The current issue and full text archive of this journal is available on Emerald Insight at:

www.emeraldinsight.com/0263-5577.htm

An Ontology-
An Ontology-based Bayesian based Bayesian
network modelling for supply network
modelling
chain risk propagation
Shoufeng Cao and Kim Bryceson 1691
School of Agriculture and Food Sciences,
Received 20 January 2019
The University of Queensland, Brisbane, Australia, and Revised 28 May 2019
Damian Hine 19 June 2019
Accepted 5 July 2019
UQ Business School, The University of Queensland, Brisbane, Australia

Abstract
Purpose – Supply chain risks (SCRs) do not work in isolation and have impact both on each member of a chain
and the performance of the entire supply chain. The purpose of this paper is to quantitatively assess the impact
of dynamic risk propagation within and between integrated firms in global fresh produce supply chains.
Design/methodology/approach – A risk propagation ontology-based Bayesian network (BN) model
was developed to measure dynamic SCR propagation. The proposed model was applied to a two-tier
Australia-China table grape supply chain (ACTGSC) featured with an upstream Australian integrated grower
and exporter and a downstream Chinese integrated importer and online retailer.
Findings – An ontology-based BN can be generated to accurately represent the risk domain of interest using
the knowledge and inference capabilities inherent in a risk propagation ontology. In addition, the analyses
revealed that supply discontinuity, product inconsistency and/or delivery delay originating in the upstream firm
can propagate to increase the downstream firm’s customer value risk and business performance risk.
Research limitations/implications – The work was conducted in an Australian-China table grape supply
chain, so results are only product chain-specific in nature. Additionally, only two state values were considered for
all nodes in the model, and finally, while the proposed methodology does provide a large-scale risk network map, it
may not be appropriate for a large supply chain network as it only follows the process flow of a single supply chain.
Practical implications – This study supports the backward-looking traceability of risk root
causes through the ACTGSC and the forward-looking prediction of risk propagation to key risk
performance measures.
Social implications – The methodology used in this paper provides an evidence-based decision-making
capability as part of a system-wide risk management approach and fosters collaborative SCR management,
which can yield numerous societal benefits.
Originality/value – The proposed methodology addresses the challenges in using a knowledge-based
approach to develop a BN model, particularly with a large-scale model and integrates risk and performance for a
holistic risk propagation assessment. The combination of modelling approaches to address the issue is unique.
Keywords Supply chain risk management, Global supply chain management, Risk performance,
Fresh produce, Ontology-based Bayesian network, Risk propagation
Paper type Research paper

1. Introduction
Companies traditionally focus risk management efforts within their boundaries
(Revilla and Saenz, 2017). As such, previous studies have mostly investigated supply
chain risks (SCRs) at a firm level (Świerczek, 2016). However, SCRs do not work in
isolation and risk outcome of one firm can be easily transformed into a risk event for
another firm (Manuj and Mentzer, 2008). Risk dependency and propagation has thus
become a critical issue in today’s supply chains, particularly in globally dispersed supply
chains where companies increasingly align key suppliers and buyers for competitive
advantage. It is thus vital to understand and manage risk dependencies across inter-firm Industrial Management & Data
Systems
Vol. 119 No. 8, 2019
This study was supported by an “Australian Government Research Training Program Scholarship”. pp. 1691-1711
The authors thank the two anonymous reviewers whose comments and suggestions greatly assisted © Emerald Publishing Limited
0263-5577
with improving the quality of the manuscript. DOI 10.1108/IMDS-01-2019-0032
IMDS operations as practical examples prove that risk consequences at the inter-firm level may
119,8 be costly for supply chains (Świerczek, 2016). Knowing how to anticipate risk propagation
within and between firms can enable predictive or proactive risk management strategies
for risk interdependence, thus creating competitive advantage for firms and the entire
supply chain (Ouabouch and Paché, 2014).
To investigate the issue of supply chain risk propagation, a BN approach is often used to
1692 develop a risk network map in a supply chain (e.g. Badurdeen et al., 2014, Garvey et al., 2015,
Qazi et al., 2017; Ojha et al., 2018). However, a significant challenge in a knowledge-based BN
for SCRs is the proper identification of risk events and risk categories that can impact a
supply chain (Lockamy and McCormack, 2012). There is also a debate around building a
suitable structural model from expert knowledge due to possible subjective judgements
(Cowell et al., 2007).
Agri-food supply chains that involve the production of raw and/or processed food
products to the consumer are more vulnerable to risk propagation due to specific product
and process characteristics, such as variable harvest and production yields and the huge
impact of weather conditions on product availability, perishability of end products and
consumer demand (Van der Vorst and Beulens, 2002). Moreover, agri-food supply chains
often have a dynamic and non-transparent structure due to the prevalence of transactional
relationships (Roth et al., 2008). As a result, firms find it more difficult to see the
detrimental effects flowing from one part of a supply chain to another (Wu et al., 2007).
Without understanding risk interactions within and between firms, food supply chain
firms may attempt to implement company-specific risk mitigation strategies, which may
not lead to the desired performance (Srivastava et al., 2015) and even negatively impact
another firm’s performance.
While increasing works have examined risk propagation in a supply chain context
(e.g. Wu et al., 2007; Shin et al., 2012; Badurdeen et al., 2014; Garvey et al., 2015;
Qazi et al., 2017; Qazi et al., 2018; Ojha et al., 2018), risk quantification in the food supply
chain has received limited attention in the literature (Rathore et al., 2017), and there is an
apparent shortage of quantitative investigation of risk propagation in global fresh
produce supply chains (GFPSCs). While some studies have quantified risk dependency
and propagation in food supply chains, their investigations are mainly focussed on the
firm level, rather than a whole chain (e.g. Diabat et al., 2012; Chaudhuri et al., 2016;
Prakash et al., 2017).
To fill these research gaps, this research developed a risk propagation modelling
framework for a two-tier GFPSC featured with an upstream Australian integrated grower
and exporter and a downstream Chinese integrated importer and online retailer to answer
the two questions:
(1) What risks would propagate within and between firms to impact the supply chain
performance along the GFPSC?
(2) What influence does risk propagation have on upstream firms’ customer value risk
and further on downstream firms’ customer value risk and business performance risk?

