Conceptualization of Social Systems:
Actors First
Camilo Olaya*
Juliana Gomez-Quintero
Department of Industrial Engineering
Universidad de los Andes
Bogota, Colombia
* Corresponding author: colaya@ uniandes.edu.co
Abstract
Social systems are formed by actors whose actions and decisions form a complex
structure of continuous inter-actions that shape the performance of those systems.
System dynamics models help to design and redesign new configurations of a
system in order to improve its performance, which for social systems requires
thus intervening and modifying actions and decision-making processes. There are
several ways for building system dynamics models. Being a heuristic, a powerful
advantage of system dynamics is that it does not come with a “recipe”; instead,
each modeler, according to purposes, particular problem, interests, etc. can build
a model in different ways. In particular, the conceptualization stages are critical
since they form the base for imagining the system and formulating models which
later on serve as tools for developing understanding and taking actions to
improve the system. However, it is not easy to find explicit guidelines that
consider a full and systematic analysis of actors (in terms of their actions and
decisions) as a source for conceptualizing the social system that dynamically
“produces” the problem to be modeled. This paper presents a methodological
guideline for conceptualizing models of social systems intended to address the
actor-driven nature of such systems. The emphasis on decision-making in social
systems serves as a heuristic that guides the model building process and leads to a
shift from “variables” to “decisions rules” and “actions”. Such heuristic favors
the creation of policies and interventions that rest on the power of actors to
change their own system.
Keywords: conceptualization, methodology, model building, social systems,
actors, agency.
1. Introduction
System dynamicists are frequently interested in modeling social systems (e.g. a firm, a corporation,
an university, a football team, etc.), that is, purposeful systems formed by decision-making actors
that act according to their own interests and goals. That is a very special characteristic of those
systems: both their elements (actors, institutions, organizational units, individuals, etc.) and the
whole system have the ability to make choices according to goals, interests and purposes, as
opposed for instance to deterministic systems (e.g. a clock) in which neither the whole system nor
the parts are purposeful, or animate systems (e.g. a person) whose parts do not display choice
(Ackoff, 2001; Ackoff & Gharajedaghi, 1996). Hence, the interactions in a social system
correspond to those purposeful decision making processes that convert information into action—
such is the definition of “decision making” of Forrester (1961)—through the exchange of resources,
materials, information, meanings, communications, etc. These systems produce problems that
interested parties intend to resolve. Consequently, the redesign and improvement of social systems
require changing the actions of their own actors, new arrangements, new ways of organizing and
doing decision processes. Such transformations can be boosted through the construction and use of
models. System dynamics (SD) allows to build explicit models of such social systems in order to
change mental models and enhance our understanding about the dynamics of a system in order to
improve its performance and accomplish desired goals. Building models of social systems requires
then explicit knowledge about the dynamics of the social interactions of the modeled system
(Vriens & Achterbergh, 2006).
The modeling process creates knowledge that relates the structure of systems with their
performance in order to improve it in desired ways. How to build these models? The heuristic mode
in which SD has traditionally approached this question has opened a variety of options and
methods, which is one of its distinct advantages since model building can be oriented according to
the purposes, goals and interests of modelers. Those methods are explicitly available and described
in the rich literature of SD and help to guide (both beginners and experts) the modeling process,
which expresses the breadth of the field. However, in spite of such available sources, “sometimes
making sense of all of the material is difficult, especially for novices in the field” (Martinez-
Moyano & Richardson, 2013, p. 105). Nevertheless, there is always a shortcut: system dynamics
models seem easy to build. And perhaps it is true. Apparently the point is to find the relevant
“variables” that describe a problem (keeping in mind a special treatment for stocks and their flows),
identify auxiliary variables and parameters, connect these variables in appropriate ways, run some
tests and develop an understanding that relates the model structure with its behavior. But there are
risks involved if we take lightly the apparent easiness with which SD models can be built. For
introducing his “Guidelines for Model Conceptualization” Randers (1980) stressed that “few
beginners resist the temptation to follow what is felt to be the ‘natural’ approach to modeling: a
headlong rush into description of the real world in the form of flow diagrams” (p. 118). Modern SD
software packages paradoxically may risk matters further since indeed it is uncomplicated to
connect in a computer screen diverse variables with causal links and build any model with them,
which echoes the old criticism of Maloney: “With a mouse, and just enough self-restraint not to try
connecting everything to everything... ba-da-bing, ba-da-boom, you’ve got a model” (p. 305). Such
simplicity may lead beginners to become overconfident and oversell their skills (Meadows, 1980),
let alone the possibility of building poor models.
The mentioned risks may become especially problematic if the modeler is addressing a social
system since it is easy to exclude its purposeful and decisional nature. For instance, let us say that
the model of Figure 1 is intended for modeling the problem of decreasing revenue in a particular
organization (a social system). The model displays a first small conceptualization of such a system
with 4 variables and 5 feedback loops (though there are more) that may help to understand the
dynamics of revenue. However, it is not easy to appreciate in the model decision making processes
attached to concrete actors, either through the variables or through the feedback loops. Those
variables are abstract concepts (bureaucratization, organizational pressure, revenue, personnel) that
apparently have causal effects on others. Let us suppose that we suspect (or indeed later find) that a
specific loop (say the one labeled as “Can’‘t take it”) or some specific variables are critical for
changing the behavior of revenue. How to actually change such causal effects in the real system?
How to implement changes that affect those variables? How to modify the “strength” of the loop?
Exactly how the system “produces” the behavior of revenue? Through which actions? If we want to
change that behavior, who can/should do what and how? Perhaps in a later modeling stage these
questions could be answered. However, the explicit orientation of a model of a social system in
terms of actors and decisions starting from the conceptualization stages may facilitate a later model
formulation that accounts for the very structure of a modeled social system: the decision processes
carried out by purposeful actors.
ad
r ressure
i.
-) 7
Self-defeating success ~ ~
+f A) =
Se wa
Economic boost / Morestressed... less productive
7
Revenue
Figure 1. A model of a social system missing explicit decision-making
Unfortunately, it is not easy to find model building guidelines that stress such decisional character
of social systems at the early conceptualization stages. Usually the language and the conversation
along a modeling processes favor terms like “sectors”, “variables”, “factors”, “equations”, etc. over
“actors”, “actions”, “decision”, etc. This bias can be seen indeed in the way in which usually system
dynamicists express how they conceive the modeling process. In the revision of best modeling
practices that Martinez-Moyano and Richardson (2013) made among top SD experts, their results
about which core activities are regarded as highly important for system conceptualization and
model formulation do not mention any suggestion to explicitly address actors or decision-making
processes within a model building process. Although that study is intended to be wide enough to
serve as a guide for building diverse types of models, using different modeling tools and for
different purposes, overlooking actors and their actions entails the risk of ending up with a model
full of variables that may hide possibilities through which the modeled system can be in fact
improved: the action of decision-makers. A further and major risk is to end up in the error
underlined by Ackoff (2001): mistaking a social system as a deterministic system (as a sort of
machine whose main challenge is reduced to find the “correct” variables to “push”) by building an
abstract model disarticulated from such agency that distorts the defining characteristic of social
systems: the purposefulness of their parts.
There are different ways to organize and “run” a modeling process. Barlas (1996) suggests 6 typical
major steps that match a wide spectrum of how the modeling process is described in SD reference
works (Table 1); these steps are usually understood in terms of iterative and cyclic regimes, see
(Martinez-Moyano & Richardson, 2013). Although such categorization varies across modelers,
tools and purposes, we will use it for placing our proposal in context. We will focus on “model
4
conceptualization”, perhaps “the most important and least understood of all modeling activities”
(Sterman, 1986, p. 76). For now we can take the classic work of Randers (1980) as a reference
point for demarcating model conceptualization: once a problem and the questions to be addressed
have been identified, the conceptualization seeks to picture basic mechanisms (in terms of feedback
loops) as powerful organizing concepts for explaining the problem to address, “to arrive at a rough
conceptual model capable of addressing a relevant problem” (p. 131).
Typical stages in model building
1. Problem identification
2. Model conceptualization (construction of a conceptual model)
3. Model formulation (construction of a formal model)
4. Model analysis and validation
5. Policy analysis and design
6. Implementation
Table 1. Typical stages in model building, from (Barlas, 1996).
We will focus on model conceptualization.
Interestingly, the study of Martinez-Moyano & Richardson found that “although what model
developers do is important, when they do it also seems to be critical” (p. 119). In this paper we
suggest that identifying explicitly actors (and their relevant decisions) in the early stages of the
modeling process—as the base for conceptualization—helps to build models of social systems, that
is, models whose structure and variables capture the purposeful, action-oriented and decisional
nature of this type of systems. We introduce a methodological guideline for conceptualizing SD
models built on the assumption that a problem to be modeled is driven by a social system
constructed through actions of actors. This consideration develops a heuristic that helps to guide
and improve the modeling process and favors to achieve relevant and action-oriented results
through the design of policies and interventions that rest on the power of actors to change their own
social system.
2. System C ptualization in Model Building
How to build an SD model? This is always a recurring question, especially for philosophers and
beginner modelers. Model building is more an art than a fixed technique. It has to do more with
creativity than with algorithmic methods. Perhaps only experience and iteration strengthen
modeling skills. However, practical guidelines have been available from the first years of SD. For
instance, Forrester in Industrial Dynamics (1961) devoted the fifth chapter to setting “principles for
formulating system dynamics models”, “principles” in the sense that there is no “magic recipe” but
only useful heuristics that might help. Some of these principles are:
e From his engineering roots he knew that models are built according to a purpose.
“Questions to be answered control the content of a model” (p. 60).
e Model building should not be restricted to include only those aspects that can be solved
analytically but also “all the facets that we should consider essential to a verbal description
of the phenomena under study” (p. 60).
e The model should account for the information-feedback structure that “gives rise to so
much of the interesting behavior” (p. 61). This aspect means for him to consider not only
closed-loops but also time delays and relevant accumulations (both physical and
information reservoirs).
e The model should correspond to real-system variables, measured in the same units.
e He suggested to identify six distinct networks that “represent the grossly different types of
variables that will be encountered” (p. 70): orders, materials, personnel, money, capital
equipment, and information channels (this latter network interconnects the other ones). He
recognized that this classification is arbitrary but it helps as a guide to identify variables.
These first guidelines point at “variables” as organizing concepts for building a model, e.g. the
importance of identifying accumulation “variables”, the correspondence to real-system “variables”,
the six networks for identifying “variables”, etc. Forrester’s work always included actions of actors
through “policies” (decision rules) and indeed he always stressed that policies (in the sense of
“decision rules”) define the logic for equations; however, in his initial methodological principles
there is no explicit suggestion to think in terms of actors and decision-making processes as a guide
for steering the modeling process. In fact an emphasis on decision-rules may lead to suppose that all
relevant actors’ actions become implicitly included in a model.
One of the first clear-cut guides for SD model building was proposed by Randers (1980) who
divides the process of modeling in four iterative stages:
1. Conceptualization: Definition of problem and question to be addressed, time horizon,
organizing concepts, model boundary and verbal description of feedback loops that may
cause the reference mode.
2. Formulation: Postulation of detailed structure, stocks, flows, parameters.
3. Testing.
4. Implementation.
In particular for the conceptualization stage he suggests to start with a recognized step by any
system dynamicist: the reference mode which accounts for the development of the situation of
interest through time. As an example let us imagine a reference mode (Figure 2) for the model of
Figure 1 regarding “Revenue” as the variable of interest.
Revenue
time
Figure 2. A reference mode for Revenue.
