Regional Dimensions of the Australian Business Cycle

Abstract

Dixon R. and Shepherd D. Regional dimensions of the Australian business cycle, Regional Studies. This paper deals with the identification of, and explanations for, co-movement in regional business cycles using employment data for Australian states and territories (regions). It shows that both raw growth rates and the deviations from a Hodrick–Prescott trend reflect noise in the series as well as any cycle, but that both can be manipulated so as to reveal their cyclical components. It is found that growth rate models are dominated by noise, whereas the Hodrick–Prescott filter allows identification of the cyclical features of interest. The extent of co-movements in cyclical fluctuations in employment amongst the regions can be explained by regional industry structure and size of region.

Dixon R. and Shepherd D. 澳大利亚商业循环的区域层面， 区域研究。利用澳大利亚国家及区域就业数据，本文考察了区域商业循环中协同运动的确认及相关解释。研究表明，尽管原始增长率及 Hodrick–Prescott 偏离趋势反映出在不同阶段以及循环中存在干扰，但可通过控制上述干扰揭示出循环的组成要素。研究发现，增长率模型受制于干扰，而 Hodrick–Prescott 过滤允许收益循环特征的出现。区域就业循环波动中协同运动的程度可以通过区域产业结构以及区域大小来解释。

区域就业 区域商业循环 协同运动

Dixon R. et Shepherd D. Le cycle économique en Australie d'un point de vue régional, Regional Studies. A partir des données sur l'emploi auprès des états et des territoires (régions) australiens, cet article cherche à identifier et à expliquer le mouvement simultané des cycles économiques régionaux. On démontre que aussi bien les taux de croissance bruts que les écarts par rapport à une tendance Hodrick–Prescott reflètent de la perturbation dans la série ainsi que pendant aucun cycle, mais que l'on peut manipuler tous les deux pour dévoiler leurs éléments cycliques. Il s'avère que les modèles des taux de croissance se voient dominer par la perturbation, alors que le filtre Hodrick–Prescott permet d'identifier les caractéristiques cycliques d'intérêt. L'importance des mouvements simultanés des fluctuations cycliques de l'emploi à travers les régions peut s'expliquer par la structure industrielle régionale et par l'étendue de la région.

Emploi régional Cycles économiques régionaux Mouvements simultanés

Dixon R. und Shepherd D. Regionale Dimensionen des australischen Geschäftszyklus, Regional Studies. In diesem Beitrag werden gleichgerichtete Bewegungen in den regionalen Geschäftszyklen mit Hilfe der Beschäftigungsdaten von australischen Bundesstaaten und Territorien (Regionen) identifiziert und Erklärungen dafür erörtert. Es wird gezeigt, dass sich sowohl in den Bruttowachstumsraten als auch in den Abweichungen von einem Hodrick–Prescott-Trend Verzerrungen in der Reihe sowie in jedem Zyklus widerspiegeln, wobei aber beide so manipuliert werden können, dass ihre zyklischen Komponenten zu Tage treten. Wir stellen fest, dass die Modelle der Wachstumsraten von Verzerrungen dominiert werden, während der Hodrick–Prescott-Filter eine Identifizierung der relevanten zyklischen Merkmale erlaubt. Das Ausmaß der gleichgerichteten Bewegungen in den zyklischen Fluktuationen der Beschäftigung in den verschiedenen Regionen lässt sich anhand der regionalen Branchenstruktur und der Größe der Region erklären.

Regionale Beschäftigung Regionale Geschäftszyklen Gleichgerichtete Bewegung

Dixon R. y Shepherd D. Dimensiones regionales del ciclo comercial australiano, Regional Studies. En este artículo identificamos y explicamos los movimientos conjuntos en los ciclos comerciales regionales usando los datos de empleo para los estados y territorios (regiones) australianos. Demostramos que tanto los índices de crecimiento bruto como las desviaciones de una tendencia Hodrick–Prescott indican una distorsión en las series así como en cualquier ciclo, pero ambos pueden ser manipulados para revelar sus componentes cíclicos. Observamos que los modelos de índices de crecimiento están dominados por las distorsiones, mientras que el filtro Hodrick–Prescott permite la identificación de las características cíclicas de interés. La extensión de los movimientos conjuntos en las fluctuaciones cíclicas del empleo de las diferentes regiones puede explicarse por la estructura de la industria regional y el tamaño de la región.

Empleo regional Ciclos comerciales regionales Movimiento conjunto

Keywords:

• Regional employment
• Regional business cycles
• Co-movement

JEL classifications:

• C22
• E32
• R11

Acknowledgement

The authors would like to thank the referees for their very helpful comments and suggestions.

Notes

Data for gross state product in Australia are reported only on an annual basis and are not available prior to 1987.

The data series commences in 1978 because that is when a regular and reliable survey of employment (the Labour Force Survey) was first introduced in Australia. For obvious reasons data are used which have been seasonally adjusted data by the Australian Bureau of Statistics (ABS). Seasonal adjustment by the ABS is carried out using a filter which is a variant of the X11 scheme. A comparison of the raw data with the seasonally adjusted data shows no evidence that seasonal adjustment has introduced serial correlation into the data set (indeed, it would be most unusual if seasonal adjustment were to introduce serial correlation at business cycle frequencies) or that it has resulted in ‘phase shifting’.

Enders (2010) provides an excellent and very accessible discussion of the relationship between shocks (noise) and the cycle.

The higher-order AR parameters may reduce the degree of persistence and alter the form of the cycle, if they are negative.

For a useful discussion of the duration of business cycle phases, see Haberler (1936).

