We're live blogging Commodity EDGE today and tomorrow. If you can't join us at the event, join us virtually on Spend Matters and MetalMiner. Or stop by tomorrow for the main event at the Intercontinental Chicago O'Hare -- it's not too late to register onsite.
In the third installment of this live blogging series on Omer Abdullah's presentation (Omer is Co-Founder and Managing Director of The Smart Cube, if you're curious), we'll step back for a minute and suggest why developing forecasting and predictive models is an essential component of well-informed category and risk management perspectives. During his presentation, Omar suggested that "the value that strong forecasting" is clear. Specifically, he argued that effective forecasting could enable:
- "Buying and inventory-level decisions based on expected seasonal price fluctuations"
- "Better understanding of commodity price movements = fruitful negotiations for consistent/predictable deliveries"
- "Understanding margin volatility of suppliers = better negotiations and sourcing decisions"
- "Better savings/risk management through hedging and cross hedging"
- "Better budgeting and supply planning accounting for events/ disruptions"
- "Use of substitutes and long term change (weaning off of a volatile commodity)"
At the end of the commodity day, Omer cautions the importance of remembering that "prices are dependent on fundamentals and not always determined by speculation" -- which is precisely why forecasting is such a valuable exercise. Specifically, "volatility is mostly a function of events -- longer term trends follow fundamentals" even when "intra day and short term price changes are driven by speculators, arbitragers and market participants." Looking at forecasting from this context, it is critical to not only use "a mix of various techniques to determine fundamental factors driving prices" but also to understand that variables and specific correlated factors and inputs can evolve overtime, changing based on "economic conditions and time," among other factors.
In other words, effective forecasting is about a process -- not about building one-off models even if you can get the R-squared correlation to a level at a point in time which provides high levels of confidence in the predictive elements of the equation.