Once we have the reference mode, how to proceed? Randers suggests that “having specified the
reference mode, the modeler should identify the fund real-world hani assumed to
produce the reference mode. He should select and describe the smallest set of feedback loops
considered sufficient to generate the reference mode” (p. 131, emphasis added). Although this
heuristic is important since it orients the modeling process to a feedback-based conceptualization,
there are no further guidelines for identifying such “mechanisms”. How to identify them? Randers
does not provide hints or possible guiding questions. Regarding the next stage (model formulation),
Randers suggests to identify first the system stocks: “the levels describe a set of independent
variables, together sufficient to describe the state of the system” (p. 134). Next, the modeler should
identify the causal influences on the rates that affect the stocks, keeping in mind that “these causal
influences should embrace the basic mechanisms that the model is supposed to include” (p. 134).
Afterwards, “the modeler should choose numerical values for table functions and time constants”
(p. 134), which allows to continue with the following stages: model testing and implementation.
We want to stress that up to this point there have been no mention of actors or decisions. Indeed the
guidelines of Randers do not mention the possibility of considering actors or decisions for
conceptualizing a system. Instead, he refers to variables: stocks, flows and time constants. But
already the modeler may have a running model, whose conceptualizations rests on feedback
mechanisms, stocks, flows and parameters that do not warrant the modeling of interactions and
variables that reflect decision making. Such an approach that overlooks actors and decision through
the modeling process is fairly common in most guidelines and proposed methods.
The guidelines of Sterman (2000) are perhaps the ones that make the strongest emphasis on action-
oriented considerations for changing a social system. He emphasizes the orientation of the modeling
process to solve a problem (“not only to gain insight”, p. 83), that is, he underlines the interest in
devising actions for improving the performance of a system, “...taking action in the real world...
The purpose is to help the clients solve their problem” (p. 85). He also underlines that the modeling
process is iterative although also clarifies that there is no cookbook recipe, there is “no procedure
7
you can follow to guarantee a useful model” (p. 87). Nevertheless, he identifies 5 activities
(problem articulation, dynamic hypothesis, formulation, testing, policy formulation and evaluation)
that “all successful modelers follow... [and that take place] in context with the ongoing activities of
the people in the system” (p. 87)—see Figure 3. Such a context gives a learning and action-oriented
direction to the modeling process:
Strategies, structures and decision mules used in the real world can be represented and
tested in the virtual world of the model. The experiments and tests conducted in the
model feed back to alter our mental models and lead to the design of new strategies, new
structures, and new decision rules. These new policies are then implemented in the real
world, and feedback about their effects leads to new insights and further improvements in
both our formal and mental models. Modeling is not a one-shot activity that yields The
Answer, but an ongoing process of continual cycling between the virtual world of the
model and the real world of action (p. 88).
Re
al
ee Would
Decisions
(Organizational Information
Experiments) 1. Problem Articulation feedback
te (Boundary Selection) ri
5. Policy Formulation 2. Dynamic
& Evaluation Hypothesis
4, Testing 3. Formulation
Strategy, Structure, es \
Decision Rules Mental Models of
Real World
Figure 3. Sterman’s placement of the iterative modeling process in the
action-context of the modeled social system (Sterman 2000)
After clarifying the problem and the purpose of the model (problem articulation, reference modes,
time horizon), the guidelines of Sterman suggest as a second step to formulate a dynamic
hypothesis, which corresponds to the conceptualization of the system to be modeled. He
8
recommends exploring the current theories about the problematic behavior as a first source for
generating an initial hypothesis keeping in mind an endogenous view based on feedback structures
as the explanation of the problematic dynamics. For mapping an initial model the advice is to use
diverse tools and data including “key variables”. Sterman is explicit on the relevance of decision
tules as central components: “An endogenous theory generates the dynamics of a system through
the interaction of the variables and agents represented in the model. By specifying how the system
is structured and the rules of interaction (the decision rules in the system), you can explore the
patterns of behavior created by those rules and that structure and explore how the behavior might
change if you alter the structure and rules” (p. 95). This is perhaps the most explicit modeling guide
regarding the relevance that the identification and formulation of decision rules has on the modeling
process. And yet, it still does not point at explicitly addressing the actors of the modeled system for
helping to deliver those decision rules in the model. However, Sterman indeed stresses a decision-
rules orientation and hence does suggest corresponding diverse mapping tools for conceptualizing
the dynamic hypothesis. From his suggestions we will highlight what are called “policy structure
diagrams”.
Policy structure diagrams were proposed by Morecroft (1982) who noticed that although causal
loop diagrams are useful for policy analysis and for representing the feedback structure of a system,
they are weak tools for conceptualizing a system for various reasons: there is little correspondence
between mental models (more oriented to component parts) and loop structure; there is also the
widely recognized limitation that causal loop diagrams do not distinguish explicitly physical flows
and accumulations as distinct from information structures; and especially for our goals here,
according to him causal loop diagrams do not explicitly represent decision-making processes.
Morecroft is particularly explicit regarding this limitation:
It is not possible to look at a causal loop diagram and deduce where decision are being
made, how responsibilities are distributed, and what information different decision
makers deem important in their part of the system. By ignoring the existence of decision-
making processes, causal loop diagrams overlook real features of organizations that can
lend precision to the generation of system linkages. Decision making processes, or
policies, are nodes of the information network. They are the points in the organization.
They are the points in the organization that information is collected, processed, and
dispersed. Recognizing their role as information processor, we can be discriminating
about the quantity and content of information that is likely to be used at any policy point.
The causal loop diagram fails to make use of decision-making features of the real system
that are valuable in conceptualization (p. 22).
The model in Figure 1 shows these limitations indicated by Morecroft. Because of these
shortcomings, he introduces two tools for conceptualization that allow to portray the structure of a
system in terms of “real decision-making processes and in organizational units that are compatible
with mental models” (p. 23): the subsystem diagram and the policy structure diagram. The latter
one is of particular interest here: it is intended to show a simplified structure of the information
network “in terms of major policies, or decision functions such as inventory, control, pricing, or
manpower planning” (p. 24) and the information network that support those policies without
excessive detail. Morecroft suggests two steps for constructing these policy structure diagrams:
1. Drawing policy symbols to delineate decision-making responsibilities. He underlines that
this step is crucial and does not occur in causal-loop diagraming since the latter is based on
intuition and brainstorming without a systematic discipline. Instead, policy structure
diagrams lead to recognize explicitly decision making points.
2. Building the information network using policies as nodes for information links. These links
generate a network of communications and they can be identified by considering, for each
policy point, this type of questions (p. 24):
- What information is available at a particular point of decision making in the
system?
- What information would be relevant to the decision-making processes in question?
- With which parts of the overall system is the area containing this policy in closest
communication?
- What information is not available at this point in the system, and why not?
- How much information is entering the policy, and is it possible to collect and
meaningfully process such information?
- What is the quality of information available at this point in the system, and what
distortions are likely to arise?
Following these steps Morecroft underlines that a “feedback structure is then created from the
orderly process of piecing together multiple decision functions, rather than emerging from the more
tenuous and ad hoc methods of postulating causal links independent of the underlying decision-
making process” (p. 24).
The emphases of Sterman and Morecroft on decision-making are important for addressing social
systems and goes in line with our paper. How to identify decision-making points? How to cover all
possible actions and decision that may be relevant for a particular problem? Is it possible to have a
systematic procedure that helps modelers to conceptualize the social system so that pertinent
decision processes are included in the model? There could be an additional step in the
conceptualization process that may help to answer these questions and advance and complement the
proposals of Sterman and Morecroft: actor analysis. The next section delineates a heuristic for
integrating a systematic analysis of actors for guiding the conceptualization stage that acknowledges
decision-making as the central feature of action for improving social systems.
3. Identifying Actions and Actors for C onceptualizing Social Systems
Here we introduce a practical tool that has been useful for us when guiding the conceptualization
stages in SD modeling based on analyzing actors and their actions in the social system that produces
the problem to be modeled.
As system dynamicists know, the importance of having a clear problem and a clear modeling
purpose is to have boundary criteria. This is not trivial and we have to be explicit in recognizing
that a problematic situation has different perspectives and angles (there are already good options for
dealing with this matter from a SD perspective, e.g. (Lane & Oliva, 1998). This is why the first
stages of modeling that deal with problem identification and setting a purpose are recognized as
crucial for having successful results (Martinez-Moyano & Richardson, 2013). Assuming that the
first stages of problem identification and model purpose have been addressed (at least on a first
initial iteration) then we can recognize that there is a social system “producing” that problem and
the goal is to build a model or various models that capture the dynamics in which such system
indeed “produces” the problem. That social system to be modeled is not necessarily easy to identify.
It does not have to match the organization in which the problem develops, it does not have to
correspond to a clearly identifiable organizational unit, etc. Indeed we should be aware that we will
model a problem, not a system as such (Sterman, 2000). More precisely, our goal is to model the
social system that produces the problem. That system is indeed a “process”: it has a temporal
dimension and a temporal unity, exists through time, it is a bundle of activities, see e.g.(Leclerc,
1953; Seibt, 2013). That process-system is created, contingently many times, through the actions of
diverse actors, perhaps from different organizations or social groups, from different subsystems, etc.
that end up together, often non-intentionally, producing together the problem through their
decisions. These problem-generating social systems are formed in action, rarely they will match a
formal, recognizable object-system. For each problem there is a distinct, dynamic, emergent social
system “producing” it. Two different problems that seemingly take place in the same organization
will have certainly two different systems producing them. The heuristic that we introduce helps to
identify such dynamic systems; hence, these guidelines can be particularly helpful when the
modeler faces a messy problem to which is not easy to associate a single, concrete, formal social
system (e.g. think of public problems, region/country level problems, etc.). We will conceptualize
the system by examining actors, actions and inter-actions (system structure) that end up generating
the problem to be tackled.
3.1. Identification of Roles and Actors
We are interested in a particular social system: the one that dynamically produces a problem. Who
is an actor in a social system? This is not necessarily an easy question to answer. The classic
definition of “stakeholder” of Freeman (2010) seems a good starting place given its impact and the
11
development that has taken place on stakeholder theory; a stakeholder is “any group or individual
who can affect, or is affected by, the achievement of a corporation’s purpose. Stakeholders include
employees, customers, suppliers, stockholders, banks, environmentalists, government and other
groups who can help or hurt the corporation” (p. vi). That definition is intended to inform corporate
management and defines the stakeholders of the firm. However, we are interested in the
stakeholders of a problem (our modeling boundary criterion). Since we are interested in the way in
which actually a social system produces such a problem, we are concemed with agency and actors
and will prefer this latter term over “stakeholder”. The term “actor” denotes an agent “who act”—
from Latin actor "an agent or doer" (Harper, 2016)—in this case as related to the problem. Not
necessarily everyone that has a stake in something, does something about it. We can thus rephrase
the definition of stakeholder and define an actor as:
Any group or individual who can affect, or is affected by, the problem to be modeled.
This definition supposes that an actor affects the problem through his own actions. An actor takes
resources and information and take actions that impact the social system in which s/he participates.
The definition also supposes that those affected by the problem re-act and do something about it.
Notice that an actor can play one or more different roles in a problem (e.g. a company can be a
competitor for firm A but a partner for firm B.). Such roles then are defined by what the actor does
and by the perspective from which that action is assessed. From this point, there are several
possibilities for attempting a systematic classification of actors. Espejo and Reyes (2011) propose a
methodology for diagnosing systems that is useful for us. We adapted it to identify different types
of roles that can be played by diverse actors (as related to the problem):
e Drivers: those that “drive” the problem; their actions have a direct impact on the problem.
