect of wRef might also dif-
fer between the two contract types. Such difference would require the
use of different utility functions for different types of contracts. In this
paper, we use identical utility functions and therefore leave examin-
ing the other possibility to future research. Following Corollary 2, the
observed differences in Experiment 1 can actually be explained by a
reference revenue share uRef .
Showing that the existence of reference prices can explain differ-
ences between wholesale price and reduced revenue-sharing con-
tracts, the detailed effects of different reference prices have not yet
been analyzed. Of special interest are contracts that theoretically lead
(5)
as risk-sharing contracts because adapting the parameters shifts
the demand and supply risk between the partners (and the profits
accordingly).
3.3.2. Effect of reference prices on risk-sharing contracts
According to expected-profit-maximizing theory, different con-
tract parameter combinations (should) lead to identical inventory

⎪
⎪ decisions, if the combinations follow Eq. (2). In this section, we
⎪
⎪ u r−w+λβ (u r−uRe f r+wRe f −w )
⎪
⎪   if B analyze whether inventory decisions remain identical if decision
⎪
⎨ u r+ w2 (β −1) +λ β (u−uRe f ) r− (w−wRe f ) (β −1)
2

 Re f
ur (u−uRe f )r
makers have reference prices, as shown in the previous section. To
F qRS = , (6) analyze the effect of uRef we set wRef = w. Using Eq. (2) in Corollary 1
⎪
⎪ u r−w+λ (u r−uRe f r+wRe f −w )
⎪
⎪
  if C yields:
⎪ (w−wRe f ) (β −1)
2

⎪ u r+ w (uβr−1) +λ u r−uRe f r+
2   
⎪
⎪ ( u−uRe f )r w
− w+λ w
− uRef r
⎪
⎪
⎩ u r−w+λ(u r−uRe f r+wRe f −w) if D q=F −1 1−CR
 1−CR
(8)
u r+ w (uβr−1) +λ(u r−uRe f r ))
2 w
1−CR
+λ w
1−CR
− uRef r
with Differentiating q for w yields:
A = uRe f ≥ u ∧ wRe f ≤ w ∨ u > uRe f ∧ wRe f ≤ w − (u − uRe f )r
dq λ uRef r (−1 + CR )2
B = uRe f > u ∧ w < wRe f ≤ w − (u − uRe f )r = f ( q ), (9)
dw ((u r (−1 + CR ) + w )λ + w )2
Ref

