MODELING THE IMPACT OF LOSS IN U.S. SOYBEAN PRODUCTION
RESULTING FROM SOY RUST DISEASE
Aldo A. Zagonel Stephen H. Conrad Paul G. Kaplan
aazagon@sandia.gov shconra@sandia.gov pgkapla@sandia.gov
Sandia National Laboratories’
Critical Infrastructure Modeling and Simulation Department
P.O. Box 5800 MS 1138 — Albuquerque, NM 87185-1138 — USA
Proceedings of the 23 International Conference of the System Dynamics Society
Boston, MA, USA, July 17-21, 2005 (Draft dated June 1*, 2005)
Our objective is to examine the consequences of soy rust to the U.S. agriculture in the next 2-5
years. In 2000, the U.S. harvested approximately 2.8 billion bushels of soybeans from almost 73
million acres of cropland, accounting for more than 50 percent of the world's production. The
crop generated $12.5 billion dollars, $6.66 billion in exports. Soy rust established itself in the
south last November and is expected to disseminate and deposit in the crops during this year’s
planting season. The extent of outbreaks depends upon climatic conditions. Early detection is
crucial since soy rust is deadly to the soy plant within 48 hours. Monitoring systems will warn
farmers of the presence of the spores and farmers are instructed on how to identify and treat it.
There is uncertainty regarding the sufficient and timely availability of fungicide. In addition to
historical data, we incorporate observations of on going planting and harvesting. Parameter
ranges in the model are narrowed as more information becomes available and existing
uncertainties dissipate. The impact of soy rust is analyzed in aggregate, looking at overall
production and market share contrasted against natural noise in the yields.
Key words: Soybean, grain, soy rust, plant disease, corn blight, agriculture
Introduction
This paper reports on-going efforts to model the U.S. agricultural infrastructure, in particular
examining the consequences of a soy rust outbreak starting in the 2005 harvest. Soybean is the
second largest crop in U.S. agriculture. In 2000, approximately 2.8 billion bushels of soybeans
were harvested from almost 73 million acres of cropland, accounting for more than 50 percent of
the world’s production. The crop generated $12.5 billion dollars in cash receipts from sales;
$6.66 billion in the form of exports. [CITE]
* Sandia is a multi-program laboratory operated by Sandia Corporation, a Lockheed Martin Company, for
the United States Department of Energy’s National Nuclear Security Administration under Contract DE-
AC04-94AL85000. The Department of Homeland Security, Science & Technology Directorate, provided
funding for this work. This draft describes on-going research involving the authors, Nancy Brodsky,
Sharon Deland, and Vanessa Vargas. We thank Andy Ford and four anonymous reviewers for their
constructive critiques and helpful comments.
Soy rust established itself in the south past November and over wintered. Now it is expected to
disseminate and deposit in the crops during this year’s planting season. [CITE] The extent of an
outbreak is heavily dependent on climatic conditions, as illustrated in Figure 1. [CITE]
Fig. 1 - USDA map of soy rust susceptibility across the United States
(Overlapped with soybean production)
There are two varieties of rust, and one is deadly to the soy plant within 48 hours. [CITE] Early
detection is crucial for adequate treatment using fungicides, but it is unclear whether the two
forms can be distinguished by farmers. Monitoring systems are being put in place to warn
farmers of the presence of the spore in their regions, and farmers are being instructed on how to
proceed to identify the disease and treat it effectively. [CITE] But, there is uncertainty regarding
the availability of fungicide and equipment to timely respond to outbreaks. [CITE]
The objective of this study is to examine the consequences of soy rust to U.S. agriculture. An
important opportunity presents itself, which is to have the modeling effort take place at the same
time that soybeans are being planted and harvested in North America, for the first time subject to
this disease. In addition to historical data, we can incorporate observations of on going planting
and harvesting. Parameter ranges in the model can be narrowed as more information becomes
available and existing uncertainties dissipate. Every week there will be new information
available, both to parameterize the model, and to contrast real data with model behavior. This
means that this modeling work will not be helpful in terms of informing this year’s soy
production. But, a good model that skillfully represents what’s going on in the fields and in the
grain market will potentially be useful for medium and long-term analyses of soy rust impact.
Although present and dangerous, soy rust is not expected to be devastating to U.S. soy
production. Overall it only marginally affected productivity in South America; moreover,
production in these countries has actually continued to increase. [CITE] The U.S. will benefit
from existing experience dealing and controlling the disease in other countries. The EPA already
granted approval to use fungicides to treat the disease. [CITE] A strong market has flourished to
provide farmers with product and equipment to treat it. [CITE] Therefore, it is likely that the
disease will be controlled adequately.
However, soy rust is bound to have some impact in U.S. agriculture. It will increase the cost of
producing soybeans (due to increased labor, purchase of fungicide and new equipment, and the
cost of application) and its risk might induce farmers to substitute other crops for soybeans or
even avoid planting. Therefore, it is possible that soy rust will cause a reduction in the U.S.
market share in the global market due to comparative disadvantage, since soybean production
will become more labor intensive, and labor in this country is more expensive than for main
competitors in developing countries, primarily Brazil and Argentina.