Given defined measurements or variables for the problem (e.g. reference modes), which
actions can be identified whose outputs impact directly those variables? Who are the agents
that execute those actions? These can be difficult questions to answer, especially if the
reference modes are abstract and aggregated (Saeed, 1992).
e Suppliers: those providing resources and relevant information for the actions of the drivers.
e Affected: those who are directly affected by the problem and that can take action for
counteracting it.
¢ Owners: those who have an overview of the problem and have the responsibility to solve it.
(Indeed the “owners” of the problem). They can take action in different ways.
e Interveners: those that belong to the context or the environment but that can provide at any
time opportunities or threats for improving or worsening the situation to be solved. These
can be regulators, partners, competitors or collaborators of any other actor. Such
opportunities or threats can be actualized through direct action or by providing resources or
information to other actors.
12
Thus, a first step is to identify the actors (individuals, groups of persons, organizational units, whole
organizations, groups of organizations, etc.) that play the previous roles. Naturally, as it was
mentioned, within a modeling process these steps are also iterative. An actor can play simultaneous
roles and also a role can be played by a group of actors; they can be mapped in a table for instance
(See Table 2).
Roles Actors
Drivers -
Suppliers -
Affected -
Owners -
Interveners | -
Table 2. Mapping actors according to the roles they play.
There are different options for completing the table (and the next one that follows), many of them
related to qualitative modeling, e.g. workshops, brainstorming sessions, scripts, focal groups, etc.
Methods from Group Model Building and Community-Based System Dynamics can be particularly
helpful (Andersen & Richardson, 1997; Hovmand, 2014; Vennix, 1999; Vennix, 1996).
3.2. Interests and goals
The next step is to establish the interests and goals of the identified actors as related to the problem.
“Interest” refers to concern or attention that an actor has on the problem to the point that he is
willing to act to satisfy those interests. A “goal” is a concrete aspiration toward which an actor is
willing to act in order to achieve it (either concrete targets or informal and abstract ambitions). This
step is helpful also for the later formulation of decision rules since it helps to identify motivations
and the reasoning behind actions for each actor. Possible gaps between desired states and current
states can be later formulated, etc. Various questions can guide the identification of interests and
goals for each actor:
e Drivers: Which interests does the actor have that prompt him/it to “drive the problem”?
Which goals does he/it pursue?
e Suppliers: Why does the actor supply specific resources or information? What is the
purpose for doing it? Which goals does he seek to accomplish by providing those resources
or information?
e Affected: Which interests (of those actors that are affected) are impacted by the problem?
Which goals are affected by the problem?
e Owners: What are the goals that an owner would want to achieve by solving the problem?
Why is s/he interested in solving it?
e Interveners: What interests may the intervener have on the problem? Which goals could he
attain by improving or worsening the situation?
Table 3 includes interests and goals in the actors and roles schema.
Roles Actors Interests and goals
Actor 1 :
Drivers AetwER :
Suppliers
Affected
Owners
Interveners
Table 3. Identifying interests and goals.
3.3. Actions
Actors seek to defend their interests and to attain goals. They take action though diverse decision-
making processes. The objective of this step is to map the relevant decision processes (as related to
the problem) attached to those interest and goals. These decisions can be mapped identifying inputs
and outputs for such actions. Both inputs and outputs can be tangible (e.g. people, things, physical
resources, raw material, widgets, etc.) or intangible as e.g. information about something. Naturally
there can be more than one action associated to a particular interest or goal. Figure 4 shows a way
to map actions in terms of one or more inputs to produce the output of the action.
Input 1
i Action 1
Tnput 2 Output 1
Input n
Figure 4. Actions take one or more inputs and produce an output
At this point we find particularly useful the “policy structure diagrams” of Morecroft previously
mentioned and can be used as a tool for mapping the identified actions. Notice also that as we
mentioned before, an action oriented model brings a dynamic view of a problem since actions occur
through time as part of cycles of action > world > information > reaction. (Figure 5).
i
i 1
f Input 2
: tee Pa 42) \ Cl pthien
i 2 :
i y
i A
Figure 5. Actions entail a dynamic view of an actor
3.4, Feedback, model and formulation.
A conceptualization process based on actions and decisions warrants a feedback view since actors
take information from the world and use it to produce actions that seek to change it. With all the
actions mapped then feedback loops can be identified since all actors either act on the problem
and/or are affected by it. In addition, several actors form subnets of information and resources
exchange—see Figure 5. This is a benefit of this heuristic given that it can be hard to conceptualize
or elicit feedback structures, especially among novices and in group model building processes
(Andersen & Richardson, 1997; Vennix, 1999). These guidelines provide a systematic way to arrive
to diverse possibilities of feedback loops that can inform the modeling process. An iterative and
fruitful process of filtering and discussing the relevant feedback mechanisms should arrive to a first
agreed set of feedback loops that are candidates for explaining the problem that the social system
produces. Further iterations within the larger modeling process (Table 1) should refine that set.
'
i Actor 1
* Input 1 Actor 2!
Input 2
Figure 5. Feedback structures created in actionin a social system
From this point the modelers can retake the question addressed by Randers (mentioned earlier) and
others: which feedback loops can be particularly relevant for the problem? And the modeling
process can go to the next stages, either building a causal loop diagram, a stock-and-flow diagram
etc. and proceeding to formulate a first version of a simulating model if appropriate. Diverse
possibilities for formulating the decision rules are available depending on further field work or
modeling assumptions regarding the rationality and the way in which the involved actors are
assumed to decide, see e.g. (GroBler et al., 2004; Sterman, 2000; Sterman, 1988). In a future work
we will develop a full applied example.
4. Outlook
We introduced a methodological guideline for the conceptualization stage in model building of
social systems. The identification of actors provides a script that helps to conceptualize a first model
that warrants it to be anchored on the networks of actions and decisions that actors take and that end
up producing the problem for which the model is built.
The identification of actors serves as a heuristic for demarcating model boundary in terms of those.
Which actors drive, influence or are affected by the problem? The heuristic can be also extended for
policy analysis and design within a modeling process (Table 1). New actions, new actors, new roles
can be conceived which may mean to change or create new feedback loops. Which actions should
this or that actor take? How? Who has the power to transform the system? Who can play new roles?
Which actors should be included? Who should be the suppliers? Who should be the owners? Who
should be the interveners? Why? This approach helps to identify normative implications for an
ethical modeling practice (Ulrich, 1987). Moreover, there can be stakeholders that are unable to act
on the problem (that is, affected stakeholders that do not have the possibility of doing something
about it). The inclusion of those groups as possible new actors opens the doors for modifying
boundaries and creating new systems.
As a final word, we do not present a “recipe”. It is just a heuristic. It can be helpful for some
modelers for some problems and not for others. It can be helpful in particular situations and not in
others. We believe that it can be particularly useful for novice modelers that face the challenge of
conceptualizing systems though modeling. It can be also helpful as the base for group model
building scripts (Andersen & Richardson, 1997). The presented schema serves as an organizing
device for thinking in terms of actors and in this way helps to avoid the risk that Ackoff warned
about when attempting to build a model of a social system: we may end up modeling it,
unknowingly, as a determinist system by ignoring that it is formed by actors who act and that
through their actions indeed generate what the system does. Our heuristic is intended to improve the
mental models of modelers (and parties involved in a modeling process) by making them to
consider actions and actors as prominent over causes and variables, the latter being more abstract
and not necessarily connected with agency.
References
Ackoff RL 2001. OR: after the post mortem. System Dynamics Review, 17(4) 341-346.
Ackoff RL,Gharajedaghi J 1996. Reflections on Systems and their Models. Systems Research and
Behavioral Science, 13(1) 13-23.
Andersen DF,Richardson GP 1997. Scripts for group model building. System Dynamics Review,
13(2) 107-129.
Barlas Y 1996. Formal Aspects of Model Validity and Validation in System Dynamics. System
Dynamics Review, 12(3) 183-210.
Espejo R,Reyes A. 2011. Organizational Systems: Managing Complexity with the Viable System
Model.Springer Berlin Heidelberg: Berlin, Heidelberg.
Forrester JW. 1961. Industrial Dynamics.Productivity Press: Cambridge, MA.
Freeman RE. 2010. Strategic Management. A Stakeholder Approach (25th anniversary
edition).Cambridge University Press: Cambridge.
GroRler A, Milling P,Winch G 2004. Perspectives on rationality in system dynamics—a workshop
report and open research questions. System Dynamics Review, 20(1) 75-87.
Harper D (Ed.) (2016) Online Etymology DictionaryOnline Etymology Dictionary.
http://www.etymonline.com/index.php?term=actor, (accessed: 20 March 2016).
Hovmand P. 2014. Community Based System Dynamics.Springer New Y ork.
Lane DC,Oliva R 1998. The greater whole: Towards a synthesis of system dynamics and soft
systems methodology. European Journal of Operational Research, 107(1) 214-235.
Leclerc I 1953. Whitehead's Transformation of the Concept of Substance. The Philosophical
Quarterly, 3(12) 225-243.
Martinez-Moyano JJ,Richardson GP 2013. Best practices in system dynamics modeling. System
Dynamics Review, 29(2) 102-123.
Meadows D. 1980. The Unavoidable A Priori. In J. Randers (Ed.), Elements of the System
Dynamics Method (pp. 23-57). Productivity Press: Cambridge, MA.
Morecroft JDW 1982. A Critical Review of Diagraming Tools For Conceptualizing Feedback
System Models. Dynamica, 8(I) 20-29.
Randers J. 1980. Guidelines for Model Conceptualization. In J. Randers (Ed.), Elements of the
System Dynamics Method (pp. 117-139). Productivity Press: Cambridge, MA.
Saeed K 1992. Slicing a complex problem for system dynamics modeling. System Dynamics
Review, 8(3) 251-261.
Seibt J. 2013. Process Philosophy. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy.
The Metaphysics Research Lab, Stanford University: Stanford, CA.
Sterman J 1986. Introduction note to ‘Methods of conceptualization". System Dynamics Review,
2(1) 76.
Sterman J. 2000. Business Dynamics. Systems Thinking and Modeling for a Complex
World.McGraw-Hill: Boston, MA.
Sterman JD 1988. Deterministic chaos in models of human behavior: Methodological issues and
experimental results. System Dynamics Review, 4(1-2) 148-178.
Vennix J 1999. Group model-building: tackling messy problems. System Dynamics Review, 15(4)
379-401.
Vennix JAM. 1996. Group Model Building John Wiley & Sons: Chichester.
Vriens D,Achterbergh J 2006. The Social Dimension of System Dynamics-Based Modelling.
Systems Research and Behavioral Science, 2006(23).
community leaders and designing support tools to engage in hands-on project experience by
connecting them with enabling resources. Both are critical, but neither is sufficient to guarantee
success. Future research will be designing these decision support tools to connect hands-on
experience with enabling resources.
References
1. National Institute of Standards and Technology, Towards a More Resilient Community: An
Overview of the Community Resilience Planning Guide for Buildings and Infrastructure Systems.
2015, Washington, DC: U.S. Department of Commerce.
2. Jaffee, D. and T. Russell, The Welfare Economics of Catastrophe Losses and Insurance. The
Geneva Papers on Risk and Insurance Issues and Practice, 2013. 38(3): p. 469-494.
3. NRC, Building Community Disaster Resilience Through Private—Public Collaboration. 2011,
Washington, DC: National Academies Press.
4. Brose, D. Developing a Framework for Measuring Community Resilience: Summary of a
Workshop. 2014.