C = u > uRe f ∧ w − (u − uRe f )r < wRe f ≤ w

D = uRe f ≥ u ∧ w − (u − uRe f )r < wRe f ∨ u > uRe f ∧ w < wRe f .
Using β = 1 as a special case simplifies the analysis. This simpli-
fication does not affect the general conclusion but highlights the ef-
fect of the reference point. Using β = 1 imposes the assumption that
losses are valued equal to gains. Therefore, the conclusions are based
solely on the effect of the reference point. When loss-aversion (β >
1) is integrated into the analysis, the results are qualitatively similar,
but the magnitude is stronger, e.g., in the domain of reference losses.
For β = 1, the right-hand side of Eq. (6) reduces to one case, and we
can derive the following Corollary:
which is greater than zero for λ > 0. Hence, stocking quantities
are higher for contracts with higher wholesale prices (and revenue
shares); e.g., a contract with w = 50, u = 1 leads to higher stocking
quantities than a contract with w = 5, u = 0.1, although CR = 50 in
both cases.
Property 2. Stocking quantities decrease for decreasing uCR (wRS ), with
the critical ratio held constant for λ > 0 for a fixed uRef .
For λ = 0, Eq. (9) equals zero, and the model reduces to the
newsvendor model with similar order quantities for different w with
the same CR.
1038 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049
u uRef
100
90
80
70
60
50
40
30
20
10
0 0
0 10 20 30 40 50 60 70 80 90 100
wRef
Fig. 2. Results of Experiment 2.
Note that this effect is contradictory to an outcome-based util-
ity function. In fact, a change in order quantities due to risk or
loss-aversion should be in the opposite direction (if there is a
change at all) because the risk decreases as w decreases. The ob-
served effect also cannot be explained with a separated valuation
of cash streams, which also leads to increasing order quantities for
risk-sharing contracts (the proof to this argument can be found in
Appendix A).
The effect of wRef can be shown similarly. Setting uRef = u and dif-
ferentiating q with respect to w yields:
dq −λw Ref
= f ( q ),
dw w2
which is smaller than zero for λ > 0. This result shows that the refer-
ence values of wholesale prices decrease order quantities for increas-
ing w, which is reasonable.
Property 3. Stocking quantities increase for decreasing uCR (wRS ), with
the critical ratio held constant for λ > 0 for a fixed wRef .
Because there are (almost) an infinite number of different struc-
tures of reference prices, we will test the structure of references
prices next.
3.4. Structure of reference points
The literature on reference-dependent valuation has primarily fo-
cused on the status-quo, external aspiration profits or decisions de-
pending expected profits (Köszegi & Rabin, 2006). To the best of our
knowledge, there is no consistent approach to determining how ref-
erence values are set for contracting parameters. For a multi-criteria
decision-making context, Stewart (1992) argues that decision mak-
ers’ reference points are not necessarily the aspiration levels for each
criterion; rather, they may be favorable options for the decision cri-
teria. Transferring this approach to supply contracts, (favorable) con-
tract parameter expectations can influence the decision maker in the
newsvendor setting facing different supply contracts.
Because we are the first to model reference points for sup-
ply contracts with more than one parameter, we conduct an ex-
periment to analyze how subjects set reference points in the
setting of a revenue-sharing contract. There are two main possibil-
ities for constructing reference values in our setting. First, we can
set fixed reference values for certain contract parameters indepen-
dent of the other parameters; e.g., uRef = 0.95 independent of the ac-
tual wholesale price. Second, we can have parameter-dependent ref-
erence values; e.g., uRef increases/decreases in the actual wholesale
100
uRef
90
uRef(w)
80 wRef
70 wRef(u)
60
50
40
30
20
10
0 10 20 30 40 50 60 70 80 90 100
w
price. We design an experiment to differentiate between these two
possibilities.
Experiment 2. Demand: U(1, 100), r = 100, N = 22.
Reference wholesale price: wRef , wRef (u = 1 ), wRef (u = 0.75 ),
w (u = 0.5 ).
Ref
Reference revenue share: uRef , uRef (w = 25 ), uRef (w = 50 ),
uRef (w = 75 ).
We asked 22 new subjects of the University of Cologne to explic-
itly state their expectations regarding contract parameters. In this ex-
(10) periment, we first explained the revenue-sharing contract and the
newsvendor setting in the briefing (screen shots can be found in
Appendix C; there were no additional instructions - everything was
displayed on the screen). To ensure that the subjects had an idea
about the effect of the contract parameters on the resulting profit, we
gave them the opportunity to test any combination of contract pa-
rameters, stocking quantities and demand realization. The subjects
inserted any value for these parameters, and the computer calcu-
lated the corresponding profit. We did not present explicit exam-
ples of profit calculation to avoid any anchor of external reference
point.
As (our) main task, we then asked the subjects about their ex-
pectations regarding the contract parameters. In the first question,
the subjects stated their expectations regarding both revenue shares
uRef and wholesale prices wRef simultaneously. We then displayed
three wholesale prices (w = 25, w = 50, and w = 75) and the sub-
jects stated their expectations regarding the corresponding revenue
shares uRef (w ). We also displayed three revenue shares (u = 1, u =
0.75, and u = 0.5) and the subjects stated the corresponding expec-
tations regarding the wholesale price wRef (u ). We randomized the
sequence of the six questions for each subject to avoid order effects.
The stated expectations were not payoff relevant for the subjects. Af-
terward, the subjects placed order quantities for four sets of contract
parameters (set by us) and received payment according to the ex-
pected profit of one randomly chosen parameter set. This part of the
experiment was designed to ensure that the subjects really thought
about the contracts and the resulting profits.
Fig. 2 shows the results of the experiment. The black circles in
the left graph show the average expected wholesale prices for given
revenue shares (u = 0.5, 0.7, 1.0), and the right graph shows the ex-
pected revenue shares for given wholesale prices. The white cir-
cle and triangle represent the unconditional reference values for
wRef and uRef . For increasing wholesale prices w, the reference rev-
enue share uRef (w ) strictly increases. The same holds for reference
 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049 1039

