Okun’s Coefficient states that there is a three to one relationship between outputand unemployment, a relationship that policy makers can exploit. Numerous studiesconducted at the national level have confirmed the overall validity of this empiricalproposition, even while calling into question some of the details. Using U.S. State leveldata, this paper tests for Okun’s coefficient over the period of 1977- 1997, and 1997 –2007. This paper finds that Okun’s coefficient does exist at the state level, though it isgenerally lower than national U.S. measures, averaging about 1.5 to 1, a result that is inline with other recent regional studies, but which differs from previous U.S. studies.Pooled data also strongly suggests that Okun’s coefficient in the U.S. is asymmetrical,negative output changes are associated with larger changes in employment than positiveones, though this result is sensitive to how one defines downturns. No strong evidence ofa structural break was found.Additionally, this paper tests the resulting coefficients against a variety of laborsupply and demand variables, to see which ones influence the existence and size of therelationship. These results support the generally accepted notion that high unionizationrates, large manufacturing sectors, and high tax rates lead to a higher Okun’s coefficient.The size and composition of the labor force is also important. As more women and youthenter the work force, unemployment rates become less responsive to output growth.Political issues did not seem to have much effect, though some government policies did.Evidence is presented of interstate migration as an explanation of dynamics in Okun’scoefficient, though no statistically significant results were found.Finally, an extended form of Okun’s coefficient, one that accounted for variancein capital stock and capital and labor utilization rates, was tested for the ten mostpopulous U.S. states. These tests gave a lower estimate for Okun’s coefficient – as mostmacro theories with diminishing marginal returns to labor, would predict. However, theydid not seem to add much explanatory power to our original estimates, and in most casesgave a less robust result.