5. Executive Office of the President, Council on Environmental Quality,, Actions to Build Resilience
to Climate Change Impacts in Vulnerable Communities. 2015, Executive Office of the President:
Washington, DC.
6. Executive Office of the President, The President’s Climate Action Plan. 2013, Executive Office of
the President: Washington, DC.
7. Rodenbush, P., HUD LAUNCHES $1 BILLION NATIONAL DISASTER RESILIENCE COMPETITION, in
Annouces Partnership with Rockefeller Foundation. 2014, U.S. Department of Housing and Urban
Development: Washington, DC.
8. Gonzalez, G.|., HUD AWARDS $1 BILLION THROUGH NATIONAL DISASTER RESILIENCE
COMPETITION, in 13 states/communities to receive funding for resilient infrastructure and
housing projects. 2016.
9. Manyena, S.B., The concept of resilience revisited. Disasters, 2006. 30(4): p. 434-450.
10. Ellemor, B.J.D.F.M.J.B.A.G.H., How do we know about resilience? An analysis of empirical
research on resilience, and implications for interdisciplinary praxis. Environ. Res. Lett., 2013. 8: p.
8.
11. White, R.K., et al., A Practical Approach to Building Resilience in America's Communities.
American Behavioral Scientist, 2014. 59(2): p. 200-219.
12. Cimellaro, G.P., A.M. Reinhorn, and M. Bruneau, Framework for analytical quantification of
disaster resilience. Engineering Structures, 2010. 32(11): p. 3639-3649.
13. Gunderson, L.H. and C.S. Holling, Panarchy: understanding transformations in human and
natural systems. 2002, Washington, D.C.: Island Press. 507.
14. Lundberg, J. and B.J.E. Johansson, Systemic resilience model. Reliability Engineering & System
Safety, 2015. 141: p. 22-32.
15. Abramson, D.M., et al., The Resilience Activation Framework: a Conceptual Model of How Access
to Social Resources Promotes Adaptation and Rapid Recovery in Post-disaster Settings. The
Journal of Behavioral Health Services & Research, 2015. 42(1): p. 15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
275
28.
29.
30.
31.
32.
33)
34.
35.
36.
375
Cutter, S.L., et al., A place-based model for understanding community resilience to natural
disasters. Global Environmental Change, 2008. 18(4): p. 598-606.
Ainuddin, S. and J.K. Routray, Community resilience framework for an earthquake prone area in
Baluchistan. International Journal of Disaster Risk Reduction, 2012. 2: p. 25-36.
Joerin, J., et al., Assessing community resilience to climate-related disasters in Chennai, India.
International Journal of Disaster Risk Reduction, 2012. 1: p. 44-54.
Rahimi, M. and A.M. Madni, Toward a Resilience Framework for Sustainable Engineered
Systems. Procedia Computer Science, 2014. 28: p. 809-817.
Labaka, L., J. Hernantes, and J.M. Sarriegi, Resilience framework for critical infrastructures: An
empirical study in a nuclear plant. Reliability Engineering & System Safety, 2015. 141: p. 92-105.
Eakin, H.C. and M.B. Wehbe, Linking local vulnerability to system sustainability in a resilience
framework: two cases from Latin America. Climatic Change, 2008. 93(3-4): p. 355-377.
Maru, Y.T., et al., A linked vulnerability and resilience framework for adaptation pathways in
remote disadvantaged communities. Global Environmental Change, 2014. 28: p. 337-350.
Wright, C., et al., A Framework for Resilience Thinking. Procedia Computer Science, 2012. 8: p.
45-52.
Lu, M., Coastal Community Climate Change Adaptation Framework Development and
Implementation, in Telfer School of Management 2013, University of Ottawa: Ottawa, Canada.
Friend, R. and K. MacClune, Climate resilience framework: Putting resilience into practice. 2012,
Boulder, CO: U.S. Agency for International Development (USAID), the Rockefeller Foundation,
Institute for Social and Environmental Transition-International.
Arup International Development, City Resilience Framework. 2015, The Rockefeller Foundation:
New York City, NY.
Scheffer, M., et al., Generic Indicators of Ecological Resilience: Inferring the Chance of a Critical
Transition. The Annual Review of Ecology, Evolution, and Systematics, 2015. 46: p. 22.
Henly-Shepard, S., et al., Quantifying household social resilience: a place-based approach ina
rapidly transforming community. Natural Hazards, 2014. 75(1): p. 343-363.
John Snow, |.J., Engaging Your Community: A Toolkit for Partnership, Collaboration, and Action,
JSI, et al., Editors. 2012, Department of Health and Human Services (DHHS): Office of Adolescent
Health (OAH).
Garzon, C., et al., Community-Based Climate Adaptation Planning: Case Study of Oakland,
California. 2012, Pacific Institutue: California.
The Field Museum of Chicago, Chicago Climate Action Plan: Engaging Chicago's Diverse
Communities in the Chicago Climate Action Plan, ed. J. Diego, et al. 2009, Chicago, IL: The City of
Chicago Department of Environment.
California Adaptation Planning Guide: Planning for Adaptive Communities, C.E.M. Agency and
C.N.R. Agency, Editors. 2012, California Governor's Office of Planning and Research: California,
USA.
100resilientcities.org. 100 Resilient Cities. 2016 [cited 2016 Aug. 17]; Available from:
100resilientcities.org.
C. Stwertka, M.A., K. White, A new systems approach for building climate resilience for
communities in the U.S. Climatic Change, 2017. in review.
Comments on “A New Product Growth for Model Consumer Durables The Bass Model”.
Management Science, 2004. 50(12_supplement): p. 1833-1840.
Norton, J.B., Frank, A Diffusion Theory Model of Adoption and Substitution for Successive
Generations of High-Technology Products. Management Science, 1987. 33(9): p. 1069-1086.
Newell, B.R., et al., Rare disaster information can increase risk-taking. Nature Climate Change,
2015.
20
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
Clayton, S., et al., Psychological research and global climate change. Nature Climate Change,
2015. 5(7): p. 640-646.
Sterman, J.D., Communicating climate change risks in a skeptical world. Climatic Change, 2011.
108(4): p. 811-826.
Adger, W.N., et al., Cultural dimensions of climate change impacts and adaptation. Nature
Climate Change, 2012. 3(2): p. 112-117.
Sterman, J.D., Learning from evidence in a complex world. Am J Public Health, 2006. 96(3): p.
505-14.
Tversky, A. and D. Kahneman, Rational Choice and the Framing of Decisions. The Journal of
Business, 1986. 59(4): p. S251-S278.
Jack, B.K., C. Kousky, and K.R. Sims, Designing payments for ecosystem services: Lessons from
previous experience with incentive-based mechanisms. Proc Natl Acad Sci U S A, 2008. 105(28):
p. 9465-70.
McLaren Loring, J., Wind Energy Planning in England, Wales and Denmark: Factors Influencing
Project Success. Energy Policy, 2007. 35(4): p. 12.
Wiener, J.G. and T.M. Koontz, Extent and types of small-scale wind policies in the U.S. states:
Adoption and effectiveness. Energy Policy, 2012. 46: p. 15-24.
Warren, C.R. and M. McFadyen, Does community ownership affect public attitudes to wind
energy? A case study from south-west Scotland. Land Use Policy, 2010. 27(2): p. 204-213.
Devine-Wright, P., Beyond NIMBYism: Towards an Integrated Framework for Understanding
Public Perceptions of Wind Energy. Wind Energy, 2005. 7: p. 125-39.
Hax, A.C. and N.S. Majluf, Competitive Cost Dynamics: The Experience Curve. Interfaces, 1982.
12(5): p. 50-61.
Wright, T.P., Factors Affecting the Cost of Airplanes. Journal of the Aeronautical Sciences, 1936.
3(4): p. 122-128.
Nelson, D.R., W.N. Adger, and K. Brown, Adaptation to Environmental Change: Contributions of
a Resilience Framework. Annual Review of Environment and Resources, 2007. 32(1): p. 395-419.
Faber, J.W., Superstorm Sandy and the Demographics of Flood Risk in New York City. Human
Ecology, 2015. 43(3): p. 363-378.
Loucks, D.P., et al., Private and Public Responses to Flood Risks. International Journal of Water
Resources Development, 2008. 24(4): p. 541-553.
Gamper-Rabindran, S. and C. Timmins, Hazardous Waste Cleanup, Neighborhood Gentrification,
and Environmental Justice: Evidence from Restricted Access Census Block Data. American
Economic Review, 2011. 101(3): p. 620-624.
Greenberg, M.R., et al., Public support for policies to reduce risk after Hurricane Sandy. Risk Anal,
2014, 34(6): p. 997-1012.
Schwab, J.C., Hazard Mitigation: Integrating Best Practices into Planning. 2010, Chicago, IL:
American Planning Asoociation.
City of Roseville, Roseville Multi-Hazard Mitigation Plan. 2005, City of Roseville: Roseville, CA.
Lantz, E. and S. Tegen, Economic Development Impacts of Community Wind Projects: A Review
and Empirical Evaluation. 2009, National Renewable Energy Laboratory: Golden, CO.
press@ceq.eop.gov, FACT SHEET: Actions to Build Resilience to Climate Change Impacts in
Vulnerable Communities. 2015.
21
determined by supply, but also the demand for smallholder maize in the form of the DAR
informal grain retailing and the miller’s demand driving the commercial assemblage flow.
Summing up, the following can be used as a rule of thumb for understanding the dynamics
behind the fluctuations in our key parameter ADESM: it usually rises in May as the new
harvest comes in, staying at a plateau value of one that indicates the full servicing of
demand in both value chains. The grain supply then dries up either because the harvest was
simply too small to satisfy total demand, or because FRA has locked up large amounts of
maize in the formal value chain. ADESM then falls to a value around 0,84 indicating that
people who would prefer to consume grain have to reduce their daily consumption and
eventually resort to buying expensive maize meal. The time when this shift occurs depends
on how much smallholder maize was channelled into the informal value chain. This in turn
depends on the availability of maize in the formal and informal value chain in the preceding
marketing year: if the supply gap in the informal value chain was greater than in the formal
chain, the initial demand to refill stocks at the beginning of the marketing year is higher and
the informal value chain therefore attracts relatively more maize.
If the total harvest in the current marketing year was smaller than total yearly demand, the
supplies in the formal value chain eventually also dry up, leaving the ADESM to fall to zero.
If it was a surplus year, ADESM stays at 0,84. As the next main harvest comes in in May, the
cycle begins again. Only when the difference between smallholder surplus harvest and FRA
purchases is big enough to allow private traders and grain retailers to satisfy demand for
grain all year round, the ADESM stays at 1 throughout the whole year.
The sales by commercial farmers later in the year and the incoming green harvest in March
and April do bring some relief in the lean season, but they generally are rather insignificant
due to their small size in comparison to total yearly harvest and demand. Furthermore, as
commercial farmers only sell to the formal value chain, their production does not help to
reduce the gap in demand in the informal value chain, and thus will not change the value of
the ADESM if it is at 0,84.
One more important determinant of the ADESM’s behaviour is the existence of carryover
maize stocks from last year. If the current year shows a structural maize deficit, the
20
consumption of stocks that have been accumulated in a better preceding year can stabilize
the ADESM.