The properties of the Butterworth filter and the relationship between the time and frequency domains are discussed by Gomez (2001), Harvey and Trimbur (2003), Iacobucci and Noullez (2005), and Shepherd and Dixon (2008). Given the similarity between the Hodrick–Prescott filter and the Butterworth high-pass filter, the application of this procedure to the Hodrick–Prescott detrended series is similar to applying a band-pass filter to extract the cyclical component, with the noise extracted by the high-pass wall of the band-pass filter.

This conclusion is based on the application of Fisher's z-transformation test for testing the equality of (independent) correlation coefficients. (All comparisons of correlations made in this paper use this test.) Using Fisher's test for the equality of the (population) correlation coefficients, one fails to reject the null that the two corresponding (population) correlation coefficients are significantly different at the 5% (and indeed at the 10% level) for any of the pairs in Tables 1 and 2.

These are the pairings: NSW:VIC, VIC:QLD, VIC:SA, VIC:WA, QLD:WA, SA:WA and SA:TAS.

While most of the (twenty-eight) correlations in the tables which report co-movement of the cyclical component (Tables 2 and 4) are significantly different from zero (sixteen in Table 2 and twenty-two in Table 4), a much smaller number of the correlations in the tables which report co-movement of the noise component (Tables 3 and 5) are significantly different from zero (nine in Table 3 and three in Table 5).

The Hodrick–Prescott filter was applied to the logarithms of the employment series, with the smoothing parameter set at the conventional 1600 value used for quarterly data. This preserves the variation in the series at cycles of approximately eight years or less.

The authors are very grateful to a referee for drawing their attention to this.

Only one of the correlations in Tables 10 and 11 are significantly different from their counterpart in Tables 7 and 8, and that is the pairing for TAS and SA in Tables 7 and 10 (which deal with the cycle).

While most of the (twenty-eight) correlations in the tables which report co-movement of the cyclical component (Tables 7 and 10) are significantly different from zero (twenty-five in Table 7 and twenty-seven in Table 10), a much smaller number of the correlations in the tables which report co-movement of the noise component (Tables 8 and 11) are significantly different from zero (eight in Table 8 and three in Table 11).

One of the original and still one of the key papers in this area, by Clark and van Wincoop (2001, p. 72), find no statistically significant relationship between co-movement and measures of similarity of industry structure. A previous study looking at gross state product and hours worked for the six states of Australia over the period 1986–2005 found ‘no evidence that differences in [co-movement across] states […] are related to differences in industrial structure, suggesting that alternative approaches which give greater weight to industrial structure are unlikely to provide much additional information’ (Norman and Walker, 2007, p. 365).

In addition to the Krugman Index, the present authors have also used other measures of structural similarity or dissimilarity including differences in the size of the manufacturing sector, differences in the size of the services sector, and differences in the size of mining and primary sectors. In each case the variables had the same sign and significance (or insignificance) as the Krugman Index, except that, as judged by explanatory power, the Krugman Index out-performed all of these measures. This is not surprising as the Krugman Index includes in the one measure differences in the size of all sectors, not just one or a few. For this reason only results will be reported where the Krugman Index was used to capture the role of industry structure.

Clark and van Wincoop (2001), Barrios et al. (2003), Barrios and De Lucio (2003), Imbs (2004), Belke and Heine (2006), and Poncet and Barthelemy (2008) provide examples of the use of this variable in the context of studying regional co-movements. The Krugman measure is related to a measure with a long history in regional studies called the ‘coefficient of regional specialisation’ (Isard, 1960, pp. 270ff.; Dixon and Thirlwall, 1975, pp. 16f.). The present empirical work finds that the results from using the coefficient of regional specialization and the Krugman Index are essentially the same (this is not surprising since one is a simple linear transformation of the other). Since it is common for researchers in this area to use the Krugman measure, the results for it will be reported herein.

Data for employment by industry are available for fifty-three industries covering all sectors in the economy. The average value of the Krugman Index for each pair of regions over the sample period is used. This gives essentially the same results as are obtained if one used the values of the Krugman Index for the middle year of the sample period (1993).

The Krugman Index indicates that the two regions most unlike all others are NT and ACT, while the two regions which are most alike are NSW and VIC. The individual industries which contribute the most to differences between regions are mining, metal manufactures, automobile production and various services.

TAS (a large island to the south of VIC) is separated from VIC by Bass Strait which is around 250 km wide. Since there is considerable sea and air traffic between TAS and VIC (and much less direct sea or air traffic between TAS and other states), the present paper has recorded TAS as being adjacent to (having a border) with VIC but not being adjacent to any other state. In terms of climate, etc., TAS is more like VIC than any other state.

The same results (industry structure and size matter, while distance or adjacency do not) are also obtained if one uses the correlations in Table 6 (the raw detrended data) or Table 10 (filtered cycles) as the dependent variable.

Dixon and Shepherd (2000) found that Australian states and territories with similar industrial structures also have similar unemployment rates.

This latter result may reflect a similarity of central place activities and not just greater trade between the regions.

This (negative) result is also obtained if one uses the correlations in Table 1 (the raw growth rates) as the dependent variable. The difference in results may explain why there is no agreement between studies of the role of industrial structure.

A referee has commented that the finding that the smaller states exhibit more independent behaviour might reflect the operation of the central limit theorem rather than anything specific about those states. In this light, small sub-regions of the major states may exhibit a similar amount of ‘independence’ from the other larger states. This implies that economic policies deemed relevant for the smaller states may also be appropriate for small sub-regions of the larger states.