Using artificial intelligence (AI) algorithms for the forecasting of financial asset prices is common for major investors in large liquid financial markets. However, such methods are less frequently used by commodity producers and consumers in niche markets such as Crude Palm Oil (CPO), who tend to rely on fundamental supply and demand dynamics in determining when to hedge. With its ability to analyze large amounts of financial and proprietary data sets to predict price movements and the underlying drivers without human bias, AI holds great potential for price prediction in these niche markets.
Adatos has developed a robust CPO price prediction model capable of forecasting CPO prices up to 6 months, which has over 18 months of track record in live markets. It incorporates over 10 years of historical financial data from commodities, commodity related equities, interest rates, foreign exchange, and other public sources. Additionally, our models utilize sentiment analysis through natural language processing, leveraging news headlines to enhance prediction accuracy. In addition the model identifies the main drivers of CPO prices in each of the 6 horizons (1 to 6 months), so that clients can understand their macro risks.
Since inception, we have tested and refined our CPO price prediction models based on its performance in live markets. The results have consistently surpassed our initial performance targets. We evaluate the models using two key metrics: normalized root mean square error (nRMSE) and hit ratio, which indicates the percentage of correct predictions for the direction from the date of forecast (up or down) of prices compared to actual values at each horizon. Lower values of nRMSE and higher hit ratios demonstrate the accuracy of the models in predicting CPO prices.
The model has provided 4 distinct forecasts since September 2021, with regularly retraining to capture recent market data and performance measured vs actual CPO front month futures prices.
Actual vs Predicted CPO prices with 6 month ahead forecasts and periodic retraining.
Forecast 1 (orange): This forecast demonstrated exceptional performance. With the model trained on a decade's worth of historical data up to September 2021, we accurately forecasted CPO prices from October 2021 to March 2022, achieving an nRMSE of 0.178 and a hit ratio of 100%. Importantly it also identified US interest rates as a major driver of CPO prices and it was clear that once the Fed began hiking this would most likely have a significant negative impact on prices.
Forecast 2 (green): During the Russia-Ukraine conflict in early 2022, the model was retrained using data up to March 2022, in order to capture the new market environment. The model identified natural gas and again US interest rates as major factors influencing CPO prices, leading to a successful prediction of a decreasing trend in CPO prices for the next 6 months projection, with an nRMSE of 0.182 and a hit ratio of 100%.
Forecast 3 (purple): The model again identified US interest rates as a key driver of CPO prices, indicating the risk of a sell-off due to continued increases in US rates. Although the model's performance was slightly divergent from actual prices, it still achieved an nRMSE of 0.313 and a hit ratio of 83%. This was caused mainly by the Fed hiking more than what was implied in the futures market.
Forecast 4 (red): Adatos retrained the model using data up to January 2023. It identified urea fertilizer as a key factor influencing the 1st, 2nd, and 3rd month horizons. For longer horizons, CPO spot and future prices, with other equities, were found to be significant drivers. This prediction performed well, with an nRMSE of 0.098 and a hit ratio of 60%.
The Adatos CPO price prediction model has consistently proven to be effective and adaptable. It is continually updated to identify key drivers of price changes and leverages a wide range of financial data and sentiment analysis to capture crucial market trends and drivers for accurate forecasts. We look forward to releasing regular updates to the model forecasts.
Please reach out to firstname.lastname@example.org to discuss your specific requirements.