Combining Behavioral Experiments and Agent-based Social Simulation

to Support Trust-aware Decision-making in Supply Chains

Diego de Siqueira Braga

, Marco Niemann

, Bernd Hellingrath

and Fernando Buarque de L. Neto

Westf

alische Wilhelms-Universit

at M

unster, M

unster, Germany

University of Pernambuco, Recife, Pernambuco, Brazil

Keywords:

Social Simulation, Trust, Bullwhip Effect, ABSS, Behavioral Experiment, Behavior Modeling.

Abstract:

Trust is seen as one of the most important dimensions in developing and maintaining fruitful business re-

lationships and has deep impact on the decision-making process in the supply chain planning. Despite its

importance, very limited research has been done in the trust-aware decision-making ﬁeld. This paper aims

to experimentally examine how trust can be assessed over different dimensions and then be used to support

decision-making in order to reduce the Bullwhip Effect, which is one of the biggest efﬁciency problems shown

by supply chains of highly interconnected organizations. As industry is generally reluctant to provide data due

to privacy concerns and trade secret protection, the authors of this paper, designed and conducted a web-based

trust behavioral experiment. The data collected was used to evaluate the proposed trust mechanism through

an Agent-Based Social Simulation. The results revealed that it is possible to infer trust relationships from

behavioral experiments and historical based data, and use these relationships to inﬂuence the procurement, or-

dering and information sharing process. Although additional research is still necessary, the preliminary results

revealed that the use of computational trust mechanisms can be helpful to lower the Bullwhip Effect.

1 INTRODUCTION

The issues of trust have been an active research area in

different disciplines including economics (Greif et al.,

1994), sociology (Bachmann, 2001), business infor-

mation systems (Ba and Pavlou, 2002; Ba et al., 2003;

Resnick and Zeckhauser, 2002), information security

(Weeks, 2001), online auctions (Houser and Wood-

ers, 2006), social relationships (Castelfranchi and Fal-

cone, 1998), multi-agent systems (Braynov and Sand-

holm, 2002b; Braynov and Sandholm, 2002a; Yu and

Singh, 2002; Zacharia et al., 2000), and supply chains

(Laeequddin et al., 2010).

Researchers so far did not agree on a commonly

accepted deﬁnition of trust. Instead there is only

some elements - like the Trustor (trusting party) and

Trustee (trusted party) concept - that most authors can

agree on. Furthermore trust can be seen the trade-off

between potential risk and expected gain (Rousseau

et al., 1998). In this context, trust is the result of

prior (positive) experiences from mutual interactions

with other parties (Ring and Van de Ven, 1994; Kim,

2009). As an effect of this dependency trust is a dy-

namic property, readjusting itself based on new inter-

action outcomes (Abdul-Rahman and Hailes, 2000).

1.1 Trust in Supply Chains

A Supply Chain (SC) is a system of semi-autonomous

business entities (i.e. suppliers, manufacturer, retail-

ers, customers), linked by material, ﬁnancial and in-

formation ﬂows across several processes and activi-

ties (Christopher, 1999). Such modern SC are typi-

cally viewed as socio-technical systems (linking so-

ciety’s social and organizations’ technical aspects)

(De Bruijn and Herder, 2009).

Organizations in a SC face a vast number of prob-

lems, such as decision making (i.e. inventory man-

agement), where interdependent decisions are com-

monly managed separately. Additionally organiza-

tions are not isolated, but inﬂuence each other (Chaib-

draa and M

uller, 2006). One of the biggest SC

efﬁciency problems, the so called Bullwhip Effect

(BWE), describes the phenomenon of increasing up-

stream order variances (Forrester, 1958). Predic-

tion and planning are aggravated by high levels of

order variances, decreasing customer service levels

and lowering SC’s competitiveness. Common cop-

ing strategies include increased stock levels and infor-

mation sharing. While the ﬁrst increases costs even

further, the second one is considered a non-simple

260

de Siqueira Braga D., Niemann M., Hellingrath B. and de L. Neto F.

Combining Behavioral Experiments and Agent-based Social Simulation to Support Trust-aware Decision-making in Supply Chains.

DOI: 10.5220/0006200802600267

In Proceedings of the 9th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2017), pages 260-267

ISBN: 978-989-758-219-6

task. Information sharing can be inﬂuenced by for-

mal (e.g. contracts) or informal (e.g. trust) interac-

tions. So SC performance analysis and improvement

requires consideration of both aspects and their inter-

dependencies (Ottens et al., 2006). Since the SC do-

main is very dynamic, trust is considered an essential

tool to develop and manage business relationships, by

lowering transaction costs and shifting to continuous

exchange relationships (Kwon and Suh, 2005; Rai-

mondo, 2000; Tykhonov et al., 2008). These effects,

predicted by theory, have been validated by empirical

analysis (

Ozer et al., 2011) and real-world environ-

ment studies (Ha et al., 2011).

