top of page

Commodity Price Prediction Using AI

Using machine learning and deep learning algorithms for the forecasting of financial asset prices is now widely used by leading investors, but less frequently by many commodity producers, who tend to rely more on fundamental supply or demand dynamics and other factors in determining when to hedge. Deploying AI is a breakthrough that can aid producers by analyzing large amounts of financial and other proprietary data sets to predict price movements and equally importantly the drivers of those movements with no bias, unlike humans. For instance at the end of 2021, Adatos had already identified US interest rates as a major factor driving Crude Palm Oil (CPO) and therefore the risk that higher US rates would cause a sell off in CPO prices.


Adatos’ CPO price prediction models use historical price data of related commodities, equities, interest rates and foreign exchange and other data to forecast prices up to 6 months in the future. They also include sentiment analysis using natural language processing.


The models are evaluated using normalized root mean square error (nRMSE) and hit ratio, the % correct predictions for the direction (up/down) of prices vs actual at each horizon. Low values of nRMSE and high values of hit ratio show that the models are efficient in predicting CPO prices. The monthly model has been trained on over 10 years of historical data and sentiment, and is able to forward predict CPO prices for 6 months with an nRMSE of 0.182 and a hit ratio of 100%. Adatos has built models using different frequencies of data to forecast at the corresponding horizon.


6 months forward prediction from Adatos' monthly CPO price prediction model with an nRMSE of 0.182 and a hit ratio of 100%.


Crucially, the models also determine the importance of various features for price prediction at each horizon. Commodities including natural gas and crude oil are important as expected, but other variables such as US interest rates also play a significant role.

Adatos believes this type of tool can add value to producers as an input in their hedging strategy and in identifying the factors that drive prices for their own scenario planning. Message us on LinkedIn to find out more.




Commentaires


bottom of page