THE APPLICATION OF SYSTEM DYNAMICS (SD) SIMULATION
TO VOLATILE SYSTEM MANAGEMENT
John F. Affeldt
Booz Allen Hamilton
8283 Greensboro Drive
McLean, VA 22102-3838, U.S.A.
ABSTRACT
One of the most volatile market environments of our
time is the energy business. Whether the energy medium
is gasoline, electricity, or natural gas, traditional market
forces do not seem to exert the same influences as in
other markets. Indeed, the behavior of the energy market
sometimes seems to defy traditional understanding of the
law of supply and demand. Management of, and survival
within, such a system requires deep understanding of the
system’s potential behaviors under many different
scenario settings. System Dynamics (SD) is posited as
the most appropriate first methodology to apply when a
system with highly volatile behavior is under scrutiny.
This paper presents the background and some of the
lessons learned from projects in which SD simulation
was applied to analyze and understand the highly volatile
energy market. Two projects are described in which
different aspects of the energy market were modeled. A
natural gas strategic acquisition simulation provides a
tool for examination of market dynamics with a focus on
acquisition strategy, while a gasoline business simulation
provides insights into the supply side of the energy
business. The simulations were used to examine system
physics, understand endogenous and exogenous
influences, validate performance measures, and to
produce a systematic process for energy strategy
development. The application of SD to volatile
environment management is not new; the scale of these
simulations, and some of the techniques used for design
and rollout potentially make the projects unique.
1, INTRODUCTION
Presented here are the results of two projects in which
SD simulation was applied to the problem of energy
acquisition and distribution strategies. While SD
simulation is not a new discipline, its application to large
scale problems is not yet as common as other simulation
methods. In fact, SD simulation is not always the most
appropriate technique. In this paper, two projects are
described, one fairly small and the other large scale. In
both cases, the system under scrutiny is part of the very
volatile energy business, one of the most dynamic,
complex, and interrelated environments of the present
time. The thesis of this paper is that for systems and
environments that are characterized by their volatility,
complexity, interdependencies, and sometimes
unpredictable dynamics, System Dynamics is the most
appropriate first step to build an appropriate framework
for management and understanding of the system. The
SD framework may suggest points within the system that
are more appropriately studied with other tools and
techniques.
2. WHY SIMULATION ?
The use of simulation was a novel approach in this
context, given the legacy tools and methods available to
energy planners and analysts. For the natural gas
application, a colleague with responsibility for that
market, with significant domain expertise and a strong
desire to expose planners to new and better ways of
conducting their business, decided to attempt an
introduction of System Dynamics to his professional
community. For the gasoline business application, the
use of simulation was less of a new approach than was
the complexity of the simulation that was produced.
We had demonstrated the usefulness of
simulation as a learning and analysis tool in several other
engagements, thus the application of the technology was
a natural extension in these markets.
Simulation brings to the analyst/planner the
ability that to alter the world quickly, and from that
altered state, view many potential alternative futures.
This was the primary selling point to use simulation as a
general approach to natural gas acquisition
strategy/management. The ability to embed a reasonably
high fidelity economic model with the more typical
supply chain constructs in the gasoline market was also a
motivator.
System Dynamics brings the unique capability
to easily integrate parameters that are generally non-
quantifiable, and thus to test impacts of those parameters
over time on a large, interdependent system. Energy
market dynamics often respond disproportionately to
these non-quantifiable parameters compared to other,
traditional market factors. This reduces the efficacy of
other analysis techniques and argues for a different
approach. For example, the rumor of a hurricane hitting
the Gulf Coast of the United States has been sufficient to
induce a significant spike in energy prices., The energy
acquisition manager may then be forced to spend far
more for the necessary gas supplies than had been
anticipated. A plan to acquire energy resources with a
strategically sound framework that anticipates events
like this is critical to reducing this cost.