1.1 Purpose
The purpose of this research was to create a risk propagation ontology-based BN
model to address the challenges in a knowledge-based approach and to quantitatively
measure the impact of dynamic risk propagation in a GFPSC. The methodology
directly used the knowledge and inference capabilities inherent in a risk propagation
ontology, which could facilitate the creation of a BN to accurately represent the risk
domain of interest.
1.2 Structure An Ontology-
The remainder of the paper is organised as follows. Section 2 provides an overview of extant based Bayesian
studies in the relevant areas. Section 3 presents the design of an ontology-based BN model for network
dynamic SCR propagation. Section 4 shows the application of the proposed model to a two-tier
GFPSC. Theoretical and managerial implications are followed in Sections 5 and 6, respectively. modelling
Finally, potential limitations and future research directions are presented in Section 7.
1693
2. Literature review
2.1 Supply chain risk propagation
Wu et al. (2007), among the first researchers, modelled how disruption risk can propagate
through a supply chain and affect its performance. However, modelling holistic interaction
between SCRs has recently gained the interest of researchers in SCRM (Garvey et al., 2015).
Ghadge et al. (2011) developed a conceptual model to identify key SCRs and their
approximate risk propagation durations for a typical natural disaster in global supply
networks. Shin et al. (2012) modelled the risk propagation by integrating Bayesian Belief
Network and supply chain network after describing risk factors and inter-relationships
between vertices. Badurdeen et al. (2014) introduced a SCR taxonomy and a risk network
map to capture the interdependencies between 105 risk events. Garvey et al. (2015)
developed an analytical framework to measure risk propagation in supply networks, where
the interdependencies among 12 different risks and supply chain network structures were
considered. Qazi et al. (2017) developed dependency based probabilistic risk measures after
extending Garvey et al.’s (2015) study to capture the interdependency between risks and
strategies for a medium-sized Turkish company. Daultani et al. (2017) modelled the firm-
level risks by explicitly integrating risk propensities of different functional divisions from
the perspective of an original equipment manufacturer. More recently, Goswami (2018)
represented the underlying SCRs in terms of logistics risk, planning risk, and sourcing risk
and examined their interdependencies in new product design decisions for the entire supply
chain of a product focussed manufacturing enterprise. Qazi et al. (2018) operationalised a
SCR network management process that captures interdependencies between risks, multiple
performance measures and risk mitigation strategies within a (risk) network setting for a
leading global technology provider. These works explored the propagation of
interconnected risks, but they only constrained their efforts in a focal firm’s supply chain
or supply chain network and did not answer how various risks propagate among various
supply chain stakeholders.
SCRs should be managed as a whole for an end-to-end supply chain as they transmit
among supply chain members (Zhu et al., 2017). However, a limited number of studies have
investigated risk propagation across firms in a multi-tier supply chain, for example, Lin and
Zhou (2011) qualitatively identified internal and external dimensions of SCRs and their
respective causes in a three-tier supply chain and Pfohl et al. (2011) examined the
relationships between 21 risks in a virtual supply chain consisting of a focal industrial
company, its first-tier supplier and the first-tier supplier’s 3PL and classified the risks based
on their driving and dependence power. Their merit is that they examined risk propagation
across firms, but only identified the risk consequences and missed the consideration of risk
probability. A latest work that focusses on cross-firm risk propagation along a supply chain
was conducted by Ojha et al. (2018), where a holistic measurement approach was developed
for predicting the complex behaviour of risk propagation in four-tier automotive SC network
for improved SCRM. However, a significant limitation is that their model only considered a
small number of risk factors and the link failure at the nodes were not fully considered.
Although increasing numbers of studies have emerged to investigate the dependency and
propagation of SCRs, diagnostics of the impact of risk propagation on performance has not
yet received much attention (Sharma and Routroy, 2016).
IMDS 2.2 Supply chain risk propagation modelling technique
119,8 Siew are examples of qualitative methods for mapping the causal relationships between SCRs.
Given that they have limited capability to compare and prioritise interdependent risks,
quantitative methods, such Petri-net modelling (Wu et al., 2007), have been used to capture
quantitative risk propagation. The quantitative approach is however not always available due
to a lack of quantitative data or inappropriate model parameters (Wang et al., 2012).
1694 As such, several hybrid approaches, such as interpretive structural modelling (Pfohl et al.,
2011; Diabat et al., 2012; Srivastava et al., 2015), Fuzzy interpretive structural modelling
(Chaudhuri et al., 2016), and BN modelling framework (Garvey et al., 2015; Qazi et al., 2017;
Qazi et al., 2018), have been used to assess the cause-effect relationships of SCRs.
BNs are the most preferred technique as it enables a systematic risk analysis
through identifying causal relationships between key factors and final outcomes of a
system (Bryceson and Smith, 2008; Pollino and Henderson, 2010). Structural models of a
BN can be developed either via existing knowledge or via a data-based approach
(Sharma and Sharma, 2015). Given that the data-based approach is subject to non-
availability of complete information or data concerning risks (Qazi et al., 2017),
knowledge-based approach is widely used to build a structural risk network (Badurdeen
et al., 2014; Sharma and Routroy, 2016; Qazi et al., 2017). However, the knowledge-based
BN often involves expert judgement in both qualitative and quantitative parts of the
model, which could make the process quite subjective (Qazi et al., 2017). Also, experts
may have incomplete knowledge of a complex system, which makes it a challenge to
identify relevant variables and their relationships (Fenz et al., 2009) particularly for a
large-scale BN.

2.3 Holistic supply chain risk assessment

A holistic risk assessment is vital for organisations in modern supply chains as risk
events often influence each other and rarely act independently (Badurdeen et al., 2014).
Aven (2016) indicated that a general research should focus on dynamic risk assessment
and management, rather than the static or traditional risk assessment. Moreover,
Chaudhuri et al. (2016) argued that an effective risk assessment approach should capture
the propagation of different SCRs and analyse their impact on the overall performance
measures. Thus, a holistic SCR assessment should link SCR sources with supply chain
performance, which could generate effective proactive risk management strategies
(Ghadge et al., 2012).
Mature companies known for their excellence have widened the focus of their SCRM
efforts to align internal efforts with key suppliers, strategically important buyers and other
relevant stakeholders (Revilla and Saenz, 2017). This is because mitigating some isolated
risks within a firm’s boundary may not only result in an undesired performance (Srivastava
et al., 2015), but can also impact on another firm’s performance (Zeng and Rogers, 2011).
As a result, a holistic model for SCRM should go beyond the individual firms to inter-firm
levels and integrate performance and risk management to create shareholder value
(Oehmen et al., 2009).

2.4 Agri-food supply chain and risk propagation

Agri-food supply chains distinguish from manufacturing and service sector supply chains
because the risk can arise from weather, biological aspects of production cycles and
perishable characteristics (Sporleder and Boland, 2011). Given the uniqueness of agrifood
supply chains, risks are pervasive for all parties (Sporleder and Boland, 2011) and the
occurrence of risks at a point can easily transmit in part of, or in whole supply chain
(Świerczek, 2014). For example, any break in any part of the fresh produce supply chain,
such as poor harvesting, post-harvest management, breaks in temperature/humidity
integrity and distribution and logistics procedures, could impact the safety, quality and An Ontology-
shelf life of the produce (Siew and Lam-Chan, 2007). Food supply chains are also exposed to based Bayesian
a new set of uncertainties and risks with global spanning (Dani and Deep, 2009). network
For example, global trade in many agricultural commodities is subject to market forces and
government policy (Sporleder and Boland, 2011), which can disturb the smooth product modelling
flow. As global agri-food supply chains represented with several actors who are exposed to
risks occurred in different countries and regions that can result in systematic risks, 1695
how to manage risk and potential risk propagation are thus really important for fresh
food supply chains.
There are a rather small number of studies that have exclusively explored risk
dependency and propagation in food supply chains given that many works, such as
Rathore et al. (2017) who delivered a risk quantification framework for food supply
chains, did not address risk propagation. Diabat et al. (2012) who are among the first
researchers determined the interdependencies between five risk categories in food supply
chains, including product/service management risk, macro level risk, demand
management risk, supply management risk and information management risk.
Srivastava et al. (2015) analysed interdependencies of 24 risks and their impact on
performance in fresh food retail supply chains and Chaudhuri et al. (2016) examined how
26 risk drivers interdepends and impact important performance measures in food
processing supply chain. Prakash et al. (2017) developed an integrated methodology to
identify, quantify, analyse and evaluate interdependencies among 17 risks and relevant
mitigation methods in perishable food supply chains. Most recently, Ali et al. (2019)
developed a sustainable framework for assessing the relationships among identified
major risks in food supply chains. However, no research has been conducted to measure
risk propagation in GFPSCs.

3. Research methodology
This research developed an ontology-based BN model using the knowledge and inference
capabilities inherent in a risk propagation ontology, rather than traditional knowledge-
based BN. Ontologies which define basic terms and relations of a domain and the rules
for terms and relations, can create a potential solution for building an objective BN
(e.g. Devitt et al., 2006; Fenz et al., 2009; Bucci et al., 2011).
An ontology-based BN approach was proposed by referring to the works of Pollino and
Henderson (2010) and Badurdeen et al. (2014) as shown in Figure 1. The essential and initial

Risk propagation
ontology-based
Bayesian network
model
development
Data collection Model evaluation
• Reuse (revise) or
Define modelling develop a risk • Collect prior and validation
propagation Data analysis
objective probabilities
ontology in the • Check acyclicity
domain of • Collect prior constraint • What-if scenario
• An objective
interest conditional analysis
agreed upon • Validate
probabilities to
between the • Automatically or conditional • Sensitivity
parameterise
modeler and end manually probabilities analysis
conditional
users translate the using sensitivity
ontology into
probability tables
(CPTs)
analysis Figure 1.
Bayesian Risk propagation
network ontology-based
• Define and Bayesian network
assign state approach
values
IMDS step is a consensual modelling objective between the modeller and end users, which guides
119,8 the development of an ontology-based BN in terms of: developing/revising a relevant risk
propagation ontology or directly reusing an existing ontology; automatically or manually
translating the risk propagation ontology into a BN with a software tool; and defining and
assigning state values to each node. This is followed by the collection of prior probability
and prior conditional probability. After that the developed model was evaluated and
1696 validated to ensure the validity and reliability of the model, where iterative revisions
marked with dash lines was conducted to remove any discrepancy. Data analysis comes as
the final step, which involves the measurement of risk propagation effect with sensitivity
analysis and what-if scenario analysis.

4. Application of ontology-based Bayesian network model

An illustrative case of the Australia-China table grape supply chain (ACTGSC) was used to
demonstrate the development of an ontology-based BN model for risk propagation in a
global food supply chain. The reasons behind this are that: table grapes are vulnerable to
cold chain risk and are Australia’s largest horticulture export (ATGA, 2012); and China is
the largest export market for Australian table grape, where Australian suppliers compete
with low-cost southern hemisphere rivals through quality and better customer service
(Scott, 2016).
The ACTGSC considered includes an Australian integrated table grape grower and
exporter (named AA) who export nearly half of their table grapes to China, and a Chinese
integrated importer and retailer (named BB), who is one of AA’s regular Chinese customers
and sells imported table grapes via both traditional wholesale and online retail channels.