1.2 Computational Trust

Trust was ﬁrst introduced as a measurable property

of an entity in computer science by (Marsh, 1994).

Since then computational trust has grown into a broad

research area, comprising lots of different trust and

reputation models (Sabater and Sierra, 2005; Yu et al.,

2013).

Despite its importance, the conducted SC trust re-

search mostly focuses on its use for partner selection,

while explicit inclusion into the decision making pro-

cess would be desirable to facilitate information and

asset sharing (Burnett et al., 2014).

Performance evaluation is typically either con-

ducted via real world datasets (scarce due to privacy

issues) or the more widely used simulation-based

evaluation (Yu et al., 2013).

2 TRUST BEHAVIORAL

EXPERIMENT

Triggered by the lack of trust related data and moti-

vated by the growing interest in behavioral research

in supply chain management (Donohue and Siemsen,

2011) a trust experiment is proposed. It is expected

that the experiment will gather information regarding

individuals’ behavior during procurement, and order-

ing decisions considering trust relations in the context

of supply chains.

The experiment aims at (i) exposing different rel-

evance proﬁles; (ii) evaluating order assessment con-

cerning trust dimensions; and (iii) validation of ad-

ditional features of the trust mechanism (see Section

3).

The participant acts as a retailer aiming to fulﬁll

the demand of a customer by ordering from a vari-

ous number of suppliers. Each supplier has a differ-

ent proﬁle (explained in Section 2). In addition, the

participant has to consider inventory and back order

costs as well as decreasing payments by the customer

for each round not delivering.

Gamiﬁcation concepts such as a high score list and

progress bars were used in order to increase the par-

ticipants motivation.

The experiment consists of three parts. The ﬁrst

part aims at exposing the relevance proﬁles regarding

the procurement and trust agents. The goal of the sec-

ond part is to expose the trust assessment. The last

part consists of a questionnaire to evaluate the deci-

sions made by the participants.

Firstly, the suppliers have different proﬁles based

on a price per product, an expected delivery time and

a trust value to evaluate the procurement decisions.

After nearly 20 rounds the experiment form is modi-

ﬁed in order to capture information regarding the pref-

erences of the participants regarding the four dimen-

sions: {Reliability, Quality, Competence, Shared Val-

ues}.

The goal of the second part is to expose the trust

assessment. Therefore, the participant has to assess

each dimension regarding the received order. Each

order has a speciﬁc delay or failure rate regarding the

delivered products. Concerning these factors, the par-

ticipant has to adjust the existing rating of each di-

mension by a number between 0 and 100. Thereby, it

is possible to analyze individuals’ behavior regarding

the assessment of each dimension – for example, how

the participant deals with an order delay of one week.

The third part is a questionnaire aiming to assess

the participants decisions. Firstly, the relevance pro-

ﬁles are recalled so that each participant has to eval-

uate his/her behavior regarding price, delivery time

and trust or the four trust dimensions. Afterwards, the

participant demonstrates his/her preferences regard-

ing indirect and direct trust by rating the utilization of

trust-relating information given by others. Finally, the

experiment evaluates which level of trustworthiness

is necessary for each participant to share information

with suppliers.

Results and Analysis

After comparing the relevance proﬁles given by the

participants with their real ordering decisions three

main segments regarding the procurement process

were identiﬁed. The biggest one (group A) having

a share of round 26% of the whole participants is

mainly focusing on a relatively low price and a short

delivery time. A trust value is not that important for

this group. To the second group (group B) the price is

the most important value for decision-making where

to order without focusing very much on a delivery

time or the trust value. About 23% of all participants

Combining Behavioral Experiments and Agent-based Social Simulation to Support Trust-aware Decision-making in Supply Chains

261

Table 1: Procurement proﬁles.

Group Price Delivery Time Trust

A 44% 41,5% 14,5%

B 63% 22,5% 14,5%

C 34% 34% 32%

Table 2: Relevance of Trust and Indirect Trust.

Group Direct Indirect

A 65% 35%

B 73,75% 26,25%

C 58% 42%

used this kind of relevance proﬁle for choosing their

suppliers. The last big group (group C) is a balanced

one. It has a share of 17% and focuses on price, de-

livery time and trust almost equally. The proﬁles with

their rounded relevance values can be seen in Table 1.

In Table 1 the different proﬁles of group A, B and

C can be seen regarding the procurement decision-

making. Each value represents the importance of

each dimension in the calculation during the decision-

making process.

After exposing the trust relevance proﬁles, these

three groups were analyzed regarding their decisions

made in the second part of the experiment. In Table 3

it is possible to see that Reliability and Quality were

the most important dimensions. Having equal num-

ber in groups A and C, and small differences when

considering the ratings for the other two dimensions.

