SATELITE PRODUCTION FORECASTS:
VAULED WITH SIMULATED FUTURES AND OPTIONS TRADING
by
Lucanus Earl Martin
A thesis submitted in partial fulfillment
of the requirements of the degree
of
Master of Science
in
Applied Economics
MONTANA STATE UNIVERSITY
Bozeman, Montana
November, 2005
© COPYRIGHT
by
Lucanus Earl Martin
2005
All Rights Reserved
ii
APPROVAL
of a thesis submitted by
Lucanus Earl Martin
This thesis has been read by each member of the thesis committee and has been found
to be satisfactory regarding content, English usage, format, citations, bibliographic style,
and consistency, and is ready for submission to the College of Graduate Studies.
Joe Atwood
Approved for the Department of Agricultural Economics and Economics
Richard L. Stroup
Approved for the College of Graduate Studies
Joseph Fedock
iii
STATEMENT OF PERMISSION TO USE
In presenting this thesis in partial fulfillment of the requirements for a master’s degree
at Montana State University, I agree that the Library shall make it available to borrowers
under rules of the Library.
If I have indicated my intention to copyright this thesis by including a copyright notice
page, copying is allowable only for scholarly purposes, consistent with “fair use” as
prescribed in the U.S. Copyright Law. Requests for permission for extended quotation
from or reproduction of this thesis in whole or in parts may be granted only by the
copyright holder.
Lucanus Earl Martin
November 1, 2005
iv
TABLE OF CONTENTS
1. INTRODUCTION ......................................................................................................... 1
2. VALUE OF INFORMATION....................................................................................... 3
3. USE OF SATELLITE IMAGERY IN YIELD PREDICTIONS................................... 12
4. FORECASTING YIELDS WITH SATELLITE IMAGES........................................... 16
THE DATA ............................................................................................................... 16
THE MODEL ............................................................................................................ 19
5. TESTING THE VALUE OF THE YIELD PREDICTIONS......................................... 33
THE DATA ................................................................................................................ 33
THE MODEL ............................................................................................................ 34
FUTURES TRADING STRATEGY................................................................................. 40
OPTION TRADING STRATEGY .................................................................................. 49
6. CONCLUSIONS AND QUALIFICATIONS................................................................ 64
REFERENCES .................................................................................................................. 67
APPENDIX A: FUTURES TRADING PROFIT .............................................................. 71
v
LIST OF TABLES
Table
Page
1. Satellite Periods and Corresponding Dates .................................................... 18
2. Percent RMSFE for Iowa Corn Yield Prediction ........................................... 23
3. Percent RMSFE for Illinois Corn Yield Prediction ...................................... 24
4. Percent RMSFE for Iowa Soybean Yield Prediction ................................... 25
5. Percent RMSFE for Illinois Soybean Yield Prediction ................................ 26
6. Results of Regressing IA and IL Production on U.S. Production .................. 38
7. Corn Profit/Loss and Standard Deviation Summary ………………………..42
8. Soybean Profit/Loss and Standard Deviation Summary…………………….42
9. Average Corn Option Trading Profits – Forecast on Aug. 26........................ 50
10. Average Corn Option Trading Profits – Forecast on Sep. 9........................... 51
11. Average Corn Option Trading Profits – Forecast on Sep 23.......................... 52
12. Average Soybean Option Trading Profits - Forecast on Aug 26.................... 53
13. Average Soybean Option Trading Profits – Forecast on Sep 9...................... 54
14. Average Soybean Option Trading Profits – Forecast on Sep 23.................... 55
15. RetursnFrom Trading November Soybean Options ....................................... 56
16. Willingness To Pay For Satellite Forecasts Based On Option Profit ............. 61
17. Option Trading Profit With Different Levels of Information........................ 63
vi
LIST OF FIGURES
Figure
Page
1. Pixel Selection Process..................................................................................... 17
2. Iowa Corn – Predictions for Period 13(Jul-1) .................................................. 29
3. Iowa Corn – Predictions for Period14 (Jul-15) ................................................ 30
4. Iowa Corn – Predictions for Period 15(Jul-29) ................................................ 30
5. Iowa Corn – Predictions for Period 16(Aug-12) .............................................. 31
6. Iowa Corn – Predictions for Period 17(Aug-26) .............................................. 31
7. Iowa Corn – Predictions for Period 18(Sep-9) ................................................. 32
8. Iowa Corn - Predictions for Period 19(Sep-23)................................................ 32
9. Relationship Between Corn Stocks-To-Use Ratio and Futures Price .............. 35
10. Relationship Between Soybean Stocks-To-Use Ratio and Futures Price....... 36
11. Corn Production By State ............................................................................... 39
12. Soybean Production By State ......................................................................... 40
13. Dollars of Corn Futures Trading Profit in Period 18
(NDVI And Crop Condition Scores- Simple Strategy`……………………..45
14. Dollars of Corn Futures Trading Profit in Period 19
(NDVI And Crop Condition Scores- Simple Strategy) …………………….45
15. Dollars of Corn Futures Trading Profit in Period 18
(NDVI And Crop Condition Scores- Selective Strategy) …………………..46
16. Dollars of Corn Futures Trading Profit in Period 19
(NDVI And Crop Condition Scores- Selective Strategy) …………………..46
vii
LIST OF FIGURES - CONTINUED
17. Dollars of Soybean Futures Trading Profit in Period 18
(NDVI And Crop Condition Scores- Simple Strategy) ............................. 47
18. Dollars of Soybean Futures Trading Profit in Period 19
(NDVI And Crop Condition Scores- Simple Strategy) ............................. 47
19. Dollars of Soybean Futures Trading Profit in Period 18
(NDVI And Crop Condition Scores- Selective Strategy) .......................... 48
20. Dollars of Soybean Futures Trading Profit in Period 19
(NDVI And Crop Condition Scores- Selective Strategy) .......................... 48
21. Dollars of Corn Option Trading Profit in Period 18
(NDVI And Crop Condition Scores- Simple Strategy) ............................. 57
22. Dollars of Corn Option Trading Profit in Period 19
(NDVI And Crop Condition Scores- Simple Strategy) ............................. 57
23. Dollars of Corn Option Trading Profit in Period 18
(NDVI And Crop Condition Scores- Selective Strategy) .......................... 58
24. Dollars of Corn Option Trading Profit in Period 19
(NDVI And Crop Condition Scores- Selective Strategy) .......................... 58
25. Dollars of Soybean Option Trading Profit in Period 18
(NDVI And Crop Condition Scores- Simple Strategy) ............................. 59
26. Dollars of Soybean Option Trading Profit in Period 19
(NDVI And Crop Condition Scores- Simple Strategy) ............................. 59
27. Dollars of Soybean Option Trading Profit in Period 18
(NDVI And Crop Condition Scores- Selective Strategy) .......................... 60
28. Dollars of Soybean Option Trading Profit in Period 19
(NDVI And Crop Condition Scores- Selective Strategy) .......................... 60
viii
ABSTRACT
Both the USDA and private firms are allocating substantial capital towards
providing accurate and timely crop production forecasts. Production forecasts based on
satellite imagery have been suggested as a means of making forecasts earlier, more
frequent, and cheaper. This thesis attempts to determine if satellite data increases
information with respect to crop condition and final production. If so, does the additional
information have value and can it be used to make profitable trades in the futures market?
These questions are answered using NDVI data for Iowa and Illinois. Jackknifed
out-of-sample crop production estimates are calculated for both corn and soybeans for the
individual states. A variety of models were used, each including different bi-weekly
periods. USDA crop condition scores are also tested in some of the models. A model
based on the current stocks-to-use ratio for each commodity is used to predict the
market’s expected production level. When the satellite forecast differed from the
market’s expectation a trade was made in the futures markets. Both futures and option
strategies were tested.
Results suggest that satellite based production forecasts may result in profitable
soybean trades, particularly when downside risk can be reduced by trading options.
Further work should focus on refining the satellite images used in the model and
exploring more complex option trading strategies.
1
CHAPTER 1
INTRODUCTION
Accurate production predictions are essential for agricultural markets. To that
effect, the USDA was allotted $128 million to fund the National Agricultural Statistics
Service (NASS) in 2005, with an expected $16.7 million increase planned for 2006
(USDA 2005). Furthermore, there are now numerous private firms that produce
production forecasts that are available to subscribers prior to USDA report release dates.
The forecast errors from these reports tend to be unbiased and highly correlated with
USDA forecast errors (Egelkraut et al. 2003).
One of the newer forecasting methods involves the use of satellite imagery. If
satellite imagery has information value then yield forecasts could be made earlier,
cheaper and more frequently than with existing methods.
This thesis explores the value of production predictions derived from satellite
image based yield forecasts. Two primary questions are examined:
1.
Can satellite data increase information with respect to crop condition
and final production?
2.
Does the additional information from satellite data have value? Can it
be used to make profitable futures trades?
The thesis is organized as follows. Chapter 2 reviews the existing work from the value of
information field. It highlights how and why the satellite yield predictions can be valued
using a willingness-to-pay test. Chapter 3 explains the technical side of remote sensing
2
and highlights the work that has previously been done in this field. In Chapter 4, the data
and the model used for developing out-of-sample crop reporting district (CRD) yield
forecasts for corn and soybeans is explained. A model is developed in Chapter 5 that
converts the CRD level yield forecasts for Iowa and Illinois into a national production
estimate. This production estimate is then compared to an estimate of the market’s
production expectation and an appropriate futures or options trade is made. Profit and
losses from these trades are evaluated to determine the value of the satellite based yield
forecasts. Conclusions and qualifications follow in Chapter 6.
3
CHAPTER 2
VALUE OF INFORMATION
It has been shown that the greatest social welfare occurs when price information
is accurate and known well in advance. When price information is known in advance
producers can adjust their output to the optimal market clearing level, and total surplus is
higher as a result. Hayami and Peterson (1972) use a two period inventory adjustment
model to illustrate that better yield and production predictions result in larger surplus.
Furthermore, a clear vision of future production helps to drive storage decisions, resulting
in more stable levels of consumption and price (Wright and Williams 1982).
For many producers futures markets are their primary source for price discovery.
There has been considerable work on the efficiency of futures markets. Fama (1965)
defines an efficient market as:
"…a market where there are large numbers of rational, profit-maximizers actively competing, with
each trying to predict future market values of individual securities, and where important current
information is almost freely available to all participants. In an efficient market, competition among the
many intelligent participants leads to a situation where, at any point in time, actual prices of individual
securities already reflect the effects of information based both on events that have already occurred and on
events which, as of now, the market expects to take place in the future. In other words, in an efficient
market at any point in time the actual price of a security will be a good estimate of its intrinsic value."
Futures markets require accurate timely information in order to be efficient and
hence useful as a price discovery tool. Any information that guides the futures market to
the equilibrium price level will increase total surplus.
There is a wide array of both public and private agricultural production
information available in the market. There is considerable work that examines the value
of this information. Much of the focus has been on testing whether USDA production
4
forecasts have value. Sumner and Mueller (1989) examined the informational content of
USDA corn and soybean domestic supply reports for 1961-1982 by measuring the
absolute change in futures prices around the release date, and comparing it to absolute
price changes on non-release days. They found that there were significant price change
differences between release and non-release days and concluded that the USDA reports
provided value to the market. However, Fortenbery and Summner, (1993) using a similar
methodology, found that the USDA forecasts did not contain value from 1985-1989. As
technology advances it becomes relatively cheaper and easier for private companies to
make accurate production forecasts. Private companies typically sell their forecasts to
subscribers, releasing their reports prior to the respective USDA report. The increase in
private predictions, made in advance of USDA reports, is one reason why USDA reports
may have a decreasing effect on markets over time.
If futures markets are efficient they should contain all available information
(Fama 1965). Therefore private production forecasts which are released prior to USDA
forecasts tend to be priced into the market (Collings and Irwing 1990). The market will
only react to the unexpected information contained in the USDA report. Past research has
used the residual between a public forecast and the private forecast as a measure of
unanticipated information (Collings and Irwin 1990; Egelkraut et al. 2003). Live hog
futures do not react to anticipated information, but prices do adjust significantly when
there is unanticipated information in Hogs and Pigs Report (Collings and Irwin 1990).
A variety of work has addressed the issue of whether USDA or private forecast
reports have informational value. Garcia et al. (1997) identified three alternative tests of
5
information. The first test is based on Baur and Orazem (1994) finding that greater value
comes from government supply forecasts when the market’s supply forecast variance
decreases as a result of the government forecast. When this happens the market price
moves closer to the equilibrium price that would exist if all traders possessed perfect
information. As such Garcia et al. compared the market’s supply forecast variance before
the release of the USDA report to the variance of the market’s supply forecast after the
USDA report. They found that the USDA report lowered the variance of the market’s
supply forecast of corn early in the season, and was therefore valuable. Later in the corn
season, and throughout the soybean season Garcia et al. found that the USDA report did
not significantly lower the market’s supply forecast variance.
The second test comes from Fama’s Efficient Market Hypothesis (1970). In an
efficient market al. new information that enters the market should quickly be factored
into the prevailing price. The market will only react to unanticipated information. Garcia
et al., following the work of Collings and Irwin (1990), use the difference between
USDA forecast and private forecasts to define unanticipated information. They regress
the unanticipated information on the change in futures price from the closing price on the
day of the USDA announcement to the first non-limit opening or closing price following
the announcement. They find that the USDA report explains approximately a third to a
half of the variance of the futures price immediately following the release of the report.
The final test by Garcia et al. is a willingness-to-pay test inspired by Carter and
Galopin’s (1993) test of the value of the quarterly hog and pig report. Carter and Galopin
found that although there is a significant movement in price following the release of these
6
reports, a trader would not be able to earn a risk-adjusted profit, after transaction costs,
by having these reports in advance. If a trader could not earn a profit, his willingness to
pay would be zero, and therefore the value of the quarterly hogs and pigs report would
also be zero.
Garcia et al. constructed a willingness-to-pay test, modeling it after a market
timing test developed by Henrikssson and Merton (1981). This test calculates the
probability of successful market timing. A trading rule that sells (buys) corn or soybeans
if the USDA production forecast is higher (lower) than the market forecast was
developed, and the probability of a profitable trade was analyzed. They were able to
strongly reject the null hypothesis of no market timing for both corn and soybean USDA
production reports from 1971-1992. While this test does not indicate what a trader would
pay to obtain the information it does suggest that a trader would have a positive
willingness-to-pay.
Other work by McNew and Espinoza (1994) has pointed out that evaluating the
information value of the USDA production report should not be based solely on changes
in the futures market price. It is possible that the mean price level of the futures market
does not change following a report, as Sumner found occurred for the USDA crop reports
after 1985. If this was the case the traditional willingness-to-pay type tests would also
indicate that the report contained no value, as a trader would be unable to trade futures
profitably by obtaining the report in advance. However, as McNew and Espinoza argue, a
trader’s total return is a function of both changes in price, and changes in risk. If a report
narrows the distribution of the futures market, decreasing risk, then the report is valuable.
7
By examining option prices McNew and Espinoza was able to conclude that USDA crop
reports reduced uncertainty in the markets, and therefore contained value. In such cases
it may be more appropriate to use options when doing a willingness-to-pay test.
Many of the previous attempts at valuing information have been event studies
whereby the authors evaluate the effects of information –usually a USDA report- as it hits
the market. An event study such as these, however, is not a feasible method of evaluating
the effects of a satellite based yield model. It is likely that there are traders using satellite
imagery to guide their decisions, however it is not possible to decipher when they are
using the imagers or how. Until satellite imagery is more prevalent, and its effects on the
market are measurable, an event study is not possible. A willingness to pay test then
becomes the next best alternative. A trading strategy based upon the satellite imagery
yield predictions must be determined and tested. In order to do this there are two primary
issues that must be examined: When would a trade by initiated, and how long would that
trade be held?
The satellite yield predictions will only have value if they contain unanticipated
information. Therefore it is necessary to determine the market’s yield expectation.
Previous willingness to pay tests on USDA reports used an average of the private market
forecasts as a proxy for market expectation. Profitable trading strategies from using
satellite yield predictions are more likely if one can trade prior to the release of the
private yield forecasts. This is because more of the information contained in the satellite
images is expected to be priced into the market after the release of the private reports.
8
With no readily available proxy for market expectation it becomes necessary to
derive the market’s expectation. This study estimates the market’s expectation by
determining a relationship between the stocks to use ratio of a given commodity and the
price of that commodity. The stocks-to-use ratio, which is total ending stocks divided by
total ending supply, both of which are quoted monthly in the World Agricultural Supply
and Demand Expectation (WASDE) report, is a commonly examined ratio that depicts
supply levels. It is the most important statistic in developing a price forecast because it
measures the surplus that will be carried forward to the next crop year (Purcell and
Koontz, 1999). McNew and Musser (2002) used this ratio to develop price forecasts for
corn. They found that the model fit relatively well with R2 ranging from .74 to .98.
