What’s the natural rate of unemployment? Answers from forecasters

ABSTRACT

For its panel survey of professional macroeconomic forecasters, the US Federal Reserve asks panelists whether they use the ‘natural rate of unemployment’ (NRU) or ‘non-accelerating inflation rate of unemployment’ (NAIRU) in formulating their forecasts and, also, to estimate the NRU/NAIRU. This paper studies responses and non-responses to those questions in order to interrogate forecasters’ views of the NRU/NAIRU. We find that most panelists in most years say they do not use the NRU/NAIRU, and they do not provide an estimate of it. Among those who do, their NRU/NAIRU estimates differ but usually do not differ significantly from the average of recent unemployment rates. We fail to find much evidence – in either their inflation and unemployment forecasts, NRU/NAIRU estimates, or revisions to those estimates over time – that panelists who use the NRU/NAIRU treat it as an unemployment rate below which inflation rises and above which inflation falls. Finally, we find that the NRU/NAIRU has no informational value for panelists who say they use it; they are no better at accurately forecasting inflation or unemployment.

KEYWORDS:

• Unemployment
• inflation
• NAIRU
• Survey of Professional Forecasters

JEL CLASSIFICATION:

• E24
• J21
• E31
• B59

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. The first recorded instance of Friedman using the NRU phrase was in comments at a conference on wage-price controls from the year before his AEA presidential address (Nelson 2018, 4). In those comments, he said ‘there is what might be termed a “natural” level of unemployment […]. This level of unemployment is the level at which real wages […] would have a tendency to behave in accordance with productivity’ (Friedman 1966, 60; his emphasis). The Council of Economic Advisers (CEA) had previously advised that productivity growth should be a guide for wage and price growth (CEA [1962] 1966).

2. The NAIRU term can be traced to at least Tobin (1980, 24) who traced it to Modigliani and Papademos (1975) who called it the ‘noninflationary rate of unemployment’ and distinguished it from Friedman’s NRU. For them, Friedman’s NRU was only one theoretical construct that implied a constant-inflation unemployment rate (Modigliani and Papademos 1975, 145).

3. On the pros and cons of this anonymity, see Croushore (1993, 8). Of note, for the years of the SPF we study, the Fed offers the following warning about their treatment of individual forecasters who change firms: ‘If a forecast seems associated more with the firm than the individual, the identification number [ID] stays with the firm’ (Federal Reserve Bank of Philadelphia 2017, 34). Thus, the same ID does not necessarily track the same forecaster over time. We will nevertheless treat each ID as a unique forecaster because we do not observe when such switches occurred and, at least in the judgment of the Fed, the forecasts were similar enough to continue to associate them with the same ID.

4. We can only speculate about why the Fed added the NRU/NAIRU questions to the SPF in 1996, but transcripts of the Federal Open Market Committee (FOMC) meetings from around the same time suggest they were debating the NRU/NAIRU (Meade and Thornton 2011). Similarly, we can only speculate about why the NRU/NAIRU questions are asked just once per year, but in a different context, the Fed justified asking about some variables annually by saying ‘these variables are not expected to change much from quarter to quarter, [so] we include [them] in the survey just once a year’ (Croushore 1993, 7, n. 2). Of note, the European Central Bank conducts a similar survey of professional macroeconomic forecasters, but they do not ask about the NRU or NAIRU, at least as of writing.

5. See Fed (2006) for an example of a survey form with those two questions.

6. It is not implausible that a forecaster could use the concept of an NRU/NAIRU in formulating their forecasts yet be unable or unwilling to provide a point estimate of its current value. Indeed, in an episode recounted by James Galbraith (2000, 182–83), Joseph Stiglitz staunchly defended the use of the NAIRU but refused to provide an exact estimate of its numerical value or even a range of estimates!

7. Hall and Sarget (2018) go so far as to claim that, among economists in recent decades, ‘the idea of the natural rate or NAIRU has become uncontroversial’ (p. 127). The above-discussed survey evidence refutes any claim of universal acceptance and even a cursory look at the universe of economists would refute such a claim, too (Eisner 1995; Galbraith 1997a,b; Isaac 1993; Madsen 1998; Sawyer 2001; Shulman 1989; Storm and Naastepad 2007, 2008 etc.). That said, the NRU/NAIRU seems to be more widely accepted among economists than forecasters.

8. Solow never told that joke in a publication, to the best of our knowledge, but said something similar when he said: ‘A natural rate [of unemployment] that hops around from one triennium [i.e. 3-year period] to another under the influence of unspecified forces, including past unemployment rates, is not “natural” at all’ (Solow 1986, S33). That the level of any actual or estimated NRU/NAIRU could depend on past or present unemployment rates due to demand- rather than supply-side effects, hysteresis effects, or other forces is not an idea unique to Solow (Baker 1998; Eisner 1997; Gordon 1988; Storm and Naastepad 2007; Thirlwall 1983 etc.), but Solow’s above-given claim is perhaps the most specific and directly testable claim about the relationship between the NRU/NAIRU and actual unemployment rates.

9. One outlier is not shown in the histogram; the estimated beta for one panelist in one year (2006) was about five with a standard error of about one third for a normalized beta of 15. That panelist forecasted that inflation would fall as the unemployment rate rose to the NRU/NAIRU from below. There were also two panelists in one year (2005) for whom Equation (1) could not be estimated because they forecasted that inflation would fall as the unemployment rate remained constant at their NRU/NAIRU estimates.

10. Recall from Figure 2 that most panelists always or never estimate the NRU/NAIRU, so it is unusual for a panelist to estimate the NRU/NAIRU in a given year and then stop estimating it in the next year if they are still in the survey.

11. Similarly, James Galbraith wrote around the same time that, ‘The present estimates and reestimates [of the NAIRU] seem largely a response to predictive failure. […] The [econometric] literature simply observes that inflation hasn’t occurred and so the previous estimate must have been too high’ (Galbraith 1998, 180).

12. We do not use the absolute value or square of a forecasting error because we wish to examine whether a panelist responds differently to over- vs. under-predicting. For the realizations, we use the official statistics that would have been available, according to the Fed’s real-time dataset. The only exception is as follows. Recall that, for a given year’s third-quarter survey, a panelist forecasts the previous quarter. We treat that as the realization of their forecasts from last year’s third-quarter survey to better capture their view of the accuracy of those forecasts.

13. Galbraith’s claim admittedly refers to only ‘some economists’ and not everyone, so perhaps that qualification reconciles his claim with our finding.