The increased use of fungicide is likely to raise concerns related with human health and
environmental degradation. Also, some inequities vis-a-vis other bean crops may come into play
since special EPA authorization applies only to soybeans, while other types of beans are equally
susceptible to the disease.
Yet another significant issue may result from an imbalance between the demand and supply of
fungicide in one season, potentially leading to market overreaction in the following season —
similarly to this winter’s shortage of the flu vaccine due to overproduction and underutilization
last winter. [CITE] Therefore, there may be a supply management problem here, resulting in
oscillations due to imbalances between the supply and demand of fungicide. This would indicate
market failure —a problem that could be aggravated to the extent that the fungicide produced for
this year may not be useful anymore next year, as the fungus builds-up resistance to the
fungicides used in the treatment of the crops— resulting in economic losses either to fungicide
producers or consumers, whoever bears the burden of the unused product. In the case of the flu
vaccine it was the manufacturers that were stuck with the bill, and some went bankrupt as a
result. [CITE] In the case of fungicide, it is more likely that those toward the end of the
distribution chain will bear the burden, since fungicide is currently being sold with a
nonrefundable, no-return policy. [CITE]
A definitive solution to the soy rust problem may take five to ten years, and it will involve
developing a genetically resistant soybean. [CITE] This hinges on the assumption that the
needed strain is available in a gene bank. Otherwise, it has to be genetically modified, raising
yet new problems since genetic modification of food is a highly controversial issue. [CITE]
Taken together, these issues suggest there is value in building a system dynamics model to look
at the consequences of soy rust, and to examine problems that may unfold within a two to five
years time horizon.
Building toward a “generic” crop model
Our soybean model builds upon the corn model reported in Conrad (2004), but it incorporates
international competition from southern hemisphere countries already exposed to and
experienced in this disease. It treats the disease itself exogenously, and relies upon expert
estimation of the overall seasonal crop losses. For now, foreign supply and demand are treated
exogenously too, but foreign production is accounted for in the computation of the price received
by U.S. soybean farmers. In this estimation we use an econometric model developed by Plato
and Chambers (2004). But, the system dynamics model will endogenously balance the supply
and demand loops, giving shape to key dynamic indicators such as: relative coverage (ratio of
supply to demand), soybean price, seasonal crop planted and on-going demand. The impact of
soy rust is analyzed in aggregate, looking at overall production and market share contrasted
against natural noise in the yields.
This year’s crop production will be monitored carefully to parameterize and calibrate the model.
By August most of the existing short-term uncertainties will have been clarified (changes in
production due to risk perception, incidence of the disease, timely diagnosis, availability of
fungicide and equipment, timely treatment, and overall seasonal crop losses). In addition, we
intend to test the model simulating the consequences of the 1970 corn blight —which affected in
average 25-30 percent of the national harvest, completely destroying as much as 80-100 percent
of the crop in some areas of the country. Thus, behavioral reproduction tests —contrasting model
output with real data, both historical and for this harvest— will help us refine, calibrate, and build
confidence in the model.
Figure 2 is a systems-level diagram of our generic crop model. The description of the diagram
and the order in which we are building the layers of complexity is more or less as follows:
Fig. 2 — Systems diagram of generic crop model
1. The stock-and-flow structure captures the processes of planting, growing, harvesting,
storing and selling the grains. Some of the key inputs to this production chain are:
farmer’s seasonal planting commitment, period of the planting season (start and
duration), time for the crop to mature (LOS in fields), duration of the harvest, yield under
normal conditions, and demand for grains
2. Weare interested in studying and contrasting two exogenous effects upon this production
chain, soy rust disease vs. noise in the system (mostly how the weather affects the yields)
a. A key parameter in the model is the net fraction of crop loss. It depends upon a
number of things, such as fraction of the crop vulnerable, disease education, crop
monitoring, timely diagnosis, treatment training, availability of fungicide and
equipment, and timely treatment. The better we measure this risk, the narrower
its range of variance. Thus, the validity of our conclusions in the comparison
between noise vs. impact of the disease relies heavily on this parameter
b. Another important exogenous element is the magnitude of the impact of weather
on the yields, captured as historical variance in the yield, controlled for advances
in productivity
3. The seasonal planting commitment by farmers depends on a number of things: the
forecasted price of grains provided by the USDA, the price of grains in the futures
market, elasticity of grain supply, government subsidies, risk perception regarding soy
rust disease, availability/cost of insurance, return on alternative investments (including
other crops), and availability of land to plant. Some of these inputs to farmers’ decisions
are easier to comprehend and synthesize than others. We’ll do our best to capture as
many as possible. Note that the subsidized price of grains, if greater than the break-even
price, constitutes a floor in terms of production, and land availability constitutes a ceiling.