4.1 Modelling objective

Upon discussion with case participants, the modelling objective was defined to assess how
risks propagate to influence customer value risk within AA, and how AA’s customer value
risk could further propagate to influence the customer value risk and final business
performance risk of BB’s online retailing sales.

4.2 Model development

4.2.1 Developing a risk propagation ontology. SCR information and the integration of
performance and risk were, respectively, identified prior to creating a comprehensive and
accurate risk propagation map.
Operational risk events were initially gathered from 496 SCRM papers retrieved from 17
quality journals of operations research and management science in 2004–2017 and a two-
month operational internship with an Australian fresh produce exporter. All the harvested
risk terms were then grouped into internal risk and external risk (Wagner and Bode, 2008;
Trkman and McCormack, 2009). The internal operational risk was further segmented into
people risk, process risk and infrastructure risk as risk can arise from people, process and
infrastructure or technology in supply chain operations (Lockamy and McCormack, 2010).
Next, all the risk events were verified and validated by ten ACTGSC actors following the
supply chain processes with the “story telling” method. Details of the participating firms,
including AA and BB, are shown in Table I. The ten firms were identified at different
tiers of the ACTGSC for developing a multi-tier chain-wide risk propagation ontology.
Integrated supply chain firms were included given the emergence of vertical integration.
Online retailers are also included given the development of online sales.
Operational performance of the ACTGSC were defined in terms of customer value and
business performance (Wieland and Wallenburg, 2012). Product availability, quality
consistency, on-time delivery and order fulfilment correctness were identified as the key
indicators of customer value as they were highly valued by customers (Gaudenzi and An Ontology-
Borghesi, 2006; Yin and Ma, 2015). The four indicators were also validated by data mining 709 based Bayesian
consumer comments on buying Australian table grapes from two popular Chinese online network
sellers. Meanwhile, sales growth, cost reduction, profitability were identified as the key
indicators of business performance by referring Baier’s (2008) work. Supply discontinuity, modelling
quality inconsistency, delivery delay and order fulfilment error were consequently identified
as the key indicators of customer value risk, while increased cost, reduced sales and profit 1697
slippage were identified as the key indicators of business performance risk.
The risk propagation ontology validated by the ten participating firms had 155 concepts,
which were represented with the Unified Modelling Language (UML) to illustrate risk
dependence and propagation to the four key indicators of customer value risk (i.e. supply
discontinuity, delivery delay, quality inconsistency and order fulfilment error) within each
actor of ACTGSC and finally to the three key indicators of business performance risk (i.e.
reduced sales, increased cost and profit slippage) of Chinese importers and online retailers.
The constructed risk propagation ontology was accordingly revised to fit the case and
the modelling objective in this research. As per discussed with the case participants (AA
and BB), the “order fulfilment error” occurring at AA and its associated concepts were
deleted as BB repack table grapes received for customised online sales and did not think
that AA’s “order fulfilment error” had a direct impact on their order fulfilment error.
Besides, two new nodes (i.e. behaviour uncertainty of cash buyer_AA and behaviour
uncertainty of cash buyer_BB) were added given the active involvement of Chinese cash
buyers in recent years.
4.2.2 Translating risk propagation ontology into a Bayesian Network (BN). Ontology
concepts were directly used to create the nodes of a BN, and relations between the concepts
were used to establish the edges between the BN nodes (Fenz et al., 2009).
GeNIe Modeller was chosen to illustrate the risk network and calculate the impact of risk
propagation as it works well with a pre-defined large-scale model (GeNIe User Manual,
2017). Although a BN structure can be generated with ontologies through either automatic
or manual creation (Devitt et al., 2006), the risk propagation ontology-based BN was
generated with manual creation as the risk propagation ontology was conceptually depicted
with UML diagrams and has not yet been implemented with machine interpretable
representations. A total of 100 nodes were depicted in the risk propagation ontology-based
BN model, where _AA and _BB were suffixed to each variable to indicate its occurrence at
either the integrated upstream firm (AA) or the integrated downstream firm (BB). Figure 2
shows the chain-wide risk propagation map where the arrows illustrate the occurrence of
risks in the upstream firm propagate to customer value risk and eventually impact the
downstream firm’s customer value risk and business performance risk.

No. of firm Title of informant Supply chain category

1 General manager Australian table grape grower and packer

2 Operations manager Australian table grape grower and exporter
3 Directors’ assistant (AA)
4 Managing director Australian merchant exporter
5 Sales and marketing manager
6 Executive director
7 General manager (BB) Chinese importer and online retailer Table I.
8 Produce technical manager Chinese importer and traditional retailer Participating firms in
9 General manager Chinese importers and wholesaler risk propagation
10 Sales and marketing manager development
 119,8
IMDS

1698

Figure 2.

GeNIe modeller
Chain-wide risk
propagation map in
Inability to meet Unexpected import Entry CIQ
Natural Production quality customer quality restriction_BB failure_BB
disaster_AA variability_AA requirement_AA

Yield Quarantine Supply

Supply Product Customer Reputation
reduction_AA inspection discontinuity_BB
discontinuity_AA stockout_BB value risk_BB risk_BB
failure_AA

Inability to
Market supply Supply Product
secure Inability to Ordering quantity Inability to meet
shortage_AA shortage_AA stockout_AA
supply_AA secure uncertainty_BB customer quality
supply_BB requirements_BB
Quality degradation
Production Quality in shipping_BB Order fulfillment
Cash buyer degradation in
capacity error_BB
involvement_AA harvest_AA
constraint_AA Customer Loss of
Quality churn_BB sales_BB
Quality degradation in
degradation in storage_BB
packhouse_AA
Quality Quality
Quality Claim
degradation in inconsistency_BB
Quality Inconsistency_AA dispute_BB
receiving_BB
degradation in
Extreme
loading_AA
weather_AA Quality
degradation in
Quality degradation repacking_BB
Quality degradation in in trucking to
Maturation Harvesting trucking to port_AA warehouse_BB Customer
uncertainty_AA delay_AA claim_BB
Reduced
Quality sales_BB
Picker Stock degradation in
Grading delay_AA Import regulation distribution_BB
shortage_AA unavailability_AA
variation_BB

Labor regulation Packer Packaging

change_AA shortage_AA delay_AA Sales
competition_BB
Document Import
Package Inspection Handling clearance
Customer order material booking delay_BB delay_BB
transmission delay_AA delay_AA
error_AA Sales price
Market volatility_BB Issued credit Profit
slippage_BB
Inspection Inspection On-farm notes_BB slippage_BB
Order
booking confirmation inspection
processing Holiday
error_AA delay_AA delay_AA
delay_AA impact_BB
Sales time
Inspection conflict_BB
notification
Customer order
delay_AA
confirmation
delay_AA
Loading
booking
Trucking to Distribution
delay_AA
warehouse delay_BB
delay_BB
Lack of integrated Loading Loading Product
Loading and trucking Delivery Customer order
order management booking confirmation Delivery delay_AA receiving
delay_AA delay_BB cancellation_BB
system_AA error_AA delay_AA delay_BB

Loading
notification
delay_AA Destination port Order preparing
Cash buyer
arrival delay_BB delay_BB
involvement_BB
Holiday Freight booking Increased
Product
impact_AA delay_AA storage
overstock_BB
cost_BB

Freight Procurement
SLI error_AA confirmation competition
risk_BB Increased
delay_AA High repacking
product
wastage_BB
Shipping delay_AA wastage_BB
Export document
Freight space
submission Increased
shortage_AA Exchange rate Increased
delay_AA procurement
variation_BB cost_BB
price_BB
Export Increased
Export regulation clearance repacking
variation_AA delay_AA cost_BB
4.2.3 Assigning state values. The state values of each node were assigned for measurable An Ontology-
communications. The trade-off between information loss and model over-complication was based Bayesian
balanced in defining state values, since having too many states can overcomplicate the model,
while defining too few states can result in information loss (Pollino and Henderson, 2010).
network
Given that ACTGSC actors lack historical risk data within and beyond their boundaries, modelling
best-guess estimations of probabilities had to be employed for our large-scale model, which
could present challenges in expert elicitation when the conditional probability tables (CPTs) 1699
were too large to handle with human intuition/experience. Although most extant works
often considers at least three risk states to avoid information loss (Goswami, 2018), two
states can be found in some previous works (e.g. Qazi et al., 2017, 2018) and the loss of
information is less crucial if the states of each node can successfully represent modelling
objectives (Pollino and Henderson, 2010). Two states were finally assigned as per discussed
with our case participants. Following this, three node types – binary, Boolean and
deterministic (Korb and Nicholson, 2010; GeNie User Manual, 2017) – were determined
based on the node attributes. Table II shows the classification of all the nodes in the
ontology-based BN. In total, 12 nodes were defined as Boolean nodes, which take the values
of “Yes” or “No” under Boolean logic. Meanwhile, three nodes were set as deterministic
nodes, which represent either constant value, taking the values of “Yes” or “No”. The
remaining 85 nodes were defined as binary nodes and took the values of “High” or “Low”.

