Academic journal article Economic Inquiry

Panel Tests of Okun's Law for Ten Industrial Countries

Academic journal article Economic Inquiry

Panel Tests of Okun's Law for Ten Industrial Countries

Article excerpt



"Okun's Law" describes an empirical regularity between unemployment and real output growth: "... each extra percentage point in the unemployment rate above four percent (i.e., full employment) has been associated with about a three percent decrement in real GNP" (Okun [1962, 9]). (1) Okun's Law has proved to be one of the more durable relationships in modern macroeconomics, and is now often presented as an element of core beliefs, as in textbooks by Hall and Taylor (1993, 77) and in policy discussions by Blinder (1997). Durability and simplicity being relatively scarce commodities in macroeconomics, Okun's Law is appealing as a shorthand guide to policy outcomes.

Whether Okun's Law travels well is another question. Cross-country comparisons of the relationship between unemployment and output have been estimated for other countries by Knoester (1986), Paldam (1987), and Moosa (1997), among others, but magnitudes vary depending on country and time period. Estimates in these papers of the increase in output growth associated with unemployment reduction of one percent range from a low of around two percent in the U.S. and Canada, to three to five percent in Europe, to more than 10 percent in Japan, with estimates for Europe tending to be higher for more recent time periods.

Because of the way that Okun's Law is framed, tests of its proposition involve two imposing empirical problems. First, as usually stated, the Law refers to the relationship between departures of real GDP (2) and unemployment from their long-run, or equilibrium trends, often referred to as "potential GDP" and the "natural rate of unemployment," respectively. These equilibrium trends must be estimated, either by time series methods as in Moosa (1997) or by construction using a modified production function approach in which potential output is estimated using fully employed inputs, as in Gordon (1984) or Adams and Coe (1989). Either of these Fapproaches invites considerable criticism of assumptions and methods; see, for example, the comments of Clark (1984) and of Baily (1984) on Gordon (1984).

Second, in formulating his law, Okun seems clearly not to be referring to a ceteris paribus relationship between changes in the unemployment rate and changes in real GDP. Writing later, Okun posited that other factors and inputs would be changing pari passu with employment: "The 3 percent result [from a reduction of one percent in the unemployment rate] implies that considerable output gains in a period of rising utilization rates stem from some or all of the following: induced increases in the size of the labor force; longer average weekly hours; and greater productivity" (Okun [1970, p 140]). Thus, a complete specification of the effect of changing unemployment rates on output would include factors such as capital inputs, labor hours and participation rates, all measured as deviations from long-run trends.

This article addresses these two obstacles by using a recent trend-cycle decomposition developed by Baxter and King (1999) to measure "gaps" in output, unemployment, capital stock and the labor force. The Baxter-King "bandpass" filter extracts the trend component of economic time series by eliminating both high frequency (those less than two years) and low frequency (more than eight years) components. It is both flexible and easy to implement. By using the bandpass filter to filter all the time series using the same methodology, this paper departs from previous research of Prachowny (1993) and Attfield and Silverstone (1997) incorporating capital and labor to augment estimates of Okun's Law.

Additionally, using the bandpass filter allows the estimates in this article to incorporate more timely data reflecting the shift to lower levels of unemployment in the U.S. and higher levels of unemployment in Europe and Japan during the 1990s. The latest year of observation for the constructed data sets of Gordon and of Adams and Coe used, for example, by Prachowny (1993) and Attfield and Silverstone (1997), is 1988. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.