The broader the range between the floor and the ceiling, the more important it is to
capture accurately these things that shape farmers’ planting decisions
4. To close the production loop, we need to capture the processes through which the future
price of grains (and/or the price of grains in the futures market) is forecasted. This
involves examining both the USDA and Chicago Board of Trade (CBT) forecasting
procedures, and perhaps reconciling them. We assume these forecasts are based upon a
number of things, such as: expected demand for grains (domestic and foreign), adequacy
of the physical inventory (grain in storage vis-a-vis grains needed to meet demand until
the next harvest, or for the following “N” months), adequacy of the upcoming harvest
(what will be the yield coming into the inventory?), and the impact of inventory coverage
upon grain prices (coverage elasticity)
5. Crop loss and subsequent reduction in harvested yields is likely to trigger a number of
compensating mechanisms:
a.
More production, provided the shortage makes prices rise, and provided there is
additional land to plant
More imports to accommodate existing demand
Rationing if the shortage is serious and imports are not available
Adjustments in demand due to rise in the price of grains. This, in turn, depends
upon the elasticities of demand (for animal feed and for other usages)
6. We will consider expanding the model boundary to treat endogenously some of the
exogenous parameters or time series:
a.
Building the interdependencies between the demand and supply of fungicides
(interdependency with chemical industry) and equipment
Building the interdependencies between irrigation (energy and water
infrastructures) and crop land availability
Including corn as a separate sector and examining the interactions between these
two commodities (both in terms of production and consumption)
Endogenizing foreign production to address global grain interdependencies (both
in production and consumption)
Coupling the soybean and corn sectors with the beef and dairy sectors; adding
poultry and hogs
This is an ambitious scope of work. For this paper, we would like to be able to conclude item 4
and as many as possible of the compensating mechanisms mentioned in item 5.
Modeling foundation
This modeling work builds upon Meadows’ (1970) hogs’ model, addressing commodity
production cycles. The basic feedback structure for production cycles proposed by Meadows is
shown in Figure 3. Inventory coverage is at the center of a pair of negative feedback loops
which act to eliminate imbalances between demand and supply; the resulting price acts as its
signal in promoting the efficient allocation of resources (production and consumption).
However, due to delays in capacity acquisition and bounded decision making by producers,
market reactions of demand and supply to price are very slow, resulting in oscillations (Sterman,
2000; Meadows, 1970). Additional instabilities and delays are introduced to the extent that
commodities are interdependent or act as substitutes. For example, grains (such as corn and
soybeans) are used in animal feed, and the price of feed influences decisions regarding animal
heard sizes, which in turn affects the consumption of feed, thus closing the loop through its
influence in grain prices. Moreover, different grains (corn vs. soybeans) can be used in the
production of animal feed, depending upon their relative prices, thus changing relative demand
and serving to balance grain prices.
Fig. 3 — Feedback loop structure of production cycles (copied from Meadows, p. 19)
Conrad (2004) described an initial crop model capturing the production cycle for corn, and how
it interacted with beef and dairy production. Figure 4 shows the corn sector. The negative
feedback loop for production is identified by the orange arrows. The total corn inventory and the
corn sales together determine the corn coverage time, which in turn determines the price of corn.
Since corn production is so strongly seasonal (planted in the spring and harvested in the fall),
seasonal effects are explicitly captured in the model. Although in reality farmers can respond to
price signals during the growing season by varying their applications of fertilizer and pesticide,
the foremost way they respond to price is in their decision about how much corn to plant in the
spring. In the model this is the only way for corn producers to respond to price. Corn
production responds to relative coverage (through price) but is confined within a range
characterized by a floor (the subsidized corn price or a break even price) and a ceiling (the
maximum acres of land available for production). Harvested corn accumulates in the fall and is
depleted over the course of months until the next harvest. Harvested corn is distributed primarily
as animal feed (~ 58 percent), [CITE] but it also goes to dry and wet mills and exports.
Consumption depends upon demand from the various types of buyers.
Fig. 4 — Corn sector diagram (copied from Conrad, p. 5)
Rasmussen and Becker (2004) did an initial stability and sensitivity analysis of Conrad’s three-
commodities model. Their assessment focused upon behavioral stability of the agricultural sub-
sectors (corn, beef and dairy) given parameter changes, particularly changes in assumptions
regarding aggregate agent reactions to prices and sector stresses, captured in the model as
elasticities’. They found model behavior to be highly sensitive to these elasticities. For corn,
they demonstrated that dramatic oscillations occur in acres of corn (planted), with small
increases in corn planting elasticity (the elasticity of supply), as illustrated in Figure 5.
Fig. 5 — Acres of corn under different assumptions for corn planting elasticity,
0.5, 0.625 and 0.75 (copied from Rasmussen and Becker, p. 6)
Rasmussen and Becker concluded that two key elements dominate the sector dynamics: (1) the
manner in which agents react to prices and sector stresses, and (2) where production delays occur
and their nature. In order to advance the modeling effort, they recommended (i) adopting model
simplifications to clarify model functionalities, (ii) further investigation of the human decisions
" Coverage, production and consumption elasticities — i.e., what effects will coverage have on price? What effects
will price have on production and on consumption?
models/curves to limit the family of resulting behaviors to the range of realistic dynamics, and
(iii) using historical time series to gain better insight into the agricultural sub-sector dynamics.