In group B the observed value for the Reliability di-

mension was higher in comparison with groups A and

C (i.e. 45,5%), which is interesting as it contradicts

some results of Table 1. There Group B valuate price

almost three times more important as delivery time.

But when assessing the trust proﬁle where price is

out of scope, Reliability measured by the on-time KPI

(see Section 3) is the most important trust property to

this group.

Furthermore, the analysis exposed how these

groups consider indirect trust values given by other

actors to decide which supplier to choose comparing

to their own direct value. The different importance

proﬁles regarding this weighting of direct or indirect

trust can be seen in Table 2.

It was also assessed the necessary trustworthiness

a supplier must have so that participants share infor-

mation. Table 4 shows the identiﬁed trust threshold

values for information sharing to happen. The ﬁrst

level of information sharing is supposed to tell a sup-

plier if one will order from him in the next round.

The second level is about sharing the order amount

one round earlier with the supplier.

Another important part of the experiment was to

Table 3: Trust proﬁles. Legend: Competence (C), Reliabil-

ity (R), Quality (Q), Shared Values (SV).

Group C R Q SV

A 22% 31% 31% 16%

B 10% 45,5% 30,5% 14%

C 25% 31% 31% 13%

Table 4: Trust threshold values for information sharing.

Group Share order intention Share demand

A 53% 67%

B 45% 71,25%

C 58,5% 62,5%

identify how participants assess different orders. The

experiment exposed that participants assess, in aver-

age, a delay of one week by decreasing the reliabil-

ity value by round 8,21 % and increase it by round

4,89 % for each order without a delay. The Qual-

ity dimension is assessed based on the percentage of

failure rate so that participants decrease it by round

0,72 % for each percentage of failure (e.g. decrease

of quality dimension by 7,2% when having a failure

rate of 10% related to a speciﬁc order). For each re-

ceived order without any defective products the Qual-

ity value is increased by round 4,76 %. For every re-

ceived order which is perceived as bad by the partici-

pants the Competence value is decreased by 3,7%, and

increased by 0,5% for every order perceived as good

(these values can be seen in Table 5). Considering the

results present in Table 5 one can see that received

orders which are perceived as negative are assessed

more strictly regarding each dimension. Because of

this higher impact in the dimension, a bad order can

not be equalized by a proportionally good order. This

observation reinforces the ﬁndings regarding the dif-

ﬁculties of mitigating damaged trust relationships al-

ready present in literature (Kim et al., 2006).

3 SOCIAL SUPPLY CHAIN

SIMULATION

The simulation allowed an examination of trust as-

sessments through different dimensions and the use

of trust in lowering the BWE.

The supply chain conceptualized for the simula-

tion experiment has been constructed with six differ-

ent, commonly used actors: suppliers, manufactur-

ers, distributors, wholesalers, retailers and customers

(Mentzer et al., 2001; La Londe and Masters, 1994).

In this model the supplier is considered to be a source

of raw materials for the manufacturer. For the sake

of simplicity it is assumed that the supplier is always

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

262

Table 5: Participants Trust Assessment.

Assessment Reliability Quality Competence

Perceived as bad -8,21 % per week -0,72% per percentage of failure -3,7%

Perceived as good +4,89 % per week +4,76% +0,5%

able to provide ordered goods. After reﬁnement by

the manufacturer the products are delivered to the

wholesalers via the distributors. The wholesalers act

as the supplying entities for the retailers who sell the

product to the customer. Beside this unidirectional

ﬂow of material the proposed supply chain model al-

lows a bidirectional ﬂow of information.

In addition to the six common actors, this paper

introduces another entity: The lying actor. It concep-

tually differs from the previous constructs, as it does

not aim at modeling a new type of SC member, but

rather represents one of the other types with the addi-

tion of lying in form of opportunistic behavior. Typi-

cal behavior exhibited by such an actor would be or-

dering new goods at two suppliers to counter potential

shipment delays. On receipt of the earliest delivery it

would automatically cancel all other orders. Doing so

these agents will increase the BWE and thus can serve

as disturbing factors in a SC that are to be ruled out

via trust-based decision-making.

Agents

A set of nine different agents was conceptualized to

model the behavior of the different SC participants.

A majority of them aims at the fulﬁllment of typical

SC tasks. For example the inventory agent is respon-

sible for managing both the incoming as well as out-

going product stocks of an actor. It makes use of the

perpetual as well as the order-up-to level policy (De-

jonckheere et al., 2003; Clark and Scarf, 1960).

Additionally a forecast agent is implemented

based on the OpenForecast library (Gould, 2011), al-

lowing an actor to predict the demand of the next pe-

riod.