wholesale price wRef (u ), which increases for increasing revenue Table 1

Optimal and mean order quantities in Experiment 3.
shares.
Interesting insights can be gained from Fig. 2. Specifically, the data w
show that the reference values for contract parameters are not fixed; u 5 20 35 50 65 80 95
rather, they depend on the values of the other contract parameters.
Note that this result is not surprising because the relations among Newsvendor quantity
1.00 95 80 65 50 35 20 5
the contract parameters are highly relevant for the payoffs. To achieve 0.85 94 76 59 41 24 6
at least nonnegative profits, it is necessary for u · r ≥ w. Therefore, 0.70 93 71 50 29 7
wRef = 50 is valid in a wholesale price contract with r = 100, but for 0.55 91 64 36 9
a revenue-sharing contract, wRef = 50 would not be appropriate for 0.40 88 50 13
0.25 80 20
low revenue shares, i.e., u < 0.5.
0.10 50
A simple form of modeling the dependency is
Mean order quantity
w+χ (r−w )
uRef (w ) = r
, (11) 1.00 80 68 63 51 47 36 14
0.85 78 61 59 52 32 25
where χ can be interpreted as the share of unit profit (r − w ) that 0.70 74 67 52 45 26
0.55 64 61 45 25
the retailer is expecting to keep. We note that this equation is merely 0.40 60 48 24
one way to model the dependency of reference points, but it has 0.25 52 24
interesting properties because 0 ≤ uRef ≤ 1 for 0 ≤ χ ≤ 1 for all 0.10 32
0 ≤ w ≤ r.4
Estimating the parameter χ for our data with a fixed-effects least-
squares estimation yields χ = 0.51, which is significantly different
from 0 and 1 (p < 0.01). Examining individual subjects, we find that
We displayed the contracts sequentially, one contract at a time. No
expectations regarding the revenue share increase in the wholesale
feedback was provided between the presentation of the different con-
price for all subjects except for five subjects, who expect uRef = 1 for
tracts to avoid demand realization effects (such as demand chasing).
all wholesale prices. Note that this expectation is a special case of
After all decisions were made, we showed the demand realizations
Eq. (11) for χ = 1. The reference values for all subjects can be found
for all rounds, and the subjects were paid according to their average
in Appendix B.
profit over all 28 rounds.
In sum, we conclude that reference points have an important ef-
At the beginning of the experiment, we spent 15 minutes brief-
fect on decisions in supply contracting settings. The subjects seem
ing the subjects on the revenue-sharing contract setting. The briefing
to have a reference value for contract parameters, which are in-
included a description of the setting, exercises to test subjects’ under-
deed contract specific. To analyze the reference point effect for the
standing of the setting (with corrections for wrong answers) and an
revenue-sharing contract in detail and to test whether our model
explanation of the software (screen shots can be found in Appendix C
holds with human decision makers, we conduct an extensive ex-
and the instructions can be found in the supplementary material).
periment for revenue-sharing contracts. We describe the experiment
The subjects then performed five test rounds to become familiar with
next.
the decision task and the software before starting with the payoff-
relevant decisions. The experiment lasted approximately 1 hour, and
4. Effect of reference prices on contract performance the subjects earned 15.34 EUR on average, including the 2.50 EUR par-
ticipation fee for each subject. There was no communication between
Having shown the effect of possible reference prices on inventory the subjects, and the subjects made their decisions independently.
decisions, we want to test the model in an extensive experiment with N = 32 students of the University of Cologne participated in this ex-
different revenue-sharing contracts. In this section, we first describe periment. It was conducted at the Cologne Laboratory for Experimen-
the experimental design and provide information on the laboratory tal Research (CLER).
protocol (Section 4.1). We then summarize the results of the exper-
iment, highlight the key observations and discuss their implications
(Section 4.2). Finally, we estimate the reference points based on our 4.2. Experimental results
dataset (Section 4.3).
We first report the results of the experiments and then analyze
the results in detail in the next section. The mean order quantities of
4.1. Experimental design and laboratory protocol
the experiment are shown in the bottom part of Table 1, and Fig. 3
visualizes the results of the 28 decisions. As expected from earlier
In the experiment (Experiment 3), we offered subjects 28 differ-
studies in newsvendor settings, mean order quantities are biased to-
ent revenue-sharing contracts, and the subjects placed an order for
ward the mean demand, which indicates that inventory decisions are
each contract. We used contract parameter values that spanned an
below the optimum for high critical ratios (CR > 0.5) and above the
equidistant grid over the feasible region of the contract parameters
optimum for low critical ratios (CR > 0.5).
for 0 ≤ w ≤ r and 0 ≤ u ≤ 1. In total, we had 23 different critical ra-
Analyzing the order quantities for CR = 0.5, we can exclude the
tios. Further, we had two contracts with CR = 0.2, two contracts with
anchoring effect as an explanation for the deviation from the optimal
CR = 0.8, and four contracts with CR = 0.5, and the 28 contracts con-
order quantity because the optimal newsvendor quantity q∗ equals
tained seven reduced contracts with u = 1. The contracts used in the
the mean demand. Therefore, the order quantities can be directly
experiment, including the corresponding newsvendor order quanti-
compared with the newsvendor prediction q∗ . Table 1 shows an in-
ties, are displayed in the upper part of Table 1.
teresting observation for contracts with different parameters but a
similar critical ratio: q̄(u = 1, w = 50 ) = 51 > 50 and q̄(u = 0.1, w =
4
The reference values for wRef (u ) could be modeled accordingly but the results of
5 ) = 32 < 50. Although the contracts induce similar order quantities
our later experiments suggest that uRef is more relevant, and we leave wRef out here for according to Eq. (1), the mean order quantity for u = 0.1, w = 5 is sig-
clarity reasons. nificantly lower than that for u = 1, w = 50 (p < 0.001).
1040 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049

u Table 2
100
Estimation results of parameters, with standard errors
1
presented in parenthesis.
90 0.85
0.7 Model 1 2 3
80 (anchoring) (uRef = χ ) uRef (w, χ )
0.55
70 1 0.4 α 0.435 0.413 0.457
0.25 (0.06)∗∗∗ (0.07)∗∗∗ (0.07)∗∗∗
Empirical 60 χ – 0.726 0.267
0.1 (0.14)∗∗∗ (0.05)∗∗∗
mean 1 0.25
50
order λ – 0.10 0.268
quantity 40 (0.02)∗∗∗ (0.04)∗∗∗
1 LL −3, 848 −3, 811 −3, 802
30 0.1 BIC 7,716 7,690 7,671
0.25 (∗∗ p < 0.05, ∗ p < 0.1). ∗∗∗ p < 0.01.
20

10

0
0 10 20 30 40 50 60 70 80 90 100 dering behavior observed in our experiment. We conduct a nonlinear
Newsvendor order quantity random-effect maximum likelihood (ML) estimation of the parame-
ters to fit the parameter to our reference model using the following
Fig. 3. Mean order quantities of Experiment 3.
equation:

qn j = (1 − α ) qRef
RS + α μ + ηn + n j , (12)
The effect can also be observed for CR = 0.2 and CR = 0.8. The or-
ders for these treatments are biased owing to the pull-to-center ef-
where ηn is a subject-specific error for subject n with ηn ∼ N(0, θ ) and
fect, but the effect for different contract parameters is similar to that
n j ∼ N (0, σ ) is a standard error term. Note that we include a mean
for CR = 0.5. Generally, the mean order quantities for contracts with
anchoring factor α in line with Bostian et al. (2008), where α is the
a high revenue share tend to be larger than the mean order quantities
weight of the mean demand.
for contracts with a small revenue share and a similar critical ratio.
Table 2 shows different estimation results for the ML estimations.
These observations are in line with Property 2 and the argu-
We test two different reference models: Model 2 uses a constant ref-
ments in Section 3.3. The observations also show that the reference
erence value, where uRef = χ is independent of the wholesale price
value of the revenue share dominates the reference wholesale price
w, whereas Model 3 uses the structure from Eq. (11), with increas-
(Property 3), which is reasonable under this type of contract. Accord-
ing reference values uRef (w, χ ). Comparing the Bayesian information
ingly, we use this type of reference value for the remainder of the
criteria (BIC), we find that Model 3 fits our data best, and the re-
paper.
sulting parameters are α = 0.452, λ = 0.268 and χ = 0.267 (all p <
Fig. 4 shows the individual order quantities for the revenue-
0.01). The results show significant anchoring (α > 0), and we find
sharing contract for the critical ratios, which were tested at least
evidence of significant reference dependency (λ > 0). Further, the
twice. For most of the subjects, the order quantities are higher for
preference for Model 2 over Model 3 provides evidence for a refer-
the u = 1 contracts than for u < 1 contracts with the same critical
ence value of the revenue share, which increases with the wholesale
ratio. This result indicates that the observed effects in Fig. 3 have no
price.
aggregation effect.
Based on our analyses, we find support for our theory that deci-
sion makers have reference values for contract parameters. The ob-
4.3. Estimation of reference points servations are in line with the predictions derived in Section 3.3,
and the model shows significantly better fit with the behavioral
Using the optimal order quantity for the reference-dependent val- model than with the normative benchmark. Moreover, note that
uation qRef
RS
(Eq. (7)), we can analyze the reference points for the or- the results from our main experiment cannot be explained by

CR=0.2 CR=0.5 CR=0.8

100 100 100

90 90 90
80 80 80
Orders u = 1, w = 80

Orders u = 1, w = 50

Orders u = 1, w = 20

70 70 70
60 60 60
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
0 50 100 0 50 100 0 50 100
Orders u = 0.25, w = 20 Orders u = 0.1, w = 5 Orders u = 0.25, w = 5

Fig. 4. Individual order quantities for contracts with identical critical ratios.
 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049 1041

1400

1200 Rational Contracts, high Ref Point

Rational Contract, low Ref Point

Manufacutrer Profit 1000
Behavioral Contracts, High Ref Point
800
Behavioral Contracts, low Ref Point
600
Non-Coordinating Behavioral Contracts

400

200

0
0 200 400 600 800 1000 1200 1400
Retailer Profit

Fig. 5. Supply chain profits with behavioral retailers with c = 50.

Maximum Manufacturer Profit

1400

1200

1000
Manufacturer Profit

800

600

400
Coordinating Behavioral Contracts
Non-Coordinating Behavioral Contracts
200

0
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
Reference Point

Fig. 6. Optimal manufacturer’s profit with behavioral contracts.