In spite of the concerns raised, they concluded that this version of the model provided “good
initial systems approximations and, in particular, a wealth of information about how to model the
detailed sub-sector price formation processes.” (p. 1)
We believe this modeling extension follows suggestions (ii) and (iii). This work aims towards
careful estimation and calibration of model parameters and table functions, within empirically
derived ranges whenever available, using both “snapshots” and time series comparisons to refine
and build confidence in the model and simulations. A better understanding of the physical and
behavioral processes captured in the crop model will allow us to narrow the range of feasible
real-world model behaviors. Hopefully this will provide not only better insights into structural-
behavioral links, but also more robust forecasts of soy rust impacts on U.S. agriculture and
economy. (Appendix | illustrates the use of parameterization and calibration spreadsheets to
substantiate and document parameter ranges, as well as available data, for the piece of the model
dealing with planting and harvesting with disease.)
On-going model refinements
Introspection and feedback from several reviewers led to the following list of model refinements
and tests:
Reformulation of the planting and harvesting processes
Reformulation of the disease scenario and crop losses
Reformulation of relative coverage
Logistic growth issue involving formation of the unsubsidized corn price (goal-seeking
vs. S-shaped adjustment)
Revisit resource allocation formulation (sales)
Parameterization and calibration of the production and consumption loops (including
coverage, production and consumption elasticities)
e “Snapshot” behavioral reproduction tests to examine model output against cross-sectional
data
e “Longitudinal” behavioral reproduction test, replicating the 1970 Corn Blight, to contrast
model behavior with actual time-series data
eyvyee
The first two items in this list were addressed and are reviewed in the next section. We are
currently working on the third item, the reformulation of relative coverage, also discussed below.
Planting and harvesting with disease
The motivation to reformulate the planting and harvesting with disease processes in the model
was due to:
e Desire to parameterize these processes drawing upon /ength of planting season, length of
stay of crop in the fields, and length of harvesting season, in addition to start of planting,
using auxiliaries to capture end of planting, start of harvesting and end of harvesting,
previously treated as parameters
e Elaborate the disease scenario to include fraction of crop vulnerable, as well as make the
formulation more flexible and robust in terms of variations in the fraction of crop loss
scenario and the timing of the disease (onset of disease and disease duration)
e Capture both physical crop loss (acres/month) and yield loss (tons/month), using a co-
flow structure
e Generate an indicator of fractional yield loss while the crop is in the field, as an early
signal about the yield of the future harvest
The resulting model diagram is depicted in Figure 6.
Fig. 6 — Planting & harvesting with disease model diagram
The results of a five year simulation —assuming a net fraction of crop loss of 25 percent during
the second season— are captured in Figure 7. Every year the crop is planted and harvested, as
illustrated in terms of crop in fields (line 1). As a function of the disease in year 2, the average
seasonal yield (line 2) falls during the time the crop is growing and maturing. When the crop is
harvested, an equivalent loss in terms of acreage is captured as crop loss (line 3). The fractional
yield loss (line 4) normalizes the average seasonal yield and constitutes the signal that there is
going to be a problem with the future harvest. This signal is perceived in the market as soon as
word is out about the effects of the disease on the crop, and it precedes the physical accounting
of soybeans that happens months later, during the harvest itself.
Fig. 7 — Disease scenario (net fraction of crop loss = 0.25 during the second season)
Relative coverage
The motivation to revisit the formulation of relative coverage was due to the issues discussed
earlier and highlighted in the systems diagram, involving USDA price of grains forecasts and the
CBT futures market. Both these forecasts take the perceived adequacy of inventory (plus
harvest) vs. expected demand (i.e. relative coverage) as the signal to adjust future prices. Thus,
we replaced the previously used table function capturing desired coverage with formulation that
computes the desired stock to use ratio from coverage needed until harvest (a combination of
demand and time remaining until the next harvest) and desired reserves (a minimum inventory
coverage level just before the new harvest begins to refill the inventory). We also incorporated
time to deplete the remaining inventory and we wish to determine if there is (or could be) a
rationing policy in the case of shortages.
As of yet, the revised model does not include the demand feedback loops. On going demand is
treated for now as constant. The revised model diagram is depicted in Figure 8.
Fig. 8 — Relative coverage model diagram
Figure 9 portrays the results of the same simulation for harvesting yield (line 1), on going sales
(line 2), harvested crop (line 3), adequacy of physical inventory (line 4), and perceived adequacy
of future inventory (line 5). During the simulation, on going demand remains constant and equal
to 20 million tons per month. The saw-shaped behavior of harvested crop (inventory level)
matches that of the desired coverage, except when there is a shortage in the harvest due to
disease (during the second season). This shortage unfolds into the following seasons.” Note that
the inadequacy of the physical inventory is not perceivable until the new harvest comes in.
