![Sample our Economics, Finance,Business & Industry journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5931/TOC-Subject-Sample-2021-Economics-Finance-Business-Industry.jpg"})
Abstract
New Zealand shares a wealth of common interests and experiences with Australia. This has tempted some to assume that these economies form an ‘Economic Club’, in which one would expect to identify common aggregate trends and growth experiences. In this paper we present results that test, and generally reject, convergence in labour productivity across Australia and New Zealand, using both aggregate and disaggregate, industry‐level data. We find that only two industries satisfy our definition of Conditional Convergence (Agriculture, Forestry and Fishing and Cultural and Recreational Services), and that the Mining and Wholesale Trade industries have particularly important roles to play in explaining the measured divergence. Cointegration‐based tests reveal more stochastic trends governing Australian productivity than in New Zealand. The evidence suggests, therefore, that the underlying growth processes of the two economies are fundamentally different, thereby questioning the relevance of aggregate comparisons between them. New evidence using industry‐level data does not, therefore, resolve the aggregate‐level ‘non‐convergence puzzle’ identified here, and elsewhere.
Acknowledgements
The views expressed in the paper do necessarily reflect those of the Reserve Bank of New Zealand. We wish to thank two anonymous referees for their helpful comments and suggestions that have improved the clarity of the arguments presented. The second author acknowledges support from Marsden Fund grant UOC108. We would like to thank Kevin Fox for useful comments on earlier drafts. Normal caveats on errors and omissions apply.
Notes
1. Lucas (Citation1988) discusses models that generate sectorally varying growth.
2. See, for example, Jones (Citation2002).
3. See Bernard & Durlauf (Citation1995) and redefined definitions below.
4. Campbell & Perron (Citation1991) show that the exclusion of deterministic components that are in the ‘true’ data generating process, which are not included in the testing procedure, can cause the power of the test to be reduced. This uncertainty about the deterministic components of the regression leads to an iterative testing procedure if the null hypothesis cannot be overturned (see Enders, Citation1995).
5. It is clear that our testing procedure is sequential in nature. Type I errors made in a particular stage of testing will, therefore, carry through to all subsequent stages of the testing procedure. This paper makes no attempt to correct for this ‘pre‐testing bias’ in the size of our tests.
6. For a detailed discussion of data issues discussed in this section, see Matheson (Citation2003), ch. 4, or contact the authors for an extended summary.
7. See www.abs.gov.au for details of this classification.
8. Available from www.stats.govt.nz.
9. Gross value added measured at basic prices is the sum of the compensation of employees, the gross operating surplus, and the consumption of fixed capital.
10. PPPs are price relatives that measure the relative price of the same bundle of goods and services in different countries. The OECD calculates aggregate PPPs on a tri‐annual basis (see www.oecd.org for a discussion of PPPs, and PPPs for Australia and New Zealand).
11. See www.eco.rug.nl/GGDC/icop/html for a detailed discussion of industry‐level price comparisons.
12. Using weekly data from a specified period in the middle of the quarter avoids the problem of varying number of pay periods quarter to quarter.
13. EViews 4.1, Quantitative Micro Software.
14. Each country’s market sector productivity series is computed using the ratio of a Tornqvist (Citation1936) output index to a Tornqvist labour index. The indexes are based to 89:Q1, with the base level of the indexes calculated as the ratio of an additive output index to an additive labour index.
15. MacKinnon (Citation1996) critical values.
16. The SBC generally chooses shorter lag lengths than other information criterion. We use this criterion because our lag length is limited by a small sample size.
17. The trace statistic tests the null hypothesis of a number of vectors less than r against a general alternative. The maximal eigenvalue statistic, however, tests the null hypothesis of a number of vectors of less than r against a specific alternative hypothesis of r+1 vectors. Given the restricted nature of the alternative hypotheses in maximal eigenvalue tests, it is not surprising that the number of cointegrating vectors revealed by the statistic is generally less than the number implied by the trace statistic. We maximize the possibility of retaining our null hypothesis by testing using trace statistics: for a further discussion of these two tests see Enders (Citation1995).
18. Stock and Watson showed that the number of stochastic trends in a system of variables is the number of variables in that system minus the number of cointegrating relationships between the variables in the system. The number of stochastic trends in a system of stochastic variables is simply the number of variables in which each variable in the system can be expressed.
19. Notice that the Zivot and Andrews testing procedure shows a change in intercept and a change in slope for testing for a unit root in the Retail Trade differential. When estimating the deterministic components of this series the changing trend is found to be insignificant at the 5% level.
20. We can say the convergent industries, Agriculture, Forestry, and Fishing and Cultural Recreational Services, have contributed little (if any) to the market sector divergence.