For many years, Wall Street investors have used
sophisticated software like artificial neural networks to gain a trading
advantage. These software tools use a range of data inputs and historical
trends to predict stock prices.
But the cattle market is a different beast. “The
software tools used to predict the stock market fail miserably if you apply them
to cattle futures,” says Jordan Baumeister. She worked the past year with
fellow computer science majors Trevor Borman and Dustin Reff to build models
that could better predict the cattle and corn markets in an effort to offer
commodity traders an edge. The team used artificial intelligence and data
science to create mathematical models to predict future market trends and
provide a comparison for anomalies, like droughts or floods, using historical
data analytics.
“Our overall goal was to optimize the risk versus
reward tradeoff that shows up when you exchange these contracts on the futures
market,” says Reff.
To achieve this goal, the students had to rely on
decades of previous work.
A long history of success
In 1993, Todd Gagne was a student at Mines
developing his own software programs when he crossed paths with Ron Ragsdale,
who ranched on 55,000 acres of rolling prairie near the confluence of the Belle
Fourche and Cheyenne Rivers.
Ragsdale came to ranching following a successful
career in law along with a background in math and statistics. He developed his
own system for predicting the cattle market using a series of equations that he
worked out by hand with a pencil and paper. The model helped him determine when
to buy and sell both corn and cattle. The two commodities are related because
cows are often fattened with corn.
“What he did was kind of genius,” says Gagne. “He
looked at the futures market for both cattle and corn and backed out all the
costs needed to fatten his claves. He used 187 variables, not just feed. He
included the costs of the lights in his barn, vaccination, fuel, everything.
This way he knew what he could pay for his calves to make a profit in the
future.” If the model showed Ragsdale that he could not profit that year, he
would lease his land to other ranchers.
Ragsdale asked Gagne to help enhance his equations
with a computer program that he and his spouse, Holly, created while they were
college students in 1993. “I was in my early 20s, and he was a guy who had been
around the block. He saw everything as statistics and math. He taught me a lot
and he was a great mentor,” says Gagne.
The software they developed was employed by Ragsdale
successfully over the coming decades. It failed to predict positive results on
only two occasions, one was on Sept. 11, 2001, the other the 2008 recession
with the collapse of Lehman Brothers. “Everything else the model held up. It
would bend, but it did not break,” says Gagne.
Gagne graduated from Mines and went on to spend a
career in software development. Today, he is an Entrepreneur
in Residence at the university. He serves as a consultant for start-up companies. But
he never fully lost touch with Ragsdale. The two stayed friends over the years
and continued to work on the project, adjusting the program and learning as
they went. Ragsdale ended up writing a long, unpublished thesis on his market
theory before he passed away in 2021 at 72 years old. Before he died, he worked with Gagne to launch
the student project.
“It’s been an intellectual curiosity that began as a
side-hustle and has evolved into something much bigger,” says Gagne.
Coming Full Circle
In the fall of 2021, Gagne shared the software that
he and his wife Holly developed as college students, nearly 30 years prior, with
a new team of Mines students. Gagne sponsored the team’s work and challenged
them to use modern tools like artificial intelligence and data analytics to
delve into decades of cattle market data and enhance the original program.
The goal was to make the software more robust to
better predict commodity prices when outside factors drive the market off its
normal course.
“If I know what the value should be in the future,
what happens when something like mad cow disease, or widespread drought, or widespread
flooding occurs, all these things can send the market into arbitrage,” says
Gagne. He tasked the students to build software that could better predict what
to do when the market gets wacky.
“We twisted and tuned this data and tried to look at
it in new ways to see anomalies or patterns that we think are tradable in the
future."
The team of students spent a full year working on
the project. “The computational complexity was enormous,” says Baumeister.
The team overcame challenges such as filtering out noise
in the data to get to the heart of the information needed to predict the
markets and homing in on key variables that make the most impact to commodity
process. They ran their model using historical numbers and worked on many
iterations of the program until it could most accurately predict the known outcome.
By the end of the year, the team developed two
different computer models to help make better commodities trades. One examines
historic trends to help determine the risk versus reward analysis. The other, a
predictor model, calculates the best times to buy and sell. “We developed a tool to help play the
commodities trading game a little bit better and to get some edge over the
competition,” says Baumeister.
The project is ongoing. Baumeister, Reff and Borman have
all graduated and began their careers, but they will be
briefing a new team of students in the Fall of 2022 to help launch the next
phase of the project. “I was very pleased, these students are all going to
different jobs, but they are willing to come back and help the next team take
up the next phase,” says Gagne.
In the coming semester, the new team will rebuild
the model and then work on a sensitivity study to understand which of the 187
variables carry the most weight in the model.
They will run more historical market data through the model to see how
it performs over time, and they will build in indicators as the model runs that
will check when the animal might be over or under valued.
Mines faculty members who oversee computer
engineering senior research projects are pleased with the progress. "As a
sponsor, Todd provided years of data, support, and a good story,” says Brian
Butterfield, a lecturer of computer science and engineering at Mines. “These
students took advantage of the opportunity by applying their skills in data
science and data analysis to advance the work. I appreciate watching what emerges by providing students with the
framework to build something and gain real world experience."