However, the inadequacy of future inventory, which takes into account the early signal from the
health of the crop in the field, is perceivable as soon as word gets out about the disease.> On-
going sales are constrained by the availability of inventory.
Fig. 9 — Relative coverage (Perceived adequacy of future inventory)
For now, we used a weighted formula to establish the adequacy of future inventory combining
both the adequacy of physical inventory and the fractional yield loss in the field. The weight
changes dynamically depending upon where the soy is found —whether in fields or silos. As we
can observe in Figure 10, this formula discounts both the disease in the field (between months
15-19) and the shortage in the silos (between months 27-33 and, later, 39-45 and 51-57). It may
not be the ideal solution to represent how real people, in this case brokers in the grain market,
combine information regarding what’s going on in the fields vs. silos.» Thus, some empirical
research is required to build confidence in this formula.
Figure 10 helps to highlight some issues that come up due to delays in this system:
Fig. 10 — Types of (in)adequacies
(How to combine fractional yield loss with adequacy of physical inventory?)
1. The fractional yield loss (line 1) resulting from a disease in the field between months 15-
19 does not translate into a problem in the inventory (inadequacy of physical inventory,
line 2) until the crop is harvested. Therefore, if folk in the silos did not have information
about what’s going on in the fields, they would be clueless that there’s going to be a
problem keeping their inventories at the levels they would like them to be;
2. Worse, the inadequacy of supply (line 5) is not truly observed until more than a year later,
between months 29-32, just before the next harvest comes through. This is because on
going demand can be met all the way up until the inventory falls below desired reserves;
? In this version of the model soy production is also constant. Thus, the part of the harvest lost in the second season
is never compensated, except for reduced sales just before the 3", 4" and 5" harvests, when inventory is below the
level of desired reserves. Normally producers would plant more following the bad season motivated by rise in
prices resulting from shortage in supply.
* There is at least one more piece of information that we may need to incorporate to establish grain prices, which is
the commitment farmers make prior to actual planting, captured and reported by USDA. [CITE] This information
may constitute an earlier signal that needs to be considered in the perception of the adequacy of future inventory.
3. In addition, the shortage caused due to the disease in the second year propagates
throughout future years unless more grains are planted in the subsequent season, to
compensate for this season’s loss. Essentially, this becomes an inventory management
problem. In fact what corrects for this future problem is exactly the upward fluctuation in
price due to the shortage. But, as the argued above, this information (of a physical
shortage) is not going to be available until more than a year after the problem occurred.
Thus, it is the perception that there will be a shortage (or surplus) that drives prices, and
not actual shortages and surpluses.
This “market failure” due to inherent delays in the system is compensated by the existence of a
futures market, which acts as a proxy for the market’s “invisible hand,” attempting to forecast
future supply and demand for grains, and provide early signals of future prices. [CITE]
Therefore, it is important to capture in the model this decision rule that real decision makers use
to govern grain prices. This calls for examining how the USDA produces its forecasts, and
checking with futures brokers how the CBT establishes futures grain prices. It might be worth
representing this forecasting process in a policy structure diagram (Morecroft 1982), capturing
the causal structure and time delays involved in this decision.
We continue to work on the relative coverage issue, and other items of the outline of
refinements. We are also planning on a number of model additions.
Model additions to adapt the crop model from corn to soybeans
e Introduce specific soy rust issues to disease scenario: timely diagnosis, availability of
fungicide and equipment, and timely treatment. These issues help shape the fraction of
crop loss scenario which, combined with fraction of crop vulnerable, determines overall
seasonal crop losses (net fraction of crop loss)
e Incorporate Plato and Chambers’s econometric estimation of the price received by U.S.
soybean farmers (accounting for foreign production) in the formulation of unsubsidized
soybeans price
e Incorporate changes in production due to risk perception (including availability of
alternative crops, break-even prices, etc.) to the production loop
e Incorporate commodity substitution in animal feed (and maybe other areas of
consumption) to the consumption loop
e On-going examination of model output vis-a-vis available data as planting and harvesting
unfolds
e Revisit parameterization and calibration of production and consumption loops (including
disease scenario and coverage, production and consumption elasticities), given
availability of new data from this year’s production
The next section addresses where we are headed in terms of model-based analysis and sought-out
insights.
Discussion
The purpose of this model is to capture in aggregate level the impact that soy rust will have in
U.S. agriculture and economy. For this iteration of the modeling effort, we can define the
problem as an attempt to identify and measure the medium and long-term consequences of soy
rust in terms of acreage planted, seasonal production, productivity, break-even prices, crop
substitutions, grain substitutions, transient or steady state imbalances between supply and
demand, volume of sales and exports, and global market share. But the measuring stick used is
not one of acres, tons or dollars. Although the model will necessarily indicate values and units
that will imply that we can measure the metric decrease in production and dollar increase in
price, etc, we are primarily interested in the magnitude represented in the soy rust problem, as
well as possibly revealing counter-intuitive insights:
e Is the potential effect of soy rust upon U.S. agriculture, consumers and economy, in the
next five years, greater or smaller than have been the historical natural effects of, for
instance, weather?
e Will the current trend in U.S. market share in this industry remain the same now that soy
rust is present in North America?
e Could a relatively minor incidence of the disease this year create a market failure that
could result in being unprepared for a more significant manifestation of the disease in
future years?