This paper uses the stocks-to-use ratio model in reverse. From the current futures
price, an expected stocks-to-use ratio is calculated. With an expected stocks-to-use ratio,
the market’s expectation of US production can be determined by comparing the expected
stocks-to-use ratio to the most recent production numbers in the World Agricultural
Supply and Demand Estimate (WASDE) report.
The market’s current expectation of production can be compared to the satellite
forecast of production. If the market expects lower production than the satellite forecast,
a short futures position would be entered into by a trader having the satellite forecast. The
futures price should fall as the market becomes aware of the higher production levels
forecasted by the satellite imagery, and the futures position would generate profit. If
consistent, risk-adjusted opportunities are possible by having access to the satellite
9
imagery, then a trader would have a positive willingness-to-pay and it can be concluded
that the satellite images have value.
The focus of this paper is on empirically measuring the value of satellite
information by assessing a trader’s ability to generate profitable trades by using satellite
based production forecasts. It is useful however, to examine the value of information
question from a wider perspective. Specifically, what would a producer be willing to pay
for a satellite based production forecast? McNew (1997) found that based on the
coefficient of variation, corn yield variability had increased from 8% to 11% from the
year 1920 to 1990. Crop yields are primarily impacted by weather events, and so this
variability in yields, and the consequent agricultural losses, might be diminished by a
more accurate satellite based yield forecast.
A model where producers make production decisions based upon price levels is
provided to illustrate willingness to pay. In this model prices levels are a function of total
aggregate output Q, with:
(2.1)
Q=f(z, ε1) and ε1~h(µ1, δσ1)
where z is a vector of variables affecting output, Q. The error term ε1 is defined by a
function h and is distributed with a mean of µ1 and a standard deviation of δσ1. Assuming
the producer has imperfect information δ=1 and δσ1 is positive. A producer’s profit
function under uncertainty can be then be defined as:
(2.2)
E[π (Y(x, ε2), P(Q)), w ] and ε2~k(µ2,σ2)
Where E is the expectation operator, π is the profit which the producer would want to
maximize by choosing the input level x, and producing output of Y with an error of ε2
10
caused by fluctuations in weather, rainfall, ect. The error, ε2 is defined by a function k
and is distributed with a mean of µ2 and a variance of σ2. The producer chooses the
optimal x based on the given price level, P and a vector of input prices is represented by
w. The key to this decision is the function h which will effect how dispersed the
producer’s aggregate price forecast will be.
The optimal level of x, assuming the implicit function theorem holds is:
(2.3)
x*(h(⋅,⋅) , k(⋅,⋅), z, w)
Where h and k are defined above as ε1 and ε2 respectively. This decision to use inputs of
x* results in E[π(Y(x*), P, h, k ], or the highest expected profit for the current level of
information on price forecasts, weather, drought conditions, rainfall ect. If a producer
were able to obtain a perfect forecast of these conditions then all uncertainty as to
aggregate quantity and price levels would disappear. In this perfect information scenario
δ=0 and the optimal level of x is:
(2.4)
x**(k(⋅,⋅), z, w).
A producer’s willingness-to-pay (WTP) for this perfect forecast would be the difference
between the two maximized profit scenarios:
(2.5)
WTP=π (Y(x**), P, k) - E[π(Y(x*),P, h, k)]
While satellite images may be able to provide producers with better production forecasts
they will not eliminate uncertainty. Therefore a theoretical model that allows for
uncertainty is more accurate.
11
With the addition of satellite imagery an improved forecast is possible and the
variance of the error term will decrease. To model this decrease, δ=.5. The producer’s
profit is:
(2.6)
E[π (P(Q), Y(x, ε2), w ]
The first order conditions are:
(2.7)
x***(h(⋅,⋅) , k(⋅,⋅), z, w)
Where x*** is the optimal level of inputs if one has access to satellite imagery. In this
scenario of less than perfect information a producer would have the following
willingness-to-pay:
(2.8)
WTP= E[π(Y(x***),P, h, k) - E[π(Y(x*),P, h, k)]
The willingness to pay will be positive if the expected profits are greater as a result of
having the increased level of information provided by the satellite images. While
expected profits from having satellite imagery may be positive on average it is not
possible to conclude that in every state of nature a producer would be better off with this
information. There would certainly be occasions where a satellite image based forecast
may encourage a producer to undertake an action that he would not have done if he had
only the first information set. The results of these actions could cause a substantial loss at
any one point in time. On average, however, if the willingness to pay is positive then the
satellite imagery has some predictive power, and the producer would be better off as
result of having it.
12
CHAPTER 3
USE OF SATELLITE IMAGERY IN YIELD PREDICTIONS
A basic understanding of the process used to generate the vegetative index scores,
which are the primary explanatory variables in the economic model, is provided below.
Chlorophylls within plant vegetation absorb light from the visible red spectrum (.9-.7um).
Mesophylls in plants will reflect 40%-60% of the near infrared (.7-1.3) (Lillesand and
Kiefer 1994). Changes in the levels of absorption and reflection of the visible red
spectrum and the near infrared can be used as indicators of the photosynthetic capacity
and efficiency of the observed plant canopy (Lillesand and Kiefer 1994).
If a plant is experiencing stress from disease, insects, or nutritional deficiencies,
the plant will lose vegetative mass. Bare ground or non vegetative matter will be visible
in the plant canopy under such stress. Bare ground and non-vegetative matter reflect
similar amounts of light from the visible red and near infrared (Lillesand and Kiefer
1994). This reflectance is distinct, and therefore a stressed plant canopy can be
distinguished from a healthy plant canopy. Even mild plant stress that does not result in
defoliation is observable because the level of chlorophylls in the plant will drop and less
visible red light will be absorbed as the photosynthetic capabilities of the plant decrease.
There are a variety of vegetative indices (VI) that convert the multi-band
observations of visible red and near infrared in to a single numerical index. Indices can be
sums, differences, ratios or another linear combination of the two wavelengths (Wiegand
et al. 1991). An extensive body of literature illustrates that VI correlate highly with the
13
photosynthetic process of non-wilted plants. They have been found to be good predictors
of plant biomass, vigor, and stress. (Tucker, 1979) Hatfield (1983) correlated VI to
canopy development and was able to quantify canopy responses and measure the amount
of photosynthetically active plant tissue. The VI is also highly correlated with the plant
stand parameters green leaf area index (L) (Wiegand et al. 1991). Tucker et al. (1981)
used a hand held radiometer to measure red and infrared spectral data. They found these
measures to be highly related (R2=.86) to the dry matter accumulation of winter wheat,
and suggested that satellite measures of this spatial data could be used to monitor plant
growth.
Due to the nature of the VI, yield predictions are most accurate if the crop’s yield
comes entirely from above ground production, mainly leaves. Alfalfa and other fodder
crop yields are directly linked with the VI score. However, for grain crops whose yield is
determined solely by the storage organs, the photosynthetic ability of the plant’s leaves to
assimilate CO2 is key to the filling rate of the storage organs. Yield on these crops is only
indirectly related to the VI score, and it is therefore essential that any model that predicts
grain yields must recognize this constraint (Benedetti and Rossini, 1993). Zhen (2001)
highlighted this fact by showing that early green up in wheat, prior to filling, can lead to
erroneously high yield predictions if precipitation is not adequate during the filling stage.
The most common VI used for yield predictions is the normalized difference
vegetation index (NDVI). NDVI is defined as the difference between near infrared and
visible red divided by their sum or:
(3.1)
NDVI=(NIR - R) / (NIR + R)
14
where NIR and R represent near infrared and visible red respectively. This is a unitless
measure that usually ranges from -1 to 1. Water, snow, clouds or any other non-vegetated
scene is represented by a negative number. Low positive numbers near zero indicate rock
and bare soil, which reflect near infrared and red at the same level. Increasingly positive
numbers indicate greener vegetation (Lillesand and Keifer 1994).
NDVI is a product of the National Oceanic and Atmospheric Administration
(NOAA) Advanced Very High Resolution Radiometer (AVHRR). The output form
NOAA satellites have been used frequently in scientific studies because it provides low
cost twice daily global coverage. The primary drawback to this data is that NDVI has a
low spatial resolution of 1km. Agricultural fields are of a much smaller size, and
therefore it can be difficult to determine how much of a 1 km NDVI score is the result of
agricultural practices and not something else such as a lake, or timber. To use the NDVI
scores this drawback must be overcome by selecting only those 1 km pixels that are
predominately agricultural land. In this paper, only those pixels that are considered at
least 80% cropland are included.
Noting the high correlation between plant biomass and NDVI scores, many
researchers have attempted to predict crop yields using NDVI over both large and small
areas. Most of this work has involved integrated or summed NDVI scores across the
growing period. One approach is to keep a running sum of NDVI scores as the season
progresses (Tucker et al. 1981; Benedetti and Rossini 1993; Quarmby et al. 1993). while
another is break the growing season into distinct periods and sum the NDVI score across
that period (Sellers 1986).
15
These integration methods have seen considerable use because a large sample of
NDVI scores is difficult to come by. With a small sample size the focus has been on
preserving degrees of freedom. The drawback, however, is that summed or integrated
models force the weight of the NDVI score in each period to be the same. Although the
two weeks of filling may be the most crucial time for determining corn yields, these
models allow no distinction between time periods.
Regardless of the problems, these integration models have been shown to produce
exceptionally large adjusted R2s. Zhen (2001) however made a strong case for
reassessing these models. Prior to Zhen’s work on Montana wheat yields most of the
studies were in sample, and the results appear misleading. Despite a high in-sample
adjusted R2 of .747 for his July wheat forecast, Zhen found that the July out-of-sample
forecasts appeared to have limited predictive power.
This paper will expand on Zhen’s findings using a large pooled sample of NDVI
scores to evaluate an out-of-sample model of sequential NDVI scores.
16
CHAPTER 4
FORECASTING YIELDS WITH SATELLITE IMAGES
The Data
The National Center for Earth Resources Observation and Science (EROS)
collects NOAA NDVI scores on a daily basis. Cloud cover can distort the NDVI score
and in order to limit the effects of cloud cover EROS uses an algorithm to select the
highest NDVI score from a two week period. This data is then changed from a -1 to 1
scale to a 0 to 200 scale. Scores that were -1, become 0, the original score of zero
becomes 100, and 1 becomes 200.
The Kansas Applied Remote Sensing Program (KARS) at the University of
Kansas obtained biweekly AVHRR NDVI from EROS. KARS processed the images
using three GIS software packages, IMAGINE, ArcMap, and MatLab. The result of their
work was a dataset that contained bi-weekly NDVI scores for each square kilometer (km)
pixel of Iowa and Illinois. For each km, or pixel, the county was listed, and the percent of
crop land in that km was listed. The percent of crop land is calculated by overlaying a
land use footprint over each square km pixel. For each km pixel there are 10,000 land-use
pixels. Only those km square pixels where at least 80% of the land-use pixels were
cropland, were included in future calculations. Figure 1 visually depicts this process. The
large squares each represent a one km NDVI pixel; while the smaller squares represent
land-use pixels. Cropland pixels are shaded in gray. Only two of the six NDVI pixels are
17
used (they are highlighted with a darker border), because 80% of the land-use pixels
within their boundaries are cropland.
Figure 1. Pixel Selection Process………………..
Figure 1. Pixel Selection Process – 80% of land use pixels
within an NDVI pixel must be considered agricultural land
From this dataset it was possible to compile average bi-weekly NDVI scores for
each of the nine crop reporting districts (CRD) in Iowa and Illinois. The bi-weekly
periods begin on January 1st. For the final data each year of NDVI data for each CRD is
stacked on top of each other. This results 135 data points for each state (15 years of data
multiplied by 9 CRDs). Various models are evaluated, and unlike similar work by Zhen
18
the bi-weekly periods are not limited to those that are perceived as being within the
growing period. Table 1 lists the last date of each two week satellite period.
Table 1. Satellite Periods and Corresponding
Dates
Satellite Period
Ending Date of Period
S1
S2
S3
S4
S5
S6
S7
S8
S9
S10
S11
S12
S13
S14
S15
S16
S17
S18
S19
S20
S21
S22
S23
S24
S25
S26
14-Jan
28-Jan
11-Feb
25-Feb
11-Mar
25-Mar
8-Apr
22-Apr
6-May
20-May
3-Jun
17-Jun
1-Jul
15-Jul
29-Jul
12-Aug
26-Aug
9-Sep
23-Sep
7-Oct
21-Oct
4-Nov
18-Nov
2-Dec
16-Dec
30-Dec
Final harvested yields1 for each CRD in Iowa and Illinois were obtained from
NASS’s online database of state and county level field crops. An estimate of the yield
1
Harvested yields are used because the WASDE reports, which are later used to derive the market’s
expectation of production, use harvested yields. It is possible that this could be problematic. If a portion of
the crop is very poor in a given region those acres may be abandoned and thus not counted in the final
harvested yield. This could cause yields to actually increase as crop conditions get worse. (Fackler and
Norwood, 2000)
19
trend line from 1950 to 1988 was calculated. A strong trend in corn and soybeans was
apparent. For corn, trend ranged from 1.58 to 2.14 bushels per year across the CRD’s.
Soybean trend ranged from .33 to .55 bushels per year for the various CRDs. Using the
trend model, deviations from trend were calculated for the years 1989 to 2003. The
deviations from trend are used as the dependent variable in the model, rather than the
actual yields. The model forecasts deviations from trend, and the previously mentioned
trend model is then used to convert the deviation forecasts into actual yield predictions.
In addition to NDVI, USDA crop conditions scores were used in the model.
During the growing season, on a weekly basis, NASS produces state level crop progress
reports based upon subjective surveys. One component of these reports is condition
scores for the various crops. The condition scores list the percentage of the crop that falls
into one of five categories: very poor, poor, fair, good and excellent. Analysts that
frequently use these reports tend to summarize the five conditions into one variable in
one of two ways. One method is to weight the percentages in some fashion and sum
them together. Another common method is to simply add the percentage of good crop
and excellent crop together. This second method is employed here.
The Model
Prior to Zhen (2001) the remote sensing literature focused on models using
summed or integrated NDVI scores. This approach forces the effect of each two week
period to be the same. Realistically the marginal importance of each bi-weekly NDVI
20
score should be allowed to differ. As mentioned before, the filling stage may be more
crucial to the final yield than the first two weeks of the growing season.
This model uses sequential NDVI scores to forecast deviations from trend for
each CRD. Forecasts are made for Iowa and Illinois separately. For each state there are
nine CRD’s and there is data for a total of fifteen years, (1989-2003) resulting in 135 data
points for each state model. Predictions are made for both corn and soybeans using the
following basic model:
(4.1)
DEVi,t=α + ΣβkSi,t,k+ εi,t
for i=1…12
for t=1…15
for k=1…19
where DEVi,t is the deviation from trend for harvested acres for region i in year t. The
period Si,t,k is the NDVI value for region i in year t, during the bi-weekly period k. As
there are only 15 years of data these data were pooled. This pooling allows β to change
for each bi-weekly period k, but forces βk to be the same across all of the regions i. For
corn, forecasts are made beginning in period 13 while forecasts are made beginning in
period 14 for soybeans.2 The forecasts are made every two weeks as an additional NDVI
score becomes available. For each forecast period each of the possible beginning dates
were tested. For example, when producing a forecast for period 15, it is possible to use all
of the satellite periods after period 1 (January 1st) or it is possible to only include those
images after period 2 (January 15th) or period 3, up to period 15. For the July 1st or the
13th period forecast, models were run beginning with all of the satellite scores, S1-S13.
2
Crop condition scores were not available for soybeans for all years prior to period 14. (July 15th).
21
Next the first period was dropped and the model was run with only S2-S13. This
continued until the model was run with only the final 13th period, S13. In the following
discussion, S refers to a satellite period. However, where a period is mention without the
S, it is assumed that it is in reference to one of the bi-weekly satellite periods. All of these
models were evaluated in an out of sample process.
Additional variables were added to the basic model and a prediction was made for
each of the forecast periods S13-S19 for corn and S14-S19 for soybeans. The same
pattern of including all of the satellite periods and then decreasing by one until only the
last available NDVI score was used in the model was employed. The additions resulted in
three variations of the basic model.
First dummy variables for each region were tested, using the model:
(4.2)
DEVi,t=α + ΣΩiDi + ΣβkSi,t,k+ εi,t
where D represents dummy variables for region 2-9. Next a model that contained the
crop condition score variable was tested:
(4.3)
DEVi,t=α + ΣβtCt + ΣβkSi,t,k+ εi,t
where C is the USDA crop condition score for year t at the time of the forecast. Condition
scores are only available at a state level, and only the most recent crop condition score
was used. The models 4.2 and 4.3 were combined to test the effects of dummy variables
and the crop condition score.
(4.4)
DEVi,t=α + ΣΩi Di + ΣβtCt + ΣβkSi,t,k+ εi,t
Finally a model using only condition scores was run.
22
(4.5)
DEVi,t=α + ΣβtCt
As the focus of this work is on testing the value of these models for futures
traders, all of the above models were run out-of-sample. While in-sample forecasts may
yield seemingly positive results, the value of any in-sample prediction will be suspect.
