The NAIRU is 4.3%

One problem with using the NAIRU as a guide to policy is that most analyses do not include a useful theory of how and why the NAIRU shifts over time. By now, everyone acknowledges that it does shift. Otherwise the late 1970s and the late 1990s (as experienced by the US) could not have happened in the same country. Unfortunately, the usual way of dealing with these shifts is to treat them as a purely empirical phenomenon. The problem with that approach is that we don’t find out about the shifts until long after they happen.

Here I will make the bold claim to have a useful theory of NAIRU shifts. It’s not a new theory, really, but it has, in my opinion, gotten a lot less attention than it deserves (and as a result, the whole NAIRU concept has gotten more abuse than it deserves). The idea is that inflationary pressure from the labor market is determined by the relationship between the number of unemployed workers and the number of available jobs. In general, the more jobs there are available, the more inflationary a given level of unemployment will be. If only a few firms want to hire, for example, then most firms will not have occasion to raise wages even if the unemployment rate gets quite low.

The specific relationship between unemployment, job vacancies, and inflation can be estimated empirically by including a job vacancy variable in a standard Phillips curve specification. The Phillips curve results can be combined with information about the Beveridge curve – the inverse relationship between unemployment and job vacancies over the business cycle – to provide an estimate of what unemployment rate is consistent with stable inflation, and voila: the NAIRU!

It so happens I’ve done just that. The whole procedure isn’t very fancy, but I think it provides a sensible first cut. My Phillips curve has no control variables. (It does have a third-order polynomial distributed lag on past inflation and a constraint that the coefficients sum to one – the latter necessary to identify the NAIRU.) The proxy for job vacancies is based on a combination of the Conference Board Help Wanted Index and the Monster Employment Index. (I won’t go into the details of how I combine them, but the Conference Board index usually dominates, because newspapers still have the lion’s share of the want ad market.) I use a purely visual process to ascertain the normal slope of the Beveridge curve, because I don’t have time right now for the fancy econometrics needed to estimate a shifting relationship between two noisy variables. (I assume that all shifts in the Beveridge curve are parallel shifts.) I use monthly data going back to 1953.

As expected, both unemployment and job vacancies are highly significant in the Phillips curve. I took a line through the data point for March 2006 to represent the current position of the Beveridge curve. Conclusion: See the title of this post.