Abstract
We follow Ewing et al. (Citation2006) and examine the US federal revenue–expenditure nexus in an Error-Correction Model (ECM) allowing for asymmetric adjustment. Symmetric adjustment is rejected by data covering 1959.3 to 2007.4. Depending on whether a Threshold Autoregression (TAR) or Momentum Threshold Autoregression (M-TAR) model is employed, the estimated asymmetry is very different. We argue that there is little justification for preferring one model over the other and the concepts of ‘worsening’ or ‘improving’ budgets are model-contingent in important ways. Taking this into account, the results across models need not be contradictory.
Acknowledgements
I thank Russ Sobel for comments on a previous version. I also thank the participants at the West Virginia University seminar series for their stimulating discussion on this topic.
Notes
1For a comprehensive review of the tax-spend debate and the related empirical literature see Payne (Citation2003).
2Young (Citation2009) allows for asymmetry in not only the adjustment to budgetary disequilibria but also in the response to expenditures to increases versus decreases in revenue.
3These data are taken from the Bureau of Economic Analysis (BEA) Table 3.2., Federal Government Current Receipts and Expenditures. Data are in billions of dollars and seasonally adjusted. It differs only slightly in coverage from Ewing et al. (Citation2006) where the data run from 1958.1 to 2003.2.
4The data are taken from the BEA Table 1.1.5., Gross Domestic Product. They are in billions of dollars and seasonally adjusted.
5Ewing et al. (Citation2006) do not report the value(s) of p used to generate their results.
6This is peculiar because the only data, beyond that of Ewing et al., included in the present analysis are the budget deficits of the later G. W. Bush administration.
7I also estimate the TAR model using (approximately) Ewing et al.’s threshold value (∼−0.0066) for both p = 2 and p = 4, and then again with my data truncated to 2003.2 to match Ewing et al. The resulting ρ1 point estimates range from −0.3125 to −0.3005 and ρ2 estimates range from −0.0412 to −0.0285. All of these are consistent with my results reported in .
8As do Ewing et al., I find that symmetry is rejected for the M-TAR model: for p = 2 the F-statistic is 16.97 and for p = 4 the F-statistic is 14.03.
9Estimation using 4 lags produced very similar results.