Parameter variation in the 'log t' convergence test

ABSTRACT

This article estimates a local linear version of the model used in the ‘log t’ convergence test. It documents the economically and statistically significant within-sample variation in the estimated value of the key parameter of that test when applied to data for 18 OECD countries during the twentieth century. This variation suggests the substantial waxing and waning of the forces driving convergence, possibly due to low-frequency shocks and changes in the level of economic integration.

KEYWORDS:

• Cross-country convergence
• σ-convergence
• transition
• globalization

JEL CLASSIFICATION:

• O11
• O47
• F63

Acknowledgments

I thank Steven Durlauf and Paul Ruud for comments on an earlier draft, Diana Henry for efficient research assistance, and Donggyu Sul for supplying the OECD dataset used in Phillips and Sul (2009). Financial support from the Lucy Maynard Salmon Research Fund is gratefully acknowledged. All errors are mine.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ This test can be interpreted as a type of σ-convergence test. See Islam (2003), Durlauf, Johnson, and Temple. (2005), and Johnson and Papageorgiou (2018) for surveys of the convergence literature and expositions of the different types of convergence tests.

² The countries are: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Italy, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, the UK and the USA. The data are available at http://doi.org/10.3886/E108421V1.

³ The results here are qualitatively robust to other choices of $K\left(s\right)$available in locfit as well as to variations in the bandwidth so that the window contains between 5% and 50% of the observations.

⁴ This article is not intended to present tests of the convergence hypothesis per se so De Long’s (1988) well-taken point about the sample selection issues involved in using data a group of successful countries in this context does not detract from the conclusions.

⁵ Repeating the calculations for Figure 1 with the raw data, so that both the trend and cyclical parts of the variation are included, yields a qualitatively similar result although, as might be expected, the estimated derivative is more variable, dipping into the growth-rate convergence zone during the early-1980s recession, and has a wider confidence interval.

Additional information

Funding

This work was supported by the Lucy Maynard Salmon Research Fund at Vassar College.