4.3 Data collection

Direct expert inputs were used to elicit posterior and conditional probability (Cowell et al.,
2007). To ensure the reliability of the risk probability elicitation from the case participants,
the elicitation process proposed by Renooij (2001) was followed to gather the data from
managers, including six stages as shown in Table III.
Respective data collection sheets were prepared to elicit posterior and conditional
probabilities from experts in the case companies (AA and BB). Table IV shows a snapshot of
data collection sheet for building CPTs.
Although it is better to have more than one expert in the data elicitation (Renooij, 2001),
one expert from AA and one from BB were chosen as they were involved in the
development and revision of risk propagation ontology and have the necessary domain
knowledge as top management. The expert from AA is the directors’ assistant assigned
with the upstream 51 nodes, while the expert from BB is the general manager
assigned with the downstream 49 nodes. For reasons of confidentiality, the posterior and
conditional probabilities were assigned with approximations and hypothesised data,
which could represent the current state of ACTGSC as per indicated by the case
participants. The responses from managers were then checked to ensure the probabilities
provided conform to observed frequencies and obey the laws of probability. Moreover, the
researchers validated certain probabilities with empirical data gathered from the industry.

Node type Node value Node name

Boolean node Yes/No Customer order transmission error_AA; Extreme weather_AA; Natural
disaster_AA; Freight space shortage_AA; Inspection booking error_AA;
Loading booking error_AA; Shipper’s letter of instruction error_AA; Labour
regulation change_AA; Production capacity constraint_AA; Order fulfilment
error_BB; Product overstock_BB; Sales time conflict_BB
Deterministic Yes/No Lack of integrated order management system_AA; Holiday impact_AA; Table II.
node Holiday impact_BB The classification of
Binary node High/Low The remaining 85 nodes node types
IMDS No. Stage Description
119,8
1 Heuristics and biases Beware of reasons for motivational bias and cognitive bias, and understand how
to avoid potential biases
2 Selection and Should select an expert who has the necessary domain knowledge; it is also
motivation better to have more than one expert involved
3 Training Should coach an expert to be familiar with the concept of probability and to
1700 express his knowledge in the format required by the elicitation method
4 Structuring Document variable definition and states, and prepare an attractive format
5 Elicitation and Should avoid coaching the expert and taking too much control, and complete
documentation elicitation in less than one hour
Table III. 6 Verification Check if the probabilities provided are well calibrated (conform to observed
A well-structured frequencies), obey the laws of probability (are coherent) and are reliable
elicitation process Source: Renooij (2001)

Table IV.
Elicitation sheet for
prior conditional
probability of supply
discontinuity

Prior marginal probabilities for all nodes are shown in Figure 3. All the conditional
probabilities were then automatically generated with CPTs in the GeNie Modeller.
The mathematical function utilised is shown with a simple case in Figure 4, where one
child risk event C is dependent on two parents A and B, the probability of node C on
parent nodes A and B is: PC ¼ P(C|A, B)P(A)P(B)+P(C|A,B’)P(A)P(B’)+P(C|A’,B)P(A’)
P(B)+P(C|A’,B’)P(A’)P(B’).

4.4 Model evaluation and validation

The risk network map generated from the risk propagation ontology in the GeNie
Modeller was checked to avoid the violation of acyclicity constraint, i.e., there are no
ways to loop back to the same node from any node (Hänninen, 2014). Sensitivity analysis
was then conducted in the GeNIe Modeller to validate the prior conditional probability
parameters elicited. The results of the sensitivity analysis are huge in nature as a
result of the large scale model. Due to space limitations, an illustrative example of setting
the node of “quality degradation in receiving _BB” as a target node with high state is
presented in Figure 5.
 Production quality Inability to meet customer quality Unexpected import Entry CIQ failure_BB
Natural disaster_AA
variability_AA requirements_AA restriction_BB High 1%
Yes 10%
High 15% High 13% High 1% Low 99%
No 90%
Low 85% Low 87% Low 99%
Supply Customer value Reputation risk_BB
Supply Product stockout_BB discontinuity_BB risk_BB
Quarantine inspection failure_AA Inability to secure supply_BB
discontinuity_AA High 64% High 67% High 70%
High 1% High 66% High 42%
High 16% Low 36% Low 33% Low 30%
Low 99% Low 34% Low 58%
Yield reduction_AA Low 84%
High 12% Product stockout_AA
Low 88% High 16%
Market supply Low 84%
shortage_AA Inability to secure supply_AA Supply shortage_AA Ordering quantity Inability to meet
High 10% Quality degradation in uncertainty_BB customer quality
High 15% High 11%
harvest_AA High 60% requirements_BB
Low 89% Low 90%
Low 85% Order fulfillment
High 18% Low 40% High 9%
error_BB
Low 82% Quality degradation in Low 91%
Cash buyer Production capacity shipping_BB Yes 5%
involvement_AA constraint_AA Quality degradation in No 95%
High 8%
High 80% Yes 5% packhouse_AA
Low 92% Quality degradation Customer churn... Loss of sales_BB
Low 20% No 95% High 1% in storage_BB High 85%
High 88%
Low 99% High 20% Low 15%
Low 12%
Quality degradation in Low 80%
Extreme weather_AA Quality Quality degradation in Quality
Yes 20% loading_AA Inconsistency_AA receiving_BB inconsistency_BB Claim dispute_BB
No 80% High 1% High 54%
High 20% High 27% High 28%
Low 99% Low 46%
Low 80% Low 73% Low 72%
Quality degradation
Maturation uncertainty_AA Harvesting delay_AA Quality degradation in trucking to in repacking_BB
Quality degradation
port_AA
High 30% High 55% in trucking to High 30%
Low 70% Low 45% High 1% warehouse_BB
Low 70% Reduced
Low 99% High 10% sales_BB
Labor regulation Picker shortage_AA Grading delay_AA Stock unavailability_AA Low 90% High 85%
Quality degradation Customer claim_BB
change_AA High 12%
High 54% High 28% in distribution_BB High 51% Low 15%
Yes 60% Low 46% Low 88% Low 72% High 40% Low 49%
No 40% Import regulation
variation_BB Low 60%
Packer shortage_AA
Packaging delay_AA High 5%
High 54%
High 13% Low 95%
Customer order Low 46%
transmission Low 87%
Sales competition
error_AA Package material delay_AA Inspection booking delay_AA Import document risk_BB Increased credit
Yes 10% High 10% High 18% submission Import clearance notes_BB
delay_BB delay_CC High 87%
No 90% Low 90% Low 82%
Low 13% Yes 56%
High 70% High 39%
Order processing Inspection booking error_AA Inspection confirmation On-farm inspection delay_AA No 44%
delay_AA Low 30% Low 61% Market volatility_BB Sales price
delay_AA
Yes 2% High 13% slippage_BB
High 18% High 5% High 85%
No 98% Low 87% Holiday impact_BB
Low 95% Low 15% High 77%
Low 82%
Yes 100% Low 23%
Customer order Inspection notification No 0% Sales time
Profit slippage_BB
confirmation delay_AA conflict_BB
delay_AA High 2% Yes 6% High 79%
Trucking to
High 15% Low 98% No 94% Low 21%
warehouse delay_BB
Low 85%
Loading booking delay_AA High 10%
High 18% Low 90% Distribution delay_BB

Low 82% High 30%

Low 70%
Lack of integrated Product receiving Customer order
order management Loading booking error_AA Loading confirmation Loading and trucking delay_AA Delivery delay_AA Delivery delay_BB
delay_BB cancellation_BB
system_AA delay_AA High 20% High 29%
Yes 2% High 25% High 34% High 20%
No 98% High 16% Low 80% Low 75% Low 71%
Yes 0% Low 66% Order preparing Low 80%
No 100% Low 84% delay_BB
Destination port
Loading notification arrival delay_BB High 10%
delay_AA High 25% Product Increased
Low 90%
High 2% Low 75% overstock_BB storage cost_BB
Low 98% Yes 26% High 35%
No 74% Low 65%
Holiday impact_AA Freight booking delay_AA
Yes 0% High 18%
No 100% Low 82% Cash buyer Procurement
High repacking Increased product
involvement_BB competition risk_BB
wastage_BB wastage_BB
SLI error_AA Freight confirmation High 90% High 83%
delay_AA Yes 45% High 56%
Yes 2% Low 10% Low 17%
High 5% No 55% Low 44%
No 98%
Low 95%
Export document
submission delay_AA Freight space shortage_AA Shipping delay_AA Increased
Exchange rate Increased
Yes 5% Yes 30% High 27% variation_BB procurement
price_BB cost_BB
No 95% No 70% Low 73% High 55%
High 57% High 73%
Low 45%
Export regulation Low 43% Increased repacking Low 27%
variation_AA Export clearance delay_AA cost_BB
High 10% High 26% High 5%
Low 90% Low 74% Low 95%
1701
network

Chain-wide risk
Figure 3.

probabilities in
propagation map

GeNIe modeller
with prior marginal
modelling
based Bayesian
An Ontology-
IMDS P(A) P(A’) P(B) P(B’)
119,8 0.8 0.2 A B 0.9 0.1

1702
C

A B P(C)

Yes Yes 0.98

Figure 4. Yes No 0.56

Example of a
simple BN with No Yes 0.75
Boolean nodes No No 0.05

Sensitivity for QualityDegradationInReceiving_BB = High

Current value: 0.265752 Reachable range: [0.25357 .. 0.277935]
0.255 0.26 0.265 0.27 0.275

1: QualityDegradationInReceiving_BB = High QualityInconsistency_AA = High, QualityDegradationInShipping_BB = Low, QualityDegradationIn...