Are the results the effect of data mining, or is the forecast error small because a large
number of explanatory variables have been added? As all of the degrees of freedom are
exhausted due to a large quantity of right-hand side regressors, a spurious perfect model
fit will occur. (Fair and Shiller 1989) To avoid these problems the model must be tested
out-of-sample.
Due to the limited years of data the standard method of time series out-of-sample
estimation is not possible. Instead a jackknife procedure is used. A given year is left out
of the model. The model parameters are then estimated using data from the remaining
years. The parameters are applied to the data from the skipped year to create an out of
sample prediction. This process of skipping a year and then using the year’s data to
generate a forecast was done for each of the years, 1989-2003, and for each of the models
described above.
Tables 2-5 display the root mean squared forecast error (RMSFE) percents for
each of the models. The prediction period is listed at the top of the sheet, and the first
satellite image included in the model is listed on the left hand side.
23
Table 2. Percent RMSFE for Iowa Corn Yield Prediction
First Satellite Period Used in The Forecast Model
NDVI and crop conditions scores
NDVI scores with
Regional Dummy NDVI scores only
Variables
Models
1
2
3
4
5
6
7
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
1
2
3
4
5
6
7
8
9
10
11
12
13
20.92
20.57
20.48
20.18
19.91
20.23
20.09
22.45
21.57
21.20
20.95
20.62
20.78
20.48
15.24
14.71
14.26
13.89
14.27
14.07
13.93
13.69
12.34
17.30
16.84
16.94
16.80
14
20.82
20.43
20.45
20.15
19.64
20.10
19.94
22.24
21.35
21.04
20.82
20.17
20.45
20.14
14.75
14.58
14.54
15.71
15.97
15.80
15.70
15.59
16.80
21.00
20.31
20.08
19.84
19.40
Forecast Periods
15
16
17
22.22
17.51
15.63
21.65
17.02
15.37
21.66
16.89
16.26
21.26
16.70
16.21
20.34
16.06
15.52
20.82
16.30
15.71
20.66
16.27
16.28
23.64
18.07
16.31
22.63
17.30
15.96
22.36
17.13
16.73
22.07
16.84
16.64
20.99
16.08
15.98
21.31
16.15
16.03
21.00
16.03
16.52
13.27
11.99
9.48
14.18
12.38
9.70
13.48
11.72
10.63
13.23
11.52
10.67
13.26
11.34
10.14
13.16
11.08
10.31
13.62
11.23
10.12
13.49
11.90
14.06
13.08
11.41
13.40
14.67
12.72
15.52
14.44
13.41
15.63
14.72
13.82
15.55
14.52
13.99
15.40
14.06
13.90
14.66
13.93
13.71
15.16
13.53
15.15
16.15
NDVI scores, Conditions scores,
and regional dummy variables
16.74
15.97
14.17
12.23
15.60
15.42
14.98
12.59
14.88
15.10
14.04
11.80
14.61
16.14
13.72
11.38
14.71
16.09
13.55
11.06
14.46
15.93
13.43
10.78
14.37
15.82
13.95
10.88
14.16
15.59
13.80
11.76
12.59
17.07
13.40
11.30
17.24
20.99
14.92
12.49
16.69
20.44
14.65
13.24
17.13
20.25
14.82
13.99
Condition Score Only
16.55
19.61
13.78
14.74
Periods in this table represent bi-weekly satellite periods beginning January 1st.
Period 13=1-Jul, 14=15-Jul, 16=12-Aug, 17=26-Aug, 18=9-Sept, 19=23-Sept
9.74
10.13
10.93
10.74
10.38
10.48
10.38
14.39
13.50
15.08
15.87
15.41
15.89
18
15.66
15.46
16.16
16.38
15.79
15.99
16.55
16.37
16.08
16.53
16.74
16.15
16.21
16.71
12.69
12.60
12.90
12.68
12.12
11.97
11.66
14.86
14.43
16.43
16.29
16.12
15.94
14.70
16.05
16.07
17.13
16.83
12.52
12.94
12.95
12.76
12.16
11.96
11.80
14.61
14.21
15.76
16.30
15.67
16.88
19
15.41
15.17
16.28
16.53
16.07
16.20
16.22
16.37
15.95
16.93
17.19
16.82
16.87
17.02
11.28
10.58
10.80
10.53
10.49
10.42
10.13
12.69
12.54
14.15
13.72
13.54
13.34
12.77
14.01
14.27
16.58
16.57
16.48
10.72
10.37
10.01
9.90
9.57
9.58
9.67
12.24
12.28
12.95
13.22
12.78
16.03
24
Table 3. Percent RMSFE for Illinois Corn Yield Prediction
First Satellite Period Used in The Forecast Model
NDVI and crop conditions scores
NDVI scores with
Regional Dummy NDVI scores only
Variables
Models
1
2
3
4
5
6
7
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
1
2
3
4
5
6
7
8
9
10
11
12
13
26.28
25.77
25.30
25.01
24.87
23.95
23.21
27.99
27.40
26.89
26.36
26.08
25.32
25.65
27.45
26.93
26.52
25.96
25.64
24.70
24.32
23.23
22.83
21.87
18.55
18.14
14
26.80
26.19
25.73
25.37
25.22
24.40
24.29
28.01
27.31
26.84
26.21
26.03
25.25
25.07
26.54
25.87
25.61
25.19
24.65
23.85
23.46
22.27
21.98
20.61
15.91
15.49
15.09
14.70
Forecast Periods
15
16
17
28.03
28.70
29.01
27.46
28.14
28.43
26.99
27.77
27.86
26.61
27.39
27.52
26.45
27.22
27.28
25.73
27.02
27.04
25.84
26.52
26.47
31.34
31.11
31.49
30.59
30.41
30.77
30.18
30.11
30.14
29.64
29.56
29.53
29.02
28.89
28.93
28.42
28.87
28.81
28.55
28.66
28.21
18.11
17.79
21.09
17.76
17.43
20.68
17.53
17.22
20.36
17.17
17.08
19.92
18.62
18.89
21.16
19.30
19.33
21.56
18.08
17.58
19.84
16.80
16.37
18.36
16.63
16.42
18.51
15.40
15.18
17.44
13.42
13.00
15.31
12.61
12.45
14.24
12.50
12.27
14.08
12.04
11.75
13.11
12.18
12.25
13.65
11.66
12.89
14.33
NDVI scores, Conditions scores,
and regional dummy variables
28.31
22.01
17.10
18.96
27.33
21.26
16.54
18.42
26.88
22.30
16.29
18.11
26.30
22.03
15.97
18.02
26.03
21.28
17.29
19.51
25.19
20.42
18.42
19.80
24.91
18.52
16.91
18.02
24.49
21.06
15.96
17.32
23.95
21.08
15.98
17.05
24.57
21.48
15.02
15.92
20.00
17.65
14.05
14.31
19.50
16.81
13.04
13.78
17.86
16.02
12.90
14.11
Condition Score Only
Periods in this table represent bi-weekly satellite periods beginning January 1st.
Period 13=1-Jul, 14=15-Jul, 16=12-Aug, 17=26-Aug, 18=9-Sept, 19=23-Sept
23.03
22.37
21.80
21.13
22.39
22.37
20.53
19.52
19.14
17.59
16.10
15.20
15.14
18
29.15
28.57
27.95
27.73
27.50
27.21
26.90
31.73
31.02
30.33
29.78
29.33
29.16
29.01
17.33
17.13
17.05
16.68
18.04
18.53
17.20
15.92
16.29
15.63
14.37
13.45
13.29
12.07
11.90
11.35
12.41
13.32
16.63
16.33
16.18
15.70
16.88
17.38
16.14
15.16
15.05
14.42
14.41
13.51
13.75
19
25.31
24.78
24.66
24.58
24.40
24.36
24.63
26.46
26.18
25.75
25.38
25.12
25.42
26.70
13.96
13.72
13.29
13.04
13.79
14.25
13.24
12.43
12.78
12.33
11.62
11.15
10.97
10.52
11.10
10.76
11.75
12.81
12.19
15.61
15.21
14.87
14.61
14.81
15.12
14.12
13.79
13.59
13.01
12.76
11.97
13.59
25
Table 4. Percent RMSFE for Iowa Soybean Yield Prediction
First Satellite Period Used in The Forecast Model
NDVI and crop conditions scores
NDVI scores with
Regional Dummy NDVI scores only
Variables
Models
1
2
3
4
5
6
7
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
1
2
3
4
5
6
7
8
9
10
11
12
14
24.11
23.69
23.39
23.12
22.54
23.00
22.45
26.08
24.46
23.91
23.72
23.55
24.23
23.74
25.00
24.13
24.24
23.97
23.20
23.26
22.91
21.63
21.13
21.18
20.92
20.59
20.31
21.40
15
23.87
23.57
23.18
23.05
22.59
23.14
22.61
25.81
24.29
23.67
23.67
23.67
24.41
23.94
24.93
24.64
24.72
24.75
23.96
24.29
24.06
22.49
22.05
21.91
21.74
21.59
21.52
21.81
21.19
Forecast Periods
16
17
23.05
20.10
22.51
19.54
22.27
19.12
22.12
18.88
21.67
18.67
21.95
18.59
21.45
18.67
24.05
19.98
22.46
19.07
21.81
18.67
22.01
18.65
22.09
18.25
22.50
18.45
22.08
18.95
22.77
22.81
22.18
22.31
22.69
21.61
22.50
21.17
21.90
21.31
22.07
20.98
21.55
20.87
20.46
19.56
20.11
18.30
19.71
17.61
19.75
17.30
19.67
17.41
19.80
17.24
20.21
19.04
19.04
17.41
18.14
15.70
15.40
NDVI scores, Conditions scores,
and regional dummy variables
27.34
28.07
24.30
23.17
25.49
26.39
22.84
22.33
24.88
25.46
22.44
22.07
24.53
25.76
22.61
21.96
24.34
25.64
22.45
21.61
24.41
26.07
22.79
21.13
23.90
25.40
22.14
21.15
22.78
24.03
21.33
20.32
22.24
23.60
20.98
18.88
22.93
23.71
20.95
18.55
22.45
23.10
20.65
18.38
21.41
22.64
20.38
18.04
20.35
19.64
17.87
15.38
Condition Scores Only
Periods in this table represent bi-weekly satellite periods beginning January 1st.
Period 13=1-Jul, 14=15-Jul, 16=12-Aug, 17=26-Aug, 18=9-Sept, 19=23-Sept
18
18.40
18.33
17.91
17.74
17.38
17.38
17.45
18.78
18.10
17.56
17.27
16.73
17.04
17.54
16.85
16.88
16.33
16.07
16.18
16.02
15.77
15.34
13.79
13.04
12.80
13.04
12.81
13.63
12.01
10.85
10.70
10.44
17.72
17.65
17.10
16.87
16.91
16.56
16.31
15.97
14.19
13.64
13.37
13.31
10.64
19
18.47
18.35
17.91
17.73
17.36
17.31
17.16
18.81
17.95
17.56
17.24
16.79
17.06
17.13
17.52
17.33
16.66
16.36
15.82
15.74
15.58
15.09
14.31
13.43
13.13
13.55
13.37
13.02
11.76
11.17
10.72
10.38
10.20
18.31
17.94
17.60
17.38
17.03
16.72
16.50
15.92
15.04
14.44
13.80
14.10
10.19
26
Table 5 Percent RMSFE for Illinois Soybean Yield Prediction
First Satellite Period Used in The Forecast Model
NDVI and crop conditions scores
NDVI scores with
Regional Dummy NDVI scores only
Variables
Models
1
2
3
4
5
6
7
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
1
2
3
4
5
6
7
8
9
10
11
12
14
15.24
15.09
14.87
14.59
14.53
14.17
13.74
16.22
16.05
15.60
15.26
15.22
14.89
14.50
16.01
15.91
15.61
15.82
15.66
15.27
14.76
14.32
14.43
15.26
13.58
13.62
13.12
12.72
15
16.00
15.84
15.58
15.15
15.04
14.65
14.17
17.13
16.93
16.47
16.00
15.90
15.54
15.12
15.91
15.77
15.55
14.87
14.82
14.54
14.08
13.68
13.78
14.92
13.41
13.18
12.99
12.43
12.34
Forcast Periods
16
17
15.05
15.16
14.76
14.89
14.54
14.68
14.29
14.31
14.20
14.23
14.05
14.07
13.47
13.52
15.56
15.75
15.03
15.25
14.66
14.85
14.54
14.61
14.45
14.52
14.42
14.45
13.78
13.85
14.43
14.84
14.36
14.98
14.27
14.80
13.92
14.67
14.19
14.80
13.85
14.55
13.17
13.98
12.95
13.62
12.80
13.46
13.81
14.44
12.41
13.27
13.19
13.86
12.89
13.61
12.11
12.59
11.91
12.16
11.29
11.66
12.14
NDVI scores, Conditions scores,
and regional dummy variables
16.38
18.03
17.35
17.45
16.21
17.85
16.94
17.09
15.59
17.59
16.74
16.70
16.09
16.57
16.09
16.56
15.94
16.47
15.98
16.48
15.57
16.08
15.34
16.06
15.15
15.60
14.29
15.31
15.10
15.30
14.18
15.21
15.39
15.49
14.04
15.09
16.08
16.14
15.02
15.89
15.04
15.18
13.89
15.13
14.28
13.74
13.89
14.54
Condition Score Only
12.55
11.85
11.67
11.85
Periods in this table represent bi-weekly satellite periods beginning January 1st.
Period 13=1-Jul, 14=15-Jul, 16=12-Aug, 17=26-Aug, 18=9-Sept, 19=23-Sept
18
14.89
14.88
14.58
14.16
14.01
13.73
13.22
14.36
14.27
14.05
13.52
13.41
13.33
13.04
11.49
11.93
11.78
11.50
11.79
11.85
11.38
11.16
11.04
11.04
10.87
12.52
12.41
11.63
11.52
11.10
11.28
11.06
11.72
12.07
12.16
11.60
11.77
11.57
11.28
11.27
11.20
11.22
11.04
13.95
11.30
19
16.14
15.47
15.11
14.69
14.45
14.05
13.56
14.57
13.62
13.39
13.13
12.96
12.89
12.89
11.74
11.31
11.22
11.07
11.02
10.91
10.49
10.82
10.66
10.51
10.86
12.79
12.60
11.91
11.74
11.32
11.28
10.90
10.55
12.38
11.41
11.56
11.42
11.20
10.95
11.03
11.38
11.33
10.94
11.03
15.43
10.89
27
Examining the RMSFE percent of each of the models it becomes apparent that the
yield predictions generally tend to improve towards the end of the season, and that the
scores improve when ground level information, in the form of crop condition scores, is
added. It does not appear that adding regional dummy variables improves the predictions.
For corn, the lowest RMSFE percent occurs during the 16th period (August 12th) in Iowa
and the 17th period (August 26th) in Illinois. The lowest RMSFE result from Model 4.3
which contains crop conditions scores and NDVI scores. This model appears to be the
most useful. Running a simple F-test on an in-sample version of this model indicates that
the crop condition score and NDVI variables are jointly significant at the 99% level, with
F statistics of approximately 30 across the different states and commodities.
There is an interesting difference between the Iowa and Illinois crop condition
scores. It appears as though the Illinois crop condition scores contain more information
than the Iowa scores. The RMSFE from model 4.5, which contains only the condition
scores is much lower in Illinois than in Iowa. Furthermore the difference between the
RMSFE in Models 4.1 and Models 4.3 is much larger for Illinois than Iowa. One of the
goals of running multiple models was to determine if the need for the inclusion of early
vs. late season satellite images varied based upon when the prediction is made. For corn
in Iowa, it appears as though the later the prediction the earlier the satellite periods that
should be included in the model. For Illinois, prior to the August 26th prediction, the
lowest RMSFE occurs when only the last satellite image and the crop condition score is
used in the model. The three predictions after August 26th have the lowest RMSFE when
28
they contain the satellite scores from period 16 (August 26th) and up, and the most recent
crop condition scores. For corn it appears that the satellite imagery scores from Iowa
contain more information than those from Illinois, as the NDVI only model (4.1) for Iowa
results in considerably lower RMSFE that Illinois.
In the case of soybeans the opposite is true. For Illinois, the NDVI only model
(4.1) results in considerably lower forecast errors than the Iowa, NDVI only model.
Illinois condition scores, for all but the final two periods contain more information than
the Iowa condition scores as the RMSFEs for model 4.5 (condition scores only) are much
lower. The forecast errors for both corn and soybeans are similar, however due to their
later growing season soybeans have very high RMSFEs for the early season predictions
compared to corn. Furthermore for soybeans in both Iowa and Illinois the forecast errors
are lowest when only the most recent satellite images are included. Based upon the
RMSFE the optimal model for predicting soybean yields would involve the use of the
most recent crop condition scores and the very last satellite image.
