Contextual AI Asset COMMODITY PRICE FORECASTING TOOL
A multinational automobile giant based in Europe, regarded as one of the biggest producers of premium cars was looking at employing a data-driven approach to forecasting commodity prices. The client’s procurement and supplier quality team forecasted the price of commodity and used it for budget planning and vendor negotiations for the subsequent years. The team used qualitative or judgmental techniques to forecast the price of commodity. They made educated guesses based on intuition, knowledge and experience. High volatility in the price of commodities due to different factors like the exchange market, political and geographical events lead to a lot of planning inefficiency and budget gaps. The client wanted to seek a data-driven approach to forecast the commodity prices.
The automobile industry, much like many other industries, has felt the pinch from rising commodity prices. The intermingling and co-dependencies of multiple global factors puts commodity markets in an extremely delicate position worldwide, where a landmark decision, a calamity or an unpredicted event is all it takes for ripple effects to translate to fluctuations in commodity markets. The resulting price fluctuations are especially problematic for businesses – as volatile market dynamics deliver a hard blow to procurement budgets. With raw materials, like steel, glass, aluminium, plastic rubber etc. accounting for nearly half of automobile manufacturing costs in the past, it has grown essential to foresee price forecasts, even in the event of sudden commodity market movements. Inaccurate forecasts can translate to steep costs for businesses, which cannot be entirely transferred or passed on to consumers. It becomes essential for businesses and specifically procurement leaders to have an accurate forecasting tool that can predict movements in the commodity market, ahead of time.
There are many factors that contribute to the volatility of commodity prices, ranging from market inflation to demand rise, supply chain hiccups, macroeconomic and geopolitical changes, and more. With investors, traders, buyers and many other players influencing commodity market movements, and endless market information available online, it’s understandable why commodity prices oscillate as often, influenced by herd mentality.
While the concept of monitoring and predicting commodity prices itself is not alien to procurement leaders in the automobile industry, the means to predict price fluctuations evades a formulaic response despite its longstanding prevalence, and what could otherwise prove to be a time-saving and cost-saving tool for procurement teams. Indeed, price predictions in many industries continue to take a conservative approach even today - from closely monitoring market movements to trusting hunches based on information in industry networks and the interwebs, intermittent insights from expert consultants, and many others. However, a major differentiator that has caused procurement managers to revisit the conservative approach is that commodity volatility is no longer confined to local or closed markets, as they extend to a global scale, making it far more difficult and unfeasible to tap into immediate supplier networks to gain accurate commodity pricing intelligence.
It’s only natural that automobile category leads have chosen to lean on a more reliable and time-saving approach to predict commodity prices over time, given the large volumes of commodities that they grapple with on a daily basis, and the small cost differences that snowball to major losses as a result. Consumer Price Risk Analytics (CPRA) has grown more and more relevant over the years. The surplus of data available today, further facilitates building pinpoint prediction tools with near real-time accuracy on price forecasts, helping procurement leaders have more effective negotiation discussions on commodity prices with suppliers, thereby saving automobile businesses anywhere between hundreds of thousands to millions in commodity purchase expenses.
A Commodity Price Forecasting Tool can be treated as a custom AI asset hat can be configured to the specifics of an automobile business – such as data, processes etc., and the accompanying industry dynamics, with the capability of generating accurate commodity price intelligence to procurement teams. With the help of our proprietary platform Co.dx, our commodity market analytics experts are able to speedily codify price forecasting algorithms using learnings from previous problem statement blueprints, to build an application that answers pertinent questions associated with price forecasting, specifically for each business and industry.
Broadly speaking, here are essential capabilities of a typical price forecasting tool that can equip category managers to make smarter procurement decisions and mitigate commodity risks
A machine learning tool can infinitely mine online data in real-time to build price prediction models, based on developments in global news and events. This would essentially be a tool that continually learns and evolves. Therefore, the longer it runs, the better it gets at forecasting commodity prices.
ML tools can collate data from reliable sources and display it an interactive real-time dashboard so the user can monitor events impacting the price of commodity on an ongoing basis. The tool can be customized to notify category managers on specific market trigger events, pivotal to their business. Early price warning systems can help users make informed decisions associated with the commodity, and the ripple effect it may have on market prices in the immediate future.
Recommendations can be treated as compasses that guide procurement teams on the most suitable action to take in the existing market situation. Here again, market data and newsfeed associated with the commodity is closely monitored by the application. These engines mine through mounds and mounds of market reports to gauge their veracity before factoring in only valid data. Configuring the engine to business data sources and simulating relevant supply-demand scenarios allows to further manage commodity price risks more effectively, by gaining a comprehensive view of cost drivers. Here again, the engine would learn iteratively to prescribe a better set of actions, over time.
In addition to price forecasting, applications can also offer overview of budget and spend profile analyses, cost deviations, cost savings and more, helping category managers to budget more efficiently by digging deep into their commodity purchase data. Factoring in commodity purchase impact on cash inflow, EBITDA etc., can further assist in commodity purchase planning and weighing associated monetary risks.
Accurate forecasts demand a distinctive solution approach, one that employs both judgmental forecasts (forecasts based on qualitative analysis, market expertise of a variety of factors and analysis of supply and demand fundamentals) and statistical quantitative model forecasts based on historical price information. A keen view and understanding of market dynamics that can be feature-engineered to build a contextual AI asset, can help in proactively predicting commodity prices.
