by Stephen K. McNees and Geoffrey M.B. Tootell
July/August 1991

This article attempts to identify precursors, or indicators, of New England employment. The predictive power of a diverse array of variables is calculated and compared. However, because no single variable is likely to contain all information of predictive value, the article then explores alternative methods of combining several variables into an index or statistical "model" of New England employment growth. The variables are separated into regional, national, and expectational in order to measure the predictive value of each type of information.

In both in-sample and out-of-sample tests, a model that included all categories of variables was the most successful. However, relative to its in-sample fit, every model performed poorly out of sample. The reason for this breakdown is clear--the models were fit during the "Massachusetts Miracle" while their predictive power was tested over the New England bust. The lack of a theoretical model and deficiencies in the available data make a purely statistical approach to predicting the New England economy suspect. Some insights can be gained, however, by surveying the New England economy in a broader historical context.

Full-text article