4.2 Production Shock Scenarios
It was shown that the most relevant adverse impacts on the maize production would come
from sudden changes (shocks) in the following parameters:
* Cultivated land
¢ Exposure to water
¢ Fertilizer Use
The most likely and relevant scenarios significantly altering these variables were found to
be the following:
Exchange rate shocks
Such shocks reduce the purchasing power of the Kwacha in relation to the US Dollar, the
currency in which fertilizer is internationally traded - which again increases the average
price for fertilizer for Zambian farmers, as most fertilizer is imported. I assume a dynamic
response of the economy; insofar steady high prices for imported fertilizer will make
domestic production more attractive and thus reduce import rates.
Floods
Large floods cause significant losses in the cultivated land area.
Fertilizer subsidy shocks
Cuts in the fertilizer subsidy program for smallholder farmers reduce the amount of
fertilizer that farmers can purchase and use.
Droughts
Droughts lead to a loss of cultivated area and lower yields due to a lack of rain and strong
heat. A scale with 5 different drought strengths was used for different scenarios.
21
4.3 Simulation Results for the Different Scenarios
Following the methodology laid out in chapter 2, I will measure the resilience of the value
chain in terms of the integral between the curves for the ADESM in the base run and the
ADESM in the respective shock scenario runs. The maximal impact of a shock would thus be
that the ADESM goes to zero for all the six years, or 72 months, simulated into the future.
This would lead to an integral of 60 between the base run’s ADESM and the shocked run’s
ADESM. The minimum difference is of course 0 when there are no adverse effects on the
ADESM in the scenario run. Using this range as a yardstick, we can thus analyse the
resilience of the value chain towards the different production shock scenarios in chapter 5.
The simulation results are ized in figure 8 below.
Scenario Description of shock ADESM integral
final value
1 Permanent Increase in Kwacha Value Towards US Dollar 4,86
by 35%
2 Permanent Increase Kwacha Value Towards US Dollar by 8,49
50%
3 Flood Loss of Cultivated Area by 10% in 2015 0,18
4 Flood Loss of Cultivated Area by 20% in 2015 0,45
5) Flood Loss of Cultivated Area by 30% in 2015 1,44
6 Flood Loss of Cultivated Area by 20% in two consecutive 1,95
years (2015-16)
Z Extreme 3-year Drought (2015-17) 20,08
8 Extreme 2-year Drought (2015-16) 12,44
9 Extreme 1-year Drought (2015) 6,17
10 Severe 2-year Drought (2015-16) 10,33
11 Severe 1-year Drought (2015) 5,02
12 Moderate 2-year Drought (2015-16) 6,71
22
13 Moderate 1-year Drought (2015) 2519
14 Extreme followed by Severe Drought (2015-16) 11,41
15 Steady Fertilizer Subsidies of 1500 Kwacha/Person/Year 0
16 Fertilizer Subsidies Permanently Cut in Half to 750 4,64
Kwacha per Person and Year
17 Fertilizer Subsidies Permanently Abandoned 13,88
18 Flood Loss of Cultivated Area by 20% and Zero Subsidies 3,96
in 2015, followed by Subsidies Cut-in-Half to 750
Kwacha/Person/Year in 2016
19 Flood Loss of Cultivated Area by 20% in 2015 and 2017, 18,48
as well as Permanently Abandoned Fertilizer Subsidies
20 Severe Droughts in 2015 and 2017 and Extreme Drought 19,77
in 2016, as well as Reduced Fertilizer Subsidies in 2015
and 2018 (750 Kwacha/Person/Year) and No Fertilizer
Subsidies in 2016-17
21 Severe Droughts in 2015 and 2017 and Extreme Drought 26,31
in 2016, as well as Permanently Abandoned Fertilizer
Subsidies
22 Extreme Droughts in 3 consecutive years (2015-17), as 27,41
well as Permanently Abandoned Fertilizer Subsidies
Figure 7: Overview of production shock scenarios
5. Scenario and Resilience Analysis
Using the final value of the integral between the scenario and base run ADESM as a metric,
we can evaluate the relative resilience of the value chain towards the different production
shock scenarios. In order to keep the analysis as concise and informative as possible, I will
group the scenarios according to the nature of the shock scenario and evaluate the
resilience of the value chain to the different types of shock scenarios.
23
5.1 Exchange Rate Shock Scenarios (No. 1-2)
While currency shocks do have a significant impact on the ADESM, their accumulated effect
is less pronounced in the medium to long term compared to the fertilizer subsidy and
drought scenarios. In terms of our methodological framework, the initial vulnerability is not
very high, but the permanence of the change undermines the adaptive capacity, so that in
the third year, the maize buffer stocks (i.e. stocks maize stocks that are carried over from
one year to the next to act as a security buffer in case of a shock) are depleted and the
structural maize (production to demand) deficit that is growing bigger over the years,
cannot be compensated any more. The ADESM therefore then breaks down to zero in 2017
(months 156 - 167) and the integral surges up.
® 1: Integral between ADESM base and scenario run
Vi 54
|
ia
1 2,54
de"
33,00 150,75 168,50 186,25 204,01
Page 1 Months
2 Integral between ADESM for base and scenario run
Figure 8: ADESM integral for scenario 1
However, due to the dynamic market response I assume, the producers compensate for the
more expensive fertilizer imports by raising domestic production. This leads to decreasing
marginal yearly impacts of the changed exchange rate value on the ADESM, as can be see by
the decreasing growth of the integral. The adaptive capacity thus becomes stronger again
over time.
24
5.2 Flood Loss Scenarios (No. 3-6)
The maize value chain in Zambia is quite resilient towards flood losses of cultivated area, as
the relatively small maximal value of 1,88 for the impact of the flood scenarios shows. As
can be seen by comparing scenarios 4 (one-year flood) and 6 (two-year flood), the adverse
effects on the food supply rise exponentially when floods occur in two consecutive years:
one year with 20% area loss has an effect of only 0,45, while the same event occurring in
two consecutive years has an effect of 1,95. The impact is thus more than four times as high
when the same flood loss shock is repeated in a consecutive year.
To investigate the reasons for this increasing impact, it we need to look at the change of
maize supply over the flood years (2015-16). Maize can be supplied to consumers either
from fresh production of the current year or from carryover stocks that were accumulated
over the last years. Comparing the graphs for SC 6 and the base run in figure 9, we can see
that the difference between yearly maize production in both actually becomes smaller in
the second shock year of 2016. Changes in the production for the current year can therefore
not explain the increasing impact.
Total yearly maize production under 20% flood loss
2.050
2.000 |
1.950 4
1.900 >
1.850 —
1.800
1.750 / f ==="2-yr flood (SC 6)
1.700
1.650 VA J
| <a
1.600 T T T
2015 2016 2017 2018 2019 2020
==Base Run
== 1-yr flood (SC 4)
Thousand tons of maize
Figure 9: Total yearly maize production in scenario 4
The answer can be found in the difference of the total maize that is stored throughout the
value chain: for this parameter, the difference between scenario 6 and the base run
25
becomes much bigger in 2016, as buffer stocks had to be used up in order to maintain a
sufficient supply in 2015 (see figure 9).
Maize stored in value chain at the end of february
1.200 \
\
800 Run
== 1-yr flood (SC 4)
Thousand tons of maize
== 2-yr flood (SC 6)
200
eee
2015 2016 2017 2018 2019 2020
Figure 10: Total maize stored in value chain at the end of February in scenario 4
Since 2016 is a year with just enough production to prevent the ADESM from dropping to
zero, stored maize stocks are consumed and reach virtually zero at the beginning of the
2017 marketing season. 2017, however, is a year with an even worse supply-to-demand
ratio where buffer stocks would be needed even more. As these are now depleted, the low
production can - other than in the base run and scenario 4, which feature enough buffer
stocks, not be offset and the ADESM drops to zero, as can be seen in figure 11a.
B® 1: ADESM Base Run Reference: ‘2: ADESM @ 1: Integral between ADESM base and scenario run
1 1 i y nT 1 #
; VV. ie
2. Ina,
3 o. | 1 1
1
A | : : : 1 a ;
133,00 15075 768.50 1ee25 204,00 733,00, T5075 Toe50 Te025 2040
age 1 Months Page 1 Months
2 Comparison ADESM scenario run to base run 2 Integral between ADESM for base and scenario run
‘igure 11a: ADESM comparison scenario 4 Figure 11b: ADESM Integral scenario 4
26
This drop of course leads to a strong increase in the integral between the scenario and base
run ADESM (cf. figure 11b), therefore explaining the big difference between the two- and
one-year flood scenarios.
These observations lead to an interesting conclusion: the initial vulnerability of the value
chain to the flood-induced area losses is not that high, but there is a certain threshold of
time with consecutive shocks, after which the system becomes very vulnerable to any
further perturbation in the production due to the depletion of buffer stocks. We can thus
attribute the resilience properties towards production shocks to two main factors: the
change in yearly production itself, and the ability to buffer the effects of production shocks
through carryover maize stocks from the preceding years.
The behaviour observed and described in the last paragraphs can be generalized across all
the scenarios simulated: while differences between the base run and scenario ADESM
normally are greatest in the years of the actual shock events, there usually is a lasting
adverse effect buffer maize stocks. Looking at these stocks and the current production is the
key to understanding the development of our resilience indicators.
The reader should note that in some scenarios, production lags behind in the years
following the shock by a small margin, e.g. the drought scenarios; while in other scenarios
yearly production actually overtakes the base run reference production due to a
compensation response. The latter is the case for the flood loss scenarios. However, these
responses are caused by dynamics in the production sector, are thus external and I
therefore will not expand on this topic.
While there is not much that actors in the value chain can do to change the production
output of maize, the finding about the buffer stocks is interesting in terms of my research
question of how resilience properties can be enhanced. If it was possible to accumulate
higher buffer stocks in the value chain, the impact of shock events could be mitigated and
the resilience properties thereby ameliorated. This will be discussed in more detail in
section 6.2.
27
5.3 Drought Scenarios (No. 7 — 14)
Drought scenarios have the highest impact of all the single-shock scenarios. In the case of
three consecutive extreme droughts in scenario 7, the final integral value of 20,08 shows a
substantial impact, which amounts to more than a third of the integral value that a complete
loss of supply would cause. We can thus conclude that the value chain is very vulnerable to
drought scenarios, mainly because the adverse effects of droughts on maize production are
very substantial compared to other scenarios.
Marginal impact on ADESM Integral per year of consecutive
extreme drought
Consecutive drought years
Figure 12: Marginal impact of consecutive extreme drought years on ADESM integral
Just like the flood scenarios, drought scenarios show an increasing marginal yearly impact
on our resilience indicator. This is due to the same reasons as discussed for the flood
scenarios, namely the progressively depleted buffer stocks. However, since the overall loss
in yearly production is much higher in these scenarios, the effect of change in current
production is so great that the effect of the buffer stock development is relatively less
important. This can be seen by the small relative growth in marginal impact compared to
the flood loss scenarios, displayed in figure 12.
5.4 Fertilizer Subsidy Scenarios (No. 15 -17)
The fertilizer subsidy shock scenarios are different from the other classes of shocks, as the
system faces a permanent change without a built-in compensation response like in the
exchange rate shock scenarios. Subsidies are cut in half (scenario 16), or abandoned
completely (scenario 17) in 2015 and then stay that way all through to 2020. This leads to
production constantly being around 6,5% lower every year compared to the base run in
28
scenario 16 and around 19,5% lower in scenario 17 throughout all six years. Looking at the
graphs for scenario 16, we can see how this translates into changing our resilience
indicators.