In all but the late season forecasts the condition scores alone are the best yield
predictors. It is apparent that the inclusion of such ground information dramatically
increases the predictive power of the forecast model. Figures 2-8 provide a graphical
representation of the importance of ground information. These charts examine Iowa corn
yield predictions for each of the predictive periods. USDA yields are compared to three
distinct forecast models: NDVI scores alone, NDVI scores with crop condition scores,
and crop condition scores alone (models 4.1, 4.3, 4.5). The NDVI scores on their own
can vary dramatically from the USDA yields, but with the addition of the crop condition
29
scores the forecast errors decrease significantly. By the end of the season the NDVI
scores with the crop condition scores forecast better than the other two models. The
findings are similar for Illinois corn, and Iowa and Illinois soybeans.
The value of satellite imagery for trading purposes appears questionable, given
that the publicly available crop conditions scores have a lower RMSFE than the satellite
imagery models. However, a final conclusion regarding the optimal model can not be
determined by RMSFE alone as it is an averaging process. Large profits or losses will
result from the tail events. The true value of the different models can only be determined
by evaluating the final profit or loss from trading on them.
185
Yield (bu/harvested acres)
165
145
125
105
USDA yield
7-14NDVI
7-14NDVI & Cond. Scores
cond. Score only
85
65
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
Figure 2. Iowa Corn – Prediction for Period 13 (1-Jul)
30
185
Yield (harvested acre)
165
145
125
105
85
USDA yield
7-14NDVI
65
7-14NDVI & Cond. Scores
cond. Score only
45
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
Figure 3. Iowa Corn – Prediction for Period 14 (15-July)
185
165
Yield (harvested acres)
145
125
105
85
USDA yield
7-14NDVI
65
7-14NDVI & Cond. Scores
cond. Score only
45
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
Figure 4. Iowa Corn – Prediction for Period 15 (29-July)
31
185
165
Yield (harvested acres)
145
125
105
85
USDA yield
7-14NDVI
65
7-14NDVI & Cond. Scores
cond. Score only
45
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
Figure 5. Iowa Corn – Prediction for Period 16 (12-August)
185
165
Yield (harvested acres)
145
125
105
85
USDA yield
7-14NDVI
65
7-14NDVI & Cond. Scores
cond. Score only
45
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
Figure 6. Iowa Corn – Prediction for Period 17 (26-August)
32
185
165
Yield (harvested acres)
145
125
105
85
USDA yield
7-14NDVI
65
7-14NDVI & Cond. Scores
cond. Score only
45
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
Figure 7. Iowa Corn – Prediction for Period 18 (9-September)
185
165
Yield (harvested acres)
145
125
105
85
USDA yield
7-14NDVI
65
7-14NDVI & Cond. Scores
cond. Score only
45
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
Figure 8. Iowa Corn – Prediction for Period 19 (23-September)
33
CHAPTER 5
TESTING THE VALUE OF YIELD PREDICTIONS
The Data
The state level yield predictions must be converted into a national production
estimate so that a comparison can be made between the market’s expected production and
the satellite imagery driven production forecasts. A number of different data sources were
used in this process.
To calculate a production estimate some measure of harvested acres must be used.
To make the production forecasts out-of-sample the measure of harvested acres must be
available prior to the actual forecast. NASS produces a supplemental acreage report on or
around June 30th of each year. This report contains estimates of planted and harvested
acres. The estimates are based on farmer surveys from the first two weeks of June, trend
predictions, and changing weather factors. There is a .8 and 1.4 percent root mean square
error for corn and soybeans respectively and the 90% confidence intervals are plus or
minus 1.4 and 2.4 percent. (NASS Crop Production – Acreage – Supplement) This report
is the earliest subjective estimate of harvested acreage from USDA. Prior WASDE
reports contain acreage estimates but they are based only on trend models.
The final state level production estimates must be turned into a national estimate.
A relationship between the sum of Iowa and Illinois production was calculated using the
historical production numbers of Iowa, Illinois and the US from the NASS website.
(USDA)
34
The market’s expectation of production is derived by finding a relationship
between stocks to use and the current futures price. The World Agricultural Supply and
Demand Expectations (WASDE) report is the main component of this model. The USDA
produces WASDE reports every month in the growing season on or near the 11th of the
month. These reports contain information on USDA’s expectation of supply (beginning
stocks, production, and imports) and demand (domestic use and exports). For each report
a stock to use ratio can be calculated as:
(5.1)
SU= (total supply- total demand) / total demand
Futures prices are used in both the calculation of the market’s production
expectation and in the final evaluation of profit or loss. The December Chicago Board of
Trade contract is used for corn and the November contract is used for soybeans.
The Model
Efficient markets should only react to unexpected information. The first step in
valuing a production estimate based upon satellite imagery, therefore, is to determine
what the market expects. Should the satellite imagery prediction suggest that the market
expectation is not in line with reality, then an opportunity for a profitable trade becomes
likely. If a trader can consistently profit by having access to satellite imagery yield
predictions then the information has value.
35
To decipher the market’s expectation of production for a given year the
relationship between price and stocks to use is determined. The following model is used
to arrive at this relationship:
(5.2)
LnP = α + β LnS
where LnP is the log of the average November futures price for the December contract
from the 13th to the 17th. This time period was selected because reports are usually
released on the 10th-12th of the month. LnS is the log of the stocks to use ratio from the
November WASDE report. Figure 9 and 10 provide a graphical display of the
relationship between the log of November stocks-to-use and the futures price.
6
5.9
Ln Nov Futures Price
5.8
5.7
5.6
5.5
5.4
5.3
5.2
5.1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
Ln Nov SUR
Figure 9. Relationship Between Corn Stocks-To-Use Ratio and Futures Price (19892003)
36
6.9
Ln Nov Futures Prices
6.8
6.7
6.6
6.5
6.4
6.3
6.2
6.1
6
1.5
2
2.5
3
3.5
4
Ln Nov SUR
Figure 10. Relationship Between Soybean Stocks-To-Use Ratio and Futures Price
(1989-2003)
The stocks to use ratio is calculated as:
(5.3)
SU=(Total Supply – Total Use) / Total Use
Total supply is the sum of beginning stocks, production, and imports. Total use is the sum
of exports and domestic use. By November the market has a good understanding of the
final condition of that year’s crop. Traders are buying and selling based on a wider set of
information, and therefore the relationship between stocks to use and price should contain
less noise and be more reliable than at points earlier in year. β is calculated for each year,
using the previous ten years of data. The intercept term tends to vary over time as market
conditions such as exchange rates shift. Rather than attempting to model all of these
potential shifters it is preferable to discard the intercept term that is calculated above, and
instead calibrate the intercept each year by using the stocks to use ratio and the futures
37
price from the previous month. For example, in order to predict the stocks to use ratio for
corn in July of 2000, data from November of 1990 to 1999 is used to determine β from
equation 5.2. In this example α was 6.503 and β equaled -.366. The intercept term from
equation 5.2 is ignored and a new intercept is calculated by using the stocks to use ratio
and futures price from the June WASDE report. For 2000 the recalibrated intercept was
6.505. Using equation 5.2 with the adjusted α it is possible to calculate the markets
expected stocks to use ratio by inserting the future price on July 1 of the December
contract into the formula. In 2000 the futures price was 191.25 which results in a stocks
to use ratio of 30.43%.
With an expected stocks to use ratio in hand it is possible to return to the most
recent WASDE report, June 10th of 2000 in this example, and calculate expected
production. The expected stocks-to-use ratio of .3043 is multiplied by reported total use,
9.985 billion bushels to arrive at estimated ending stocks of 3,038 billion bushels. Ending
stocks are then added to total use in order to calculate total supply, 13.023 billion bushels.
Total supply is the sum of beginning stocks, 1.621 billion bushels and imports, 15 million
bushels and production. Subtracting beginning stocks and imports from total supply we
are left with the markets estimate of production, 9,360 billion bushels.
After securing the market’s production expectation it is necessary to derive a
national forecast from the state level satellite imagery forecasts. Using simple OLS the
relationship between National production and the sum of Iowa and Illinois production is
established.
(5.4)
USprodt = α + β(IAprodt+ILprodt) + εt
38
In order to simulate the most realistic scenario, and to avoid in-sample errors, the alpha
and beta coefficients for time t were estimated using production data from the previous
19 years. For example, the relationship in 1989 was calculated using production data
from 1970-1988. For each additional year the 19 years of data rolled forward by one year.
The sum of Iowa and Illinois production explains a great deal of the variation in US
production. Table 6 displays the results of equation 5.4.
Table 6. Results of Regressing IA and IL Production on U.S. Production
Corn
Soybeans
2
Year
Coeficient
Adj R
Coeficient
Adj R2
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2.5316
2.4945
2.4903
2.4646
2.4526
2.3689
2.3837
2.3544
2.4045
2.5283
2.5342
2.6091
2.6377
2.5693
2.5502
0.9446
0.9379
0.9426
0.9442
0.9548
0.9504
0.9716
0.9736
0.9634
0.9432
0.9469
0.9580
0.9634
0.9354
0.9388
3.2005
3.1049
2.9780
2.8043
2.7027
2.5901
2.5410
2.2419
2.2812
2.4741
2.5207
2.5195
2.5911
2.7273
2.7309
0.8834
0.8586
0.8257
0.8139
0.8265
0.7622
0.8150
0.7619
0.7833
0.8350
0.9052
0.9188
0.9248
0.9498
0.9486
All of the coefficients were significant at the 99% level, and the adjusted R2 from
the corn model ranged from .9379 to a high of .9736. The relationship is weaker for
39
Soybeans with adjusted R2 ranging from .7619 to .9498. It appears that the relationship
has strengthened over the past five years.
It is not surprising that the total corn production in Iowa and Illinois explains so
much of US corn production. Figures 11 and 12 illustrate where corn and soybeans are
produced. Iowa and Illinois combined produce 32% of US corn, and no other state
produces as much corn as Iowa or Illinois. Furthermore, the remainder of the corn
production is in nearby states with similar climates and growing conditions. The adjusted
R2 for soybeans is noticeably lower, despite that fact that on average 35% of soybean
production is in Iowa and Illinois. This is likely due to the fact that soybeans are grown in
more geographically diverse regions of the United States.
IA+IL
32%
Other
32%
WI
4%
NE
11%
MN
9%
IN
9%
MI
3%
Figure 11. Percent of Total Corn Production by State
40
All Others
28%
IA + IL
35%
SD
2%
NE
4%
MO
7%
MN
8%
KS
2%
AR
IN 5%
9%
Figure 12. Soybean Production by State
.
Arkansas, for instance accounts for 5% of production and because of differences
in climate Iowa and Illinois production is not likely to be a good predictor of production
in Arkansas.
Futures Trading Strategy
The market’s production expectation can be compared with satellite based
production expectations. If these two predictions differ sufficiently a trade could be
initiated. If the satellite imagery production prediction is higher than the market’s
expectation and if it is correct then there will be a larger supply and price will fall. A
short position in futures would be taken in order to profit from the satellite imagery
41
model. Conversely, if the satellite production predictions indicate that production will be
lower than the market expects a long position in futures would be taken.
The simplest type of trade is to sell when it appears that supply will be higher than
expectation and buy when supply appears lower than expectation. For each of the models
the profit or loss from such a trade was calculated. The trades were entered into the day
after the last day of each two week satellite imagery period, beginning July 1st for corn
and July 15th for Soybeans. If there was no trading on that day the price from the next
available trading day was used. Short term and long term trading scenarios were
examined. For the short term, the positions were closed after two weeks, and in the long
term scenario the positions were closed on November 15th.
An alternative, more selective, trading strategy was also tested. If the difference
between the satellite imagery production estimate and the market’s production estimate
was greater than 5% of the market’s production estimate then a trade was made. If the
difference was not greater than 5% then no trade was made for that year. This method
should have decreased the number of losing trades, because a trade would only be entered
into if there was a very strong signal that the market was out of line.
Tables 18-30 in the appendix display detailed profit and loss information from the
various trading strategies. As there are multiple strategies and multiple models, an
examination of this data is worthwhile. Table 7 and 8 shows the profit and loss from
trading the non-selective strategy, using the model containing NDVI scores and crop
condition scores. The use of period 13 as the beginning period in these tables is
somewhat arbitrary, but it allows for a comparison between corn and soybeans. At
42
different points in time models with a later or earlier beginning period were more
profitable. It is also important to note that the model that contained the lowest RMSFE
was very rarely the model that resulted in the largest profits. Due to the convex
relationship between stocks to use and price it is possible that if a yield model is
primarily wrong when the stocks to use ratio is high, then the price impact of this error
may be small. However, even small errors, when supply is tight may result in large price
impacts.
Table 7. Corn Profit/Loss and Standard Deviation Summary
Prediction Time Model-w/crop cond. scores - held 2 weeks Average Profit
1-Jul
15-Jul
29-Jul
12-Aug
26-Aug
9-Sep
23-Sep
Period 13:13
Period 13:14
Period 13:15
Period 13:16
Period 13:17
Period 13:18
Period 13:19
-192
19
56
-90
103
-378
188
Standard Deviation
831
867
537
277
364
545
400
Table 8. Soybean Profit/Loss and Standard Deviation Summary
Prediction Time
15-Jul
29-Jul
12-Aug
26-Aug
9-Sep
23-Sep
Model-w/crop cond. scores – held 2
weeks
Period 13:14
Period 13:15
Period 13:16
Period 13:17
Period 13:18
Period 13:19
Average
Profit
417
258
-343
255
278
395
Standard Deviation
1736
1397
759
990
1007
1304
For corn, when the beginning satellite period used is the 13th, the results are
mixed. Larger profits and losses occur at the end of the year. When examining all
43
possible starting periods (see appendix ) profit levels tend to be higher later in the year,
but as in the case of the September 9th forecast period, in table 5.2 above, some periods
still had negative profits. The standard deviation of the profit is so large that the trader
would have to risk significant losses if he were to make trades at this time. Soybeans also
display positive profits. Surprisingly, some of the largest profits appear early in the
season with trades made on July 15th. Profit vacillates during the season, and is again
high at the end of the season. On average, if there is a possibility to make profitable
trades in both corn and soybeans using satellite imagery it would likely be at the end of
the growing season. Average profit from September 23rd corn predictions from all of the
different possible models and beginning time periods ranges from approximately $200$800. For soybeans profits, examining all of the possible models and beginning time
periods, are as high as $1100 during the same time period. While the standard deviation
of profit is still high the risk-return tradeoff might be bearable if risk was managed using
options.
The results indicate that there is a higher possibility of profit from trading
soybeans than corn. This is unexpected, because U.S. soybean production is a smaller
percent of the world production than is corn.
One hypothesis was that the short term trading strategy would be more profitable
than the long term strategy. If the satellite imagery predictions are close to the USDA’s
forecasts, then a trade that is entered near the first of the month, before the WASDE
report becomes available, would see some positive price movement when similar
production numbers are released two weeks later. Late season droughts, disease, or even
44
shifts in demand could not be forecasted with the early satellite prediction. Therefore,
holding a position through November would increase the probability of losing money on
the trade. For corn, this proves true for all but the last two prediction period of September
9th and September 23rd. For soybeans the opposite appears true. For most periods there is
more profit if positions are held until November.
Examining the average profit and loss and the standard deviation of corn and
soybeans does not provide a complete picture of the possible risk and returns from such
an investment. A look at the distribution of the profit and loss is useful. F display
histograms of profit from trading futures in corn and soybeans for the last two trading
periods. (September, 9 and September, 23)
The distribution of trading profit for corn and soybeans illustrates that trading
soybeans could be more profitable, but also with higher risks. It is interesting that the
selective strategy appears to under perform the simple strategy. Using the selective
strategy decreases the number of profitable trades without significantly reducing the
number of large losses.
It is possible that if there was plentiful rain early in the season, and a late drought
after the NDVI images are examined, the GIS forecast could be off considerably, causing
large losses. By using the selective strategy this could exacerbate the problem because
there would be greater reliance on these losing trades.