2: QualityInconsistency_AA = High QualityDegradationInHarvest_AA = High, QualityDegradationInPackhouse_AA = Low, QualityDegradationI...

3: QualityDegradationInHarvest_AA = High ExtremeWeather_AA = Yes

4: ExtremeWeather_AA = Yes

5: QualityDegradationInReceiving_BB = High QualityInconsistency_AA = Low, QualityDegradationInShipping_BB = Low, QualityDegradationIn...

6: QualityDegradationInTruckingToWarehouse_BB = High

7: QualityDegradationInReceiving_BB = High QualityInconsistency_AA = Low, QualityDegradationInShipping_BB = High, QualityDegradationIn...

8: QualityDegradationInShipping_BB = High
Figure 5.
Sensitivity analysis of 9: QualityDegradationInReceiving_BB = High QualityInconsistency_AA = Low, QualityDegradationInShipping_BB = Low, QualityDegradationIn...
“quality degradation
in receiving_BB”
10: QualityDegradationInReceiving_BB = High QualityInconsistency_AA = High, QualityDegradationInShipping_BB = Low, QualityDegradationI...
(being high state)

The results are shown through the tornado graph, where the influential risk event nodes on
“quality degradation in receiving _BB” (being high state) exposed by the downstream
integrated Chinese importer and online retailer (BB) were sorted in terms of the most to the
least sensitive. It shows that the probability of “quality degradation in receiving _BB” being
high can be changed from between 0.25357 and 0.277935 when the probabilities of all other
nodes were changed by 10 per cent. This indicates that this variable is not sensitive to
changes in other variables compared with the initial probability value in the high state
(0.265752). All other outputs of the sensitivity analysis were validated with the two case
participants to precisely reflect the risk influences they had assumed.
4.5 Data analysis An Ontology-
The impact of risk propagation was analysed in the GeNIe Modeller using sensitivity based Bayesian
analysis and what-if scenario analysis. Sensitivity analysis was used to create a critical path network
of influence in the risk network map for the backward-looking traceability of risk root
causes (Amundson et al., 2012), while what-if scenario analysis was undertaken to examine modelling
the impact specific risk events have on the key performance measures (Amundson et al.,
2012) and to predict forward-looking risk propagation from one risk event to another, 1703
eventually to the performance measures. Colour coding and line width were utilised to
indicate the impact of risk influence generated from the sensitivity analysis as shown
in Figures 6 and 7. It should be noted that the darker red it is or the thicker the line is, the
stronger influence is.
The critical path of risk influence to customer value risk within the upstream firm (AA)
is illustrated in Figure 6, which shows that different risks interact to result in
supply discontinuity, quality consistency and delivery delay. It was found from the
backward-looking traceability that the upstream firm’s (AA) supply discontinuity is
mainly caused by the inability to meet customer quality requirement, which has its root
cause in natural disasters. Quality inconsistency is the result of four key direct risk events
with roughly equal influence, while quality degradation in harvest is significantly

Natural disaster_AA Production quality Inability to meet customer quality

variability_AA requirements_AA
Yes 10%
No 90% High 15% High 13%
Low 85% Low 87%
Quarantine inspection failure_AA Supply
discontinuity_AA
High 1%
High 16%
Low 99%
Yield reduction_AA Low 84%
High 12% Product stockout_AA
Low 88% High 16%
Market supply Low 84%
shortage_AA Inability to secure supply_AA Supply shortage_AA
High 10% Quality degradation in
High 15% High 11% harvest_AA
Low 85% Low 89% Low 90%
High 18%
Low 82%
Cash buyer Production capacity
involvement_AA constraint_AA Quality degradation in
High 80% Yes 5% packhouse_AA
Low 20% No 95% High 1%
Low 99%
Extreme weather_AA Quality degradation in Quality
Yes 20% loading_AA Inconsistency_AA
No 80% High 1% High 20%
Low 99% Low 80%
Maturation uncertainty_AA Harvesting delay_AA Quality degradation in trucking to
High 30% High 55% port_AA
Low 70% Low 45% High 1%
Low 99%
Labor regulation Picker shortage_AA Grading delay_AA Stock unavailability_AA
change_AA High 54% High 12% High 28%
Yes 60% Low 46% Low 88% Low 72%
No 40%
Packer shortage_AA Packaging delay_AA
High 54%
Customer order High 13%
Low 46%
transmission Low 87%
error_AA Package material delay_AA Inspection booking delay_AA
Yes 10% High 10% High 18%
No 90% Low 90% Low 82%
Order processing Inspection booking error_AA Inspection confirmation On-farm inspection delay_AA
delay_AA delay_AA
Yes 2% High 13%
High 18% High 5% Low 87%
No 98%
Low 82% Low 95%
Customer order Inspection notification
confirmation delay_AA
delay_AA High 2%
High 15% Low 98%
Low 85%
Loading booking delay_AA
High 18%
Low 82%
Lack of integrated
order management Loading booking error_AA Loading confirmation Loading and trucking delay_AA Delivery delay_AA
system_AA Yes 2% delay_AA High 20% High 25%
No 98% High 16% Low 80% Low 75%
Yes 0%
No 100% Low 84%
Loading notification
delay_AA
High 2%
Low 98%
Holiday impact_AA Freight booking delay_AA
Yes 0% High 18%
No 100% Low 82%

Yes 2%
SLI error_AA Freight confirmation
delay_AA Figure 6.
High 5%
No 98%
Low 95% Critical path of risk
Export document

Yes 5%
submission delay_AA Freight space shortage_AA
Yes 30%
Shipping delay_AA
High 27%
influence to three key
No 95% No 70% Low 73% indicators of AA’s
Export regulation
variation_AA Export clearance delay_AA
High 26%
customer value risk
High 10%
Low 90% Low 74%
 119,8
IMDS

1704

Figure 7.