other existing behavioral operations theories developed in previous value because any deviation from the value reduces the expected
studies. supply chain profit. If the reference point of the retailer is known,
We can use insights derived from these results to improve contract the contract parameters can be chosen accordingly to achieve full
design, and we do so in the following section. supply chain efficiency. The drawback is that the profit alloca-
tion is fixed; thus, we lose the allocative power of coordinating
5. Structural insights and implications for contract design contracts.
To regain (at least parts of) this power, behaviorally coordinat-
The existence of reference points has an important implica- ing contracts can be used. For the solid lines in Fig. 5, we use
tion for contract design. Fig. 5 shows the effect of reference points behavioral contracts accounting for the reference points. We use
on the profits of members of the supply chain. For the produc- Eq. (8) to achieve supply chain optimal inventory decisions (for c =
tion cost c = 50, we determine the coordinating contracts accord- 50 q∗SC = 50) for uRef = 0.1 and uRef = 0.75. We find that a broader
ing to Eq. (2) and calculate the resulting profits for the manufac- range of parameter values coordinates the supply chain. However, we
turer and the retailer. For rational retailers, the profits are on the cannot distribute the optimal supply chain profit arbitrarily for ev-
efficiency frontier (gray dashed line), distributing the optimal sup- ery reference point. Indeed, very high or very low reference points
ply chain profit between the manufacturer and the retailer. In con- allow for only a limited range of contract parameters and profit
trast, profits do not show such a pattern if retailers have reference distributions.
points. In an alternative setting the manufacturer might act as a Stack-
The dashed black line shows the expected profits under coor- elberg leader and wants to maximize his own profit. Regarding
dinating contracts if the retailer has a high reference point (we this objective, Fig. 5 shows that the maximum profit for the man-
use uRef = 0.75 and λ = 1 for illustrative purpose), and the dot- ufacturer is far below the optimal profit for high reference points
ted line shows the equivalent values for a low reference point (uRef = 0.75). When traditional coordinating contracts are used, the
(uRef = 0.1). The graph shows that supply chain coordination can maximum profit is approximately 36 percent of the supply chain
be achieved only if the revenue share is equal to the reference optimum, and using behavior-coordinating contracts increases the
1042 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049
Optimal Revenue Share
1 100
0,9
0,8
0,7
Revenue Share
0,6
0,5
0,4
0,3
0,2
0,1
0 0
0 0,2 0,4 0,6 0,8
Reference Point
Fig. 7. Behavioral contracts that maximize the manufacturer’s profit.
maximum profit to approximately 50 percent. For low reference
points, the manufacturer is able to extract the full supply chain profit.
Fig. 6 shows the maximum profits achievable for the manufacturer
for different reference points and the contract parameters needed
to achieve that level of profit. The graphs show that up to a cer-
tain reference point, the manufacturer can extract the total supply
chain profit. If the reference point is above a certain threshold, ex-
tracting the total supply chain profit is no longer possible. The black
lines refer to behaviorally coordinating contracts (inducing the sup-
ply chain optimal order quantity), and the gray solid line refers to
contracts that maximize manufacturer profits but not necessarily the
supply chain profit, i.e., non-coordinating behavioral contracts (such
contracts are also included as the dash-dotted line for uRef = 0.75 in
Fig. 5). Fig. 6 shows that the manufacturer can increase his expected
profit by manipulating the reference point of the retailer toward low
values.
Fig. 7 shows the contract parameters that optimize the manufac-
turer’s profits. The star shows the optimal parameters for rational
retailers, i.e., the manufacturer’s optimal supply chain coordinating
contract. We find that optimal behavioral contracts significantly de-
viate from traditional coordinating contracts. We also find that for
higher reference points, the manufacturer’s goal of maximizing his
profit conflicts with the goal of supply chain optimization because the
manufacturer can be better off by using non-coordinating behavioral
contracts. In the classical contracting literature, the optimal manu-
facturer’s profit is achieved by coordinating the supply chain and al-
locating the total profit to the manufacturer (using specifically pa-
rameterized coordinating contracts). Note that the optimal contract
parameters depend on the reference points χ and the importance of
the reference value λ.
Summarizing the analyses, we conclude that we can increase per-
formance substantially by using our behavioral model instead of a
model that assumes rational retailers. The model is applicable to
different goals, e.g., optimizing supply chain performance, or opti-
mizing manufacturers profit.
6. Conclusion and outlook
Following the growing literature in behavioral operations, we
challenge profit-maximizing theories to explain decision making in
the newsvendor context. As the first to compare identical contracts
in detail (Katok and Wu, 2009 did this just for one parameter set-
ting), we observe a new ordering pattern in which a significant dif-
ference between contract types arises. Specifically, order quantities
Optimal Wholesale Price
90
80
70
Wholesale Price
60
50
40
30 Coordinating Behavioral
Contracts
20
Non-Coordinating
10 Behavioral Contracts
1 0 0,2 0,4 0,6 0,8 1
Reference Point
differ between reduced revenue-sharing contracts (u = 1) and the
corresponding wholesale price contracts. We show that this differ-
ence between the contract types can be explained by reference-
dependent valuation of the outcomes, where the reference point is
set on the contract parameters.
Accounting for this difference, we derive guidelines for contract
designers in determining contract parameters to achieve channel-
optimal or manufacturer-optimal solutions.
An interesting question for future research is whether the refer-
ence values can be manipulated. To test this possibility, one could
design an experiment in which subjects face contracts with very
high or very low revenue shares first. If the reference points are af-
fected by the observed parameter values, the order quantities for
certain contract parameters tested afterward should be lower for
subjects who have been previously exposed to high parameter val-
ues than for those who have been previously exposed to low rev-
enue shares. Tong (2011) observe such effects with previous whole-
sale prices in a simple newsvendor setting and we also find some
indication of this effect in our first experiment. Although the or-
der quantities are significantly above 50 in the first round for the
reduced revenue-sharing contracts, this effect disappears over the
10 rounds in the experiment. In addition to demand chasing, learn-
ing and other potential explanations, one explanation for this ef-
fect might be that subjects get used to the revenue share u = 1. If
this is the case, the reference point uRef moves over time toward
1, and as a result, order quantities would adjust toward the opti-
mal order quantities. Disentangling these possible explanations re-
quires a different experimental setup, and we leave this task to future
research.
If reference points are experience sensitive, this feature has an im-
portant implication for contract designers. To avoid negative effects,
contract designers should be careful when offering contracts with
high buyback values or high revenue shares because retailers might
get used to them and not react appropriately to later changes in the
parameter values. Furthermore, contract designers can use this ef-
fect to adapt retailers’ expectations such that they choose contract
parameters that are more favorable to the contract designer’s ob-
jective. Another way of influencing the effect of reference points is
to manipulate the salience of certain contract parameters. Under a
revenue-sharing contract, the wholesale price is small relative to that
under a wholesale price contract or buyback contract. Addressing this
difference (e.g., focusing the relevance of wRef > w) would engender
a different effect from focusing on the revenue share. This is also an
interesting topic for future research.
 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049 1043

Acknowledgment erence value functions for u = uRef :

  ymax Ref
 β ymin π f (y ) ∀ wRef ≤ w
We gratefully acknowledge the financial support of the German
Science Foundation (DFG) through the research unit “Design & Be- v(π Ref ) =  y (A.6)
havior” (FOR 1371). A previous version of this paper was part of the
y max
ymin π Ref f (y ) ∀ wRef > w,
Ph.D. thesis of one of the authors (Becker-Peth, 2012).
For u < uRef the reference profit is always negative for wRef ≤
w and is always positive for wRef > w − (u − uRef )r. For w < wRef ≤
w − (u − uRef )r it is negative for demand realizations above d =
Appendix A. Proofs w−wRef
q. This leads us to the following reference value functions for
(u−uRef )r
A1. Proof of Property 1 u< uRef :