A procurement agent is added as a supplier selec-

tion entity. It aggregates trust information from both

trust agent as well as indirect trust agent (these con-

cepts are going to be introduced later in this Section)

and combines it with price and delivery time informa-

tion obtained from the delivery agent. Based on three

weighting parameters (one per criterion) the agent se-

lects the best supplier for the follow up transaction.

If the parent actor is a lying actor the delivery agent

additionally selects the second best supplier as well.

The actual ordering of goods is handled by the or-

der agent. It combines the forecast, a potential back-

log as well as inventory orders into one ﬁnal order

amount. For the order fulﬁllment the order agent uses

the best supplier suggested by the procurement agent.

If the parent actor is a lying actor, the same order will

be placed with the second best supplier. Whenever the

agent receives goods they are checked for complete-

ness and integrity to sort out defect ones.

In order to handle the downstream ﬂow of ma-

terial of an actor, the delivery agent has been intro-

duced. It receives the orders of the downstream part-

ner and is responsible for their shipment. Whenever

a full delivery is not possible, the available parts are

shipped while the remaining items are added to the

agents backlog. A secondary function of the agent is

the delivery of historical transaction data for the fore-

cast agent.

Logically in-between the order agent and the de-

livery agent the production agent is settled. They are

transforming incoming goods into a new product. The

exact design of this transformation is speciﬁed by so-

called production policies. For each supply chain ac-

tor such a policy speciﬁes the used input, production

time and a speciﬁc failure rate.

While the prior six agents carry out core SC tasks,

the following two deal with trust related computa-

tions needed for trust-aware decision-making. The

ﬁrst, the trust agent, aims at the evaluation of the di-

rect trust score. Direct in this domain implies that

only the agent’s own set of information is used. De-

spite, as (Pinyol and Sabater-Mir, 2013; Sabater and

Sierra, 2005) observe, most models still only con-

sider a single trust dimension, it was deemed beneﬁ-

cial to use a multi-dimensional model instead. This

supports the paper’s goal to enable a better under-

standing of trust to use it for trust-aware decision-

making. A direct implication is that analyzing trust in

a less aggregated fashion is desirable. Consequently,

trust assessment is conducted considering the quadru-

ple {Reliability, Quality, Competence, Shared Val-

ues} (Haghpanah and DesJardins, 2010; Lin et al.,

2005; Handﬁeld, 2003; Morgan and Hunt, 1994).

Reliability is assessed based on the On-Time KPI

(ServiceNow, 2016), which represents the share of

ordered goods arriving on time. Similarly, Quality

is measured based on the Undamaged Goods KPI

(ServiceNow, 2016), marking the percentage of un-

damaged goods after inspection. In order to evalu-

ate the Competence of a supplier, the trust agent as-

sesses the order history. Beside the prior three fac-

tors only being based on the perceptions of a single

agent, Shared Value is added to integrate agent simi-

Combining Behavioral Experiments and Agent-based Social Simulation to Support Trust-aware Decision-making in Supply Chains

263

larity into the model. This measure allows to evaluate

the degree of equality between two agents ( j ∈ {1, 2})

in their weightings w

i j

of the quadruple {Reliability,

Quality, Competence, Shared Values} (Morgan and

Hunt, 1994). It is assessed via the euclidean distance

between the quadruples of both agents.

The computed value henceforth will be consid-

ered as the OrderFul f illment (OF). In combination

with the trust dimension weightings of the parent user

this value allows to update the trust score assigned to

a speciﬁc supplier. As a ﬁrst step the weighted av-

erage (again weighted with w) of the difference be-

tween OrderFul f illment and weightings w will be

computed to obtain a trust update value.

To compute the new trust value the weighted av-

erage from above is weighted by and added to the old

trust value.

The inﬂuence of the impact of the update value

can be regulated by a learning rate α. Assigning a low

α value will smooth the shift, high values will speed

it up.

Additionally an indirect trust agent is proposed to

account for trust computation from witness informa-

tion as a second traditional source of trust information

(Sabater and Sierra, 2005). It collects all direct trust

values other agents assign to suppliers. From there the

agent is able to compute a mean trust value for a spe-

ciﬁc supplier. The degree to which this value will be

considered depends on the preferences of the parent

actor.

Depending on the degree of trust assigned to a

supplier by the evaluating trust agents the actor might

share additional information or even assign an open

order earlier (referred to as information sharing from

now).

4 VERIFICATION &

VALIDATION

In order to verify the correctness of the implemented

agents and actors the commonly used JUnit unit test-

ing framework was used. It appeared to be the

most reasonable choice as it was already integrated in

the Repast framework (Argonne National Laboratory,

2015).

A special focus of the tests have been the classes

used for trust assessment. This way it should be en-

sured that each dimension was computed correctly to

guarantee correct supplier selection.