2.1 WHY SYSTEM DYNAMICS ?
Once simulation was accepted as a viable and desirable
alternative to traditional energy planning and analysis
methods, introducing the value of SD simulation was
straightforward.
The three basic questions of SD were
introduced within the context of assessing potential
futures: what is flowing in a system, where does it
collect, and what causes it to flow. Knowing the answers
to these basic questions permitted the development of a
simulation. Just the development of the answers added
significantly to the knowledge base within the
organization because they forced planners to focus on
the problem from a different perspective.
Every volatile system environment is a
relational environment, with highly interdependent
activities throughout the structure. Causal relationships
are often understood yet overlooked in the daily grind of
energy production and acquisition management. The
value of applying SD as the starting point for deeper
analysis becomes clear when the relational aspects of the
system are clearly demonstrated to analysts. A simple,
yet relatively robust, causal loop diagram (CLD) that
addresses a portion of the energy market can introduce
subject matter experts to the three principal effects that
this type of simulation illuminates for its users: systemic
feedback loops, systemic delays, and unintended
consequences. This usually convinces the experts that
SD is the right tool for initial analyses of volatile
management environments.
3. THE NATURAL GAS PROJECT
The United States Government is a sizeable user of
natural gas. The dynamics of that market are as much
influenced by speculation as actual supply and demand.
A market space driven by intangibles is, at best, a
difficult environment for managers who must acquire
large amounts of product at the best possible price.
Traditional forecasting tools and methods appeared to be
inadequate to assist analysts in devising consistent
strategies for product acquisition that would satisfy
demand as well as meet taxpayer expectations of lowest
possible cost.
In this project, the design team endeavored to
capture the essence of the market dynamics for this
commodity to produce a simulation useful to support
analyses for product acquisition. The focus was on
presenting various market scenarios, and several
acquisition options with the resultant predictions of
annual cost of procurement and estimated savings
against the standard purchase strategy.
3.1 MODEL STRUCTURE
The simulation was designed to represent three distinct
sites, each site being either a physically different
location, or a single location with three distinct buying
strategies. Each of the sites can be described by
parameters such as firm and interruptible ceiling prices
by month, any alternate fuels that may be desired should
the supply of natural gas be interrupted, firm and
interruptible gas budget levels, and the desired buying
strategy for the site. With this approach, three different
strategies for a single site may be simulated
simultaneously, permitting rapid analysis of alternatives.
The figure shows an example input screen for one of the
sites.
The analysis screen, shown below, allows direct
comparison of site physics, either during a model run
(using the pause function) or at the conclusion of the
simulation.
Navigation and Feedback [Ewes
Pee tape Serve
Notice that in this example, the strategy for site A was
flawed, resulting in a net projected loss for the year.
Sites B and C had strategies that project a net savings
across the year. These were three acquisition strategies
for the same site, allowing rapid comparison of the
relative merits of each. The simulation permits the user
to build an acquisition strategy for the year, described
month by month. Using a single site with these three
views, three completely different strategies for the year
may be constructed and played, with immediate
comparative feedback, as shown above.
The analysis capabilities provided with the
simulation are fairly standard for this type of tool,
focusing primarily on behaviors over time. The figure
shows a composite display of prices paid at each site
over time.
att
Causal Tracing - Avg Gas Price Paid{The City Gates}
———__— F ]
See || op pice eer nia
fa] paw Sa | hearer
age Reena br a |
| SS :
snennicrtgrg TPB
The natural gas model was designed to be an analysis
tool for the planner, providing an experimental platform
for scenario selection and detailed alternative courses of
action analyses. For example, some of the scenario
selections available are shown below. Provision has been
made to select random events, build interruption profiles
for each site, and build an alternate fuel profile, to name
a few of the options provided.
Scenario Set-Up
ee
state det 02 See
fx]
surat cortet O40 a
ntact O10 Sythe
Ca]
System Dynamics has demonstrated its value in this
decision support context convincingly because the value
proposition centers strongly on rapid analysis of
alternatives, the ability to compare multiple scenarios
simultaneously, and the ability to compare three distinct
sites or three strategies for a given site.
