Following our recent article on long term (6 month) forecasting of CPO prices targeted at producers, we have had a number of enquiries from traders and other market participants for much shorter horizon forecasts.
Leveraging this experience and recognizing that a number of the input variables are not available on a daily basis, we have adapted the model to be less data intensive and target higher accuracy on short horizons.
Our daily model is trained on a comprehensive set of features from 10 years of financial market data, each of them providing the model with a unique set of information. Not all features are created equal. For instance, soybeans have shown consistent relevance in the daily model, which should be no surprise as it is a related commodity. However soybean prices alone are not sufficient to make reliable predictions in a volatile market with a large number of variables. Our daily model incorporates more than 40 direct features from interest rates, exchange rates to security prices and over 100 financial derivatives.
Model Performance
We have tested the daily results and analyzed the performance of our model based on different scenarios. Not only do we want our model to focus on the direction of the move in prices, we are also concerned regarding model performance in different percentage change scenarios.
Actual vs. forecasted CPO price for daily tests
Model performance in various % moves since 2022
Scenario | % Change Range | nRMSE* | Hit Ratio** |
Overall | All Range | 0.029 | 0.55 |
Minimal | 0-1% | 0.005 | 0.56 |
Small | 1-4% | 0.020 | 0.63 |
Moderate | 4-8% | 0.051 | 0.71 |
Significant | >8% | 0.088 | 0.80 |
* nRMSE: normalised root mean square error
** Hit ratio: % of correct forecast of market move up vs. down
Good model performance is defined as nRMSE <0.22 and hit ratio >55% and the model outperformed these benchmarks in all scenarios.
We also observe that in larger market moves, the hit ratio of our model increases significantly and reliably. Our model is able to predict direction of movement correctly 80% of the time when there is a large movement in price >8%.
Connect With Us
If you're interested in harnessing the power of our daily forecasting model for your trading decisions, don't hesitate to reach out to us. Send us a message on LinkedIn to explore how our insights can enhance your market strategies.
Comments