Our working hypothesis is that the effect of soy rust will be smaller vis-a-vis normal fluctuations
observed due to weather and noise in this system. If, in the process of conducting this study, we
find evidence to reject this hypothesis, then, soy rust presents a BIG problem to the U.S.
agriculture and economy. To this end, we will have also shed some light on the following
questions:
e Which of the issues related to soy rust present themselves as significant vulnerability
issues to this industry and, subsequently, to the U.S. economy?
e What management issues do we need to consider (leverage or sensitive points in the
system)?
e What are the infrastructure intra-dependencies between soybean and other agricultural
commodities?
e What are the interdependencies between soybean (and grains in general) and other
national infrastructures (e.g. chemical industry)?
If we fail to reject our working hypothesis, then, in the absence of better evidence, we must
conclude that while the fungus is present in this country, and constitutes a serious danger to soy
production, soy rust is not expected to inflict severe aggregate harm to the nation (provided some
safeguards are taken!). However, it will likely cause many changes in the way of doing business.
These changes will generate new costs and benefits that will reflect in specific gains and losses
to specific sub groups within this industry (agriculture), as well as in other industries (e.g.
chemical). In addition, we may be able to characterize these sectorial gains and losses, and
suggest means by which they might be minimized or compensated provided due cause.
Limitations and future research
There are many limitations to this preliminary effort. First and foremost the fact that soy rust
disease is not modeled endogenously, but considered as a scenario in the simulations. We
believe other methods and tools are better equipped to model the spread and development of the
disease due to spatial considerations that are not well handled in system dynamics models. A
number of other aspects in the production of this crop and interactions with other commodities
(including other crops) fall outside of the boundary of this study. We attempted to list
endogenous, exogenous and excluded variables in a preliminary model boundary chart,
illustrated in Table 1.
Table 1 — Preliminary model boundary chart
References
Conrad SH. 2004. The dynamics of agricultural commodities and their responses to disruptions
of considerable magnitude. Proceedings of the 2004 International Conference of the System
Dynamics Society. Oxford, England (July 25-29).
Meadows DL. 1970. Dynamics of Commodity Production Cycles. Waltham, MA: Pegasus
Communications.
Morecroft JDW. 1982. A critical review of diagramming tools for conceptualizing feedback
system models. Dynamica 8(1): 20-29.
Plato G and W Chambers. April 2004. “How does Structural Change in the Global Soybean
Market Affect the U.S. Price?” Economic Research Service. United States Department
of Agriculture (USDA). OCS 04D-01.
Rasmussen S and N Becker. June 2004. “Initial Stability and Sensitivity Studies of the CIP/DSS
Agricultural Sector Models.” Project Report. Los Alamos National Laboratory.
Sterman JD. 2000. Business Dynamics: System Thinking and Modeling for a Complex World.
Boston, MA: The McGraw-Hill Companies, Inc.
Fig. | - USDA map of soy rust susceptibility across the United States
(Overlapped with soybean production)
Percentage of years out of 30 that climatic conditions are expected
to support establishment of soybean rust
be
Percentage
of years
L 0-10 51-60
11-20 61-70
21-30 71-80
31-40 81-90 ‘Soybean production
41-50 91-100 “G]__ 01 dot = 300,000 bushels
‘Sources: USDA's Animal and Plant Health Inspection Service and National Agricultural
Statistics Service.
Fig. 2 — Conceptual systems diagram of the generic crop model, displaying interactions between grains (corn vs. soybeans), between
U.S. and foreign production, between grains and other agricultural sectors, between agriculture and chemical industry, energy
and water infrastructures, and portraying the impact of weather on seasonal yields
Crop loss scenario _
Crop Timely aa ~
imontering > —™ diagnosis Inerdependency
* between agriculture
\ . « pre
and chemical .
\ awa’ ihsties Generic conceptual crop model
= k poy
\ tating a (Corn / Soybean) , .
‘ /, Fraction of erop SS Inerdependency |
Availability / ees sesnat between grains and
oftungicide —pvalablty of , Seasonal ‘animals (best captured
‘equipment weather a pier -aaal
oma
demand: ley
0S of erp
- Crop oss In flds <
od a ae Length of ss ji non
~ ge fh planting season /, \ Longs ot
planting { \ harvesting season
Irigation (energy & \ sf Beh
‘water infrastructures) = = Domestic crop - Grain in storage
Planing \ Harvesing
i ~
is ry . Min stocksonuse
— 2) | ame
Pant? Compensating oss through production some
“Analysis of risk & alternatives ‘commitment perenne oe vita
| | Forecasted
_. rs _ / os
2 ae re screage plant ie es Le
. Efecto isk & Seasonal Adequacy of 4] * Time to form
Hittin stem opportunities | veld or physical inventory sales expectation?
cal 4 pig | | ten /
/ alate — eal invent
sateen hg EY Ai “em / wi
alternative crops: id om 'roduction loop + JUSDA & Futures market ist "policies"
Information .