4. THE GASOLINE BUSINESS PROJECT
The gasoline business is indeed complex,
interdependent, and central to much of the global
economy. Understanding the physics of the business
permits decision makers to have the insights necessary to
formulate business strategies that meet their long term
objectives, while being accountable to the public at large
for providing personal energy resources at an affordable
cost and consistent rate.
This project revolved around the need for a
major United States west coast oil company to fully
address the intricacies of the business environment. The
business case involved multiple regions, multiple
channels of trade, multiple layers of competition, and a
set of intricate business rules that were masking some of
the true effects in the system.
In the geographic area of interest, there were
three sources of retail gasoline: major oil companies,
independent distributors, and the client, a mid-range
producer of petroleum products. The client’s pricing
strategy at the pump was simple — basically an average
of the majors and the independents. This strategy would
always place the pump price below the majors and above
the independents. The simulation quickly demonstrated
the flaw in this strategy.
The company in question had a single refinery
in the region under study, and purchased the remainder
of their product from the spot market. Since this
company was a large supplier of petroleum products in
the region, its shift to the spot market would invariably
trigger a rise in spot prices, thereby adversely affecting
the independents. A typical behavior of independent
distributors when the spot price rose was to immediately
raise their street price, not to recoup cost but to shift
volume. Given our company’s pricing strategy, they
invariably gained the volume shed by the independents,
causing more demand on the spot market, and again
causing the spot price to rise. Since the majors had
excess refining capacity, they sold fuel to the spot
market, effectively capturing both ends of the market.
4.1. MODEL STRUCTURE
The basic simulation is structured to include three
regions, three refineries, three levels of competition, and
three channels of trade. Known and unplanned
interruptions in refining capacity can be simulated. Each
competitor’s pricing strategies may be independently set,
and all elements of the business economics are directly
accessible.
The basic dependence on the spot price in this
market is depicted in the causal loop diagram.
ee Raia,
$C ie
St = +e note ee
Demand ~
Aen ate
sett
Price =
=“ “
ah ig,
A
is
sextet,
qi
Note the lag in major’s pricing in response to the spot
market’s fluctuations, and the lack of a lag with the
independents. An immediate increase of price at the
pump by independents as a tactic to shift volume results
in an immediate increase in demand at the minor’s
pumps (not shown in the diagram), which will eventually
influence spot price again. The spot market’s dynamics
are shown in the figure.
A spot price emulation algorithm was developed for the
simulation to provide the option of using historic spot
price data, or allowing the simulation to generate a spot
price profile. The blue line in the graph is the simulated
spot price. Note the similarity in behavior compared to
the actual spot price. To generate a simulated spot price,
the market’s reaction to planned and unplanned
interruptions, as well as documented market dynamics
were incorporated into the algorithm. The simulation
permits the use of historical interruption data or will
generate interruptions of random length and intensity.
The graphic below shows the simulation’s
results of demand over time, allowing a direct
comparison with the spot price (above), demonstrating
the shift in volume caused by the independent’s selling
strategy. The fictitious company, My Business, is
considered to be a minor oil company, and the company
of interest described in the introductory description of
the project. Note that demand spikes coincide with spot
price low values, and the demand drops occur when the
spot price spikes.
WI | i
ik nat
The same display for the independent’s demand profile,
below, demonstrates this shift.
TT. Es
[Alin
Profitability is always a concern in any business. The
energy business, especially the gasoline business, has an
intricate set of market dynamics in play. The basic
profitability causal loop diagram shows some of the
interplay.
ed ==Come Cot Das
Tota Prom
“4
sme Spot Price
| sept Con
aa 4
:
The profitability of the business has been condensed into
four metrics, as shown below.
