Campaign for Wool NZ’s chair Ryan Cosgrove has used machine learning, a branch of artificial intelligence, to identify what strong wool attributes are key to maximising returns.
Cosgrove, who is also the head of sourcing and materials for Mons Royale, and co-founder of Fusca, a wool sales data platform, recently published a paper outlining his methods and findings.
The paper, A Statistical Analysis of New Zealand Strong Wool Sold at Auction Between 2022 and 2024, showed that colour, micron and length are critical factors affecting wool prices.
Colour was, however, shown to be the most deterministic feature of price by far, accounting for 48% of price variability, regardless of other external economic factors.
For the paper, data from January 1st 2022 to June 30th 2024 was analysed.
The periods are significant and affect what attribute of strong wool is the most important, he said.
When Cosgrove moved the lens to 2013-2015, micron was the biggest driver of value.
“I chose the most recent years because it’s what we’re exposed to now. If you say 48% of the variance in price comes from colour, that’s something a grower can use right now.”
Cosgrove said as a wool trader he was always frustrated at the lack of sophistication used to analyse data.
“Every week we used pen and paper to value and put a price on wool. We’d invoice by hand. None of the data was analysed. I thought, if only I could find a method to do that, so I turned to machine learning to have a look.”
He used two machine learning methods, Random Forest Regression (RFR) and Gradient Boosting Analysis (GBA) to build accurate models predicting wool prices, across short periods of consistent economic conditions.
He wrote the code for the models himself.
Machine learning can be used to constantly update growers on what aspects are more important to focus on at farm level, he said.
Efforts should be concentrated on achieving consistent colour and optimal micron through genetic selection, improved grazing management, and careful handling practices, he said.
Continued focus on reducing vegetable matter contamination will further enhance wool quality and value, the methods showed.
“The findings allow farmers to make more informed choices about farm management practices and to focus production and handling where it matters most.”
The GBA model was able to create a model where 91% of the variability of the wool price could be explained.
“Analysis aims to uncover how controllable variables, such as wool colour, length, vegetable matter, and micron, as well as uncontrollable variables like month and the NZD:USD exchange rate impact the market price of wool and to use these insights to optimise practices both on the farm and during wool preparation.”
While most brands looked at wool through the lens of increasing wool value or offering a new product and changing demand to try to improve the value for farmers, the machine learning methods are ignorant of economic factors and market demand.
But they use data to “tell growers, here are things you can do to make sure you’re at the top end of the bell curve, that you’re always maximising the value regardless of the market. Instead of just hoping the market is good next year.
“The research provides a powerful new tool to analyse wool markets. By understanding and leveraging these predictive models, markets may be stabilised, quality improved and profitability boosted across the industry.”
In future Cosgrove aims to include other variables, such as environmental conditions and market trends, to improve the predictive capability of the models.