Apart from these static tests the simulation re-

sults have been veriﬁed based on already existing

prior studies by (Dejonckheere et al., 2004), (Chat-

ﬁeld et al., 2004) and (Chatﬁeld and Pritchard, 2013).

The used measure of comparison is the BWE modeled

by the total variance ampliﬁcation (TV

Amp

), which

compares an upstream node’s (k > 0) order variance

with the demand variance of a customer node (k = 0)

(Chatﬁeld and Pritchard, 2013).

To achieve comparable values the proposed model

has been slightly adjusted: First, the SC was restricted

to have one actor per tier, neglecting lying actors. Pro-

duction has been simpliﬁed to consider only one input

and one output. Apart from production time, no ad-

ditional temporal effort is considered. The customer

demand is normally distributed with N (10, 1).

Both models - with and without information shar-

ing - can be assumed to be veriﬁed, since the results

match those of prior studies as expected. Looking

at the results in Tables 6 and 7 values are roughly

the same. Especially with regard to the outcomes of

(Dejonckheere et al., 2004) the differences are con-

siderably small. The larger gaps with regard to the

results of (Chatﬁeld et al., 2004) and (Chatﬁeld and

Pritchard, 2013) may be partly based on their use of a

custom simulation system called SISCO.

Given this veriﬁcation, as a next step the effect of

the single components (e.g. trust agent or informa-

tion sharing) have to be evaluated, to understand their

inﬂuence on the BWE. The used metric was again

Amp

The data acquired through the Trust Behavioral

Experiment was used to initialize the different agents

in the simulation model. This includes the discovered

proﬁles as well as the identiﬁed information sharing

threshold values.

Looking at the results presented in Table 8 and

Figure 1 several interesting observations can be made.

Number one is the fact that for all actors except the

Retailer the TVAmp value is smallest when all trust

and information sharing components are active, but

no Lying Actors are present. Retailers require the

presence of Lying Actors to that point. On the con-

trary the worst values come up whenever all trust

mechanisms are deactivated. Distributors and Man-

ufacturers reach the bottom under the presence of

Lying Actors, while Retailers and Wholesalers hit it

without their appearance.

Evaluating both the Table 8 and the Figure 1 fur-

ther, it becomes apparent that the TVAmp value is ris-

ing for each trust mechanism getting deactivated. So

(as described above) the Bullwhip effect is the weak-

est during the presence of trust, indirect trust and in-

formation sharing. Deactivating information sharing

results in the biggest leap of the TVAmp values. For

the experiment without Lying Actors it on average

across all actors increases by 162%, which to a large

degree is based on the 202% increase of the TVAmp

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

264

0,00

0,50

1,00

1,50

2,00

2,50

3,00

3,50

4,00

4,50

Retailer

Info.-

Sharing

Indirect

Trust

Direct

Trust

Lying ActorNormal Actor

(a) TVAmp Retailer.

0,00

5,00

10,00

15,00

20,00

25,00

30,00

35,00

Wholesaler

Info.-

Sharing

Indirect

Trust

Direct

Trust

Lying ActorNormal Actor

(b) TVAmp Wholesaler.

Info.-

Sharing

0,00

20,00

40,00

60,00

80,00

100,00

120,00

Distributor

Indirect

Trust

Direct

Trust

Lying ActorNormal Actor

0,00

50,00

100,00

150,00

200,00

250,00

Manufacturer

Info.-

Sharing

Indirect

Trust

Direct

Trust

Lying ActorNormal Actor

(d) TVAmp Manufacturer.

Figure 1: Bullwhip-Effect visualized for the different Actors.

Table 6: Veriﬁcation based on average bullwhip values with deactivated information sharing.

Retailer Wholesaler Distributor Manufacturer

(Dejonckheere et al., 2004) 1.67 2.99 5.72 11.43

(Chatﬁeld et al., 2004) 2.22 5.21 11.51 23.77

(Chatﬁeld and Pritchard, 2013) 2.23 5.21 11.39 23.31

Results 1.49 2.35 3.41 11.70

value for the Wholesaler. In the presence of Lying Ac-

tors the shift is less extreme, with an average increase

of 48% which is much more evenly distributed across

the separate actors. Disabling the direct trust and in-

direct trust further exhibits further negative inﬂuence

on the TVAmp score, however, the inﬂuence degree

is much smaller. So the overall increase of the Bull-

whip effect is at about 199% without Lying Agents

and about 65% for cases with such agents.

Comparing Figures 1(a) and 1(b) to the Figures

1(c) and 1(d) further reveals that the BWE impacts

Retailer and Wholesaler worse with no Lying Actor

included. Distributor and Manufacturer, however, are

mostly affected whenever such liars are present.