45
4.5
4
3.5
Frequency
3
2.5
2
1.5
1
0.5
0
-1000 -750 -500 -250
0
250
500
750 1000 1250 1500 1750 2000 2250 M ore
Profit
Figure 13. Dollars of Corn Futures Trading Profit (modeled using NDVI and crop
cond. scores, periods 15-18) December Contract Held to Nov 15 - Simple Scenario
4.5
4
3.5
Frequency
3
2.5
2
1.5
1
0.5
0
-1000 -750 -500 -250
0
250
500
750 1000 1250 1500 1750 2000 2250 M ore
Profit
Figure 14. Dollars of Corn Futures Trading Profit (modeled using NDVI and crop
cond. scores, periods 15-19) December Contract Held to Nov 15- Simple Scenario
46
4.5
4
3.5
Frequency
3
2.5
2
1.5
1
0.5
0
-1000 -750 -500
-250
0
250
500
750
1000 1250 1500 1750 2000 2250 More
Profit
Figure 15. Dollars of Corn Futures Trading Profit (modeled using NDVI and crop
cond. scores, periods 15-18) December Contract Held to Nov 15- Selective Scenario
3.5
3
Frequency
2.5
2
1.5
1
0.5
0
-1000 -750 -500 -250
0
250
500
750 1000 1250 1500 1750 2000 2250 M ore
Profit
Figure 16. Dollars of Corn Futures Trading Profit (modeled using NDVI and crop
cond. scores, periods 15-19) December Contract Held to Nov 15- Selective Scenario
47
6
5
Frequency
4
3
2
1
0
-3000
-2000
-1000
0
1000
2000
3000
4000
5000
6000
7000
M ore
Profit
Figure 17. Dollars of Soybean Futures Trading Profit (modeled using NDVI and crop
cond. scores, periods 15-18) November Contract Held to Nov 1- Simple Scenario
3.5
3
Frequency
2.5
2
1.5
1
0.5
0
-3000
-2000
-1000
0
1000
2000
3000
4000
5000
6000
7000
More
Profit
Figure 18. Dollars of Soybean Futures Trading Profit (modeled using NDVI and crop
cond. scores, periods 15-18) November Contract Held to Nov 1- Simple Scenario
48
6
5
Frequency
4
3
2
1
0
-3000
-2000
-1000
0
1000
2000
3000
4000
5000
6000
7000
More
Profit
Figure 19. Dollars of Soybean Futures Trading Profit (modeled using NDVI and crop
cond. scores, periods 15-18) November Contract Held to Nov 1- Selective Scenario
3.5
3
Frequency
2.5
2
1.5
1
0.5
0
-3000
-2000
-1000
0
1000
2000
3000
4000
5000
6000
7000
More
Profit
Figure 20. Dollars of Soybean Futures Trading Profit (modeled using NDVI and crop
cond. scores, periods 15-19) November Contract Held to Nov 1- Selective Scenario
49
Option Trading Strategy
For both corn and soybeans much of the average profit is made up of tail events in
the distribution, large profits or large losses. A trading strategy that involves buying a put
or call option would eliminate a large portion of the downside risk, as the maximum loss
would be limited to the price of the call or put option. To that effect a simple option
trading strategy was also tested. The summary of profits and losses as well as the
histograms indicate that the only likely time that options would be profitable would be
late in the season. Therefore the strategy was tested on the last three periods, August 26th,
September 9th and September 24th for both corn and soybeans. Using the same
information as the futures scenario, where previously a futures contract was shorted, a put
option was purchased. Conversely, if the data suggested purchasing a futures contract, a
call option was purchased instead. For both calls and puts the nearest out-of-the-money
option was purchased. The profit from selling the options in either two weeks or on
October 15th was calculated. Tables 9-14 list the profits from the option trades.
As expected, simulation option trades rather than futures trades significantly
reduced the negative tail events. However, in the case of corn, it still does not appear that
it would be profitable to trade based on the information. Profits were mostly negative,
and when profits were positive, their magnitudes were not sufficient to offset transaction
costs. Soybeans appear more promising. Profits were almost always positive, even as
early as Aug 26th. The occurrence of profits increased on September 9th and Sept 24th but
the magnitudes declined. This is primarily a function of the declining cost of the options
as they move closer to expiration.
50
Table 9. Average Corn Option Trading Profit - Forecast on Aug 26
NDVI scores only
-83
-83
-83
-83
-83
-83
-83
176
176
176
176
176
176
176
-44
-44
-44
-44
-44
-44
-44
476
476
476
476
476
476
476
-93
-93
-99
-105
-105
-105
-117
187
187
198
190
190
190
195
-106
-106
-105
-103
-103
-103
-103
395
395
433
412
412
412
433
NDVI scores with
Regional Dummy
Variables
1
2
3
4
5
6
7
-85
-83
-85
-83
-83
-83
-83
175
176
175
176
176
176
176
-145
-44
-145
-44
-44
-44
-44
381
476
381
476
476
476
476
-93
-93
-77
-31
-45
-45
-53
187
187
194
135
135
135
141
-106
-106
-149
-56
-59
-59
-55
395
395
429
333
314
314
333
NDVI and crop conditions scores
Std. Dev.
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
-37
-37
-37
-37
12
12
12
10
12
10
10
10
10
-8
-58
-58
-8
200
200
200
200
216
216
216
217
216
217
217
217
217
221
200
200
221
-10
-10
-10
-10
-57
-57
-57
-158
-57
-158
-158
-158
-158
-141
-95
-95
-141
485
485
485
485
483
483
483
386
483
386
386
386
386
383
397
397
383
-125
-74
-71
-68
-68
-19
-63
-54
-85
9
-41
-56
-1
27
-1
-4
2
208
199
192
205
205
211
231
207
216
149
212
219
146
127
146
138
148
-227
-201
-213
-192
-192
-70
-114
-119
-220
-56
-148
-127
-53
-5
-53
-53
-83
414
376
344
378
378
463
512
428
383
337
432
453
335
322
335
316
255
1
2
3
4
5
6
7
8
9
10
11
12
12
12
12
-15
-15
-15
-18
10
7
-20
-20
-20
216
216
216
212
212
212
212
217
218
212
212
212
-57
-57
-57
-99
-99
-99
-118
-158
-176
-218
-218
-218
483
483
483
475
475
475
478
386
387
363
363
363
-124
-80
-73
-73
-73
-17
8
-2
45
45
1
15
246
214
221
221
221
227
145
156
202
182
245
259
-219
-222
-213
-213
-213
-73
30
-20
-58
-107
-167
-117
487
401
404
404
404
500
373
342
339
322
415
446
-141
383
-24
215
-104
432
Condition Score
-8
221
Beginning Period of The Forecast Model
1
Selective Trades
Oct-15th
2 Weeks
Std. Dev.
Profit
Std. Dev.
Profit
NDVI scores, Conditions
scores, and regional dummy
variables
Simple Scenario
Holding Period
2 Weeks
Oct-15th
Week1 Profit
Std. Dev.
Profit
51
Table 10. Average Corn Option Trading Profit - Forecast on Sept 9
NDVI scores only
-37
-37
-37
-75
-37
-37
-37
334
334
334
328
334
334
334
-95
-95
-95
-194
-95
-95
-95
405
405
405
253
405
405
405
-40
-150
-150
-168
-188
-150
-150
319
232
232
205
202
232
232
-145
-111
-111
-182
-209
-111
-111
250
407
407
249
250
407
407
NDVI scores with
Regional Dummy
Variables
1
2
3
4
5
6
7
-75
-75
-75
-75
-37
-37
-37
328
328
328
328
334
334
334
-194
-194
-194
-194
-95
-95
-95
253
253
253
253
405
405
405
-188
-150
-150
-150
-188
-150
-150
202
232
232
232
202
232
232
-209
-111
-111
-111
-209
-111
-111
250
407
407
407
250
407
407
NDVI and crop conditions scores
Std. Dev.
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
-74
-74
-62
-74
-74
-161
-141
-161
-161
-161
-161
-161
-161
-127
-74
-40
-127
-161
322
322
323
322
322
208
217
208
208
208
208
208
208
243
322
337
243
208
-110
-110
-149
-110
-110
-112
-105
-112
-112
-112
-112
-112
-112
-80
-110
-79
-80
-112
393
393
383
393
393
394
393
394
394
394
394
394
394
405
393
404
405
394
-136
-150
-138
-138
-150
-62
-130
-130
-150
-150
-150
-150
-150
-184
-96
-184
-62
-184
219
232
237
237
232
335
239
239
232
232
232
232
232
190
318
190
335
190
-98
-111
-149
-149
-111
-109
-104
-104
-111
-111
-111
-111
-111
-143
-141
-143
-109
-143
392
407
398
398
407
407
407
407
407
407
407
407
407
394
393
394
407
394
1
2
3
4
5
6
7
8
9
10
11
12
-161
-161
-161
-161
-161
-161
-141
-161
-170
-100
-208
-208
208
208
208
208
208
208
217
208
203
241
163
163
-112
-112
-112
-112
-112
-112
-105
-112
-128
18
-226
-226
394
394
394
394
394
394
393
394
395
553
220
220
-150
-150
-150
-150
-150
-150
-130
-130
-140
-113
-117
-150
232
232
232
232
232
232
239
239
235
205
245
232
-111
-111
-111
-111
-111
-111
-104
-104
-95
-66
-63
-111
407
407
407
407
407
407
407
407
406
378
488
407
-112
394
-178
207
-193
254
Condition Score
-161
208
Beginning Period of The Forecast Model
1
Selective Trades
2 Weeks
Nov-15th
Std. Dev.
Profit
Std. Dev.
Profit
NDVI scores, Conditions
scores, and regional dummy
variables
Simple Scenario
Holding Period
2 Weeks
Nov-15th
Week1 Profit
Std. Dev.
Profit
52
Table 11. Average Corn Option Trading Profit - Forecast on Sept 23
NDVI scores only
-3
-3
-12
-12
-12
-12
-3
185
185
186
186
186
186
185
10
10
11
11
11
11
10
302
302
301
301
301
301
302
-51
-46
-72
-60
-74
-74
-57
134
129
126
136
133
133
136
-30
-40
-101
-99
-108
-108
-38
251
243
154
171
162
162
269
NDVI scores with
Regional Dummy
Variables
1
2
3
4
5
6
7
-3
-3
-12
-12
-12
-3
-19
185
185
186
186
186
185
184
10
10
11
11
11
10
-11
302
302
301
301
301
302
305
-17
-17
-6
11
11
-6
-6
196
196
212
215
215
212
212
8
8
33
54
54
33
33
324
324
346
358
358
346
346
NDVI and crop conditions scores
Std. Dev.
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
2
2
2
2
3
3
3
3
3
-7
-7
3
3
3
-8
-47
-46
-46
-46
179
179
179
179
178
178
178
178
178
181
181
178
178
178
181
125
125
125
125
34
34
34
34
18
18
18
18
18
12
12
18
18
18
28
-35
-50
-50
-50
292
292
292
292
303
303
303
303
303
305
305
303
303
303
295
230
238
238
238
-44
-27
-44
-15
12
11
-48
-37
-24
-21
-21
-21
-21
-44
-32
-19
-15
-13
-13
132
135
132
150
141
142
144
124
127
137
137
137
137
142
129
128
124
128
128
-10
2
-10
30
48
34
-10
-22
1
-13
-13
-13
-13
-23
-72
-54
-55
-10
-10
276
263
276
286
303
314
295
283
293
324
324
324
324
302
179
171
171
270
270
1
2
3
4
5
6
7
8
9
10
11
12
-7
-7
-7
-7
-7
-7
-16
-7
-7
-7
-7
-7
181
181
181
181
181
181
186
181
181
181
181
181
12
12
12
12
12
12
-2
12
12
12
12
12
305
305
305
305
305
305
309
305
305
305
305
305
-19
-19
-19
-19
-5
-21
-19
-53
9
1
-59
-60
141
141
141
141
144
134
142
122
199
220
116
131
26
26
26
26
42
16
15
-99
53
-19
-98
-95
268
268
268
268
281
260
276
174
323
325
166
177
-64
239
-13
128
-10
270
Condition Score
-55
130
Beginning Period of The Forecast Model
1
Selective Trades
2 Weks
Nov-15th
Std. Dev.
Profit
Std. Dev.
Profit
NDVI scores, Conditions
scores, and regional dummy
variables
Simple Scenario
Nov-15th
Holding Period
2 Weks
Std. Dev.
Profit
Week1 Profit
53
Table 12. Average Soybean Option Trading Profit - Forecast on Aug 26
NDVI scores only
81
125
125
31
31
31
31
628
626
626
560
560
560
560
-81
377
377
382
382
382
382
1055
1943
1943
1941
1941
1941
1941
-133
-133
-133
-37
-37
-37
-12
461
461
461
370
370
370
387
154
154
154
301
301
301
397
1172
1172
1172
1141
1141
1141
1180
NDVI scores with
Regional Dummy
Variables
1
2
3
4
5
6
7
-25
-25
-25
3
125
3
153
523
523
523
528
549
528
545
381
381
381
328
535
328
482
1945
1945
1945
1956
1904
1956
1920
-41
-41
-37
-37
-37
-37
-37
349
349
370
370
370
370
370
118
118
301
301
301
301
301
1222
1222
1141
1141
1141
1141
1141
NDVI and crop conditions scores
Std. Dev.
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
81
81
81
120
120
81
120
165
165
165
136
92
92
214
258
258
258
628
628
628
624
624
628
624
619
619
619
623
626
626
642
630
630
630
-81
-81
-81
-138
-138
-81
-138
320
320
320
373
-86
-86
387
845
845
845
1055
1055
1055
1074
1074
1055
1074
1967
1967
1967
1956
1077
1077
1261
1947
1947
1947
-119
-119
-119
-119
-119
-119
34
34
34
34
34
34
34
318
318
316
316
488
488
488
488
488
488
668
668
668
668
668
668
668
639
639
677
677
172
172
172
172
172
172
176
176
176
176
176
176
176
369
369
506
506
1242
1242
1242
1242
1242
1242
1171
1171
1171
1171
1171
1171
1171
1141
1141
1119
1119
1
2
3
4
5
6
7
8
9
10
11
12
-41
-70
-41
-41
-70
-70
53
92
92
92
92
92
520
512
520
520
512
512
628
626
626
626
626
626
-130
-78
-130
-130
-78
-78
-29
-86
-86
-86
-86
-86
1058
1061
1058
1058
1061
1061
1056
1077
1077
1077
1077
1077
2
-127
-119
-119
-119
-119
-119
-10
-13
-10
-10
34
584
438
488
488
488
488
488
386
416
386
386
668
-36
1
172
172
172
172
172
339
248
339
339
176
1229
1222
1242
1242
1242
1242
1242
1214
1281
1214
1214
1171
845
1947
316
677
506
1119
Condition Score
258
630
Beginning Period of The Forecast Model
1
Selective Trades
2 Weks
Nov-15th
Std. Dev.
Profit
Std. Dev.
Profit
NDVI scores, Conditions
scores, and regional dummy
variables
Simple Scenario
Nov-15th
Holding Period
2 Weks
Std. Dev.
Profit
Week1 Profit
54
Table 13. Average Soybean Option Trading Profit - Forecast on Sept 9
NDVI scores only
138
2
49
93
93
2
-59
618
653
623
586
586
653
627
810
429
514
861
861
429
323
1788
1699
1653
1753
1753
1699
1687
-38
19
-71
-74
-118
-27
-58
301
599
585
583
608
556
641
288
497
200
463
115
548
221
912
1857
1668
1876
1697
1832
1722
NDVI scores with
Regional Dummy
Variables
1
2
3
4
5
6
7
94
94
94
94
94
94
2
656
656
656
656
656
656
653
463
463
463
463
463
463
429
1679
1679
1679
1679
1679
1679
1699
35
-118
-118
-118
-118
-118
-25
653
608
608
608
608
608
630
255
115
115
115
115
115
150
1707
1697
1697
1697
1697
1697
1684
NDVI and crop conditions scores
Std. Dev.
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
94
2
2
49
49
2
2
49
49
49
49
93
93
29
29
29
29
29
656
653
653
623
623
653
653
623
623
623
623
586
586
610
610
610
610
610
463
429
429
514
514
429
429
514
514
514
514
861
861
863
863
863
863
863
1679
1699
1699
1653
1653
1699
1699
1653
1653
1653
1653
1753
1753
1752
1752
1752
1752
1752
4
-25
-118
-71
-118
-71
-118
-71
-118
-118
-118
-74
-118
-182
-118
-118
-118
-118
643
630
608
585
608
585
608
585
608
608
608
583
608
608
608
608
608
608
210
150
115
200
115
200
115
200
115
115
115
463
115
118
115
115
115
115
1730
1684
1697
1668
1697
1668
1697
1668
1697
1697
1697
1876
1697
1696
1697
1697
1697
1697
1
2
3
4
5
6
7
8
9
10
11
12
94
2
2
-19
-19
2
2
2
2
2
2
93
656
653
653
656
656
653
653
653
653
653
653
586
463
429
429
343
343
429
429
429
429
429
429
861
1679
1699
1699
1703
1703
1699
1699
1699
1699
1699
1699
1753
2
-25
-118
-138
-118
-138
-118
-118
-118
-118
-118
-28
586
630
608
606
608
606
608
608
608
608
608
577
584
150
115
29
115
29
115
115
115
115
115
568
1810
1684
1697
1684
1697
1684
1697
1697
1697
1697
1697
1900
863
1752
-182
608
118
1696
Condition Score
29
610
Beginning Period of The Forecast Model
1
Selective Trades
2 Weeks
Nov-15th
Std. Dev.
Profit
Std. Dev.
Profit
NDVI scores, Conditions
scores, and regional dummy
variables
Simple Scenario
Nov-15th
Holding Period
2 Weeks
Std. Dev.
Profit
Week1 Profit
55
Table 14. Average Soybean Option Trading Profit - Forecast on Sept 23.
NDVI scores only
-80
-80
-80
-80
-80
-80
26
531
531
531
531
531
531
585
95
95
95
95
95
95
369
1141
1141
1141
1141
1141
1141
1378
17
-143
-159
-159
-159
-143
-143
604
240
246
246
246
240
240
298
-85
-114
-114
-114
-85
-85
1352
521
539
539
539
521
521
NDVI scores with
Regional Dummy
Variables
1
2
3
4
5
6
7
-50
57
57
57
57
57
57
579
624
624
624
624
624
624
70
343
343
343
343
343
343
1154
1394
1394
1394
1394
1394
1394
-217
-174
-209
-14
-195
-174
-156
237
229
247
612
232
229
255
-185
-109
-148
253
-144
-109
-68
521
542
530
1363
563
542
606
NDVI and crop conditions scores
Std. Dev.