Total Profit Refinery Revenue and Cost
Retail Revene and Cost Supply Revene and Cost
cot BOIS eee
From this display, it is apparent that the initial conditions
used to drive this particular model run resulted in a
profitable refinery operation, but a somewhat
unprofitable retail operation. The retail profit graph is
shown below in a larger format.
Retsil Reveme an Cost
Note that costs exceed revenues during periods when
spot price is elevated. This particular minor dealer has a
single refinery in the region. When the refinery is in full
production, there is rarely a need to use the spot market
to supplement stocks. When that single refinery
experiences an interrupt, dependence on spot increases,
and spot price will increase accordingly. Comparison of
the graph above and the spot price graph demonstrates
the linkage.
In use, the simulation can be used to manipulate
the many parameters describing the business in order to
produce profitability results more in line with business
desires, and then analyses can reveal the practicality of
the steps taken to achieve the change. The simulation
contains many data input screens, with the following as a
representative example.
As modifications are made, the user records them. After
a suitable solution is reached, the feasibility of the
modifications is determined. It is often discovered that
the real cost of modification of any complex system is
time rather than finances or other resources.
5. LESSONS LEARNED
Projects of the magnitude described here will always
generate many lessons. A few of the most notable are
listed.
When SD is applied to large scale simulation
projects, demonstrated results are often counter-intuitive.
This effect demonstrates the need to invest the time at
the outset to develop causal loop relationships with the
client, and to invest client staff in an understanding of
large system physics through their use. This new
perspective on organizational physics and
interdependence is fundamental to the use of the
simulation, but more importantly affords the
organization the opportunity to devise and test new
methods of management almost immediately.
It is critical to have access to a knowledgeable
member of the client organization on a routine and
consistent basis throughout design and validation of the
simulation. This facilitates developing a reasonable set
of assumptions, and lends instant credibility to the
simulation among organizational users.
A simulation, especially an SD simulation, has
its value in forcing people to learn about their enterprise,
not in deriving answers. In fact, if properly constructed,
simulations of the type described in this paper should
generate far more questions than answers, compelling
users to dig deeper into organizational structures,
broadening their individual and collective understanding,
and generating new behavior at personal and corporate
levels.
6. SUMMARY
These projects proved to be an exceptional application of
SD methodology in a non-traditional venue. The typical
toolset heavily favors regression analyses, optimization
techniques, and massive spreadsheets to solve problems.
In the approaches described here, we have sought to
demonstrate that a credible tool can be devised that will
enable decision makers to solve problems, and at the
same time provide significant value at a much wider
horizon. For example, both projects demonstrated the
value of the simulation for training others in how the
markets work and the businesses function. Individual
and collective behaviors can be influenced through
judicious application of tools such as described here.
Corporate performance measures can be placed under
scrutiny by simulation outcomes, forcing an evaluation
of metric validity. The introduction of System Dynamics
to the energy simulation environment resulted in learning
outcomes consistent with what would be expected from
more limited applications of the methodology.
Volatile systems can best be understood
through the application of SD simulation, as
demonstrated in this paper. Their very volatility, realized
through the rapidly changing nature of the industry and
its impact on the general public, argue convincingly for a
methodology that will provide system-level visibility and
understanding in order to gain a long-term view of the
impacts of management decisions. Indeed, System
Dynamics may not be the final tool for a particular
project of this type, but it is arguably the best starting
point in gaining an understanding for the proper
application of other tools, methodologies, and
techniques.
AUTHOR BIOGRAPHY
JOHN F. AFFELDT is a Senior Associate at Booz
Allen Hamilton, a global strategy and technology
consulting firm, headquartered in McLean, Virginia. He
holds a B.S. in Physics from the University of Scranton,
and an M.S. in Operations Research from the Naval
Postgraduate School. His primary interest in simulation
is organizational dynamics and the development of
corporate performance measures. Mr. Affeldt has been
involved with SD simulation and its practical application
to real world problems for more than 15 years. His
clients are from government and private industry, with
applications ranging from the simulation of personnel
training systems to large scale financial operations.