A1.1. First order condition: v(π Ref )
For β = 1 the expected utility is y
⎧ ymax Ref
 ⎪
⎪ ymin π f (y ) ∀ wRef > w − (u − uRef )r
EU(q ) = v((q, y )) ⎪
⎪  ymax Ref
(A.1) ⎪
⎨β ymin π f (y ) ∀ wRef ≤ w
y
= (A.7)
 (w−w
Ref
 ⎪
⎪ u−uRef )r
q
π Ref f (y )
+λ v((q, y ) − (uRef , wRef , q, y )) f (y ) dy. ⎪
⎪ ymin 
(A.2) ⎪
⎩ +β ymin
π Ref f (y ) else.
y w−wRef
q
(u−uRef )r

The first part is:

Adding both terms and differentiating the full function for q yields
  w
q  q the first order condition:
ur
v(π (q, y )) = f (y )β (u r y−w q )dy+ f (y )(u r y−w q ) ⎧ u r−w+λβ (u r−uRef r+wRef −w)
ymin w
ur q ⎪
⎪ ∀A
 ⎪ u r+ w (uβr−1) +λβ (u r−uRef r )
2
ymax ⎪
⎪
+ f (y )(u r − w )q, (A.3) ⎪
⎪
⎪
⎪ u r−w+λβ (u r−uRef r+wRef −w )
q
⎪
⎪   ∀B
 ⎨ (w−wRef ) (β −1)
2

u r+ w (uβr−1) +λ β (u−uRef ) r−
2
w w
where ur q is the break even demand with 0 ≤ ur ≤ 1. F qRef = (u−uRef )r
, (A.8)
In the right side of the equation (the reference part) is: RS
⎪
⎪ u r−w+λ (u r−uRef r+wRef −w )
⎪
⎪   ∀C
⎪
⎪ (w−wRef ) (β −1)
2

(q, y ) − (uRef , wRef , q, y ) = (u r min(q, y ) − w q ) ⎪

⎪ u r+ w (uβr−1) +λ u r−uRef r+
2

⎪
⎪
(u−uRef )r

−(uRef r min(q, y ) − wRef q ) ⎪

⎩ u r−w+2 λ(u r−uRef r+wRef −w) ∀D
u r+ w (uβr−1) +λ(u r−uRef r ))
= (u−uRef )r min(q, y ) − (w−wRef )q.
with
For notational simplicity we denote the reference part of the util-
ity function as: A = uRef ≥ u ∧ wRef ≤ w ∨ u > uRef ∧ wRef ≤ w − (u − uRef )r (A.9)
π Ref = (q, y ) − (uRef , wRef , q, y ). (A.4)
B = uRef > u ∧ w < wRef ≤ w − (u − uRef )r (A.10)
The sign of this term is not unique, but depends on the reference
parameters. It can be negative for all demand realizations, be positive
or have a break-even demand. Therefore, we differentiate between
C = u > uRef ∧ w − (u − uRef )r < wRef ≤ w (A.11)
different cases of uRef .
For u ≥ uRef the reference profit is always negative for wRef <
w − (u − uRef )r and is always positive for wRef ≥ w. For w > wRef ≥
D = uRef ≥ u ∧ w − (u − uRef )r
w − (u − uRef )r it is negative for demand realizations below y =
w−wRef
q. This leads to the following reference value functions for < wRef ∨ u > uRef ∧ w < wRef . (A.12)
(u−uRef )r
u> uRef :
 A1.2. Proof of concavity of the utility function:
v(π Ref ) Assuming β = 1 the expected utility is:
y
⎧  ymax Ref  q  ymax
⎪ β ymin π f (y ) ∀ wRef < w − (u − uRef )r EU(q ) = f (y )(u r y − w q )dy + f (y )(u r − w ) q dy
⎪
⎪
⎪
⎪  ymax ymin q
⎪
⎪ ymin π Ref f (y ) ∀ wRef ≥ w 
⎨ q 
= (A.5) +λ f (y ) (u r − uRef r )y − (w − wRef )q dy
⎪  w−wRef
q ymin
⎪
⎪β ymin π Ref f (y )
(u−uRef )r
 
⎪
⎪ ymax
⎪
⎪  max + f (y )(u r − uRef r − (w − wRef ))q dy . (A.13)
⎩ + yw−w Ref π Ref f (y ) else, S
q
(u−uRef )r

The second derivative is:

where π Ref is the reference profit and f(y) is the density function of
demand y. The integral splits at y = q are due to the sales, which are δ EU(q )
2
= f (q ) r (−u − λ (u − uRef )), (A.14)
min (q, y). δ q2
For u = uRef the reference profit is always negative for wRef ≤ w
and is always positive for wRef > w. This leads us to the following ref- which is smaller than zero for uRef ≤ 1+λ
λ u, as 0 < u ≤ 1.
1044 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049

For β = 1 the second derivative of the first part (value of profit) is: The utility of separate income streams for an order quantity q
  is:
w 
− f (q ) u r + (β − 1 ) w , (A.15) q
ur U (q ) = −β v(w q ) + ( f (y ) v(u r y ))dy
ymin
which is non-positive, for β ≥ 0 and w ≤ u r which is the case for the  ymax
revenue-sharing contract. + ( f (y ) v(u r q ))dy, (A.19)
For u > uRef the second derivative of the right part is: q
⎧ Setting the derivative by q equal to zero and solving this for q
⎪β f (q )(−u + uRef )r ∀ wRef < w − (u − uRef )r
⎨ yields (for q > 0):
f (q ) r (−u + uRef ) ∀ wRef ≥ w (A.16)  
⎪
⎩ ∗ −1 ( u r )α − β wα
f (q ) (β −1)(w−w ) +(u r−uRef r )2
Ref 2
else. q =F . (A.20)
(−u+uRef ) r ( u r )α
For the first and the second case this is clearly smaller than zero Using Eq. (2) for coordinating contract parameters reduces the
for β > 0, as u > uRef > 0. The same holds for the third case, because equation to:
the numerator is positive for β = 0 (with w − r (u − uRef ) < wRef < w) 
and increasing in β , because ddZβ = (w − wRef )2 . q∗ = F −1 −(1 − CR )−α β + 1 , (A.21)

