Assumptions Drive Forecasting Success and Error

If you have ever been called upon to build a financial model or market forecast for any sort of product, you know such models are exceedingly sensitive to the assumptions used to create the model. 

So difficult are such projections that clients might well--n volatile or brand-new markets--be satisfied with forecasts that miss within an order of magnitude (10 times eventual reality). Any forecaster missing by that much in an established market will have problems, as divergence from reality should never reach two to three times the actual circumstances. 

But that can happen--even in mature markets--when forecasters use accurate data without considering the assumptions about that data. 

Cable TV operators in many markets, for example, sell multiple products, but send a single bill. And that can lead to false assumptions about household spending, for example. Is “average” U.S. household spending on linear video closer to $200 per month or $80 per month? It matters. 

Some use the higher figure, but the actual figure is closer to the latter. The mistake is easy to make. A household purchasing linear video and internet access plus either fixed network voice or mobile service could spend, “on average,” $200 or more per month. 

Using either a median or mean approach to generating “average,” the point is that it is reasonable for an “average” video charge to be in the $60 per month to $80 per month range, before bundling discounts. 

source: S&P Global Market Intelligence; Fierce Video 

The point is that all forecasts are extremely sensitive to assumptions made about future developments, but also about current behavior. 

Ignore for the moment the assumption that growth rates are predictable or that no unexpected macroeconomic events will occur. It is quite easy to make faulty assumptions based on an incorrect understanding of user behavior, costs or potential revenues. 

Risk of this sort arguably is lessened when the modeler has access to multiple different data sources; different methodologies for estimating the current size of the market for any product and  domain knowledge.