influence to the

BB’s online sales

Critical path of risk

customer value risk of

Production quality Inability to meet customer quality Unexpected import Entry CIQ failure_BB
Natural disaster_AA
variability_AA requirements_AA restriction_BB High 1%
Yes 10%
High 15% High 13% High 1% Low 99%
No 90%
Low 85% Low 87% Low 99%
Supply Customer value Reputation risk_BB
Supply Product stockout_BB discontinuity_BB risk_BB
Quarantine inspection failure_AA Inability to secure supply_BB
discontinuity_AA High 64% High 70%
High 42% High 67%
High 1% High 66%
High 16% Low 36% Low 33% Low 30%
Low 99% Low 34% Low 58%
Yield reduction_AA Low 84%
High 12% Product stockout_AA
Low 88% High 16%
Market supply Low 84%
shortage_AA Inability to secure supply_AA Supply shortage_AA Ordering quantity Inability to meet
High 10% Quality degradation in uncertainty_BB customer quality
High 15% High 11%
harvest_AA High 60% requirements_BB
Low 85% Low 89% Low 90%
Order fulfillment
High 18% Low 40% High 9%
error_BB
Low 82% Quality degradation in Low 91%
Cash buyer Production capacity shipping_BB Yes 5%
involvement_AA constraint_AA Quality degradation in No 95%
High 8%
High 80% Yes 5% packhouse_AA
Low 92% Quality degradation Customer churn... Loss of sales_BB
Low 20% No 95% High 1% in storage_BB High 85%
High 88%
Low 99% High 20% Low 15%
Low 12%
Low 80%
Extreme weather_AA Quality degradation in Quality Quality degradation in Quality
Yes 20% loading_AA Inconsistency_AA receiving_BB inconsistency_BB Claim dispute_BB
No 80% High 1% High 54%
High 20% High 27% High 28%
Low 99% Low 46%
Low 80% Low 73% Low 72%
Quality degradation
Maturation uncertainty_AA Harvesting delay_AA Quality degradation in trucking to Quality degradation in repacking_BB
High 30% High 55% port_AA in trucking to High 30%
Low 70% Low 45% High 1% warehouse_BB
Low 70% Reduced
Low 99% High 10% sales_BB
Labor regulation Picker shortage_AA Grading delay_AA Stock unavailability_AA Low 90% High 85%
Quality degradation Customer claim_BB
change_AA High 54% High 12% Low 15%
High 28% in distribution_BB High 51%
Yes 60% Low 46% Low 88% Low 72% Low 49%
Import regulation High 40%
No 40%
variation_BB Low 60%
Packer shortage_AA
Packaging delay_AA High 5%
High 54%
Customer order High 13% Low 95%
Low 46%
transmission Low 87%
Sales competition
error_AA Package material delay_AA Inspection booking delay_AA Import document risk_BB Increased credit
Yes 10% High 10% High 18% submission Import clearance notes_BB
delay_BB delay_CC High 87%
No 90% Low 90% Low 82%
Low 13% Yes 56%
High 70% High 39%
Order processing Inspection booking error_AA Inspection confirmation On-farm inspection delay_AA No 44%
Low 30% Low 61% Market volatility_BB Sales price
delay_AA delay_AA
Yes 2% High 13% slippage_BB
High 18% High 5% High 85%
No 98% Low 87% Holiday impact_BB High 77%
Low 82% Low 95% Low 15%
Yes 100% Low 23%
Customer order Inspection notification No 0% Sales time
Profit slippage_BB
confirmation delay_AA conflict_BB
delay_AA Yes 6% High 79%
High 2% Trucking to
High 15% No 94% Low 21%
Low 98% warehouse delay_BB
Low 85%
Loading booking delay_AA High 10%
High 18% Low 90% Distribution delay_BB
Low 82% High 30%
Low 70%
Lack of integrated Product receiving Customer order
order management Loading booking error_AA Loading confirmation Loading and trucking delay_AA Delivery delay_AA Delivery delay_BB
delay_BB cancellation_BB
system_AA Yes 2% delay_AA High 20% High 25% High 29%
High 34% High 20%
No 98% High 16% Low 80% Low 75% Low 71%
Yes 0% Low 66% Order preparing Low 80%
No 100% Low 84% delay_BB
Destination port
Loading notification arrival delay_BB
High 10%
delay_AA High 25% Low 90% Product Increased
High 2% Low 75% overstock_BB storage cost_BB
Low 98% Yes 26% High 35%
No 74% Low 65%
Holiday impact_AA Freight booking delay_AA
Yes 0% High 18%
No 100% Low 82% Cash buyer Procurement
High repacking Increased product
involvement_BB competition risk_BB
wastage_BB wastage_BB
SLI error_AA Freight confirmation High 90% High 83%
Yes 2% delay_AA Low 10% Yes 45% High 56%
Low 17%
No 98% High 5% No 55% Low 44%
Low 95%
Export document
submission delay_AA Freight space shortage_AA Shipping delay_AA Increased
Exchange rate
procurement Increased
Yes 5% Yes 30% High 27% variation_BB
price_BB cost_BB
No 95% No 70% Low 73% High 55%
High 73%
Low 45% High 57%
Export regulation Increased repacking Low 27%
Low 43%
variation_AA Export clearance delay_AA cost_BB
High 10% High 26% High 5%
Low 90% Low 74%
Low 95%
influenced by extreme weather. Delivery delay is also the result of four direct risk events, An Ontology-
while the main cause is shipping delay. based Bayesian
The critical path of risk influence to the “customer value risk” of the downstream firm’s network
(BB) online sales is illustrated in Figure 7. Supply discontinuity, quality inconsistency and
order fulfilment error have stronger influence on customer value risk than delivery delay. modelling
In terms of how AA’s customer value risk propagates to influence BB’s customer value risk,
the former’s “supply discontinuity” does not have significant impact on the latter’s “inability to 1705
secure supply” and “supply discontinuity”. This is because that BB could buy from other
importers and wholesalers to ensure supply continuity in the case of AA’s “supply
discontinuity”. In contrast, the “quality inconsistency” from AA is a significant cause of BB’s
“quality inconsistency in product receiving” and “quality inconsistency” to the end consumers,
although any break in transportation, shipping, storage and distribution could result into
quality degradation. Moreover, AA’s delivery delay is a main cause of BB’s “product receiving
delay”, which has stronger influence on BB’s “delivery delay” to the consumers.
A series of what-if scenario analyses were implemented to show the forward-looking
propagation from AA’s “customer value risk” to BB’s “customer value risk” and “business
performance risk”. Table V shows the impact of risk propagation from AA’s three key
indicators of “customer value risk” to BB’s “customer value risk” and “profit slippage”,
which were generated in the GeNie modeller by changing the state of supply discontinuity,
quality inconsistency and delivery delay for AA.
When changing the state of AA’s “supply discontinuity” from low to high, the
probability of BB’s “customer value risk” being high remains unchanged at 67 per cent,
while the probability of BB’s “profit slippage” being high increases from 78 to 80 per cent.
These results indicated that while AA’s high “supply discontinuity” does not directly
increase BB’s “customer value risk”, since BB could buy from back-up suppliers to reduce
the effect of AA’s “supply discontinuity”; however, it could lead to BB’s “profit slippage” as
a result of the higher back-up procurement costs. When changing the state of AA’s “quality
inconsistency” from low to high, the probability of BB’s “customer value risk” being high
increases significantly from 65 to 73 per cent, while the probability of BB’s “profit slippage”
being high increases from 78 to 81 per cent. These results indicated that AA’s “quality
inconsistency” can increase BB’s “customer value risk” and “business performance risk”.
Meanwhile, when changing the state of AA’s “delivery delay” from low to high, the
probability of BB’s “customer value risk” being high increases from 66 to 68 per cent, while
BB’s profit slippage remains unchanged at 79 per cent. These results revealed that AA’s
“delivery delay” could increase BB’s “customer value risk” but that it does not have a
significant influence on BB’s “profit slippage”.

5. Theoretical implications
The research developed an ontology-based BN for measuring the impact of dynamic risk
propagation within and between integrated firms in GFPSCs. This study could contribute

Customer value risk_BB Profit slippage_BB

High (P) (%) Low (P) (%) High (P) (%) Low (P) (%)

Supply discontinuity_AA Low 67 33 78 22

High 67 33 80 20 Table V.
Quality inconsistency_AA Low 65 35 78 22 Conditional
High 73 27 81 19 distributions of BB’s
Delivery delay_AA Low 66 34 79 21 customer value risk
High 68 32 79 21 and profit slippage
IMDS for a holistic and dynamic risk propagation analysis in a multi-tier GFPSC by integrating
119,8 performance and risk. It could also contribute to the SCRM literature by revealing that risks
originating in the upstream firm can propagate to increase the downstream firm’s customer
value risk and business performance risk. This indicates the need of collaborative risk
management from the downstream firm’s perspective to creating more values to customer.
Our research might be the first to generate powerful practical insights of risk
1706 propagation effect with an ontology-based BN in a GFPSC. Our study addressed the
challenges in using a knowledge-based approach to develop a BN model, particularly with a
large-scale model and integrated risk and performance for a holistic SCR propagation
assessment. The combination of modelling approaches to address the issue would advance
intelligent risk analysis.

6. Managerial implications
The risk ontology-based BN model can assist managers in tracking risk root causes and
predict risk propagation, thus leading to the following managerial implications.
The upstream firm’s “supply discontinuity” does not have a significant influence on the
downstream firm’s “customer value risk” but does result in increased “profit slippage” for
the downstream firm. As such, Australian table grape suppliers should make efforts to
ensure supply continuity which would keep customers satisfied (Yin and Ma, 2015). As
natural disaster is the root cause of supply discontinuity, Australian suppliers should
operate geographically diversified farms or work with third-party growers to mitigate
natural disasters.
Quality consistency must be kept from the farm to the final destination as product
demand within a market is reliant on the quality of the delivered product (Rodrigue, 2014).
Since extreme weather is the main cause of on-farm “quality inconsistency”, Australian
suppliers should take proactive mitigation measures for extreme weather conditions,
such as using wind machines to reduce cold injury and/or plastic covers to protect the
fruits on the vines. An adequate procedure, protocol and standardisation should also be
followed to ensure cold chain integrity as quality degradation could occur in any stage of
the supply chain.
Improving delivery responsiveness is also important to create customer value as it can
enhance customer satisfaction (Srinivasan et al., 2011). While the upstream firm’s delivery
delay does not have a noticeable influence on the downstream firm’s business performance
risk, it does result in increased “customer value risk” for the downstream firm. Given that
the main cause of the upstream firm’s delivery delay is “shipping delay”, which has its root
cause in freight-space shortage and export clearance delay, Australian suppliers should
work closely with freight-forwarders to secure shipping space, and submit correct customs
declaration documents on time so as to reduce export clearance delay.
Risks can propagate within and between firms, thus increasing customer value risk
and business performance risk across the chain. Individual risk mitigation efforts to
minimise customer value risk and business performance risk within either the upstream or
the downstream firm may not generate desired performance (Srivastava et al., 2015).
Chinese customers thus should make collaborative efforts with Australian suppliers to
mitigate risks for creating more values to customers. Such collaborative endeavours could
help improve product availability, consistent product quality and delivery performance
along the chain, which, in turn, will improve business performance of individual actors
and the whole chain.

