A forecast based NAIRU measure of the US

Abstract

This paper proposes a method to estimate the NAIRU for the U.S. It shares the notion of Estrella and Mishkin (1999) that defines the NAIRU as a leading indicator of inflation changes over the policy horizon. Our alternative construction offers a more theoretically sound and practically useful estimate of the NAIRU.

Acknowledgements

I would like to thank the editor of the journal. This research was supported by Yonsei University Research Fund of 2004. The usual disclaimer applies.

Notes

¹ See, for example, King and Watson (1994) and Staiger et al. (1997).

² See Banerjee et al. (1993, pp. 164–168) for a discussion on ‘unbalanced regression’.

³ The exogeneity of unemployment also has some theoretical grounds. For example, it is embodied in the typical Keynesian model which assumes that the unemployment rate is essentially an indicator of demand. Real business cycle models, at the other extreme, also agree on this condition, but with the different interpretation that movements in real variables such as the unemployment rate are unaffected by nominal variables such as inflation.

⁴ By comparison, the unconditional mean of Δ u is 0.0113, which is close to its sample average of 0.0099.

⁵ The NAIRU estimates of EM are reproduced using quarterly data with a 4-quarter-ahead, 4-quarter horizon, and 4 lags of both explanatory variables. They end at 1999:Q4 as the procedure requires future inflation data for a two-year period.

⁶ One of the main reasons is that such ratios of random variables are known to be fat-tailed in finite samples, while the delta method approximates them by a normal distribution. Further to this, the EM procedure uses a multi-horizon prediction regression in which fat tails are not unusual as large residuals are frequent. In contrast, the bootstrap method approximates the exact sampling distribution of NAIRU estimates without recourse to such a specific distribution being normal. The Monte Carlo evidence by Li and Maddala (1999) confirms that it gives more accurate estimates than the delta method in obtaining confidence intervals for ratios of parameter estimates.