@ 1: ADESM Base Run Reference 2: ADESM 3: Total maize on store
io 1
2000000 a
al 1
od
1000000
6 J N\
33,00 150,75 168,50 186,25 204,
Page 1 Months
2 Comparison ADESM scenario run to base run
Figure 13: Comparison of ADESM and maize stocks scenario 16
# 1: integral betwe...e and scenario run 2: Yearly non subsistence demand 3: Yearly SH surplus production
54
1600000
2,8
e000] 7
4 |
Of
33,00 150,75 168,50 186,25 204,0
Page 1 Months
2 Integral between ADESM for base and scenario run
Figure 14: Development of ADESM integral against yearly demand/production scenario 16
29
The logic behind the behaviour of the resilience indicators is similar to the one explained in
the preceding sections: there is only a relatively small excess original demand? (i.e. demand
exceeding production) in 2015, so that the production deficit can be buffered by the
consumption of carryover stocks. The size of the carryover stocks is represented by the
local minima of the purple line in figure 13. In 2016, however, the difference between
production and original demand rises and the buffer stocks are now lower than the year
before. The growing gap in 2016 cannot be redeemed by consuming the already reduced
buffer stocks and the ADESM drops to zero later in the 2016-17 marketing year. The
permanently low production and the rising population lead to an ever-growing gap in
demand vs. production that does not allow carryover stocks to be built up. This leads to the
breakdowns in ADESM becoming progressively bigger in every consecutive year’s lean
season. The only reason why the growth of the integral slows down in 2019-20 (months
181-204) is that the base run also performs worse over time.
While the initial vulnerability is quite low, as indicated by the shallow initial growth of the
integral, due to the permanence of the effect, the adaptive capacity of the system is
undermined as buffer stocks are depleted. The shock effects therefore accumulate to a
significant level in the long run. If the shocks were only to occur in one or two consecutive
years, the impact on the ADESM would be comparatively small, probably comparable to
what we have seen for the flood loss scenarios. I therefore conclude that resilience of the
value chain towards shocks in the fertilizer subsidies is relatively high compared to other
shock types when they feature the same number of impact years.
Before moving on to discuss the combined scenarios, | would like to draw the reader's
attention to a phenomenon that is important in understanding the resilience analysis. There
is effectively a “threshold” behaviour for the ADESM in my model: since there are only
effectively two compensation mechanisms in terms of demand adjustment when maize
becomes scarce (eating less per day and changing to other carbohydrate sources), the
ADESM either stays at 0,84 where both mechanisms are at play and the consumption is
sufficiently reduced to not exceed supply - or it collapses to zero very quickly as all maize
stores in the value chain are depleted. Whenever this sometimes-fine threshold is crossed
3 » Original demand“ refers to the demand before it is adjusted for dynamic consumer responses to scarcity
30
and the ADESM thus falls to zero in the scenario run, but just manages to stay at around
0,84 in the base run, the integral surges up.
5.5 Combined Scenarios (No. 18-22)
The combined scenarios have - except for scenario 18 - a very strong impact on the
ADESM. The impact of the combined scenarios reflects what we have found out about the
resilience of the value chain to the different single-shock scenarios: the lower the resilience
of the value chain is to the single shocks that make up the combined scenario, the greater is
the impact of the combined scenario as well. The underlying dynamics of the translation of
production shocks to changes in the ADESM are essentially the same as described in the
preceding sections and | will thus not go into detail about them again.
An interesting observation is that the combined scenarios have a lower impact on the
ADESM than the sum of the two single-shock scenarios. For example, the 3-year extreme
drought leads to a final value of the integral of 20,08 and the abandonment of fertilizer
subsidies to an integral of 13,88. Yet, the impact of the combined shock scenario 22,
featuring both of these developments, does not amount to an integral value of 33,96, but
instead only 27,41. The reason for this is that the production sector shows a decreasing
marginal impact on yearly maize production when shocks are added up.
5.6 Conclusions Resilience and Scenario Analysis
To close this part of my analysis, | want to sum up the most central findings from this
chapter:
* The value chain is quite resilient towards flood events causing loss of cultivated
area, as well as towards exchange rate shocks.
¢ The value chain is moderately resilient towards fertilizer subsidy shocks. The
moderately strong effect of these scenarios is mostly attributable to the permanence
of the change. The effect can be expected to be rather small when assuming that the
shock only lasts one or two seasons, as the initial vulnerability of the value chain
towards fertilizer subsidy shocks was shown to be low.
* The value chain is vulnerable towards a prolonged drought. While a drought
lasting only one year still has only limited impact and its effects on the ADESM can
31
be mitigated through the consumption of carryover stocks, already a second
consecutive medium to extreme drought year depletes the buffer stocks and unfolds
increasingly strong impacts on the maize supply.
Even though there is a decreasing impact on the ADESM when combining two
shocks, the value chain is generally very vulnerable towards a combination of shocks
hitting it simultaneously or consecutively.
In general, the resilience of the value chain towards a one-time shock (only
occurring in one production season) is quite good and it exhibits a low initial
vulnerability. However, as soon as it is faced with consecutive shocks, the adaptive
capacity quickly wears off as buffer stocks are soon depleted after one or maximum
two years, and the impacts on the ADESM become very significant.
There are two main determinants for the effect that a shock has on the ADESM in the
value chain: the change in the current year’s production, and the availability of
carryover stocks that can act as a buffer. The policy analysis in the next chapter will
therefore focus on how buffer stocks can be used to enhance the resilience
properties of the value chain.
5.7 Impacts of Model Structure on Resilience Properties
Trying to keep the analysis of the different resilience responses of the model as concise as
possible, I focused on the most important impact factors that actually change in between
the scenarios and therefore explain the differences observed. These are, as we learned in
the preceding sections, the current year’s maize production and buffer stocks. The model
structure itself did not change across the scenarios, wherefore | did not explicitly mention
its effects on the ADESM in the previous sections.
However, the feedback structure of the value chain model naturally has a significant
influence on the results of the resilience analysis. Running different “structural scenarios”
by turning switches and loops on and off revealed the following:
The demand spill-over loop (C2) represents a very strong mechanism to cope with
food insecurity. The fact that consumers change to other crops, as well as the fact
that informal consumers buy around 20% less maize meal (due to higher prices)
than grain when being forced to change does, significantly lowers overall demand.
32
This reduces pressure on the maize market and leads to longer coverage of demand
with existing maize volumes. The importance of this mechanism can be seen by the
fact that the ADESM integral for scenario 16 would rise by 232% if one deactivated
this loop.
Less important, but still significant is loop C1 that represents how consumers in the
informal value chain reduce their demand in times of dwindling grain supply / rising
prices. By reducing demand, this loops has similarly beneficial effects on the ADESM
as loop C2. Even though the lowered consumption leads to an ADESM of around 0,84
instead of 1 (and thus a rising integral), the fact that consumers can eat maize at a
reduced level for longer before the supply breaks down completely leads to a
smaller overall integral - meaning that resilience is higher due to the effects of C1.
Another important coping mechanism that lowers the ADESM integral for a given
shock scenario is loop C3: FRA’s decision to offload buffer stocks in years of maize
deficit helps to stabilize the maize supply and thus food security. However, the
strength of this loop’s impact depends on the size of FRA’s reserves: i.e. if the
preceding year’s maize production was so bad already that FRA offloaded most of its
reserves, very little impact can be achieved through this mechanism.
6. Policy Analysis
After having evaluated the resilience properties of the value chain in chapter 5, in the
following section I want to explore how the can endogenously be improved. With results
showing that buffer maize stocks play a crucial role in determining the resilience of the
value chain towards production shocks, I designed policy interventions that aim to promote
the creation of those stocks.
6.1 Policies Under Base Run Assumptions
The policy I want to test is rather straightforward and relies on existing structures that are
already well established. Namely, I want to see what happened if FRA would try to fulfil its
original mandate: increasing food security for Zambians by buying and keeping strategic
maize reserves as buffer stocks. The stocks would only be released in times of maize
33
deficits. To test the effect of such storage policies under different circumstances, I simulated
them in a low-impact scenario with permanent change (scenario 16), as well as a high-
impact scenario featuring a shock of limited duration (scenario 8).
The results show that in an environment of consecutive structural maize deficit years, it
does simply not make sense to accumulate maize stocks in one year and release it in
another, since all one achieves with that policy is to improve the food security situation in
one year by worsening it in another. This policy has no significant positive effect on the
resilience metrics and is furthermore hardly economically feasible: it is hard to imagine that
people would be okay with taking maize out of the market in a deficit year for the sake of
storing for eventual future use in worse years.
6.2 Policies Under Changed Scenario Assumptions
Having thus established that policies relying on storing domestically produced maize are
not promising in an environment of constant structural maize deficits, | want to explore the
effects of storage policies in an environment with occasional bumper harvests - which, as
we know from historical data, do regularly occur in Zambia (cf. appendix A.1).
To investigate the effects of the policy, I will use a new scenario (No. 23), which features
base run production in 2015-16, then a high production year in 2017, followed by two
shock years in 2018-19 and base run production in 2020 again.
The policy that I want to test is for FRA to accumulate large buffer stocks in years with
bumper harvests and lock up the excess production (the amount of yearly production that
exceeds yearly demand) in their storages with the intention to keep these stocks constant at
that level, unless they need to release it in case of emergency. An emergency is defined as a
time when the ADESM would fall below a value of 0,8 without policy intervention.
In the case of scenario 23, this policy will cause FRA to keep 515.000 tons of their purchases
in the bumper harvest year of 2017 as a strategic reserve, and then release 390.000 tons in
the first shock year of 2018 to keep the ADESM over 0,8. This leaves them with 125.000
tons to spend in the second year. The uneven distribution over the two drought years is due
to the assumption that they cannot foresee the second drought coming and need to keep the
ADESM from collapsing to less than 0,8 in the first year. This assumption seems credible, as
34
it would hardly be justifiable for FRA to not release these emergency relief stocks in the
first drought year, just by pointing at the vague possibility of second shock coming up next
year. Running the simulation with and without the policy intervention, we get the following
results for our resilience indicator:
@ 1: integral ADESM with policy 2: ADESM Integral without policy
4 8,405
3 f
y
y 4.204
Ke
4 F
2: 0,00 fmm 1
33,00 15075 768,50 Tees 204.0
age 1 Months
5 Integral between ADESM for base and scenario run
Figure 15: Integral between ADESM for base and scenario run in scenario 23
Looking at figure 15, we can see that the ADESM performs significantly better when the
policy is in place. The “no policy” run of the scenario featured a final integral between the
scenario ADESM and the base run ADESM of 8,39 - while the integral in the “with policy”
run only amounted 5,56. This is a reduction by more than one third. Note that the total
production over the years is exactly the same in both runs; the only change is the policy of
FRA to store bigger amounts of maize in surplus years and not export the excess maize at
the end of the respective surplus year. We can thus conclude that an intelligent storage
policy by FRA that exploits the frequent occurrence of surplus harvest years can
significantly enhance the resilience of the value chain to production shocks - without any
exogenous help or inputs from outside Zambia.
6.3 Feasibility of the Proposed Policy
To conclude the policy analysis part, I want to address the feasibility of the proposed policy.
While I have argued that it is neither desirable nor politically feasible to accumulate maize
buffer stocks in years of structural maize deficit, the policy proposed and tested in section
6.2 appears to be useful and feasible, as I will show in this section. Possible frictions that
35
might hinder the implementation of the proposed policy can arise from political opposition,
storage capacity problems leading to high losses, and funding shortfalls for FRA. I will
discuss these problems in turn below.
Looking at political pressures that always have a big influence on the decisions taken in
politically controlled organizations like FRA, I can see no reason why an accumulation of
excess maize in surplus years should trigger political resistance or public outrage.
Loss of maize in FRA storages is actually a very valid concern when trying to implement a
policy that requires storing large amounts of maize for long times, potentially over years.