7. Conclusions
A risk propagation ontology-based BN model was proposed for a holistic dynamic risk
propagation assessment and for addressing the challenges in building a knowledge-based
BN, especially with a large-scale model. The ACTGSC was used to illustrate the application An Ontology-
of the proposed model in investigating risk propagation in a GFPSC. The method developed based Bayesian
can be deployed for food and especially the fresh food supply chains. When borrowing the network
developed method, risk events may need to be adjusted or verified given the variation of
operational activities and process flow in different supply chains. modelling
While this study presented several important findings and contributed to dynamic SCR
modelling, it is subject to some limitations. For example, the results are product chain- 1707
specific in nature as the work was conducted in an ACTGSC. Meanwhile, loss of information
would occur as only two state values were considered for all the nodes presented in the risk
propagation ontology-based model. Furthermore, the developed model may not be feasible
for a large SC network as the method follows the exact process flow of the supply chain.
Another limitation relates to the lack of historical risk data, which, in this case, resulted in
the use of approximations and hypothesised data.
Potential research directions are presented as follows. First, future development would
allocate three or more risk states and preferably involve a machine learning algorithm to
elicit real-time risk probabilities from industry settings, which could yield more accurate
estimations of probabilities (Badurdeen et al., 2014). Second, the inclusion of risk mitigation
strategies is suggested to evaluate the effects of specific risk mitigation strategies on risk
influence within and between firms. Third, it could be useful to extend the model to address
risk propagation in complex supply chain networks, given that supply chains have evolved
into complex and dynamic networks that involve an increased degree of interconnectedness
between supply chain parties and SCRs (Urciuoli and Hintsa, 2016).

