ABSTRACT
This study investigates the impacts of airport activities on regional economies using annual data on all regions and 22 airports in New Zealand from 2001 to 2016. In addition to fixed effects estimation, the system generalized method of moments approach and the dynamic common correlated effects estimator are used to account for cross-sectional dependence, cross-regional heterogeneity and feedback effects. The study shows clear and consistent evidence that aviation activities positively affect regional economies, and that it is beneficial for policy-makers and airport owners in a region to promote aviation activities.
ACKNOWLEDGEMENT
The authors are grateful to three anonymous reviewers and Ben Derudder for their detailed and thoughtful reviews during the publication process. All errors are the authors’ own.
DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.
Notes
1. Where new commercial airports are built or converted from military airports (e.g., Western Europe) during the sample period, it is challenging to design a good estimation strategy due to the endogeneity caused by the interdependence between airport activities, capital investments and economic growth.
2. A total of 22.15% of the shares of Auckland Airport are held by Auckland City Council; the New Zealand Superannuation Fund and New Zealand Accident Compensation Commission are major shareholders. We are grateful to an anonymous referee for drawing our attention to the special features of New Zealand’s airport system.
3. A similar ‘light-handed regulation/threat of regulation’ system is used in Australia (Yang & Fu, Citation2015).
4. The definition of region in this study follows that used by Statistics New Zealand. There are 15 regions in New Zealand: Northland, Auckland, Waikato, Bay of Plenty, Gisborne, Hawke’s Bay, Taranaki, Manawatu-Wanganui, Wellington, West Coast, Canterbury, Otago, Southland, Marlborough and Tasman/Nelson (Statistics New Zealand, Citation2016b).
5. The New Zealand government is currently the major shareholder of Air New Zealand, controlling 52% of its shares. Thus, the government obtains financial benefits if Air New Zealand’s profitability improves. However, although the New Zealand government retains vote power over some of the airline’s actions (Bollard & Pickford, Citation1998), it does not interfere in the airline’s daily operations, such as what fares should be charged to passengers or to which airports Air New Zealand should fly.
6. That is, it is uncorrelated with the error term. In this context, the error term represents shocks to economic development.
7. Carefully selected lags of endogenous and predetermined regressors (see Roodman, Citation2009, for more details).
8. The HHI of scheduled airline seats cannot be used as an external instrument in the dynamic system GMM model because its time variation is captured by the yearly dummy variables. As such, HHI is only included in the CCE model.
9. HHI is calculated as , where si is the market share of airport i’s scheduled airline seats in New Zealand. Therefore, it measures the level of competition between airports (see Hirschman, Citation1945; and Herfindahl, Citation1950, for more details on the index development and calculation).
10. The dynamic system GMM model uses lags of the first differences of the endogenous variables as internal instruments under the assumption of sequential exogeneity (Blundell & Bond, Citation1998). As such, for consistency, the lags of the first differences of the endogenous variable are also used as instruments in the CCE specification.
11. See Chudik and Pesaran (Citation2015) and Ditzen (Citation2018) for details of the asymptotic distribution of the group means.
12. The presence of serial correlation would render the lags of the endogenous variable invalid. As such, the Arellano and Bond (Citation1991) test for serial correlation is important in the dynamic panel GMM model. This test is performed on the first differences of the errors. For example, a reported serial correlation of order 2 in the differenced errors indicates serial correlation of order 1 in the levels equation. As such, autocorrelation of order 1 in the differenced errors is acceptable and uninformative, as adjacent differences in errors are mathematically related. Hence, only AR(2) is reported in .
13. The CCE estimator is consistent, though inefficient, if the errors are cross-sectionally independent. The Pesaran (Citation2015) test for weak cross-sectional dependence is applied to all model specifications to ascertain the suitability of the CCE framework. We conclude that the errors are cross-sectionally dependent. Moreover, the Hausman test is applied to the FE, dynamic GMM and CCE models. The results suggest that there is a significant difference in model estimates between the CCE and the more restrictive specifications. This suggests that the presence of cross-sectional dependence will result in all the specifications being biased, except for the CCE estimator. The results of the diagnostic tests are available from the authors upon request.
14. The first-stage regressions are reported in Table A1 of Appendix A in the supplemental data online. The F-statistic and proposed instruments are statistically significant, indicating a strong instrument set.