For u = uRef
the second derivative of the right part is zero. which is the same for different contract parameter combina-
For u < uRef the second derivative of the right part is: tions with the same CR. This also holds for risk-neutrality
⎧  (α = 1).
⎪
⎪ f (q ) −u + uRef r ∀ wRef > w (ii) Assuming constant absolute risk aversion, using v(x ) = 1 − e−α x ,
⎨ −(u − uRef )r with β as loss-aversion factor:
⎪ f (q ) β (−u + uRef ) r ∀ wRef ≤ w , For this utility function, assuming uniform demand between 1
⎪
⎩ f (q ) (β −1)(2 w wRef −w2 −w2Ref )+β (r2 u2 −2 r2 u uRef +r2 u2Ref ) else and 100, the first derivative is:
(−u+uRef ) r
dU (q )
(A.17) = α (u (100 − q ) e−100α u q − β w e−α w q ). (A.22)
dq
which is greater or equal than zero. For a concave function λ has
to be small enough, such that the reference part does not dom- For uC (w ) = r (1−CR
w
) the optimal order quantity for CR = 0.5 is:
inate the profit part (which is the case for non-negative order
LambertW (50αβ w e100α w )
quantities): q∗ = 100 − , (A.23)
⎧ u r+(β −1) w w αw
⎪ ( ur )
∀w >w
Ref
⎪ (−u+uRef ) r
⎪ where LambertW is the Lambert-W-function, also called the
⎪
⎪ Omega function. As the right part increases in w for β > 0, the or-
⎨ −(u − uRef )r
λ ≤ (u r+(β −1) w uwr ) der quantity decreases in w, e.g., q∗ is smaller for w = 50, (u = 1 )
⎪ β (−u+uRef ) r ∀ wRef ≤ w , (A.18)
than for w = 5, (u = 0.1 ).
⎪
⎪ (u r+(β −1) w uwr )
⎪
⎪ else
⎩ (β −1)(2 w wRef − w2 − w2Ref )+β (r2 u2 −2 r2 u uRef +r2 u2Ref )
(−u+uRef ) r

Appendix B. Additional data

A2. Proof of risk sharing effect for separate valuation of cash streams
B1. Individual reference points of experiment
We split the proof into two parts.
Here we plot the individual reference points of the subjects. The
(i) Using the standard risk-averse utility function v(x ) = xα of behav-
numbers in the boxes denote the subjects.
ioral economics, with β as loss-aversion factor:
M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049 1045
1046 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049

Appendix C. Experiment briefing and screen shots of the

software

Translated screen shot of Experiment 1:

 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049 1047

Translated screen shots of Experiment 2:

