Sunday, September 7, 2008

Forecast Error is a Fact of Life

Having spent some time doing market research and making forecasts about how much money is going to be spent, on what, in the future, one learns humility. Sometimes forecasters underestimate a growth trend, but that is fairly rare. The biggest misses I can recall in my professional career were in the early days of the Internet, where I think it is fair to say nearly everybody underestimated growth, usage and revenue potential. 

These days we have the more-typical problem, which is overestimating growth. Most of the time, forecasters are wrong about the magnitude of change, even when the basic trend is correct. Sometimes we get the trend wrong, as well, but the big variances normally are on the "magnitude of change" dimension. 

There are lots of reasons for this apparent built-in bias. In economics, a discipline too few of us pay serious attention to, statements are formally or implicitly made "ceteris paribus" (all other things being equal). In the real world, almost nothing is every really going to remain equal. 

Forecasters cannot account for the timing of economic expansions and slowdowns, wars, oil shocks, technological or business breakthroughs and other exogenous variables that shape real-world trends. And as basic as our spreadsheets have become, we can't really model more than two dimensions of change. 

I have long suspected that there is another bias working, though. Some people fund and buy research because they really want to know what might, or will, happen. It tends to be my experience, though, that many research buyers spend money for other reasons. Sometimes a forecast is a useful way of "covering your ---" when a decision already has been made. 

Often, it is an argument for pursing one avenue of investment versus others that have competing stakeholders. In such cases, more robust numbers are better. The sets of stakeholders will look for validation that their approach offers the bigger financial return. 

In some cases, robust growth statistics are needed to attract or retain capital investments. Small number do not help in that case. I'm not saying numbers routinely are "cooked." But when a client clearly wants to hear good news, one always can make reasonable assumptions at the upper end of reasonable bounds, rather than the lowest or median reasonable assumptions.

The other issue is that most forecasts cannot take into account other competing demands on resources. A given product or service could reasonably be expected to grow at certain rates if it proves popular with buyers. What cannot typically be modeled is the impact of alternate products or services that real-world buyers spend money on. 

Appetites are infinite. Budgets are not. If all growth forecasts for all products and services are tallied at one time, the result typically is that projected demand that far outstrips reality. Not all the goods and services everyone models can be bought at the predicted levels because there isn't that much aggregate demand in the whole economy. 

About all one can say is that, providing these competing claimants for spending do not "suck up all the oxygen in the room," demand for a particular product, over a specific time frame, could reach certain levels.

It appears a garden variety "ceteris paribus" explanation is at work with Internet advertising forecasts that researchers now are suddenly revising in a lower direction. Without explaining why, researchers at SNL Kagan recently downgraded U.S. cable industry advertising growth rates for 2009, but then assume growth at the former rates after 2010. 

Likewise, JP Morgan analyst Imran Khan now sees domestic online display advertising growing only 14 percent to $8.2 billion in 2008 (compared with his prior forecast of 20 percent growth to $8.6 billion), and growing 16 percent to $10 billion in 2009. Search ad spending also will grow at a slower 27 percent rate, down from an originally forecast 32 percent. International online ad-spending rates also have been similarly lowered.

Nobody can adequately incorporate macroeconomic variables, or demand shifts, into forecasts. Should forecasters permanently lower uptake expectations? No. The shift will continue, away from legacy formats and towards "targeted" and quantifiable online subsitututes. But one cannot adequately take into account unplanned real-world events, especially of the growth-stunting sort. It's simply a hazard of the business.

No comments:

Many Winners and Losers from Generative AI

Perhaps there is no contradiction between low historical total factor annual productivity gains and high expected generative artificial inte...