Written in collaboration with Dr. Ralph Finos
In an earlier post, we talked briefly about near-term and long-term forecasting – Why Do We Forecast?
Near-term Forecasting and Planning
In near-term planning we’re trying to be as accurate as possible about the outcome – regardless of how we get there. The premium is on prediction vs. explanation.
“What are my sales for established products likely to be in the next 6 months?”
Or demand pull considerations in the supply chain:
“How many toilet seat covers is Walmart going to need from me in the next 2 weeks?”
This type of forecast enables short-term resource deployment and benchmarks to facilitate near-term decision making “now decisions.” Recent history is a key factor because where you have recently been is likely to be highly correlated with where you will be in the near future. Methods that project from recent tracking data are the most useful. Moving averages and cyclical factors (i.e., seasonality) are important. Theories of market behavior are generally less important (unless you’re forecasting emerging products or markets.
Near-term planning is also vital for long-term planning. If you’re not predicting the near-term effectively, then assumptions in your long-term forecast might be incorrect. Near-term forecasting can serve as a benchmark/hypothesis test for a long-term forecast.
Long-term Forecasting and Planning
In most industries, confident forecasting can typically be accurate for 6-18 months. It should draw on the recent past, but must focus on prediction related to likely customer behavior (i.e., adoption behavior) and market context (i.e., substitute technology, competition, economic factors). As opposed to near-term forecasting, the forecaster must create a theory or set of assumptions about market adoption in the context of value as well as market context (i.e., other products and services).
For example, technology markets are subject to evolution and possible revolution in technology itself that can redraw the landscape of a market – or create entirely new ones. The technology market forecast confidence horizon doesn’t typically go much beyond 12-18 months. Obviously, the more inclusive the market segment being forecasted (i.e. all application software), the more stable and amenable to forecasting a market will be.
However, the interesting forecasting problems (new products, new markets, new technologies – and their interplay) often involve narrow segments as entry points and complex factors that govern adoption, making prediction a challenge. Therefore, any forecast going out 2 years or more requires solid theories of adoption as well as solid metrics that are correlated with adoption allowing the forecaster to create predictive models that lave a longer half-life and are robust, yet flexible.
Useful (i.e., good) forecasts must be:
- Robust enough to enable repeat calibration of core market drivers
- Flexible enough to adapt to new market realities and drivers
These conditions are easier to state than to accomplish. Any forecaster who has created and then validated his or her forecasts against actual events can testify to the challenges.
|Join Our Mailing List|