Predictive Analytics and Time-Series-Forecasting of Complex Commodities: a concrete case-study

Have you ever taken a moment to think of how much your daily life depends on aluminum? Yes, I’m talking about the foil you used this morning to wrap your sandwich for lunch today, or perhaps the slick casing of the smartphone or laptop you’re probably reading these lines from right now. Aluminum is everywhere. This versatile metal is a key element in nearly every aspect of our modern life: it is in our electronic gadgets, in any vessel we use for locomotion, be it plane, car, trains or bikes and even in the packaging that keeps our food fresh. It is fair to say, that Aluminum is an integral element to a vast number industries and applications that sustain our modern lifestyle.

Now, imagine the ripple effects if the price of aluminum suddenly went up or plummeted. What would happen to the cost of manufacturing, the price of goods, or the stability of entire industries? As you can probably imagine, the price of a commodity such as aluminum is dependent on a vast web of interconnected factors, including extraction costs, global politics, market demand, and even environmental considerations.

Therefore, predicting the price of such a complex commodity can be very valuable but also challenging at the same time, especially considering that the price of aluminum doesn’t necessarily follow a distinct type of seasonality (like say Christmas Sale every year, for example). For these type of complex cases, traditional forecasting methods often fall short. This is where Predictive Analytics and Time-Series-Forecasting (TSF) comes into play. By using sophisticated algorithms, we can narrow down the uncertainty and make informed predictions. In this article, we will explore how TSF can be used to estimate the future price of aluminum, compare a popular algorithm used in TSF with our proprietary solution, and demonstrate how our expertise at diconium data can help your business exploit the power of predictive analytics.

Time-Series-Forecasting for Aluminum Prices

If you ask ChatGPT for the five most popular algorithms for time-series forecasting, chances are high that ARIMA will top the list. ARIMA, which stands for AutoRegressive Integrated Moving Average, is widely used in time-series forecasting due to its ability to model different types of data patterns. However, this algorithm has its drawbacks, particularly in handling long-term forecasts or sudden shifts.

Since a figure often tells more than words, let’s see how ARIMA predicts the price of aluminum. We trained an ARIMA model using a dataset containing 5 years and 3 months of aluminum price data with daily granularity. We performed a train/test split with the intention of forecasting the last 3 months (Jan 1 to March 1, 2024). Here’s how ARIMA performed:

As you can see, while ARIMA (displayed in blue) provides a basic forecast, it struggles to follow the pattern of the real data (displayed in black). Additionally, we observe that the confidence interval increases dramatically over time and fails to align with the structure of the original data.

The team from the data department has spent the last year fine-tuning our time-series forecasting solutions to address such challenges. Clients have come to us with challenging datasets like this one, requesting for example an estimate prediction on various time-series-data, such as daily number of orders. To demonstrate how one of our predictive analytics algorithms perform, we invested a couple of hours tailoring one of our models specifically for this very same aluminum price dataset and wanted to see how it compared to ARIMA. The following gif illustrates a direct comparison of the prediction step by step:

 

As can be seen in the animation, our predictive model closely mirrors the original dataset by capturing the most important trends and fluctuations with a decent accuracy. More importantly, the 95% confidence interval—a range within which we can be 95% confident that the actual values will fall—aligns closely with the original data. Unlike ARIMA, whose predictions become increasingly uncertain over time, our model maintains a consistent and more reliable confidence range that naturally follows the natural ups and downs of the dataset.

If you are ready to elevate your business with cutting-edge predictive analytics, let us empower you with precise, reliable forecasts that drive informed decision-making. Contact us today and discover how our advanced solutions in predictive analytics and other fields can transform your data into actionable insights.