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
55
-98
-66
-66
55
55
88
-66
88
-66
-66
88
88
88
237
237
237
237
237
632
491
481
481
632
632
616
481
616
481
481
616
616
616
569
569
569
569
569
343
40
90
90
343
343
393
90
393
90
90
393
393
393
499
499
499
499
499
1393
1028
1015
1015
1393
1393
1372
1015
1372
1015
1015
1372
1372
1372
1319
1319
1319
1319
1319
-40
26
26
26
-76
-76
26
20
25
45
45
-53
-53
-6
98
96
96
96
96
528
501
501
501
392
392
501
457
477
495
495
370
370
360
473
449
449
449
449
120
184
184
184
-128
-128
184
201
186
231
231
-35
-35
-95
244
323
323
323
323
1169
1148
1148
1148
527
527
1148
1049
1095
1136
1136
524
524
359
1126
1100
1100
1100
1100
1
2
3
4
5
6
7
8
9
10
11
12
25
-98
88
88
88
88
88
88
88
88
88
55
638
491
616
616
616
616
616
616
616
616
616
630
293
40
393
393
393
393
393
393
393
393
393
286
1413
1028
1372
1372
1372
1372
1372
1372
1372
1372
1372
1374
-148
-148
-148
-102
-148
-148
-40
13
13
33
35
-66
409
409
409
406
409
409
528
499
499
521
552
388
-198
-198
-198
-160
-198
-198
120
188
188
238
241
-149
531
531
531
548
531
531
1169
1148
1148
1198
1270
379
196
978
96
449
323
1100
Condition Score
84
474
Beginning Period of The Forecast Model
1
Selective Trades
2 Weeks
Nov-15th
Std. Dev.
Profit
Std. Dev.
Profit
NDVI scores, Conditions
scores, and regional dummy
variables
Simple Scenario
Nov-15th
Holding Period
2 Weeks
Std. Dev.
Profit
Week1 Profit
56
Table 15. Returns From Trading November Soybean Options
Average Option Cost
Aug. 26
700
Purchase Date
Sept. 9
650
Sept. 26
465
Average Profit - Held to Oct. 15
845
863
499
Less Assumed Trading Costs
50
50
50
114%
125%
97%
Average Return
Modeled using NDVI scores starting in period 15 and current crop condition scores
Looking at the histograms of profit and loss it would appear that more complex
option trading strategies may even result in larger profits, or lower risk. Figures 21-28
display profit histograms from trading options. The dispersion of profits would indicate
that a bear or bull spread trading strategy might be useful.
If the satellite images indicated that the option price would increase then, a call
option at a high strike price would be sold, and a call option at a lower strike price would
be purchased. Selling the call would cap profits, but the proceeds from the sale would
finance the purchase of the other option, limiting the potential loss. A similar strategy
could be used if it appeared that the market price would fall. A put option could be sold
with a low strike price and a put option with a slightly higher strike price would be
purchased.
A trader’s willingness to pay is based on the additional profits that he would hope
to earn by obtaining the satellite information. The trader is makes his trading decisions in
a world of uncertainty, and while the satellite information can potentially decrease this
uncertainty, it can by no means eliminate it.
57
7
6
Frequency
5
4
3
2
1
0
-750
-500
-250
0
250
500
750
1000
1250
More
Profit
Figure 21. Dollars of Corn Option Trading Profit (modeled using NDVI and crop cond.
scores, periods 15-18) December Contract Held to Oct 15- Simple Scenario
10
9
8
Frequency
7
6
5
4
3
2
1
0
-750
-500
-250
0
250
500
750
1000
1250
More
Profit
Figure 22. Dollars of Corn Option Trading Profit (modeled using NDVI and crop cond.
scores, periods 15-19) December Contract Held to Oct 15- Simple Scenario
58
7
6
Frequency
5
4
3
2
1
0
-750
-500
-250
0
250
500
750
1000
1250
More
Profit
Figure 23. Dollars of Corn Option Trading Profit (modeled using NDVI and crop cond.
scores, periods 15-18) December Contract Held to Oct 15- Selective Scenario
6
Frequency
5
4
3
2
1
0
-750
-500
-250
0
250
500
750
1000
1250
More
Profit
Figure 24. Dollars of Corn Option Trading Profit (modeled using NDVI and crop cond.
scores, periods 15-19) December Contract Held to Oct 15- Selective Scenario
59
4
Frequency
3
2
1
0
-500
-250
0
250
500
750
1000
1250
1500
2000
3000
More
Profit
Figure 25. Dollars of Soybean Option Trading Profit (modeled using NDVI and crop cond.
scores, periods 15-18) November Contract Held to Oct 15- Simple Scenario
4.5
4
Frequency
3.5
3
2.5
2
1.5
1
0.5
0
-500
-250
0
250
500
750
1000
1250
1500
2000
3000
More
Profit
Figure 26. Dollars of Soybean Option Trading Profit (modeled using NDVI and crop cond.
scores, periods 15-19) November Contract Held to Oct 15- Simple Scenario
60
7
6
Frequency
5
4
3
2
1
0
-500
-250
0
250
500
750
1000
1250
1500
2000
3000
More
Profit
Figure 27. Dollars of Soybean Option Trading Profit (modeled using NDVI and crop cond.
scores, periods 15-18) November Contract Held to Oct 15- Selective Scenario
3.5
3
Frequency
2.5
2
1.5
1
0.5
0
-500
-250
0
250
500
750
1000
1250
1500
2000
3000
More
Profit
Figure 28. Dollars of Soybean Option Trading Profit (modeled using NDVI and crop cond.
scores, periods 15-19) November Contract Held to Oct 15- Selective Scenario
61
The profit the trader derives in the initial state of uncertainty can be empirically estimated
by calculating the profits from trading options, based upon model 3.5, which only
contains the most current crop condition scores released by the USDA. This information
is widely available and can be obtained at little or no cost to the trader. The average profit
a trader could earn, per option, using this model, is displayed in table 16.
Table 16. Willingness to Pay for Satellite Forecast, Based on Option Profit
Corn
Forecast Period 17
-Satellite Based
-Crop Condition Scores
Willingness to Pay
Forecast Period 18
-Satellite Based
-Crop Condition Scores
Willingness to Pay
Forecast Period 19
-Satellite Based
-Crop Condition Scores
Willingness to Pay
Average Profit
Soybeans
-95
-141
0
845
845
0
-79
-112
0
863
863
0
28
-64
92
499
196
303
The crop cond. score plus NDVI model, held to Oct 15th, simple trading scenario was used.
For the satellite based profit figure, the optimal time frame model was used.
The trader can improve his production forecast by including NDVI scores in his
production forecasting model. The inclusion of satellite information is costly and the
trader would only be willing to incur this cost if the information results in a profitable
trade that at least offsets these costs. Table 16 lists the optimal level of profit that could
be obtained by buying puts or calls and holding them until October 15th. It is apparent
that under these circumstances a trader would only be willing to pay for satellite imagery
62
that allows him to forecast production in period 19. For corn the estimated willingness to
pay is $92. However, this figure is questionable. While the satellite imagery results in a
more profitable trade, the profit of $28 would likely be consumed by trading costs
ranging from $15 to $50. Trading costs would make the trade unprofitable and so despite
the fact that the model results in less of a loss, a trader would not be willing to pay for the
information. In the case of soybeans the willingness to pay for a risk neutral trader would
be positive even after paying transaction costs.
The willingness to pay is calculated in per option terms. Obviously a trader that
could risk more capital and trade in higher volumes would be willing to pay more for the
data as their profit potential will increase linearly with the number of options they trade.
Table 17 illustrates what a trader would be willing to pay to move between states
of imperfect information. Perfect information is not a possibility, but it is useful to model
the trading results of having perfect production information. USDA final production
estimates can be used as a proxy for perfect production information. Assuming a trader
knew final Iowa and Illinois production numbers at time periods 17-19, the trader would
be able to calculate final U.S. production numbers using model 5.4. The U.S. production
estimate could be compared to the market’s expectation derived from the stocks-to-use
ratio.
This perfect production information scenario was tested. It appears that having
perfect information does not result in exceptionally larger average trading profit. Table
5.33 compares option trading profits under the three different levels of information.
63
Profit, while still negative, is slightly higher for corn trades in period 17 and 18. In period
19 trading profit is actually highest from using satellite imagery based forecasts. For
soybeans, using USDA final production numbers results in an additional $62 of profit. In
periods 18 and 19 average trading profit is the same using satellite imagery or USDA
production numbers.
Table 17. Option Trading Profit With Different Levels of Information
Corn
Average Profit
Soybeans
Forecast Period 17
Imperfect Information - Crop Condition Scores
Improved Imperfect Information - Satellite Based
Perfect Information - Final USDA Production
-141
-95
-28
845
845
907
Forecast Period 18
Imperfect Information - Crop Condition Scores
Improved Imperfect Information - Satellite Based
Perfect Information - Final USDA Production
-112
-79
-73
863
863
861
Forecast Period 19
Imperfect Information - Crop Condition Scores
Improved Imperfect Information - Satellite Based
Perfect Information - Final USDA Production
-64
28
-41
196
499
499
The crop cond. score plus NDVI model, held to Oct 15th, simple trading scenario was used.
For the satellite based profit figure, the optimal time frame model was used.
64
CHAPTER 6
CONCLUSIONS AND QUALIFICATIONS
Considerable resources are being spent on crop production forecasts. Many of the
forecasts rely on subjective surveys or expensive ground research. A systematic and
unbiased forecast that can be produced earlier and cheaper than other forecasting methods
would be valuable. Remote sensing is one such forecast, and there have been a variety of
studies that have indicated that satellite base forecasts contain valuable information.
Natural resource scientists have been making yield forecasts based on satellite
images for some time. While they strive to improve the scientific accuracy of their
forecasts by testing different vegetative indexes and cloud filters, there may be some
statistical improvements that could result in an improved forecast.
This thesis used the established technology to generate out of sample predictions
based upon non-cumulative, bi-weekly NDVI scores. In order to assess the value of the
satellite information, the production predictions were used to implement a trading
strategy. If trades based on satellite information were profitable then a trader would have
a positive willingness to pay for the information. Assuming that corn and soybean futures
and options markets are efficient then only unexpected information will influence the
market. To determine the market’s current production expectation a simple model
establishing the relationship between futures price and the stocks-to-use ratio was used.
This model resulted in the market’s expectation of production. If satellite based forecasts
65
indicated that the markets expectations were out of line then an appropriate futures
position was entered into.
The results of this trading strategy indicate that trading futures in both corn and
soybeans based on this information would be extremely risky. While mean profit levels
are positive late in the season the standard deviation is very large. It is important to note
that this risk becomes apparent when analyzing the production forecasts in an out of
sample forecast. Previous in sample work may have made diminished the possibility for
large losses, making a trading strategy appear more profitable than it really was.
In order to eliminate some of the downside risk a simple option trading strategy
was tested. With options the losses are mitigated and the possibility for a profitable trade
increases. It is not surprising, considering the efficiency of the corn market, that even
with options there was very little chance for profit in corn. While there was profit late in
the season it would have been eroded by transaction costs, making the value of satellite
information to a trader low. Profits did appear feasible in soybeans late in the season.
These profits were on the same level as those that could be had by making trades based
on final state level production numbers, or in other words near perfect information.
One qualification to this work is the fact that while soybeans appear to be the
more profitable commodity to trade in, the recent relationship between the futures price
and the stocks to use ratio is much weaker than that of corn. The breakdown of this
relationship is likely due to South American soybean production becoming a much larger
66
share of world production in the past several years.3 However when a South American
stocks to use variable was added to the regression the coefficient was insignificant and R2
decreased. Assuming that the relationship between soybean futures prices and the U.S.
stocks to use ratio will only get weaker, the model will have to be modified in order to
accurately estimate the futures market’s production expectation.
Another qualification and avenue for additional research is the option trading
strategy used. The option trading strategy was very simple. It involved buying a call or
put based on the production forecast being higher or lower than market expectation. A
more complex options strategy involving spreads may have made the trades even more
profitable, and more research on different option strategies is warranted.
These findings indicate that the satellite images contain valuable information and
they will be useful to traders, assuming appropriate risk controls are in place. While
satellite image based forecasts can not replace USDA forecasts; they deserve more
attention and research. Their main value may be found in developing nations where an
accurate USDA type forecast is unavailable. Considering that the images contain
information, it is now a question of how to best harness and use this information
productively.
3
R2 dropped from .74 in 1999 to .16 in 2003. Removing trading profit and loss from 2000 to 2003 still
resulted in profitable trades for soybeans.
67
REFERENCES
Baur, R. F., Orazem, P. F. (1994) “The rationality and price effects of USDA forecasts of
oranges.” Journal of Finance, 49: 681-696.
Benedetti, R., Rossini, P., (1993): “On the Use of NDVI Profiles as a Tool for
Agricultural Statistics: The Case Study of Wheat Yield Estimate and Forecast in Emilia
Romagna,” Remote Sensing of Environment, 45: 311-326.
Carter, C. A., Galopin, C. (1993) “Informational content of government hogs and pigs
reports,” Journal of Agricultural Economics, 75: 711-718.
Colling, P. L., Scott, I. H. (1990): “The Reaction of Live Hog Futures Prices to USDA
Hogs and Pigs Report,” American Agricultural Economics Association, 72: 84-94.
Egelkraut, T. M., Garica, P., Irwin, S. H., Good, D. L (2003): “An Evaluation of Crop
Forecast Accuracy for Corn and Soybeans: USDA and Private Information Agencies”
Journal of Agricultural and Applied Economics, 35: 79-95.
Fackler, P. L., Norwood, B. (2000): “Forecasting Crop Yields with Condition Indices,”
Working Paper, North Carolina State University.
Fair, R. C., Shiller, R. J. (1989): “The Information Content of Ex Ante Forecasts,”
Review of Economics and Statistics, 71: 325-331.
Fama, E. F. (1965): "Random Walks in Stock Market Prices," Financial Analysts Journal,
September/October
Fama, E. F. (1970): “Efficient Capital Markets: A Review of Theory and Empirical
Work,” Journal of Finance, 25: 383-417.
Fortenbery, T. R., Summner, D. A. (1993): “The Effects of USDA Reports on in Futures
and Options Markets,” The Journal of Futures Markets, 13: 157-174.
Garcia, P., Irwin, S. H., Leuthold, R. M., Yang, L. (1997) “The Value of Public
Information in Commodity Futures Markets,” Journal of Economic Behavior and
Organization, 32: 559-570.
Hatfield, J. L. (1983): “Remote Sensing Estimators of Potential and Actual Crop Yield,”
Remote Sensing of Environment, 13: 301-311.
Hayami, Y., Peterson, W. (1972): “Social Returns to Public Information Services:
68
Statistical Reporting of U.S. Farm Commodities,” American Economic Review, 62: 119130.
Henriksson, R. D., Merton, R. C. (1981) “On Market Timing and Investment
Performance II. Statistical Procedures for Evaluating Forecasting Skill,” Journal of
Business, 54: 513-533.
Just, R. E., Rausser, G. C. (1981): “Commodity Price Forecasting with Large-Scale
Econometric Models and the Futures Market,” American Journal of Agricultural
Economics, 63: 197-208.
Lillesand, T. M., Kiefer, R. W. (1994): Remote Sensing and Image Interpretation, 3rd ed.
John Wiley and Sons, Inc., New York.
McNew, K. P., Espinosa, J. A. (1994): “The Informational Content of USDA Crop
Reports: Impact on Uncertainty and Expectations in Grain Futures Markets,” The Journal
of Futures Markets, 14: 475-492.
McNew, K. P., Musser, W. N. (2002): “Farmer Forward Pricing Behavior: Evidence
from Marketing Clubs,” Agricultural and Resource Economics Review, 31: 200-210.
McNew, K.P. (1997): “Valuing El Nino Weather Forecasts In Storable Commodity
Markets: The Case of U.S. Corn” A preliminary report submitted to NOAA/U.S.
Department of Commerce
Quarmby, N. A., Milnes, M., Hidle, T. L., Silleos, N. (1993): “The Use of Multi-temporal
NDVI Measurements from AVHRR Data for Crop Yield Estimation and Prediction,”
International Journal of Remote Sensing, 14: 199-210.
Sellers, P. J. (1986): “Canopy Reflectance, Photosynthesis and Transpiration,”
International Journal of Remote Sensing, 6: 13-35.
Summner, D. A., Mueller, R. A. E. (1989): “Are Harvest Forecasts news? USDA
announcements and Futures market reactions,” American Journal of Agricultural
Economics, 71: 1-8
Tucker, C. J. (1979): “Red and Photographic Infrared Linear Combinations for
Monitoring Vegetation,” Remote Sensing of Environment, 8: 127-150.