Analyzing the effects within one (actor to actor)

test case indicates that the strongest increase of the

BWE occurs between Retailer and Wholesaler. There

the TVAmp value rises by 590% on average. Between

Wholesaler and Distributor the average increase with

332% is still high. In the presence of Lying Agents

increase is about 497%, whereas cases without them

only increase by 168%. This difference in TVAmp

value impact caused by the Lying Agents can simi-

larly be observed looking at the supply chain from

beginning to end: SCs free of liars (test cases 1-4) ex-

hibit an average BWE increase of 403%, while SCs

affected by liars rise about 985% (∼ twice as strong).

5 CONCLUSIONS

Based on an experimental trust examination over dif-

ferent dimensions, this paper was successful in using

created insights to support decision making in order to

reduce the BWE. Due to the scarcity of trust related

supply chain data a gamiﬁed web-based trust behav-

ioral experiment was devised.

The conducted three stepped experiment lead to

the following observations: Step one was able to un-

cover three major procurement proﬁles regarding the

three dimensions price, delivery time and trustworthi-

ness. In step two the trust proﬁles for the previously

identiﬁed groups were established. Beside the insight

that for most people the dimensions of Competence,

Reliability and Quality were of equal importance, it

proved to be a useful cross-validation tool as well:

One group valuated Delivery Time lowest in it’s pro-

curement proﬁle. Once price was out of consideration

(e.g. in the trust evaluation), the same group valu-

ated Reliability highest. This contradicts their low as-

sessment of Delivery Time since both are very similar

constructs (see Section 3). Furthermore a set of trust

information sharing thresholds has been discovered.

The acquired data was used to initialize the dif-

ferent agents in the developed simulation model. The

special focus has been to align the newly created sim-

ulation model with existing analytical SC models. For

a more realistic scenario the lying actors concept was

introduced, enabling existing actors to exhibit oppor-

tunistic behavior.

Furthermore it was possible to show the im-

portance of trust in the procurement and ordering

decision-making. The simulation experiment results

revealed that under the inﬂuence of trust, indirect trust

and information sharing the BWE is weakened. This

ﬁnding conﬁrms the already existing theoretical ideas

from the supply chain literature where these concepts

have been identiﬁed as countermeasures for the BWE

(Moyaux et al., 2007). Additionally it was possible to

conﬁrm existing believes about the difﬁculties of mit-

igating damaged trust relationships (Kim et al., 2006).

The Trust Behavioral Experiment has some limita-

Combining Behavioral Experiments and Agent-based Social Simulation to Support Trust-aware Decision-making in Supply Chains

265

Table 7: Veriﬁcation based on average Bullwhip values with activated information sharing.

Retailer Wholesaler Distributor Manufacturer

(Dejonckheere et al., 2004) 1.67 2.61 3.83 5.32

(Chatﬁeld et al., 2004) 2.22 3.89 5.76 7.62

(Chatﬁeld and Pritchard, 2013) 2.23 3.91 5.78 7.65

Results 1.38 2.11 2.92 10.89

Table 8: Bullwhip-Effect Values of the Validation Test

Cases. Legend: Test Case (TC), Retailer (R), Wholesaler

(W), Distributor (D) and Manufacturer (M).

Bullwhip-Effect Values

TC R W D M

1 2.6654 8.4928 19.8690 67.1102

2 3.6357 25.6233 71.5803 167.4304

3 3.9001 28.1745 79.3702 187.4566

4 4.1932 29.1924 81.1145 191.8513

5 1.2600 10.1519 54.5839 140.3891

6 1.9001 14.5047 90.5963 190.2219

7 2.0574 16.0391 96.8988 194.5419

8 2.1879 16.1255 99.9479 201.4702

tions. First results show that some participants seem

to assess received orders more randomly or rate di-

mensions without considering delay or quality. Am-

biguous explanations of that assessment or missing

background knowledge could be reasons for that. Po-

tential ﬁxes would be raising demographic data, giv-

ing personal, introductory brieﬁngs or experimenting

with practitioners. Furthermore the sample size is rel-

atively small, yet sufﬁcient to gain initial results. Al-

though not incorporated in this paper, current research

is taking place in order to increase participation by

improving the gaming experience. Additional dimen-

sions is also being considered.

Despite the fact that additional research is still

necessary, the preliminary results presented here re-

vealed that the use of computational trust mechanisms

can be helpful to reduce the Bullwhip Effect.

REFERENCES

Abdul-Rahman, A. and Hailes, S. (2000). Supporting trust

in virtual communities. In System Sciences, 2000.

Proceedings of the 33rd Annual Hawaii International

Conference on, pages 9–pp. IEEE.

Argonne National Laboratory (2015). Repast.