+ Va
da = *
diay ef parson saa og cmg| wee }
we ero invesnents bees diay ofimentry ols Compensating 6S trough imports
Og y rexthanes) (and raring)
oA OL x , E
. + Preuss
. . ain pice
2 se ee ‘i
- re Foreign '
pa a: con r
pr <a
ba) "Normal" price * ete
jinieadonerd Easy a ff
yetween corn. 2 8 ” ( Global grain
soybean (bestcaphred Plato & Chambers (2009) economvatic estimation? Nes lierepar Gane
in a two-sector model) * sacl
MN te -
Shee ok a“
a oe
Fig. 3 — Feedback loop structure of production cycles (copied from Meadows, p. 19)
Inventory
Production Inventory
capacity coverage
Consumption
com crop losses
disease
_
onsetofdisease
losses from
com disease
Fig.
domesti
time to begin
harvesting normal com
acres of com at exports
beginning of harvest
ic al
sales
Yi
com sales
elasticity
a
4—Corn sector diagram (copied from Conrad, p. 5)
time to form
expectation about
dairy feed
expected dairy
feed requirements
desired inventory
coverage at dairies
normal other com
inventory wentory shortage
feed sales ry shoriags
correction at dairies
at dairies
acres of com’at time to complete ib to ore Joey eae
inventory at dairies
onset of disease natvesting | —-favesting citer com A
55] acres of caren loonie exports feed sales annem
- consumption
__— planting corm |_SOm com sales to feeding dairy:
dairi mem cows “+ ——______—-in/6613
tna bog { average oom ss APS. | inventory at
harvested di
en svat es ofan ont a _
Harvest comsales to L_feedlo' feeding —? ee Sail
desired acres rarves hs eef cattle corn
cattle feediots cal
‘ofcorn ~¥—Mmaxacres of ; consumption
a ‘anticipated
invento
normal total
: com sales inventory coverage
oral acres at feediots
price effecton — fo0™ desired com sales to sales to time to correct
planting corn coverage f corn coverage dry mills wet mills inventory at feediots
__ bane desired com
coverage? relative corn ¥ inventory correction ventory shortage
at feedio
coverage atfeedots
relative expected lite 6 fam lastly
‘com prieefor, exPeCiationabout —_timetogetto “taste
ade com price normal com price desired inventory
9 ae coverage at feediots,
[me
expected coverage effect
com price Som [mee "9 am price expected feed
{for growers requirements
com price, nsubsid x
com price time to form
subsidized com expectation
price
expected
initial com price com price for
consumers
relative expected corn
price for consumers,
Fig. 5 — Acres of corn under different assumptions for corn planting elasticity,
0.5, 0.625 and 0.75 (copied from Rasmussen and Becker, p. 6)
acres of com
a acres of com acres of com
100 100 T
8 S
-
g
~4e-005 22-005 | |
0 4&2
% 1 144 18 192 216 240 om 28S Das dessa is 240 024 2 «8 anda eB O92
Tine (Mion) “Tine (Month) Tin (Month)
sctes of com: 040614-¢-ref Marre acres of com :040614c-elast065, Manes artis of com :0a0614-¢-smooth
Fig. 6 — Planting & harvesting with disease model diagram
Net fac
TIME STEP:
tion
7
Resetting the
seasonal yield lost
season “¥—___suantof
crop in fel ? {
= Seasonal yield lost
ry field is Jost
panies | mani
Noval yield 7
pep in ets Ve
Fatal Pani Yeu Hanestng Ved
yell bss ]
’
‘saci
eu
Time avaiable
to harvest Length of
r harestrg season
inal far rng we PT ewer itl
_ me ei | 7 } ie
Se: to plant
tatot \ of fa
patie ee L cote ha
Start of
may season
Planting
a
<p» Actes of crop that
‘can be planted
Max acres of
annual crop
Avg historical acres
‘of anmal erop pric eteet
‘on planing
Planting
elasticity
+ —_L0S of erop
L\!