Tucker, C. J., Holben, B.N., Elgin, J. H., McMurtrey, J. E., (1981): “Remote Sensing of
Total Dry Matter Accumulation in Winter Wheat,” Remote Sensing of Environment, 11:
171-189.
69
U.S. Department of Agriculture, Office of Budget and Program Analysis(2005): FY 2006
Budget Summary. Washington DC.
U.S. Department of Agriculture, statistical database.
http://www.nass.usda.gov/QuickStats/
Wiegand, C. L., Richardson, A. J., Escobar, D. E., Gerbermann, A. H. (1991):
“Vegetation Indices in Crop Assessments,” Remote Sensing of Environment, 35: 105-119.
Wright, B. D., Williams, J. C., (1982): “The Economic Role of Commodity Storage” The
Economic Journal, 92: 596-614.
Zhen, C. (2001): “Celestial Satellite and Earthly Crop yield: Informational Content of
Satellite-Based Crop Yield Forecasts,” A Master Thesis in the Department of Applied
Economics, Montana State University, Bozeman, MT.
70
APPENDIX A:
FUTURES TRADING PROFIT
71
Table 18. Average Corn Futures Trading Profit - July 1
NDVI scores only
-242
-242
-242
-242
-242
-242
-63
817
817
817
817
817
817
852
-458
-458
-458
-458
-458
-458
-78
2012
2012
2012
2012
2012
2012
2066
-238
-238
-295
-295
-195
-195
-236
816
816
837
837
860
860
851
-777
-777
-920
-920
-633
-633
-706
1925
1925
1967
1967
2095
2095
2119
NDVI scores with
Regional Dummy
Variables
1
2
3
4
5
6
7
-242
-242
-242
-242
-242
-242
-63
817
817
817
817
817
817
852
-458
-458
-458
-458
-458
-458
-78
2012
2012
2012
2012
2012
2012
2066
-295
-295
-295
-295
-295
-249
-277
837
837
837
837
837
808
819
-920
-920
-920
-920
-920
-694
-575
1967
1967
1967
1967
1967
2011
2057
NDVI and crop conditions scores
Std. Dev.
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
10
11
12
13
-150
-150
-150
-150
-242
-242
-242
-242
-242
-348
-192
-192
-192
841
841
841
841
817
817
817
817
817
775
831
831
831
-556
-556
-556
-556
-458
-458
-458
-458
-458
-756
-381
-381
-381
1985
1985
1985
1985
2012
2012
2012
2012
2012
1913
2029
2029
2029
-123
-123
-123
-123
-123
-123
-23
-75
-75
-239
-313
-313
-313
805
805
805
805
805
805
829
854
854
916
895
895
895
-365
-365
-365
-365
-365
-365
-270
-370
-370
-851
-886
-886
-886
2029
2029
2029
2029
2029
2029
2204
2297
2297
2350
2219
2219
2219
1
2
3
4
5
6
7
8
9
10
11
12
-150
-150
-150
-150
-150
-150
-242
-242
-242
-323
-100
-192
841
841
841
841
841
841
817
817
817
787
849
831
-556
-556
-556
-556
-556
-556
-458
-458
-458
-951
-479
-381
1985
1985
1985
1985
1985
1985
2012
2012
2012
1818
2007
2029
-210
-210
-210
-210
-210
-210
-23
-205
-147
-297
-289
-253
839
839
839
839
839
839
829
856
872
834
878
928
-543
-543
-543
-543
-543
-543
-270
-594
-520
-851
-879
-829
2060
2060
2060
2060
2060
2060
2204
2197
2289
2106
2218
2345
-381
2029
-296
907
-985
2117
Condition Score
-192
831
Beginning Period of The Forecast Model
1
Selective Trades
2 Weeks
Nov-15th
Std. Dev.
Profit
Std. Dev.
Profit
NDVI scores, Conditions
scores, and regional dummy
variables
Simple Scenario
2 Weeks
Nov-15th
1
Profit
Std. Dev.
Profit
Week
72
Table 19. Average Corn Futures Trading Profit - July 15
Simple Scenario
2 Weeks
Holding Period
Week
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
-229
-229
-229
-289
-289
-289
-289
834
834
834
814
814
814
814
-396
-396
-396
-216
-216
-216
-216
1609
1609
1609
1646
1646
1646
1646
-288
-288
-249
-203
-249
-249
-129
905
905
939
977
939
939
896
-375
-375
-438
-625
-438
-438
-461
1673
1673
1740
1713
1740
1740
1832
NDVI scores with
Regional Dummy
Variables
1
2
3
4
5
6
7
-289
-289
-289
-289
-289
-289
-289
814
814
814
814
814
814
814
-216
-216
-216
-216
-216
-216
-216
1646
1646
1646
1646
1646
1646
1646
-249
-249
-203
-203
-203
-249
-103
939
939
977
977
977
939
881
-438
-438
-625
-625
-625
-438
-470
1740
1740
1713
1713
1713
1740
1822
NDVI and crop conditions scores
1
2
3
4
5
6
7
8
9
10
11
12
13
14
-171
-171
-171
-171
-156
-216
-216
-351
-351
-174
-41
19
19
-136
849
849
849
849
852
838
838
787
787
848
866
867
867
856
-384
-384
-384
-384
-194
-14
-14
-44
-44
-659
-453
-633
-633
-261
1612
1612
1612
1612
1649
1661
1661
1660
1660
1514
1593
1526
1526
1639
-182
-129
-129
-215
-265
-133
-133
-339
-339
-339
-116
2
-116
-215
868
896
896
1035
988
951
951
860
860
860
1017
1065
1017
997
-391
-461
-461
-569
-369
-388
-388
-864
-864
-864
-731
-711
-731
-675
1754
1832
1832
1807
1818
1927
1927
2039
2039
2039
1925
1930
1925
1808
1
2
3
4
5
6
7
8
9
10
11
12
-211
-18
-211
-211
-211
-18
-78
-256
-256
-176
18
-41
839
867
839
839
839
867
863
826
826
848
867
866
-456
-429
-456
-456
-456
-429
-249
-86
-86
-468
-441
-453
1592
1600
1592
1592
1592
1600
1641
1658
1658
1589
1597
1593
-175
-175
-268
-218
-178
-178
-95
-84
75
-84
-115
-115
841
841
881
919
886
886
887
722
702
722
992
992
-333
-333
-243
-429
-249
-249
-251
-59
88
-59
-630
-630
1928
1928
1908
1925
1810
1810
1908
851
915
851
1711
1711
33
866
-251
1640
-115
992
-630
1711
Condition Score
1
Nov-15th
NDVI scores, Conditions
scores, and regional dummy
variables
1
Selective Trades
2 Weeks
Nov-15th
Beginning Period of The Forecast Model
73
Table 20. Average Corn Futures Trading Profit - July 29
Simple Scenario
2 Weeks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
23
23
39
39
39
39
39
540
540
539
539
539
539
539
-167
-167
73
73
73
73
73
1365
1365
1374
1374
1374
1374
1374
10
10
106
121
106
106
106
546
546
456
478
456
456
456
-88
-88
-189
-423
-189
-189
-189
1366
1366
1385
1210
1385
1385
1385
1
2
3
4
5
6
7
39
39
39
39
39
39
39
539
539
539
539
539
539
539
73
73
73
73
73
73
73
1374
1374
1374
1374
1374
1374
1374
121
121
121
121
121
121
76
478
478
478
478
478
478
451
-423
-423
-423
-423
-423
-423
-368
1210
1210
1210
1210
1210
1210
1303
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
39
39
39
39
39
39
39
39
39
39
39
56
56
56
159
539
539
539
539
539
539
539
539
539
539
539
537
537
537
515
-517
-517
-517
-517
-517
-517
-517
-517
-517
-517
-517
-277
-277
-277
-102
1268
1268
1268
1268
1268
1268
1268
1268
1268
1268
1268
1346
1346
1346
1372
70
82
132
158
132
82
101
158
183
183
214
214
214
214
214
583
540
537
559
537
540
562
559
537
587
619
619
619
619
619
-396
-565
-544
-419
-544
-565
-452
-419
-293
-747
-763
-763
-763
-763
-763
1400
1337
1400
1409
1400
1337
1341
1409
1400
1135
1212
1212
1212
1212
1212
1
2
3
4
5
6
7
8
9
10
11
12
39
39
39
101
39
39
39
39
39
39
101
118
539
539
539
530
539
539
539
539
539
539
530
526
-517
-517
-517
-412
-517
-517
-517
-517
-517
-517
-412
-172
1268
1268
1268
1309
1268
1268
1268
1268
1268
1268
1309
1365
158
158
158
158
132
132
88
158
85
61
183
156
559
559
559
559
537
537
538
559
464
501
587
560
-419
-419
-419
-419
-544
-544
-467
-419
-74
-436
-747
-696
1409
1409
1409
1409
1400
1400
1280
1409
1079
746
1135
1082
Condition Score
168
512
-18
1376
214
619
-763
1212
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weeks
Nov-15th
1
Beginning Period of The Forecast Model
74
Table 21. Average Corn Futures Trading Profit - Aug 12
Simple Scenario
2 Weeks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
30
30
30
30
30
30
30
291
291
291
291
291
291
291
11
11
11
11
11
11
11
408
408
408
408
408
408
408
50
50
39
44
22
63
19
280
280
241
255
236
266
250
-23
-23
-93
347
68
246
-145
1671
1671
1613
866
1620
1663
1552
1
2
3
4
5
6
7
30
30
30
30
30
30
80
291
291
291
291
291
291
280
11
11
11
11
11
11
23
408
408
408
408
408
408
407
8
8
11
5
26
26
22
251
251
266
241
250
250
265
-225
-225
5
133
328
328
289
1592
1592
784
621
735
735
769
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
-33
-33
-33
-22
-13
-13
-33
-2
-22
-2
-70
-90
-90
-90
-90
-90
290
290
290
291
292
292
290
292
291
292
283
277
277
277
277
277
-16
-16
-16
1
54
54
-16
71
1
71
-16
-86
-86
-86
-86
-86
407
407
407
408
404
404
407
401
408
401
407
398
398
398
398
398
-10
-52
-92
-63
-63
-63
-10
-52
-64
-80
-69
-52
-113
-117
-104
-86
271
270
275
283
283
283
271
270
253
325
292
270
291
269
295
265
-488
-639
-738
-621
-621
-621
-488
-639
-494
-55
-721
-639
-821
-509
-264
43
1850
1736
1631
1726
1726
1726
1850
1736
1659
1061
1887
1736
1743
1839
1007
1100
1
2
3
4
5
6
7
8
9
10
11
12
-27
-47
-47
-35
-27
-77
-47
-15
-35
-15
-83
-103
291
288
288
290
291
281
288
292
290
292
279
272
-79
-26
-26
-9
-79
-161
-26
61
-9
-63
-149
-96
399
407
407
408
399
372
407
403
408
403
377
395
-91
-91
-52
-10
-10
-10
-64
-91
-102
-102
-92
-92
273
273
270
271
271
271
260
273
293
293
256
256
-817
-817
-639
-488
-488
-488
-664
-817
-355
-355
-789
-789
1684
1684
1736
1850
1850
1850
1664
1684
1148
1148
1578
1578
Condition Score
-33
290
-16
407
-117
269
-653
1683
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weks
Nov-15th
1
Beginning Period of The Forecast Model
75
Table 22. Average Corn Futures Trading Profit - Aug 26
Simple Scenario
2 Weeks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
-97
-97
-97
-97
-97
-97
-97
366
366
366
366
366
366
366
4
4
4
4
4
4
4
1487
1487
1487
1487
1487
1487
1487
-109
-109
-73
-94
-94
-94
-113
360
360
355
360
360
360
345
-38
-38
-77
45
45
45
4
1389
1389
1255
1348
1348
1348
1339
1
2
3
4
5
6
7
-105
-97
-105
-97
-97
-97
-97
363
366
363
366
366
366
366
-221
4
-221
4
4
4
4
1470
1487
1470
1487
1487
1487
1487
-109
-109
-23
26
5
5
-14
360
360
300
228
244
244
231
-38
-38
-190
79
201
201
160
1389
1389
1156
451
636
636
627
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
10
10
10
10
112
112
112
103
112
103
103
103
103
62
-40
-40
62
379
379
379
379
361
361
361
364
361
364
364
364
364
374
377
377
374
-138
-138
-138
-138
-364
-364
-364
-589
-364
-589
-589
-589
-589
-346
-119
-119
-346
1481
1481
1481
1481
1439
1439
1439
1357
1439
1357
1357
1357
1357
1444
1482
1482
1444
-28
-53
-19
-22
-22
19
-18
-6
-30
61
13
-8
48
31
48
51
63
245
240
263
246
246
250
253
276
260
203
288
275
192
183
192
192
214
-209
-373
-347
-274
-274
-173
-261
-217
-308
-100
-369
-248
-48
-59
-48
-18
-33
1110
1149
1160
1132
1132
1166
1152
1185
1131
708
1237
1172
592
589
592
606
584
1
2
3
4
5
6
7
8
9
10
11
12
112
112
112
45
45
45
40
103
98
32
32
32
361
361
361
376
376
376
377
364
365
378
378
378
-364
-364
-364
-454
-454
-454
-516
-589
-651
-741
-741
-741
1439
1439
1439
1411
1411
1411
1388
1357
1326
1275
1275
1275
-48
-55
-24
-24
-24
17
30
40
91
120
63
25
231
239
246
246
246
250
161
191
267
269
347
334
-251
-404
-305
-305
-305
-204
8
22
-92
-283
-483
-372
1093
1131
1116
1116
1116
1154
485
523
686
843
1278
1223
Condition Score
62
374
-346
1444
33
306
-273
1212
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weeks
Nov-15th
1
Beginning Period of The Forecast Model
76
Table 23. Average Corn Futures Trading Profit - Sept 9
Simple Scenario
2 Weeks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
-100
-100
-100
-192
-100
-100
-100
663
663
663
641
663
663
663
101
101
101
-116
101
101
101
1417
1417
1417
1416
1417
1417
1417
-134
-347
-347
-421
-438
-347
-347
671
567
567
509
495
567
567
82
-271
-271
-406
-488
-271
-271
1419
1393
1393
1339
1328
1393
1393
1
2
3
4
5
6
7
-192
-192
-192
-192
-100
-100
-100
641
641
641
641
663
663
663
-116
-116
-116
-116
101
101
101
1416
1416
1416
1416
1417
1417
1417
-438
-347
-347
-347
-438
-347
-347
495
567
567
567
495
567
567
-488
-271
-271
-271
-488
-271
-271
1328
1393
1393
1393
1328
1393
1393
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
-192
-192
-162
-192
-192
-378
-327
-378
-378
-378
-378
-378
-378
-287
-192
-100
-287
-378
641
641
650
641
641
545
580
545
545
545
545
545
545
602
641
663
602
545
-148
-148
-224
-148
-148
-476
-476
-476
-476
-476
-476
-476
-476
-191
-148
138
-191
-476
1413
1413
1402
1413
1413
1333
1333
1333
1333
1333
1333
1333
1333
1407
1413
1414
1407
1333
-318
-347
-317
-317
-347
-160
-295
-295
-347
-347
-347
-347
-347
-438
-252
-438
-160
-438
585
567
586
586
567
651
598
598
567
567
567
567
567
495
619
495
651
495
-433
-271
-348
-348
-271
58
-271
-271
-271
-271
-271
-271
-271
-556
-228
-556
58
-556
1348
1393
1374
1374
1393
1420
1393
1393
1393
1393
1393
1393
1393
1299
1401
1299
1420
1299
1
2
3
4
5
6
7
8
9
10
11
12
-378
-378
-378
-378
-378
-378
-327
-378
-402
-243
-493
-493
545
545
545
545
545
545
580
545
527
622
436
436
-476
-476
-476
-476
-476
-476
-476
-476
-533
-91
-749
-749
1333
1333
1333
1333
1333
1333
1333
1333
1309
1418
1190
1190
-347
-347
-347
-347
-347
-347
-295
-295
-323
-287
-280
-347
567
567
567
567
567
567
598
598
582
537
606
567
-271
-271
-271
-271
-271
-271
-271
-271
-214
-54
-46
-271
1393
1393
1393
1393
1393
1393
1393
1393
1403
1152
1420
1393
Condition Score
-378
545
-476
1333
-415
516
-431
1349
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weeks
Nov-15th
1
Beginning Period of The Forecast Model
77
Table 24. Average Corn Futures Profit - Sept 23
Simple Scenario
2 Weks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
160
160
135
135
135
135
160
413
413
423
423
423
423
413
449
449
484
484
484
484
449
810
810
788
788
788
788
810
114
102
105
163
119
119
164
404
389
401
430
423
423
423
340
375
515
559
454
454
503
714
695
608
561
612
612
597
1
2
3
4
5
6
7
160
160
135
135
135
160
197
413
413
423
423
423
413
396
449
449
484
484
484
449
546
810
810
788
788
788
810
744
191
191
230
275
275
230
230
427
427
456
454
454
456
456
594
594
650
754
754
650
650
765
765
748
701
701
748
748
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