1048 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049
Translated screen shot of Experiment 3:
Supplementary material
Supplementary material associated with this article can be found,
in the online version, at 10.1016/j.ejor.2015.06.007.
References
Arrow, K., Harris, T., & Marschak, J. (1951). Optimal inventory policy. Econometrica,
19(3), 250–272.
Becker-Peth, M. (2012). Behavioral supply chain contracting: Decision biases in behavioral
operations management. Kölner Wissenschaftsverlag (Ph.D. thesis).
Becker-Peth, M., Katok, E., & Thonemann, U. (2013). Designing buyback contracts for
irrational but predictable newsvendors. Management Science, 59(8), 1800–1816.
Benzion, U., Cohen, Y., Peled, R., & Shavit, T. (2008). Decision-making and the newsven-
dor problem: An experimental study. Journal of the Operational Research Society,
59(9), 1281–1287.
Bolton, G., & Katok, E. (2008). Learning by doing in the newsvendor problem: A labora-
tory investigation of the role of experience and feedback. Manufacturing & Service
Operations Management, 10(3), 519–538.
Bolton, G., Ockenfels, A., & Thonemann, U. (2012). Managers and students as newsven-
dors. Management Science, 58(12), 2225–2233.
Bostian, A., Holt, C. A., & Smith, A. (2008). Newsvendor pull-to-center effect: Adaptive
learning in a laboratory experiment. Manufacturing & Service Operations Manage-
ment, 10(4), 590–608.
Cachon, G. (2003). Supply chain coordination with contracts. In S. Graves, & T. D. Kok
(Eds.), Handbooks in operations research and management science: 46 (pp. 229–340).
North-Holland: Elsevier.
Cachon, G., & Lariviere, M. (2005). Supply chain coordination with revenue-sharing
contracts: Strengths and limitations. Management Science, 51(1), 30–44.
Chauhan, S. S., & Proth, J.-M. (2005). Analysis of a supply chain partnership with rev-
enue sharing. International Journal of Production Economics, 97(1), 44–51.
Cui, T., Raju, J., & Zhang, Z. (2007). Fairness and channel coordination. Management
Science, 53(8), 1303–1314.
Edgeworth, F. Y. (1888). The mathematical theory of banking. Journal of the Royal Sta-
tistical Society, 51(1), 113–127.
Fischbacher, U. (2007). z-tree: Zurich toolbox for ready-made economic experiments.
Experimental Economics, 10(2), 171–178.
Giannoccaro, I., & Pontrandolfo, P. (2004). Supply chain coordination by revenue shar-
ing contracts. International Journal of Production Economics, 89(2), 131–139.
Hämäläinen, R. P., Luoma, J., & Saarinen, E. (2013). On the importance of behavioral op-
erational research: The case of understanding and communicating about dynamic
systems. European Journal of Operational Research, 228(3), 623–634.
Hardie, B., Johnson, E., & Fader, P. (1993). Modeling loss aversion and reference depen-
dence effects on brand choice. Marketing Science, 12(4), 378–394.
Ho, T., Lim, N., & Cui, T. (2010). Reference dependence in multilocation newsvendor
models: A structural analysis. Management Science, 56(11), 1891–1910.
Ho, T., & Zhang, J. (2008). Designing pricing contracts for boundedly rational cus-
tomers: Does the framing of the fixed fee matter. Management Science, 54(4), 686–
700.
Kahneman, D. (1992). Reference points, anchors, norms, and mixed feelings. Organiza-
tional Behavior and Human Decision Processes, 51(2), 296–312.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk.
Econometrica, 47(2), 263–291.
Kalyanaram, G., & Winer, R. (1995). Empirical generalizations from reference price re-
search. Marketing Science, 14(3), 161–169.
Katok, E., & Pavlov, V. (2013). Fairness in supply chain contracts: A laboratory study.
Journal of Operations Management, 31(3), 129–137.
Katok, E., & Wu, D. (2009). Contracting in supply chains: A laboratory investigation.
Management Science, 55(12), 1953–1968.
Köszegi, B., & Rabin, M. (2006). A model of reference-dependent preferences. The Quar-
terly Journal of Economics, CXXI(4), 1133–1166.
Kremer, M., Minner, S., & Van Wassenhove, L. (2010). Do random errors explain
newsvendor behavior? Manufacturing & Service Operations Management, 12(4),
673–681.
Kunter, M. (2012). Coordination via cost and revenue sharing in manufacturer–retailer
channels. European Journal of Operational Research, 216(2), 477–486.
Lariviere, M. (1998). Supply chain contracting and coordination with stochastic de-
mand. In S. Tayur, R. Ganeshan, & M. Magazine (Eds.), Quantitative models for supply
chain management (pp. 233–268). Boston: Kluwer.
Li, S., & Hua, Z. (2008). A note on channel performance under consignment contract
with revenue sharing. European Journal of Operational Research, 184(2), 793–796.
Lim, N., & Ho, T. (2007). Designing price contracts for boundedly rational customers:
Does the number of blocks matter? Marketing Science, 26(3), 312–326.
Loch, C., & Wu, Y. (2008). Social preferences and supply chain performance: An experi-
mental study. Management Science, 54(11), 1835.
Loewenstein, G., & Issacharoff, S. (1994). Source dependence in the valuation of objects.
Journal of Behavioral Decision Making, 7(3), 157–168.
 M. Becker-Peth, U.W. Thonemann / European Journal of Operational Research 249 (2016) 1033–1049
Ockenfels, A., & Selten, R. (2014). Impulse balance in the newsvendor game. Games and
Economic Behavior, 86(2014), 237–247.
Pan, K., Lai, K., Leung, S., & Xiao, D. (2010). Revenue-sharing versus wholesale price
mechanisms under different channel power structures. European Journal of Opera-
tional Research, 203(2), 532–538.
Ren, Y., & Croson, R. (2013). Overconfidence in newsvendor orders: An experimental
study. Management Science, 59(11), 2502–2517.
Schweitzer, M., & Cachon, G. (2000). Decision bias in the newsvendor problem with a
known demand distribution: Experimental evidence. Management Science, 46(3),
404–420.
Spengler, J. (1950). Vertical integration and antitrust policy. The Journal of Political Econ-
omy, 58(4), 347–352.
Stewart, T. (1992). A critical survey on the status of multiple criteria decision making
theory and practice. Omega, 20(5–6), 569–586.
Su, X. (2008). Bounded rationality in newsvendor models. Manufacturing & Service Op-
erations Management, 10(4), 566–589.
Thaler, R. (1985). Mental accounting and consumer choice. Marketing Science, 4(3),
199–214.
1049
Thaler, R. (1999). Mental accounting matters. Journal of Behavioral Decision Making,
12(3), 183–206.
Tong, J. (2011). Reference prices and transaction utility in inventory decisions. Durham,
North Carolina: Duke University Working paper.
Tversky, A., & Kahneman, D. (1991). Loss aversion in riskless choice: A reference-
dependent model. The Quarterly Journal of Economics, 106(4), 1039–1061.
Van der Veen, J. A. A., & Venugopal, V. (2005). Using revenue sharing to create win–win
in the video rental supply chain. Journal of the Operational Research Society, 56(7),
757–762.
Winer, R. (1986). A reference price model of brand choice for frequently purchased
products. Journal of Consumer Research, 13(2), 250–256.
Yao, Z., Leung, S., & Lai, K. (2008). Manufacturer’s revenue-sharing contract and retail
competition. European Journal of Operational Research, 186(2), 637–651.
Zhang, Y., Donohue, K. Cui, T. (2012). Contract preferences for the loss averse supplier:
Buyback versus revenue sharing.