Ba, S. and Pavlou, P. A. (2002). Evidence of the Effect

of Trust Building Technology in Electronic Markets:

Price Premiums and Buyer Behavior. MIS Quarterly,

26(3):243–268.

Ba, S., Whinston, A. B., and Zhang, H. (2003). Build-

ing Trust in Online Auction Markets Through an Eco-

nomic Incentive Mechanism. Decision Support Sys-

tems, 35(3):273–286.

Bachmann, R. (2001). Trust, Power and Control in

Trans-Organizational Relations. Organization Stud-

ies, 22(2):337–365.

Braynov, S. and Sandholm, T. (2002a). Contracting With

Uncertain Level Of Trust. Computational Intelli-

gence, 18(4):501–514.

Braynov, S. and Sandholm, T. (2002b). Incentive Compat-

ible Mechanism for Trust Revelation. In Proceed-

ings of the First International Joint Conference on

Autonomous Agents and Multiagent Systems: Part 1,

pages 310–311, Bologna, Italy. ACM Press.

Burnett, C., Norman, T. J., Sycara, K., and Oren, N. (2014).

Supporting trust assessment and decision making in

coalitions. IEEE Intelligent Systems, 29(4):18–24.

Castelfranchi, C. and Falcone, R. (1998). Principles of Trust

for MAS: Cognitive Anatomy, Social Importance, and

Quantiﬁcation. In Proceedings International Con-

ference on Multi Agent Systems (Cat. No.98EX160),

pages 72–79. IEEE Comput. Soc.

Chaib-draa, B. and M

uller, J. (2006). Multiagent based sup-

ply chain management, volume 28. Springer Science

& Business Media.

Chatﬁeld, D. C., Kim, J. G., Harrison, T. P., and Hayya,

J. C. (2004). The Bullwhip Effect-Impact of Stochas-

tic Lead Time, Information Quality, and Information

Sharing: A Simulation Study. Production and Opera-

tions Management, 13(4):340–353.

Chatﬁeld, D. C. and Pritchard, A. M. (2013). Returns and

the bullwhip effect. Transportation Research Part E:

Logistics and Transportation Review, 49(1):159–175.

Christopher, M. (1999). Logistics and supply chain man-

agement: Strategies for reducing cost and improving

service ﬁnancial times: Pitman publishing. london,

1998 isbn 0 273 63049 0 (hardback) 294+ 1× pp.

Clark, A. J. and Scarf, H. (1960). Optimal Policies for a

Multi-Echelon Inventory Problem. Management Sci-

ence, 6(4):475–490.

De Bruijn, H. and Herder, P. M. (2009). System and actor

perspectives on sociotechnical systems. IEEE Trans-

actions on Systems, Man, and Cybernetics-Part A:

Systems and Humans, 39(5):981–992.

Dejonckheere, J., Disney, S., Lambrecht, M., and Towill, D.

(2003). Measuring and avoiding the bullwhip effect:

A control theoretic approach. European Journal of

Operational Research, 147(3):567–590.

Dejonckheere, J., Disney, S., Lambrecht, M., and Towill,

D. (2004). The impact of information enrichment on

the Bullwhip effect in supply chains: A control engi-

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

266

neering perspective. European Journal of Operational

Research, 153(3):727–750.

Donohue, K. and Siemsen, E. (2011). Behavioral opera-

tions: applications in supply chain management. Wi-

ley Encyclopedia of Operations Research and Man-

agement Science.

Forrester, J. W. (1958). Industrial dynamics: a major break-

through for decision makers. Harvard business re-

view, 36(4):37–66.

Gould, S. (2011). OpenForecast.

Greif, A., Milgrom, P., and Weingast, B. R. (1994). Co-

ordination, Commitment, and Enforcement: The Case

of the Merchant Guild. Journal of Political Economy,

102(4):745–776.

Ha, B.-C., Park, Y.-K., and Cho, S. (2011). Suppliers’ affec-

tive trust and trust in competency in buyers: Its effect

on collaboration. International Journal of Operations

& Production Management, 31(1-2):56–77.

Haghpanah, Y. and DesJardins, M. (2010). Using a Trust

Model in Decision Making for Supply Chain Manage-

ment. In Proceedings of the 3rd AAAI Conference on

Interactive Decision Theory and Game Theory, pages

25–29. AAAI Press.

Handﬁeld, R. (2003). Trust in Supply Chain Relationships:

What Does It Mean to Trust? - Part I.

Houser, D. and Wooders, J. (2006). Reputation in Auctions:

Theory, and Evidence from eBay. Journal of Eco-

nomics & Management Strategy Management Strat-

egy, 15(2):353–369.

Kim, P. H., Dirks, K. T., Cooper, C. D., and Ferrin, D. L.