While FRA does possess large shed capacities, maize stored in sheds is subject to excessive
losses after just a few months of residence time: after one year, we can expect a loss ratio of
more than 50% and after 1,5 years even more than 80% (cf. appendix C.3). However, Bou
Schreiber (2015) expects FRA to keep on increasing their silo construction so that by the
end of 2019, they will have silo capacities of nearly 250.000 tons. This means that a great
portion of the maize can be stored in a way that produces almost no significant losses (less
than 3% even after 1,5 years of storage time). Furthermore, if FRA keeps up a steady flow
of maize through their storage by mixing and selling maize from last year while stocking up
fresh maize, they can limit the residence time and thus the loss ratio to reasonable amounts,
even in the sheds. | therefore believe that the storage loss problem can be adequately
addressed and will ultimately not hinder the implementation of the proposed policy.
The biggest threat for implementation is in my opinion the fact that FRA would need steady
and significant funding over a long time in order to properly execute the storage policy
proposed. Funding for FRA has been fluctuating quite a lot over the decades and was
always subject to often-arbitrary discretionary political decisions (N. Mason, 2011).
Furthermore, funding a storage programme will not immediately bring benefits that can be
presented to the electorate and there are opportunity costs of allocating funds to the
proposed policy, since that money then cannot be used for other, maybe more popular
programmes like consumer price subsidies or other poverty reduction measures. I
therefore see a real danger that policymakers might, especially in pre-election periods, re-
allocate the funds from FRA’s buffer stock programme to other measures that reap instant
4 For details on the storage loss ratios, see appendix C.3
36
benefits for the population. However, in the end the funding decisions depend on the
government’s will to follow through with a policy and it therefore can work if there is
sufficient political will to do it.
6.4 Change in FRA’s Sales Policy
Another change in policy I strongly want to suggest is that FRA should start selling maize to
grain retailers supplying the informal value chain. The current policy of just selling to big
commercial millers actually “locks up” maize in the formal value chain, which is eventually
often either exported under unfavourable terms or lost in inappropriate FRA storages,
while customers in the informal value chain at the same time cannot satisfy their demand
for cheap grain and have to reduce consumption. This obviously inefficient policy leads to
the ADESM taking on a value of around 0,84 instead of 1 even in surplus years, as can be
seen for the years 2010-12 (months 73 - 96) in figure 16.
This problem can easily be avoided by a simple policy change requiring FRA to open their
sales to grain retailers as well. This policy would actually help them fulfil their original
mandate - increasing food security for the Zambian population as a whole - much better
and more efficient. To illustrate the effects of this policy, I simulated the ADESM for the
months 73-96 with and without the proposed FRA sales policy. The results in figure 16
show that the performance of the ADESM would have been enhanced significantly.
@ 1: ADESM Base Run Reference 2: ADESM with new FRA sales policy
ere 2 Teo
4
at 0,504
:
a °,
3,00 79,00 85,00 91,00 97.0
Page 1 Months
2 Comparison ADESM scenario run to base run
Figure 16: Comparison ADESM development of old and proposed FRA sales policy
37
Concerning feasibility of this policy change, I do not see any big obstacles to
implementation. Since FRA has maize storages all over the country, FRA officials could just
go there during a number of fixed sales days and administer the exchange of maize against
money - much like they do when they buy maize grain from smallholders, just the other
way around. Moreover, such a policy change can be expected to be popular in the electorate,
as helps the majority of consumers to gain better access to their preferred form of maize.
7. Conclusion
7.1 Overview of Results
The general result was that the resilience of the value chain towards one-year shocks is
quite good, as the ADESM exhibits a low initial vulnerability due to the existence of buffer
stocks in the chain that can be consumed as a substitute for lacking fresh production.
However, as soon as the value chain is faced with several consecutive shocks, resilience is
low: the adaptive capacity quickly wears off as buffer stocks are soon depleted after 1-2
years, and the food supply breaks down. We furthermore found that there are two main
determinants for the effect that a shock has on the ADESM in the value chain: the change in
the current year’s production, and the availability of carryover stocks that can act as a
buffer.
Yet, as discussed in chapter 2, resilience can only be understood as resilience towards a
specific shock, and we therefore needed to disaggregate the results into the different types
of shocks. In doing so, we learned that the value chain is quite resilient towards exchange
rate shocks due to the dynamic response of the production system assumed, as well as
towards flood events causing a loss of cultivated area. In the case of shocks affecting the
fertilizer subsidies, the value chain is rather vulnerable when these changes become
permanent, but can be expected to show a relatively small vulnerability if the changes only
last 1-2 years.
However, the value chain is very vulnerable towards a prolonged drought. While a one-year
drought still has only limited impact and its effects on the ADESM and can be mitigated
38
through the consumption of carryover stocks; already a second consecutive medium to
extreme drought year depletes the buffer stocks and leads to increasingly strong impacts on
the maize supply. Finally, in the case of two different types of shocks hitting the value chain
simultaneously or consecutively, resilience has proven to be low. The adaptive capacities in
the form of buffer stocks are insufficient to alleviate the effects on the ADESM and the food
supply quickly falls to threateningly low levels - even though the marginal impact of a given
shock on the maize production decreases as shocks accumulate in a combined scenario.
Concerning the relation between the model’s structure and the resilience exhibited towards
production shocks by the value chain, we found out that the two consumption adjustment
loops have an especially strong impact on the resilience properties due to their direct
influence on the computation of the ADESM via changes in demand. The reduction of
demand in response to the changing availability of grain in the informal value chain, as well
as the behaviour of informal consumers to change to other crops and roller meal when
supply in the informal value chain dries up, both significantly improve the resilience in
response to a given shock. These coping mechanisms reduce demand for a given supply and
therefore improve the ratio that determines the ADESM.
Moreover, the non-FRA smallholder sales switch and the information feedback structure of
the informal and formal value chain were shown to affect the distribution of maize between
the two value chains. And since consumers respond differently to supply changes in the
informal value chain due to the two consumption adjustment loops, this distribution in turn
affects the ADESM. Furthermore, the FRA reserves switch structure and the feedback
structure of the value chain determine which actors build up how much storage stocks
throughout the value chain - which in turn influences the extent of buffer stocks available
to mitigate the impact of a given shock on the ADESM.
Analysing policies that can improve the resilience properties, we learned that the key to
endogenously improve performance was the creation and maintenance of buffer stocks.
However, it became clear that building up carryover stocks in an environment of
permanent structural maize deficits was neither desirable in terms of its effect on the
resilience metrics, nor politically feasible. Yet, | showed that a smart storage policy, using
FRA’s infrastructure and exploiting the frequent occurrence of surplus production years,
39
could significantly improve the resilience towards production shocks. Furthermore, the
feasibility of such a policy seemed promising under a few conditions, which were that FRA
keeps expanding its silo capacity as predicted by Bou Schreiber (2015), keeps a steady flow
of maize through its storages and that the storage programme is backed up by sufficient
political will. Lastly, I showed that the often inefficient distribution of maize in between the
formal and informal value chain, which can lead to supply shortages even in bumper
harvest years, could easily be remedied if FRA changed its sales policy in a way that also
allowed sales of maize into the informal value chain.
7.2 Discussion of Methodological Framework
Apart from the main goal of yielding insights about the structure, dynamics and resilience
properties of the maize value chain in Zambia, my work also served as a test for the
usefulness of my framework for quantified measurement of resilience in an SD simulation
model. I therefore shortly want to evaluate how the framework has performed.
Comparing using this framework to the “usual ways” of analysing the behaviour of SD
models by graphically comparing the development of a host of variables, I feel that the firm
focus on one metric helped to get a much clearer picture of the value chain’s capacity to
maintain a sufficient food supply in response to the different production shocks. Using the
ADESM integral as a metric, we received a fine relative scale that allowed comparing the
strengths of the impact between the different scenarios more precisely. Moreover, the
distinction between initial vulnerability and adaptive capacity helped to add further clarity
to the discussion of the shock responses.
The major drawback of using this method is probably the lack of an absolute scale - the
values of the integral only make sense in relation to each other and cannot be compared to
some form of general metric. The next step in developing a System Dynamics resilience
measurement framework would be to define an upper bound of a shock’s effect on the
integral of the FOM and compute the respective actual shock’s magnitude as a ratio of that.
This would allow the comparison of different system’s resilience towards a given shock.
40
7.3 Limitations and Areas for Further Work
Maize alone, as overwhelmingly important as it is for the food supply in Zambia, does not
determine the food security situation on its own. Even though agricultural productivity is
probably correlated between different crops, as their yields are determined by similar
parameters, one can imagine a year with a bad maize harvest and a good harvest for other
crops that may act as a substitute. In that case, a low ADESM for maize might not be so
much of a problem, as consumers could relatively easy change to other food sources. To
reflect the situation in Zambia more holistically, it would therefore be necessary to model
the value chains for other crops as well - something I unfortunately did not (yet) have the
time and resources to do. However, the literature I consulted suggested that the
distribution channels for other important crops in Zambia are structured in a similar way to
the maize value chain, so that future research could build on the basic model structure that
I carved out for maize, and adapt it to represent the value chains for other crops.
Another interesting avenue to expand this work would be to investigate the access
dimension in the model in greater detail. However, this would most probably require to
explicitly model prices. Since there is hardly enough comprehensive information about
prices at the different stages of the value chain, as well as their seasonal fluctuations that
drive the demand dynamics, further work in that direction would require field research.
Having only investigated the effects of production shocks, it would be interesting to also
look at the resilience of the value chain towards energy and transportation shocks. The
latter would require including spatial dimensions into the model, as the impact of shocks
affecting the transportation capacity of a given physical flow in the model would depend on
the distances covered in that link. A way to go about this could be to compute averages for
the distances maize typically travels from stage A to stage B in the value chain. This average
could then be used to model the degree of impact that the shocks disturbing the
transportation capacity of the flow would unfold. The means of transportation that are
typically used in that flow would probably also have to be accounted for in such an effect
variable. However, I did not find appropriate information about this in the secondary data
or literature, so that researchers looking at this phenomenon would probably need to go to
Zambia for first-hand data collection.
41
Bibliography
Barlas, Y. (1996). Formal aspects of model validity and validation in system dynamics. System
Dynamics Review, 12(3), 183-210. doi:10.1002/(SICI)1099-1727(199623)12:3<183::AID-
SDR103>3.0.CO;2-4
Bertelsmann Transformation Index. (2012). Zambia Country Report. Gitersloh.
Bou Schreiber, E. E. (2015). Maize Losses During Storage: A System Dynamics approach to the
Food Reserve Agency Case in Zambia. University of Bergen.
Carpenter, S. R., & Brock, W. a. (2008). Adaptive capacity and traps. Ecology and Society, 13(2).
doi:40
Chapoto, A., Chisanga, B., Kuteya, A., & Kabwe, S. (2015). Bumper Harvests a Curse or a
Blessing for Zambia : Lessons from the 2014 / 15 Maize Marketing Season, (93).