Let’s take a look at steps that go into building a well-rounded forecasting solution:
-The accuracy and value of recommendations depend on the input data. From internal data to historic prices, market drivers and more – both unstructured and structured data is captured to configure a forecasting application. The data is subjected to statistical forecasting and judgemental forecasting techniques to build insights.
-Data such as commodity & exchange rates, inventory, supply & demand, countries GDP & PMI data is used in the statistical forecasting approach.
-News from different reliable sources is scraped and using text mining and Natural Language Processing (NLP) sentiment of the news is developed. This sentiment helps in understanding whether an article suggests that the price of a commodity will increase, decrease, or remain constant. This forms a good proxy for judgemental forecasting approach.
-The results of the model are captured in an interactive, scalable application, that can be easily consumed by any executive audience. An overview of current and probable future prices, combined with strategic execution insights, newsfeed alerts for different commodities, etc., offers buyers a holistic view of commodity movements in an easy-to-use dashboard.
Let’s now do a deep-dive into a sample solution that was built by TheMathCompany to solve the commodity price forecasting problem for the premier automobile manufacturer to understand how the application is built and mobilized, with a real-life example.
The multinational automobile manufacturer, which was one of the biggest producers of premium cars, was looking at employing a data-driven approach to price forecasting commodity prices, to combat high-volatility in the market. A major challenge was with the battery-grade commodities like Lithium and Cobalt as their prices have been extremely volatile. The other challenge was the unavailability of import/export data, global inventory levels, and other commodity-related prices.
The figure above indicates the different challenges in the project
Commodity price forecasting requires an in-depth domain knowledge of commodities and a thorough understanding of both qualitative and quantitative approach of forecasting.
Here’s the approach we followed to get a holistic view of commodity price levers:
The figure indicates the step-by-step approach in solving the problem
• Historical data study helped in identifying price trends, benchmarks and impact of various global events on the price of the commodity.
• Commodity prices are affected by external events, which can be captured by identifying peaks or change in trend/pattern in price data.
• Key drivers of commodity price can be identified by understanding the evolution of commodity into the market.
• Impact of events and drivers on commodity price can be understood by using statistical techniques likes correlation, hypothesis testing and generalized linear model (GLM) and the impact, in turn, can be used as a reference when such events occur in future.
• Quantitative forecasting is broadly classified into univariate, multivariate and advanced techniques. Due to the limited amount of data and monthly level granularity, it is ideal to use univariate or multivariate forecasting techniques.
• Although there are many methods used for accuracy calculation, MAPE (Mean Absolute Percentage Error) is used in time series forecasting because it is independent of the scale of the output variable. MAPE for Naïve method and other methods was calculated across different time stamps and finally, the model which showed better performance across time was selected.
• ARIMA (Autoregressive Integrated Moving Average)
• SARIMA (Seasonal Autoregressive Integrated Moving Average)
• ARCH/GARCH (Autoregressive Conditional Heteroskedasticity)
• Simple Exponential Smoothing
• Holt-Winters Double Exponential Smoothing
• Holt-Winters Triple Exponential Smoothing
• Univariate Unobserved Component Model
• SETAR (Self-Exciting Threshold Autoregressive)
• ARIMAX (Autoregressive Integrated Moving Average with Exogenous Variable)
• SARIMAX (Seasonal Autoregressive Integrated Moving Average with Exogenous Variable)
• M-UCM (Multivariate Unobserved Component Model)
• VARIMA (Vector Autoregressive Integrated Moving Average)
• VECM (Vector Error Correction Model)
• By leveraging our proprietary AI engine Co.dx, we were ablet to utilize pre-built ML workflow for commodity price forecasting and cut down build time by 60%. Customizations were made in the data ingestion phase to account for multiple external data sources and additional hybrid forecasting models were added to existing time series techniques. The entire workflow was automated by Co.dx’s code modules, and the data was configured into the workflow to generate analytical results.
• The results generated on Co.dx were pushed to Tableau to create a dashboard. The dashboard was created so that the procurement and supplier quality team can easily consume the output. The dashboard had multiple forecasts, comparison of the forecasts with the expert judgemental forecasts and insights on the past price trends.
• While the forecasts were relatively better than the expert forecasts, procurement managers still had to rely on observing the market behaviour to look out for any geopolitical events, economic events or any other news that would directly impact the price of the commodity. To account for this, a new feature was developed in the dashboard that actively lists the news from various websites and displays the headline, which on clicking, would redirect to the news article. This feature would help the procurement manager to understand whether the price of the commodity will increase or decrease and based on the judgement he can modify the forecast.
After applying different univariate and multivariate techniques, the best model was selected and used for forecasting. The forecasting application helped improve accuracy by 10 – 30 percent basis points across commodities, with ability to explain fluctuations with external events.
Organizations can improve their commodity price forecasting capability by incorporating AI and Machine learning which can not only predict accurate prices for the commodities but also recommend when to purchase the commodity. A Contextual AI Price Forecasting application can provide a holistic view of the commodity, allowing for informed purchase negotiations and decisions, in automobile and other industries, by customizing data and analyses to the specifics and dynamics of these domains. The availability of purchase data over time, can offer insights on key price drivers and make way for smart budgeting by tuning into future price risks well in advance. We call it having the contextual AI edge.