Okun’s law revisited: a threshold in regression quantiles approach

ABSTRACT

This article proposes, for the first time, a threshold in regression quantiles approach to the analysis of Okun’s law. By applying to US data over the 1948Q1–2016Q4 period, we have three major findings. First, a single threshold is detected for both multiple and individual quantiles cases. However, the effect of threshold nonlinearity is only present in the middle to upper quantiles of the conditional unemployment distribution in the individual quantiles case. Second, the first-order autoregressive coefficients of unemployment are significantly larger in the lower-growth regime, indicating that shocks to unemployment appear to be more persistent during recessions. Finally, the Okun’s coefficients are all negative across the recessionary and expansionary regimes, confirming the validity of Okun’s law. Moreover, the Okun’s coefficients are smaller (more negative) in the lower-growth regime, suggesting that the effect of differenced output on differenced unemployment is asymmetric, and is more pronounced in recessions.

KEYWORDS:

• Okun’s Law
• Threshold
• Quantile

JEL CLASSIFICATION:

• C22
• E24
• E32

Acknowledgement

The authors are grateful to the helpful comments and suggestions by anonymous referees. Any remaining errors are our own responsibilities.

Notes

¹ We thank a referee for suggesting these alternative possibilities.

² Another is concerned with asymmetry in the Okun’s relationship, i.e. identical increases and decreases in cyclical output lead to different changes in cyclical unemployment. Moreover, There are also some studies that explore Okun’s law by age-cohort, e.g. Hutengs and Stadtmann (2013), and Zanin (2014).

³ In our application below, we select the minimum number of lags determined by the AIC and BIC criteria, which is 1 according to BIC. In addition, the differenced unemployment and output are stationary according to ADF and PP unit root tests. Moreover, in line with Pierdzioch, Rülke, and Stadtmann (2011), we do not consider the potential asymmetries of Okun’s law over the business cycle.

⁴ In an earlier version, both cyclical components of unemployment rate and real output are extracted from the raw data via the Hodrick and Prescott (1997, hereafter HP) filtering technique. The use of HP-filtered data provides qualitatively similar results. However, a referee points out that the HP-filtering technique produces a series of spurious dynamic relations that have no basis in the underlying data-generating process, e.g. Hamilton (2017), and, as a result, the outcomes of using the HP-based data are omitted but available upon request.

⁵ Suppose that , then .

⁶ http://research.stlouisfed.org/fred2/