Volatility and global resin prices are a fact of life. From a historical perspective, volatility is baked into pricing, fueled by various factors centred on feedstock and upstream commodity prices. Yet, despite historical price instability, existing price forecasts tend to be relatively linear.
The founders of Singapore-based MLT Analytics were convinced that, with the machine learning (MLT) and artificial intelligence (AI) technologies available today, combined with their extensive industry knowledge, it should be possible to introduce the volatility observed in historical data into resin price forecasting.
They came up with a two-pronged solution: develop a series of hypothetical feedstock price scenarios based on mega-trends, such as peak oil, vehicle electrification, natural gas-fired electricity proliferation, and regulatory developments. Then, use machine learning and artificial intelligence to pinpoint the key influencers of resin pricing. Once identified, these influencers can be correlated with historical prices to generate real-world forecasts.
“Long-term oil and gas price forecasts, from the U.S. Energy Information Administration, for example, lack volatility. They rise or fall, depending on the scenario, in a relatively linear fashion. What we are doing is introducing volatility into our forecasts based on past feedstock trends and assumptions of future market developments such as peak oil that, in turn, introduces volatility into the resin price forecasts,” said Stephen Moore, co-founder and CEO of MLT Analytics. “We analyse multiple feedstock and use machine learning to explore their correlations with resins by type and region or country where they are sold.”
While plastics are the starting point for MLT Analytics, pricing for any type of commodity, including non-plastic materials, theoretically can be forecast once historical prices are correlated with data for key influencers.
Once the forecasting model has been set up, the latest historical data is fed into the model as it becomes available, which serves to further refine the forecast. Further, “back-casting,” as indicated in the graphic by the portion of the blue line overlapping the red historical line, is a means of verifying the validity of the forecast. A close overlap of the historical and back-casted data is proof that the modelling is working from a statistical perspective.
One important caveat: “Unless the data you are feeding into the forecasting model makes sense from industry and economic perspectives, your forecast will, unfortunately, look like an unlikely outcome,” cautioned Moore. “That’s where the decades of industry expertise our team has accumulated becomes of ultimate importance.”