CSO Zambia. (2015). Database of the Central Statistical Office. Retrieved from
http://zambia.africadata.org
Dalziell, E. P., & Mcmanus, S. T. (2004). Resilience, Vulnerability, and Adaptive Capacity:
Implications for System Performance. Jnternational Forum for Engineering Decision
Making, 17. Retrieved from http://ir.canterbury.ac.nz/handle/10092/2809
Dorosh, P. a., Dradri, S., & Haggblade, S. (2009). Regional trade, government policy and food
security: Recent evidence from Zambia. Food Policy, 34(4), 350-366.
doi:10.1016/j.foodpol.2009.02.001
FAO. (2014). Food security Indicators. Rome. Retrieved from http://bit.ly/l4FRxGV
FAO. (2015a). FAOSTAT Databse. Retrieved from http://faostat3.fao.org/home/E
FAO. (2015b). Food Balance Sheet Zambia. Retrieved March 30, 2015, from
http://faostat.fao.org/site/368/DesktopDefault.aspx?PageID=368#ancor
Gerber, A. (2015). Agricultural Theory in System Dynamics. In Proceedings of the 33rd
International Conference of the System Dynamics Society. Cambridge, Mass.
Henry, D., & Emmanuel Ramirez-Marquez, J. (2012). Generic metrics and quantitative
approaches for system resilience as a function of time. Reliability Engineering & System
Safety, 99, 114-122. doi:10.1016/j.ress.2011.09.002
42
Hodges, R. (AHPLIS), & Bernard, M. (2014). APHLIS — Postharvest cereal losses in Sub-
Saharan Africa , their estimation , assessment and reduction. doi:10.2788/19582
Janssen, M., & Anderies, J. (2013). A multi-method approach to study robustness of social—
ecological systems: the case of small-scale irrigation systems. Journal of Institutional ....
Retrieved from http://journals.cambridge.org/abstract_S1744137413000180
Jayne, T. S., & Jones, S. (1997). Food marketing and pricing policy in Eastern and Southern
Africa: A survey. World Development, 25(9), 1505-1527. doi:10.1016/S0305-
750X(97)00049-1
Jayne, T. S., Mason, N., Myers, R., Ferris, J., Mather, D., Lenski, N., ... Boughton, D. (2009).
Patterns and Trends in Food Staple Markets in Eastern and Southern Africa: Toward the
Identification of Priority Investments and Strategies for Developing Markets and Promoting
Smatlholder Productivity Growth.
Kaplinsky, R., & Morris, Mi. (2001). A Handbook for Value Chain Research. Brighton.
Kuteya, A., Sitko, N., & Inn, K. (2014). Review of the effects of FRA on Zambia ’ s maize
market : High prices despite bumper harvests.
Leathers, H. 209); Transaction Costs Analysis of Maize and Cotton Marketing in Zambia and
Te inable Devel Publication Series. College Park, Maryland.
Mason, N. (2011). Marketing Boards, Fertilizer Subsidies, Prices, & Smallholder Behaviour:
Modelling & Policy Implications For Zambia. Michigan State University.
Mason, N., & Jayne, T. S. (2009). Staple Food Consumption Patterns in Urban Zambia: Results
from the 2007/2008 Urban Consumption Survey (Vol. 2009). Lusaka.
Mason, N. M., & Myers, R. J. (2011). The Effects of the Food Reserve Agency on Maize Market
Prices in Zambia (No. 60) (Vol. 2011). Lusaka.
Nyanga, P. H. (2015a). Personally Interviewed by Conrad Steinhilber, 18.05. Bergen.
Nyanga, P. H. (2015b). Personally Interviewed by Conrad Steinhilber, 27.04. Bergen.
Olsson, L., Jerneck, A., Thoren, H., Persson, J., & O’Byrne, D. (2015). Why resilience is
unappealing to social science: Theoretical and empirical investigations of the scientific use
of resilience. Science Advances, 1(4), €1400217—e1400217. doi:10.1126/sciadv.1400217
Stockholm Resilience Centre. (2012). Applying Resilience Thinking. Seven Principles for
Building Resilience in Social-Ecological Systems. Retrieved from
43
http://www.stockholmresilience.org/download/18.10119fe11455d3c557d6928/13981507997
90/SRC+Applying+Resilience+final.pdf
UNDESA. (2012). World Population Prospects: The 2012 Revision. Retrieved from
http://esa.un.org/wpp/unpp/panel_indicators.htm
UNDP. (2014). Human Development Report 2014. Sustaining Human Progress: Reducing
Vulnerabilities and Building Resilience. New York City.
Zulu, B., Jayne, T. S., & Beaver, M. (2007). Smallholder Household Maize Production and
Marketing Behaviour in Zambia and its Implications for Policy (Vol. 2007). Lusaka.
44
APPENDIX
A.1: Maize Production Data
Maize production in metric tons
Total maize Commercial Smallholder Smallholder Smallholder
production farmers total production surplus subsistence
2004 1213599 48579 1165021 343878 821143
2005 866187 65613 800574 178803 621771
2006 1424439 84960 1339479 564348 775131
2007 1366158 97099 1269059 647863 621196
2008 1211566 85578 1125988 522033 603955
2009 1887010 229893 1657117 613356 1043761
2010 2795483 331960 2463523 1062010 1401513
2011 3020380 233484 2786896 1663043 1123853
2012 2852687 220521 2632166 1362812 1269354
2013 2532800 195793 2337008 1215244 1121764
2014 3350671 259016 3091655 1550346 1541309
2015 1821490 66219 1755271 761513 993758
2016 1823090 64551 1758539 742338 1016201
2017 1843830 64258 1779572 738971 1040601
2018 1877020 64792 1812228 745113 1067115
2019 1917240 65768 1851472 756334 1095138
2020 1961300 66966 1894334 770112 1124222
Sources for maize production data:
¢ All data projections from 2015-2020 are based on simulations from Gerber (2015)
¢ Total maize production:
o 2004 - 2013: CSO Zambia (2015: data sheet "maize production timeline")
o 2014: Chapoto et al. (2015)
¢ Total smallholder production:
o 2004-2011: N. M. Mason & Myers (2011)
o 2012 - 2014: Triangulated from total maize production assuming a steady
relation between commercial and smallholder production
¢ Smallholder surplus:
o 2004-2011: N. M. Mason & Myers (2011)
o 2012-2014: Triangulated from other sources as: Smallholder surplus
= total maize traded (Kuteya, Sitko, & Inn, 2014) - commercial production
* Commercial farmers:
o 2004 - 2011: Triangulated from other sources as:
Commercial production = total production - total smallholder production
o 2012-2014:
Triangulated from total maize production assuming a steady relation
between commercial and smallholder production
¢ Smallholder subsistence:
o 2004 - 2012: Triangulated from other sources as:
SH subsistence = total smallholder production - smallholder surplus
o 2013: Triangulated from smallholder production assuming ratio of
subsistence consumption staying steady for two years.
o 2014: Triangulated from smallholder production with information about
subsistence ratio from (Chapoto, Chisanga, Kuteya, & Kabwe, 2015)
A.2: FRA Data
F.R.A. Parameter in metric tons
Yearly purchase Desiredreserves Shedcapacity Silocapacity Imports
2004 105279 0 539200 55720 9400
2005 78667 0 567020 55720 38950
2006 389510 0 566430 55270 119700
2007 396450 0 566410 55060 1458
2008 73876 0 566460 54960 1015
2009 198630 0 566480 54930 42027
2010 878570 131786 566480 54930 5704
2011 1579891 236984 590950 57370 290%:
2012 1044998 156750 652990 63540 0
2013 422391 63359 737320 71920 10)
2014 1031303 154695 833910 82460 0
2015 380757 76151 932780 99130 10)
2016 371169 74234 1026640 125610 0
2017 369485 73897 1111110 164140 0
2018 372556 74511 1179570 212660 0
2019 378167 75633 1204760 244410 10)
2020 385056 77011 1211838 248011 0
Sources for FRA Data:
¢ Yearly purchase:
2004 - 2010: N. M. Mason & Myers (2011)
2011 - 2013: Kuteya et al. (2014)
2014: Chapoto et al. (2015)
2015 - 2020: Assuming FRA wants to purchase 50% of smallholder surplus
oo 0 0
production
¢ Desired reserves:
Derived from the yearly purchase under the assumption that FRA wants to keep
20% of their yearly purchase as reserves.
¢ Silo capacity:
Bou Schreiber (2015)
¢ Shed capacity:
Bou Schreiber (2015)
¢ Imports:
o 2004-2014: FAO (2015a: Timeline under "Trade -> Crops & livestock ->
Zambia")
o 2015 - 2020: Assuming no imports take place
A.3: Storage Loss Data
Ac lated storage Loss in per cent
Residence Silo Loss Shed Loss Slab Loss
time (months)
1 0,00% 0,72% 5,72%
Z 0,28% 1,61% 6,61%
3 0,81% 1,83% 6,83%
4 0,82% 4,47% 9,47%
5 0,83% 6,50% 11,50%
6 1,36% 10,12% 15,12%
Z 1,66% 14,97% 19,97%
8 1,86% 20,61% 25,61%
9 2,04% 26,86% 31,86%
10 2,20% 34,09% 39,09%
11 2,34% 42,22% 47,22%
12 2,46% 51,23% 56,23%
13 2,56% 61,14% 66,14%
14 2,64% 71,93% 76,93%
15 2,70% 83,62% 88,62%
16 2,74% 83,62% 88,62%
17 2,76% 83,62% 88,62%
18 2,76% 83,62% 88,62%
Sources for storage loss data:
¢ All data from Bou Schreiber (2015)
A.4 Roller Meal to Consumer Made Hammer Meal Price Relation Data
Price August (ZMK/kg)
Lusaka Kitwe Mansa Kasama Mean
Breakfast meal (25 kg bag) 1391,0 1421,0 1505,0 1373,0 1422,50
Roller meal (25 kg bag) 915,0 975,0 1093,0 1000,0 995,75
Ratio breakfast of commercial meal 0,90 0,87 0,66 0,93 0,84
Composite commercial meal 1342,0 1362,2 1363,8 1348,4 1354,09
Consumer made meal (hammer 1063 942 910 941,00 956,50
mill)
Relation roller meal to grain price 1,42
Price February (ZMK/kg)
Lusaka Kitwe Mansa Kasama Mean
Breakfast meal (25 kg bag) 1536,0 1562,0 1750,0 1706,0 1638,50
Roller meal (25 kg bag) 1188,0 1261,0 1408,0 1408,0 1316,25
Ratio breakfast of commercial meal 0,92 0,88 0,62 0,96 0,84
Composite commercial meal 1506,9 1525,1 1620,1 1694,3 1586,62
Consumer made meal (hammer 1185 1138 1336 1455 1278,50
mill)
Relation roller meal to grain price 1,24
Source: Nicole Mason & Jayne (2009: tables 10 & 11)
Price relation values plotted over the year assuming steady change from lean to plenty
season yields this final relation:
Yearly Counter Price relation
127
1,24
1,42
12 1,30
oN
A.5 Urban Consumption of Meal Types Data
Commercial meal Hammer mill meal
Lusaka 0,899 0,101
Kitwe 0,855 0,145
Mansa 0,606 0,394
Kasama 0,317 0,683
Average 0,669 0,331
Source: Nicole Mason & Jayne (2009: tables 10-11)
Categories “consumer made maize meal via taking grain to grinding mill” and “maize meal
made at grinding mill and sold by a vendor/retailer” were aggregated to “hammer mill
meal” and the categories “samp” and “green maize” were excluded from the calculation to
obtain the final ratio visible in the table above.
A.6 Production Inputs for Scenario 23
Year Smallholder Ci ial bsi e Data taken from
surplus farmers Production scenario
2015 761513,4 66218,6 993758,0 Base Run
2016 761513,4 66218,6 993758,0 Base Run
2017 1370691,08 248154,7 1291558,5 High Production
2018 395346,1 34377,9 704086,0 Shock (SC8)
2019 363404,6 31600,4 691625,0 Shock (SC 8)
2020 754799,3 65634,7 1111506,0 Base Run