(2006). When more blame is better than less: The

implications of internal vs. external attributions for the

repair of trust after a competence-vs. integrity-based

trust violation. Organizational Behavior and Human

Decision Processes, 99(1):49–65.

Kim, W.-S. (2009). Effects of a trust mechanism on com-

plex adaptive supply networks: An agent-based social

simulation study. Journal of Artiﬁcial Societies and

Social Simulation, 12(3):4.

Kwon, I.-W. G. and Suh, T. (2005). Trust, commitment

and relationships in supply chain management: a path

analysis. Supply chain management: an international

journal, 10(1):26–33.

La Londe, B. J. and Masters, J. M. (1994). Emerging Logis-

tics Strategies: Blueprints for the Next Century. Inter-

national Journal of Physical Distribution & Logistics

Management, 24(7):35–47.

Laeequddin, M., Sahay, B. S., Sahay, V., and Waheed, K. A.

(2010). Measuring trust in supply chain partners’ rela-

tionships. Measuring Business Excellence, 14(3):53–

69.

Lin, A. F.-r., Sung, Y.-w., and Lo, Y.-p. (2005). Effects of

Trust Mechanisms on Supply-Chain Performance: A

Multi-Agent Simulation Study. International Journal

of Electronic Commerce, 9(4):91–112.

Marsh, S. P. (1994). Formalising Trust as a Computational

Concept. Disseration, University of Stirling.

Mentzer, J. T., DeWitt, W., Keebler, J. S., Min, S., Nix,

N. W., Smith, C. D., and Zacharia, Z. G. (2001).

Deﬁning Supply Chain Management. Journal of Busi-

ness Logistics, 22(2):1–25.

Morgan, R. M. and Hunt, S. D. (1994). The Commitment-

Trust Theory of Relationship Marketing. Journal of

Marketing, 58(3):20–38.

Moyaux, T., Chaib-draa, B., and D’Amours, S. (2007). In-

formation sharing as a coordination mechanism for re-

ducing the bullwhip effect in a supply chain. IEEE

Transactions on Systems, Man, and Cybernetics, Part

C (Applications and Reviews), 37(3):396–409.

Ottens, M., Franssen, M., Kroes, P., and Van De Poel, I.

(2006). Modelling infrastructures as socio-technical

systems. International Journal of Critical Infrastruc-

tures, 2(2-3):133–145.

Ozer,

O., Zheng, Y., and Chen, K.-Y. (2011). Trust in

Forecast Information Sharing. Management Science,

57(6):1111–1137.

Pinyol, I. and Sabater-Mir, J. (2013). Computational trust

and reputation models for open multi-agent systems:

a review. Artiﬁcial Intelligence Review, 40(1):1–25.

Raimondo, M. A. (2000). The measurement of trust in mar-

keting studies: a review of models and methodologies.

In 16th IMP-conference, Bath, UK. Citeseer.

Resnick, P. and Zeckhauser, R. (2002). Trust Among

Strangers in Internet Transactions: Empirical Analy-

sis of eBay’s Reputation System. In Baye, M. R., edi-

tor, Advances in Applied Microeconomics, number 11,

pages 127–157. Emerald Group Publishing Limited.

Ring, P. S. and Van de Ven, A. H. (1994). Developmental

processes of cooperative interorganizational relation-

ships. Academy of management review, 19(1):90–118.

Rousseau, D. M., Sitkin, S. B., Burt, R. S., and Camerer,

C. (1998). Not so different after all: A cross-

discipline view of trust. Academy of management re-

view, 23(3):393–404.

Sabater, J. and Sierra, C. (2005). Review on Computational

Trust and Reputation Models. Artiﬁcial Intelligence

Review, 24(1):33–60.

ServiceNow (2016). % of undamaged goods after ship-

ping/transportation.

Tykhonov, D., Jonker, C., Meijer, S., and Verwaart, T.

(2008). Agent-based simulation of the trust and trac-

ing game for supply chains and networks. Journal of

Artiﬁcial Societies and Social Simulation, 11(3):1.

Yu, B. and Singh, M. P. (2002). An Evidential Model

of Distributed Reputation Management. In Proceed-

ings of the First International Joint Conference on

Autonomous Agents and Multiagent Systems: Part 1

- AAMAS ’02, pages 294–301, Bologna, Italy. ACM

Press.

Yu, H., Shen, Z., Leung, C., Miao, C., and Lesser, V. R.

(2013). A survey of multi-agent trust management

systems. IEEE Access, 1:35–50.

Zacharia, G., Moukas, A., and Maes, P. (2000). Collab-

orative reputation mechanisms for electronic market-

places. Decision Support Systems, 29(4):371–388.

Combining Behavioral Experiments and Agent-based Social Simulation to Support Trust-aware Decision-making in Supply Chains

267