References
Ali, S.M., Moktadir, M.A., Kabir, G., Chakma, J., Rumi, M.J.U. and Islam, M.T. (2019), “Framework for
evaluating risks in food supply chain: implications in food wastage reduction”, Journal of
Cleaner Production, Vol. 228, pp. 786-800.
Amundson, J., Faulkner, W., Sukumara, S., Seay, J. and Badurdeen, F. (2012), “A Bayesian Network
based approach for risk modelling to aid in development of sustainable biomass supply chains”,
paper presented at the 22nd European Symposium on Computer Aided Process Engineering,
London, 17-20 June.
ATGA (2012), “Australian table grape industry strategic investment plan 2012-2017”, available at:
http://horticulture.com.au/wp-content/uploads/2015/12/HortInn-SIP-Table-Grape.pdf (accessed
24 June 2015).
Aven, T. (2016), “Risk assessment and risk management: review of recent advances on their
foundation”, European Journal of Operational Research, Vol. 253 No. 1, pp. 1-13.
Badurdeen, F., Shuaib, M., Wijekoon, K., Brown, A., Faulkner, W., Amundson Jawahir, I.S., Goldsby, T.J.,
Iyengar, D. and Boden, B. (2014), “Quantitative modeling and analysis of supply chain risks using
Bayesian theory”, Journal of Manufacturing Technology Management, Vol. 25 No. 5, pp. 631-654.
Baier, C. (2008), “A conceptual model of the alignment-performance link in PSM”, in Jahns, C. (Ed.), The
Alignment Performance Link in Purchasing and Supply Management, Wiesbaden, Gabler Verlag,
pp. 51-121.
Bryceson, K.P. and Smith, C. (2008), “Abstraction and modelling of agri-food chains as complex
decision-making systems”, European Association of Agricultural Economists 110th Seminar,
Innsbruck-Igls, 18-22 February.
Bucci, G., Sandrucci, V. and Vicario, E. (2011), “Ontologies and Bayesian networks in medical diagnosis”,
Proceedings of the 44th Hawaii International Conference on System Sciences, Kauai, HI, 4-7 January.
Chaudhuri, A., Srivastava, S.K., Srivastava, R.K. and Parveen, Z. (2016), “Risk propagation and its
impact on performance in food processing supply chain: a fuzzy interpretive structural
modelling based approach”, Journal of Modelling in Management, Vol. 11 No. 2, pp. 660-693.
IMDS Cowell, R.G., Verrall, R.J. and Yoon, Y. (2007), “Modelling operational risk with Bayesian Networks”,
119,8 Journal of Risk and Insurance, Vol. 74 No. 4, pp. 795-827.
Dani, S. and Deep, A. (2009), “Investigating risk management capability within UK food supply
chains”, paper presented at 2009 EUROMA Conference, Göteborg, 14-17 June.
Daultani, Y., Goswami, M., Vaidya, O.S. and Kumar, S. (2017), “Inclusive risk modeling for
manufacturing firms: a Bayesian network approach”, Journal of Intelligent Manufacturing,
1708 November, pp. 1-15.
Devitt, A., Danev, B. and Matusikova, K. (2006), “Constructing Bayesian networks automatically using
ontologies”, Second Workshop on Formal Ontologies Meets Industry (FOMI 2006), Trento,
14 December.
Diabat, A., Govindan, K. and Panicker, V.V. (2012), “Supply chain risk management and its
mitigation in a food industry”, International Journal of Production Research, Vol. 50 No. 11,
pp. 3039-3050.
Fenz, S., Tjoa, A.M. and Hudec, M. (2009), “Ontology-based generation of Bayesian Networks”,
paper presented at the International Conference on Complex, Intelligent and Software
Intensive Systems, Fukuoka, 16-19 March.
Garvey, M.D., Carnovale, S. and Yeniyurt, S. (2015), “An analytical framework for supply network risk
propagation: a Bayesian Network approach”, European Journal of Operational Research,
Vol. 243 No. 2, pp. 618-627.
Gaudenzi, B. and Borghesi, A. (2006), “Managing risks in the supply chain using the AHP method”,
The International Journal of Logistics Management, Vol. 17 No. 1, pp. 114-136.
GeNie User Manual (2017), Available at: http://genie-manual.geniesolutions.com.au/manual/Genie_
Manual.pdf (accessed 10 April 2017).
Ghadge, A., Dani, S. and Kalawsky, R. (2011), “Systems thinking for modelling risk propagation in
supply networks”, paper presented at the 2011 IEEE International Conference on Industrial
Engineering and Engineering Management, Singapore, 6-9 December.
Ghadge, A., Dani, S. and Kalawsky, R. (2012), “Supply chain risk management: present and future
scope”, The International Journal of Logistics Management, Vol. 23 No. 3, pp. 313-339.
Goswami, M. (2018), “Supply chain centric product line selection: a functional risk focused approach”,
International Journal of Production Research, Vol. 56 No. 20, pp. 6678-6700.
Hänninen, M. (2014), “Bayesian networks for maritime traffic accident prevention: Benefits and
challenges”, Accident Analysis & Prevention, Vol. 73, pp. 305-312.
Korb, K.B. and Nicholson, A.E. (2010), Bayesian Artificial Intelligence, CRC Press, Boca Raton, FL.
Lin, Y. and Zhou, L. (2011), “The impacts of product design changes on supply chain risk: a case
study”, International Journal of Physical Distribution & Logistics Management, Vol. 41 No. 2,
pp. 162-186.
Lockamy, A. III and McCormack, K. (2010), “Analysing risks in supply networks to facilitate
outsourcing decisions”, International Journal of Production Research, Vol. 48 No. 2,
pp. 593-611.
Lockamy, A. III and McCormack, K. (2012), “Modeling supplier risks using Bayesian networks”,
Industrial Management & Data Systems, Vol. 112 No. 2, pp. 313-333.
Manuj, I. and Mentzer, J.T. (2008), “Global supply chain risk management”, Journal of Business
Logistics, Vol. 29 No. 1, pp. 133-155.
Oehmen, J., Ziegenbein, A., Alard, R. and Schonsleben, P. (2009), “System-oriented supply chain risk
management”, Production Planning & Control, Vol. 20 No. 4, pp. 343-361.
Ojha, R., Ghadge, A., Tiwari, M.K. and Bititci, U.S. (2018), “Bayesian network modelling for supply
chain risk propagation”, International Journal of Production Research, Vol. 56 No. 17, pp. 1-25.
Ouabouch, L. and Paché, G. (2014), “Risk management in the supply chain: characterization and
empirical analysis”, Journal of Applied Business Research, Vol. 30 No. 2, pp. 329-340.
Pfohl, H.-C., Gallus, P. and Thomas, D. (2011), “Interpretive structural modelling of supply chain An Ontology-
risks”, International Journal of Physical Distribution & Logistics Management, Vol. 41 No. 9, based Bayesian
pp. 839-859.
network
Pollino, C. and Henderson, C. (2010), “Bayesian Networks: a guide for their application in natural
resource management and policy”, Technical Report No. 14, Landscape Logic, Canberra. modelling
Prakash, S., Soni, G., Rathore, A.P.S. and Singh, S. (2017), “Risk analysis and mitigation for perishable
food supply chain: a case of dairy industry”, Benchmarking: An International Journal, Vol. 24 1709
No. 1, pp. 2-23.
Qazi, A., Dickson, A., John Quigley, J. and Gaudenzi, B. (2018), “Supply chain risk network
management: a Bayesian belief network and expected utility based approach for managing
supply chain risks”, International Journal of Production Economics, Vol. 196, February,
pp. 24-42.
Qazi, A., Quigley, J., Dickson, A. and Ekici, Ş.Ö. (2017), “Exploring dependency based probabilistic
supply chain risk measures for prioritising interdependent risks and strategies”, European
Journal of Operational Research, Vol. 259 No. 1, pp. 189-204.
Rathore, R., Thakkar, J.J. and Jha, J.K. (2017), “A quantitative risk assessment methodology and
evaluation of food supply chain”, The International Journal of Logistics Management, Vol. 28
No. 4, pp. 1272-1293.
Renooij, S. (2001), “Probability elicitation for belief networks: issues to consider”, The Knowledge
Engineering Review, Vol. 16 No. 3, pp. 255-269.
Revilla, E. and Saenz, M.J. (2017), “The impact of risk management on the frequency of supply chain
disruptions: a configurational approach”, International Journal of Operations & Production
Management, Vol. 37 No. 5, pp. 557-576.
Rodrigue, J.-P. (2014), “Reefers in North American cold chain logistics: evidence from western Canadian
supply chains”, The Van Horne Institute, University of Calgary, Calgary.
Roth, A.V., Tsay, A.A., Pullman, M.E. and Gray, J.V. (2008), “Unraveling the food supply chain:
strategic insights from China and the 2007 recalls”, Journal of Supply Chain Management, Vol. 44
No. 1, pp. 22-39.
Scott, J. (2016), “An Insight to the Chinese Table Grape Industry Part 3”, available at: www.horticulture.
com.au/globalassets/laserfiche/assets/project-reports/tg14700/tg14700-final-report-710.pdf
(accessed 10 April 2017).
Sharma, S. and Routroy, S. (2016), “Modelling information risk in supply chain using Bayesian
Networks”, Journal of Enterprise Information Management, Vol. 29 No. 2, pp. 238-254.
Sharma, S.K. and Sharma, S. (2015), “Developing a Bayesian Network model for supply chain risk
assessment”, Supply Chain Forum: An International Journal, Vol. 16 No. 4, pp. 50-72.
Shin, K.S., Shin, Y.W., Kwon, J.-H. and Kang, S.-H. (2012), “Risk propagation based dynamic
transportation route finding mechanism”, Industrial Management & Data Systems, Vol. 112
No. 1, pp. 102-124.
Siew, P. and Lam-Chan, L. (2007), “Cold chain management for vegetable supply to Singapore market”,
Acta Horticulturae (ISHS), Vol. 804, pp. 577-584.
Sporleder, T.L. and Boland, M.A. (2011), “Exclusivity of agrifood supply chains: seven fundamental
economic characteristics”, International Food and Agribusiness Management Review, Vol. 14
No. 5, pp. 28-51.
Srinivasan, M., Mukherjee, D. and Gaur, A.S. (2011), “Buyer–supplier partnership quality and supply
chain performance: moderating role of risks, and environmental uncertainty”, European
Management Journal, Vol. 29 No. 4, pp. 260-271.
Srivastava, S.K., Chaudhuri, A. and Srivastava, R.K. (2015), “Propagation of risks and their impact on
performance in fresh food retail”, The International Journal of Logistics Management, Vol. 26
No. 3, pp. 568-602.
IMDS Świerczek, A. (2014), “The impact of supply chain integration on the ‘snowball effect’ in the
119,8 transmission of disruptions: an empirical evaluation of the model”, International Journal of
Production Economics, Vol. 157, pp. 89-104.
Świerczek, A. (2016), “The ‘snowball effect’ in the transmission of disruptions in supply chains: the role
of intensity and span of integration”, The International Journal of Logistics Management, Vol. 27
No. 3, pp. 1002-1038.
1710 Trkman, P. and McCormack, K. (2009), “Supply chain risk in turbulent environments – a conceptual
model for managing supply chain network risk”, International Journal of Production Economics,
Vol. 119 No. 2, pp. 247-258.
Urciuoli, L. and Hintsa, J. (2016), “Differences in security risk perceptions between logistics companies
and cargo owners”, The International Journal of Logistics Management, Vol. 27 No. 2,
pp. 418-437.
Van der Vorst, J.G.A.J. and Beulens, A.J.M. (2002), “Identifying sources of uncertainty to generate
supply chain redesign strategies”, International Journal of Physical and Distribution & Logistics
Management, Vol. 32 No. 6, pp. 409-430.
Wagner, S.M. and Bode, C. (2008), “An empirical examination of supply chain performance along
several dimensions of risk”, Journal of Business Logistics, Vol. 29 No. 1, pp. 307-325.
Wang, X., Li, D. and Shi, X. (2012), “A fuzzy model for aggregative food safety risk assessment in food
supply chains”, Production Planning & Control: The Management of Operations, Vol. 23 No. 5,
pp. 377-395.
Wieland, A. and Wallenburg, C.M. (2012), “Dealing with supply chain risks”, International Journal of
Physical Distribution & Logistics Management, Vol. 42 No. 10, pp. 887-905.
Wu, T., Blackhurst, J. and Ogrady, P. (2007), “Methodology for supply chain disruption analysis”,
International Journal of Production Research, Vol. 45 No. 7, pp. 1665-1682.
Yin, Z. and Ma, S. (2015), “Incentives to improve the service level in a random yield supply
chain: the role of bonus contracts”, European Journal of Operational Research, Vol. 244 No. 3,
pp. 778-791.
Zeng, F.G. and Rogers, J. (2011), “Assessing supply chain risk and energy efficiency simultaneously via
network DEA”, paper presented at the PICMET 11: Technology Management in the Energy
Smart World, Portland, OR, 31 July-4 August.
Zhu, Q., Krikke, H. and Caniëls, M.C. (2017), “Integrated supply chain risk management: a systematic
review”, The International Journal of Logistics Management, Vol. 28 No. 4, pp. 1123-1141.

About the authors

Shoufeng Cao received his Doctoral Degree from School of Agriculture and Food Science at the
University of Queensland (UQ). His dissertation focussed on breeding strategic competition through a
system-wide risk analysis and management in global fresh produce value chains. He currently works
in the Australian Institute for Business and Economics (AIBE) and liaises with industry peak bodies,
government agencies and Horticulture Innovation Australia to identify strategic development actions
and direct investments across three identified industries – mangoes, avocadoes and lychees, thus
delivering high valued horticultural products from Northern Australia to affluent Asian markets.
Shoufeng Cao is the corresponding author and can be contacted at: [email protected]
Kim Bryceson is Professor of Agribusiness at the University of Queensland (UQ) School of
Agriculture and Food Sciences. Her main research expertise is in supply/value chain analysis
across whole of agrifood industry chains, and in using agent-based modelling to construct supply
chain risk assessment and performance management scenarios. She has a background in industry
and government research specialising in the use of information technology in agricultural
production and agribusiness – particularly in relation to the value adding potential for electronic
technologies across agri-industry chains in production, electronic supply chain, food traceability
systems and cold chain logistics. She is also Director of the Agricultural Remote Sensing Laboratory
at UQ’s Gatton Campus, which is involved in the collection of real-time streaming Big Data
through an Internet of Things infrastructure for environmental monitoring, agricultural research An Ontology-
and teaching. based Bayesian
Damian Hine is Associate Professor in Innovation and Commercialisation at the University of
Queensland Business School. His work focusses on research commercialisation and technology network
management. As an evolutionary economist, his main areas of interest lie in the crossroads of modelling
innovation and strategy, in the field of dynamic capabilities. He is Author of two books and multiple
articles in some of the world’s top innovation and science journals. He has also written papers for
ASX200 firms, State and Federal agencies, the OECD and UNESCO. He has led projects on innovation 1711
and commercialisation for governments in Chile, Brazil, Panama, Vietnam, Fiji and France. He is
currently leading a component of a World Bank project in Indonesia and the Philippines on business
development and innovation.

For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: [email protected]