harvesting DS ofa
Net faction
of erop boss
SS Start of planting
Fraction of crop
vyunearable
Fraction oferop
loss scenario Disease Time
active
Disease
duration
Onset of
disease
Fig. 7 — Disease scenario (net fraction of crop loss = 0.25 during the second season)
Disease Scenario
100 M acres
6 M tons/M acres
40 M acres/Month
1 Dm
M acres
M tons/M acres
M acres/Month
Dmnl
ocoooo
0 6 12 18 24 30 36 42 48 54 60
Time (Month)
Crop in fields : base 4 4 4 4 4 4 4 4 i— Macres
Avg seasonal yield : base 2 2 2 = = 2 = 2 M tons/M acres
Crop loss : base —=> = = = = = at M acres/Month
Fractional yield loss : base Dmnl
Fig. 8 — Relative coverage model diagram
On going
demand
Time to form
Harvested crop _asales expectation
<Harvesting’ Yield> On going sales» aed ae
ofsupply monthly sales
Timeframe to
deplete inventory’
rationing policy
Minimum stock Expected
/ useralo annual sales Months in
Info delay oe Ps
Health of er ye <Time of year>
iat Or rep: Adequacy of
and inventory physical inventory Suiich
adequat f
“uy Dested sinck _ Coverage needed wor
ff ‘use rato a 2 reserves nat have
Adequacy of <Fractional ~*— Montns 7
i <+——jield loss
ae future inventory yield loss> unflhanest
adequacy of —__—— <Harvested crop> <End of
future inventory harvesting>
Weight on physical vopin feldee
inventory smoohed “<7 in fel
Smoothing ime
Fig. 9 — Relative coverage (Perceived adequacy of future inventory)
Relative coverage
160 M tons/Month
360 M tons
1 Dmol
80 M tons/Month
180 M tons
0.5 Dmnl
0. M tons/Month
0 M tons =
0 Dn Ed Tt INST) TNT ONT,
0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54 57 60
Time (Month)
Harvesting’ Yield : base 4 =e 4 4 4 4 + + M tons/Month
On going sales : ba M tons/Month
Harvested crop : bas M tons
Adequacy of physical inventory : base = Dmnl
Perceived adequacy of fiture inventory : base Dmnl
Fig. 10 — Types of (in)adequacies
(How to combine fractional yield loss with adequacy of physical inventory?)
(In)adequacies
Fe a nas -
0.75
0.5
0.25
0
0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54 57 60
Time (Month)
Fractional yield los 4 4 4 4 1 A 4 4 Dmnl
Adequacy of physical inventor Dmnl
Adequacy of future inventory : Dmnl
Perceived adequacy of fiture inventory : base # Dmnl
Adequacy of supply : base Dmnl
Table | — Preliminary model boundary chart
Issue type
Endogenous variables
Exogenous variables
Absent variables
Pathogen
and fungicides
Economic
Non-economic vulnerabilities
Appendix 1. Parameterization and calibration spreadsheet to substantiate and document parameter ranges, as well as
available data (planting and harvesting with disease)
Parameters, auxiliaries, rates 8 As defined in the model Range ‘Actual value [Sources, references & observations:
accumulations:| Min. Value: Max. Unis: Width: | Magnitude: | SourceA [Source B
(Parameter
To Dnt IRalio of "expecta rain pie eave to an Ta
lor-normar pie. Research annual ime series data
lon seasonal grain pies (om & soybean) forthe
fast 10 years. Obtain monthly data for lowest and
highestriced years
Plating aaa os Dyan [For com, assumed relatively inelastic. Research fr
_ a ah com & soybean
‘Aug historical acres of annual | 700 Tilion acres
[Research annual time series data on seasonal
700 Tiion ares Jcres of cop planed forthe last 10 yeas com &
Max acres of annual rp 700.0 Nilion acres loyoean)
700 Wilion ares
Start of planting} a ion “or simplicity sake, assumed planting is uniform
za: Merit anting begins in the south and ends inthe north
350 Tlfon acres per say pave
‘month
Normal yield 380 Tlfon tons per [Research annual ime sees dala On average
nilion acres |seasonalyeld productivity) measured in weigh
tons, bushes) per unit of land acres)
‘Onset af aca Month
Disease duration Month oss scenario (Paul: Provide estimate and range
Fracion of oop vumerable Dal or ration of crop vunerabe and faction of eop
Fraction of rp loss scenario Dal Joss scenario foreach year (2005-2009)
Oh Da
LOS of erp in eds] 45 Months |Research for both comm & soybean
TH Month For simply sake, assumed harvesting is uniform
Length of harvesting season] 200 Months lauing plating season. Research dates when
925 Month Jarvesting begins inthe south and ends inthe
[eth (om & soybeen)
ry Tlion aos per
‘month Loss in productivity due to disease is aooounted for
30 Tilfon acres per 2s crop loss and subtract rom the harvest
‘month
35 Tilo tons per |Fluctuales around normal ye, Mina Tor Tis
milion aces larameter wl be wider than for normal eld due to
eect of weather. Research variance in yield for
lor & soybean (due to weather?)
Research annual ime seties data On Te Tal
|seasonalyeld production) measured in weight
lions, bushes)
% Tlion tons Research annual line sees Gas on US. Sales
2 Tilion tons per nd amount of grain cari over fom one season
month fo another. The stocktouse ratio isthe amount of
70% Dal lorans cared over relative tothe seasonal sales.
lat isthe desired value forthe stockto-use ratio
[com and soybean)?