185
185
185
185
188
188
188
188
188
158
158
188
188
188
155
67
70
70
70
402
402
402
402
400
400
400
400
400
414
414
400
400
400
415
440
439
439
439
521
521
521
521
379
379
379
379
379
346
346
379
379
379
488
238
96
96
96
763
763
763
763
847
847
847
847
847
863
863
847
847
847
786
901
929
929
929
144
103
144
145
218
318
234
233
208
355
355
355
355
266
179
167
117
218
218
423
420
423
464
450
400
440
385
404
344
344
344
344
407
385
393
383
382
382
435
319
435
458
316
445
648
363
308
479
479
479
479
600
219
260
117
278
278
660
722
660
737
667
661
564
564
577
593
593
593
593
648
621
569
578
627
627
1
2
3
4
5
6
7
8
9
10
11
12
158
158
158
158
158
158
128
158
158
158
158
158
414
414
414
414
414
414
425
414
414
414
414
414
346
346
346
346
346
346
294
346
346
346
346
346
863
863
863
863
863
863
883
863
863
863
863
863
115
115
115
115
163
196
193
192
141
258
185
223
444
444
444
444
449
406
430
390
457
445
378
389
408
408
408
408
472
496
508
291
422
465
272
531
705
705
705
705
725
649
687
557
861
827
547
636
Condition Score
40
443
44
933
218
382
278
627
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weks
Nov-15th
1
Beginning Period of The Forecast Model
78
Table 25. Average Soybean Futures Trading Profit - July 15
Simple Scenario
2 Weeks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
532
532
532
532
532
218
218
1702
1702
1702
1702
1702
1774
1774
2098
2098
2098
2098
2098
1229
1229
4283
4283
4283
4283
4283
4630
4630
356
356
356
443
561
418
418
1894
1894
1894
1905
1965
1818
1818
878
878
878
845
1098
899
899
3018
3018
3018
3013
3051
2879
2879
1
2
3
4
5
6
7
218
218
218
218
218
218
218
1774
1774
1774
1774
1774
1774
1774
1229
1229
1229
1229
1229
1229
1229
4630
4630
4630
4630
4630
4630
4630
289
370
370
370
370
370
370
1993
1896
1896
1896
1896
1896
1896
1413
1384
1384
1384
1384
1384
1384
3101
2925
2925
2925
2925
2925
2925
1
2
3
4
5
6
7
8
9
10
11
12
13
14
417
417
417
417
417
103
103
417
417
103
103
103
417
417
1736
1736
1736
1736
1736
1786
1786
1736
1736
1786
1786
1786
1736
1736
1993
1993
1993
1993
1993
1124
1124
1993
1993
1124
1124
1124
1993
1993
4336
4336
4336
4336
4336
4659
4659
4336
4336
4659
4659
4659
4336
4336
356
356
356
248
248
418
418
269
418
250
250
561
561
417
1894
1894
1894
1888
1888
1818
1818
1829
1818
1949
1949
1965
1965
2095
878
878
878
596
596
899
899
1106
899
889
889
1098
1098
1416
3018
3018
3018
3054
3054
2879
2879
2924
2879
3133
3133
3051
3051
2805
1
2
3
4
5
6
7
8
9
10
11
12
103
103
103
103
103
103
103
103
103
103
103
103
1786
1786
1786
1786
1786
1786
1786
1786
1786
1786
1786
1786
1124
1124
1124
1124
1124
1124
1124
1124
1124
1124
1124
1124
4659
4659
4659
4659
4659
4659
4659
4659
4659
4659
4659
4659
356
418
418
269
370
269
269
370
531
492
523
523
1894
1818
1818
1829
1896
1829
1829
1896
1938
1898
1869
1869
878
899
899
1106
1384
1106
1106
1384
1699
1175
1133
1133
3018
2879
2879
2924
2925
2924
2924
2925
2917
3008
2897
2897
Condition Score
417
1736
1993
4336
289
1998
1072
2819
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weeks
Nov-15th
1
Beginning Period of The Forecast Model
79
Table 26. Average Soybean Futures Trading Profit - July 29
Simple Scenario
2 Weeks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
153
258
346
263
258
451
451
1414
1397
1377
1396
1397
1344
1344
-214
-224
1603
-349
-224
1593
1593
4343
4342
4020
4333
4342
4024
4024
220
133
133
133
133
314
314
1623
1622
1622
1622
1622
1504
1504
723
716
716
716
716
1014
1014
2270
2141
2141
2141
2141
2078
2078
1
2
3
4
5
6
7
153
346
346
451
451
346
346
1414
1377
1377
1344
1344
1377
1377
-214
1603
1603
1593
1593
1603
1603
4343
4020
4020
4024
4024
4020
4020
398
466
466
389
398
398
466
1577
1514
1514
1673
1577
1577
1514
440
543
543
385
440
440
543
2626
2514
2514
2779
2626
2626
2514
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
258
263
263
263
258
258
258
258
258
258
258
258
258
258
258
1397
1396
1396
1396
1397
1397
1397
1397
1397
1397
1397
1397
1397
1397
1397
-224
-349
-349
-349
-224
-224
-224
-224
-224
-224
-224
-224
-224
-224
-224
4342
4333
4333
4333
4342
4342
4342
4342
4342
4342
4342
4342
4342
4342
4342
220
171
269
296
296
314
314
314
314
317
466
392
392
345
345
1623
1522
1588
1594
1594
1504
1504
1504
1504
1471
1514
1472
1472
1511
1511
723
740
1060
1022
1022
1014
1014
1014
1014
313
543
220
220
913
913
2270
2004
2198
2204
2204
2078
2078
2078
2078
2328
2514
2321
2321
2106
2106
1
2
3
4
5
6
7
8
9
10
11
12
-1
153
153
263
258
258
258
258
258
258
258
258
1423
1414
1414
1396
1397
1397
1397
1397
1397
1397
1397
1397
-424
-214
-214
-349
-224
-224
-224
-224
-224
-224
-224
-224
4326
4343
4343
4333
4342
4342
4342
4342
4342
4342
4342
4342
398
466
466
398
398
398
398
466
466
424
424
356
1577
1514
1514
1577
1577
1577
1577
1514
1514
1450
1450
1409
440
543
543
440
440
440
440
543
543
576
576
280
2626
2514
2514
2626
2626
2626
2626
2514
2514
2400
2400
2222
Condition Score
451
1344
1593
4024
445
1472
2068
4323
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weeks
Nov-15th
1
Beginning Period of The Forecast Model
80
Table 27. Average Soybean Futures Trading Profit - Aug 12
Simple Scenario
2 Weeks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
-434
-304
-304
-284
-384
-383
-183
728
796
796
804
758
759
835
-352
-53
-53
0
-73
-72
212
1068
1127
1127
1128
1126
1126
1107
-778
-778
-778
-785
-785
-778
-778
745
745
745
667
667
745
745
635
635
635
385
385
635
635
3346
3346
3346
3054
3054
3346
3346
1
2
3
4
5
6
7
-413
-213
-213
-453
-453
-453
-253
742
827
827
716
716
716
815
-297
-13
-13
-288
-288
-288
-5
1086
1128
1128
1088
1088
1088
1128
-735
-778
-778
-742
-748
-735
-673
675
745
745
571
617
675
714
458
635
635
383
270
458
606
3024
3346
3346
2617
2805
3024
2993
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
-496
-366
-283
-283
-343
-283
-283
-343
-343
-343
-343
-343
-343
-383
-383
-383
685
767
804
804
779
804
804
779
779
779
779
779
779
759
759
759
-452
-153
2
2
-80
2
2
-80
-80
-80
-80
-80
-80
-72
-72
-72
1027
1117
1128
1128
1125
1128
1128
1125
1125
1125
1125
1125
1125
1126
1126
1126
-617
-639
-639
-773
-773
-773
-773
-581
-581
-483
-564
-402
-299
-502
-373
-502
681
680
680
610
610
610
610
783
783
789
844
802
811
810
833
810
2
202
202
498
498
498
498
422
422
167
314
159
189
450
447
450
2705
2728
2728
2804
2804
2804
2804
2605
2605
2554
2795
2731
2556
2813
2604
2813
1
2
3
4
5
6
7
8
9
10
11
12
-413
-413
-413
-413
-453
-413
-413
-413
-413
-413
-413
-343
742
742
742
742
716
742
742
742
742
742
742
779
-297
-297
-297
-297
-288
-297
-297
-297
-297
-297
-297
-80
1086
1086
1086
1086
1088
1086
1086
1086
1086
1086
1086
1125
-617
-773
-765
-742
-742
-742
-773
-773
-773
-494
-550
-483
681
610
667
571
571
571
610
610
610
729
783
789
2
498
725
383
383
383
498
498
498
-180
397
167
2705
2804
3001
2617
2617
2617
2804
2804
2804
3116
2598
2554
Condition Score
-383
759
-72
1126
-502
810
450
2813
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weeks
Nov-15th
1
Beginning Period of The Forecast Model
81
Table 28. Average Soybean Futures Trading Profit - Aug 26
Simple Scenario
2 Weeks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
173
320
320
107
107
107
107
1009
970
970
1019
1019
1019
1019
-1028
396
396
199
199
199
199
3764
3890
3890
3906
3906
3906
3906
-38
-38
-38
72
72
72
156
864
864
864
839
839
839
856
163
163
163
422
422
422
769
2492
2492
2492
2496
2496
2496
2426
1
2
3
4
5
6
7
18
18
18
50
155
50
187
1025
1025
1025
1023
1012
1023
1006
599
599
599
409
889
409
699
3862
3862
3862
3888
3802
3888
3844
86
86
72
72
72
72
72
792
792
839
839
839
839
839
-251
-251
422
422
422
422
422
3174
3174
2496
2496
2496
2496
2496
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
173
173
173
287
287
173
287
433
433
433
402
255
255
363
510
510
510
1009
1009
1009
981
981
1009
981
921
921
921
937
990
990
953
878
878
878
-1028
-1028
-1028
-1109
-1109
-1028
-1109
314
314
314
504
-919
-919
213
1636
1636
1636
3764
3764
3764
3739
3739
3764
3739
3898
3898
3898
3876
3794
3794
3905
3526
3526
3526
53
53
53
53
53
53
250
250
250
250
250
250
250
601
601
574
574
865
865
865
865
865
865
1027
1027
1027
1027
1027
1027
1027
872
872
920
920
113
113
113
113
113
113
91
91
91
91
91
91
91
801
801
958
958
2638
2638
2638
2638
2638
2638
2488
2488
2488
2488
2488
2488
2488
2067
2067
2128
2128
1
2
3
4
5
6
7
8
9
10
11
12
-97
-128
-97
-97
-128
-128
142
255
255
255
255
255
1020
1016
1020
1020
1016
1016
1014
990
990
990
990
990
-1014
-824
-1014
-1014
-824
-824
-838
-919
-919
-919
-919
-919
3768
3817
3768
3768
3817
3817
3814
3794
3794
3794
3794
3794
208
-15
53
53
53
53
53
188
261
188
188
250
902
823
865
865
865
865
865
817
854
817
817
1027
-348
-426
113
113
113
113
113
398
243
398
398
91
3106
3066
2638
2638
2638
2638
2638
2667
2841
2667
2667
2488
Condition Score
510
878
1636
3526
574
920
958
2128
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weeks
Nov-15th
1
Beginning Period of The Forecast Model
82
Table 29. Average Soybean Futures Trading Profit - Sept 9
Simple Scenario
2 Weeks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
355
193
312
278
278
193
50
981
1028
997
1007
1007
1028
1046
1451
298
314
1184
1184
298
76
3264
3580
3578
3377
3377
3580
3592
130
-40
-83
-235
-202
-117
-58
680
1047
1044
1019
1027
1041
1046
200
1008
-129
724
-146
741
76
2075
3438
3590
3514
3590
3510
3592
1
2
3
4
5
6
7
388
388
388
388
388
388
193
967
967
967
967
967
967
1028
581
581
581
581
581
581
298
3542
3542
3542
3542
3542
3542
3580
137
-202
-202
-202
-202
-202
-7
1038
1027
1027
1027
1027
1027
1048
359
-146
-146
-146
-146
-146
138
3573
3590
3590
3590
3590
3590
3590
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
388
193
193
312
312
193
193
312
312
312
312
278
278
227
227
227
227
227
967
1028
1028
997
997
1028
1028
997
997
997
997
1007
1007
1021
1021
1021
1021
1021
581
298
298
314
314
298
298
314
314
314
314
1184
1184
1338
1338
1338
1338
1338
3542
3580
3580
3578
3578
3580
3580
3578
3578
3578
3578
3377
3377
3315
3315
3315
3315
3315
59
-7
-202
-83
-202
-83
-202
-83
-202
-202
-202
-235
-202
-253
-202
-202
-202
-202
1055
1048
1027
1044
1027
1044
1027
1044
1027
1027
1027
1019
1027
1014
1027
1027
1027
1027
252
138
-146
-129
-146
-129
-146
-129
-146
-146
-146
724
-146
8
-146
-146
-146
-146
3697
3590
3590
3590
3590
3590
3590
3590
3590
3590
3590
3514
3590
3593
3590
3590
3590
3590
1
2
3
4
5
6
7
8
9
10
11
12
388
193
193
132
132
193
193
193
193
193
193
278
967
1028
1028
1039
1039
1028
1028
1028
1028
1028
1028
1007
581
298
298
56
56
298
298
298
298
298
298
1184
3542
3580
3580
3592
3592
3580
3580
3580
3580
3580
3580
3377
27
-7
-202
-263
-202
-263
-202
-202
-202
-202
-202
-188
1047
1048
1027
1012
1027
1012
1027
1027
1027
1027
1027
1041
1178
138
-146
-388
-146
-388
-146
-146
-146
-146
-146
785
3380
3590
3590
3570
3590
3570
3590
3590
3590
3590
3590
3638
Condition Score
227
1021
1338
3315
-253
1014
8
3593
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weeks
Nov-15th
1
Beginning Period of The Forecast Model
83
Table 30. Average Soybean Futures Trading Profit - Sept 23
Simple Scenario
2 Weeks
Holding Period
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
Profit
Std. Dev.
NDVI scores only
1
2
3
4
5
6
7
-28
-28
-28
-28
-28
-28
217
1366
1366
1366
1366
1366
1366
1348
3
3
3
3
3
3
929
3059
3059
3059
3059
3059
3059
2904
389
18
-59
-59
-59
18
18
1401
1006
1026
1026
1026
1006
1006
660
3
-50
-50
-50
3
3
2910
1894
1988
1988
1988
1894
1894
1
2
3
4
5
6
7
75
320
320
320
320
320
320
1365
1326
1326
1326
1326
1326
1326
169
1096
1096
1096
1096
1096
1096
3054
2841
2841
2841
2841
2841
2841
-133
-36
-94
208
-128
-36
100
971
1044
1009
1459
1063
1044
1113
177
70
133
694
18
70
244
1811
1983
1892
2910
2096
1983
2059
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
290
-140
-35
-35
290
290
395
-35
395
-35
-35
395
395
395
743
743
743
743
743
1333
1359
1366
1366
1333
1333
1304
1366
1304
1366
1366
1304
1304
1304
1130
1130
1130
1130
1130
936
32
104
104
936
936
1008
104
1008
104
104
1008
1008
1008
1111
1111
1111
1111
1111
2901
3059
3057
3057
2901
2901
2876
3057
2876
3057
3057
2876
2876
2876
2835
2835
2835
2835
2835
90
202
202
202
39
39
202
260
251
286
286
152
152
315
468
495
495
495
495
1208
1187
1187
1187
1113
1113
1187
1096
1144
1193
1193
1079
1079
912
985
939
939
939
939
647
517
517
517
-126
-126
517
583
519
759
759
250
250
339
1000
1022
1022
1022
1022
2792
2819
2819
2819
1942
1942
2819
2584
2688
2680
2680
1761
1761
1829
2710
2572
2572
2572
2572
1
2
3
4
5
6
7
8
9
10
11
12
215
-140
395
395
395
395
395
395
395
395
395
292
1349
1359
1304
1304
1304
1304
1304
1304
1304
1304
1304
1333
1024
32
1008
1008
1008
1008
1008
1008
1008
1008
1008
843
2869
3059
2876
2876
2876
2876
2876
2876
2876
2876
2876
2932
-85
-85
-85
-19
-85
-85
90
223
223
259
200
90
1117
1117
1117
1164
1117
1117
1208
1196
1196
1254
1316
1116
16
16
16
-67
16
16
647
626
626
901
942
151
1950
1950
1950
2050
1950
1950
2792
2792
2792
2781
2947
1824
Condition Score
313
1328
208
3051
495
939
1022
2572
NDVI scores, Conditions
scores, and regional dummy
variables
Week
NDVI scores with
Regional Dummy
Variables
Nov-15th
NDVI and crop conditions scores
1
Selective Trades
2 Weeks
Nov-15th
1
Beginning Period of The Forecast Model