This is the season of economic forecasts, for which there are many uses beyond their pure entertainment value. Forecasts are most often used to assemble budget estimates and to aid in developing changes to public policy.  They are used by businesses, social service providers, educators and the like to plan infrastructure, personnel investments and other important things.  While economic forecasts are almost always wrong in the strictest sense, they can be useful, if only to give everyone a point at which to begin their arguments.  I also think forecasts provide some interesting insights into the way economic research is conducted and how it is used to formulate policy.

Almost all forecasts are constructed using long historical data from a variety of sources, like the Bureau of Labor Statistics or the Institute for Supply Management.  These historical data are then tortured to reveal relationships that can be used to make predictions about future levels of economic activity.  The mathematical waterboarding of the data can be done in two different ways. The oldest of these involves building very carefully constructed equations that relate one variable to another.  An example might be that consumers will buy more cars as population and income increases, fuel prices drop and borrowing rates decline.  How much each will affect car sales is estimated using historical data and statistical tools.

Another method involves allowing historical data to explain all these relationships.  This second method requires a lot of statistical training (most economic doctoral programs look a lot like applied mathematics degrees), and enough economic theory to figure out what variables must be included. The forecasting models my center employs (the Indiana Econometric Model) uses both approaches to predict future economic performance at the county, regional and state level. 

Given the vitriol over economic policy debates, it would seem that different economic ideology drives these forecasts.  It does not.  A Marxist or libertarian economist would not really disagree on the criterion for selecting the best model: its past performance.  The real difficulties in economic forecasting are the fact that, unlike gravity, human behavior changes and so constancy in some relationships is elusive.  Luckily we are fairly good at predicting these changes, too.  It is data that gives us our biggest headaches.

For some economic variables we have monthly data, for others we have only decennial census data.  Also, the government collects data that is convenient, not data that is theoretically appropriate.  So, for example, we are never quite certain about the size and product line of the 7.4 million U.S. businesses or the technology skills of 170 million workers.  But the deeper problem is the frequency of economic phenomenon we might wish to study.  There have been fewer recessions since 1950 than there are hurricanes in a typical year, and far fewer data elements.

As a consequence economic forecasters are bad at predicting turning points in recessions and recoveries.  These are fortunately infrequent, but a well thought out economic forecast offers a great deal of useful information.