Part 1 - AL-Tax
Part 1 - AL-Tax Part 1 - AL-Tax
Chapter 3and its consensus-democracy score (Lijphart, 1999). 14 For each country, Hays(2003) used the average tax rate, i.e. the average of the dependent variable, y, inthe N 1 other countries as the spatial lag. In other words, all the off-diagonalelements of the spatial weighting matrix from equation (3.8) are set to 1/(N 1).For Hays’s original purposes, this spatial lag controls for the possibility that theobserved changes in capital taxation are being driven by tax competition betweencountries. 15 Hays estimated the model using OLS and reported panel-correctedstandard errors (PCSEs).For their part, Basinger and Hallerberg (2004) estimated spatial-lag models totest the following hypotheses derived from their theoretical model of tax competition:(1) Countries will undergo tax reform more frequently if the political costsof such reforms are low and/or the decisiveness of reforms in determining thepatterns of investment flows is high; (2) Countries will engage in tax reform whenthe political costs of reform in competitor countries is low; (3) The domesticpolitical costs of reform and the decisiveness of reform will determine the sensitivityof countries’ tax policies to tax changes in their competitors. Basinger andHallerberg (2004) included both spatially weighted X variables and spatiallyweighted Y variables (i.e. spatial lags) on the right-hand side of their regressionmodels. Hypothesis 1 is operationalized with a set of domestic X variables; Theytested Hypothesis 2 using a set of spatially weighted X variables and Hypothesis3 with the spatial lags interacted with domestic X variables.The dependent variable in their empirical analysis is the change in the capitaltaxrate. In addition to the Mendoza et al. (1997) capital-tax rates, the same variableHays (2003) used, Basinger and Hallerberg (2004) considered also the top marginalcapital-tax rates (of both central government and overall). They identifiedtwo kinds of domestic political costs as independent variables: Transaction andconstituency costs. Ideological distances between veto players were used to measuretransaction costs. The greater the ideological distance between politicalactors that can block policy change, the harder is altering the status quo (in thiscase, adopting capital-tax reform). Partisanship was used to measure constituencycosts; The constituency costs associated with capital-tax reform will be higherwhen left governments are in power. A third independent variable of interest, thedegree of capital mobility, was measured using capital controls on outflows(based on Quinn’s data). The degree of capital mobility determines the decisivenessof capital taxes in determining the location of international investments.Basinger and Hallerberg (2004) used four different spatial-weighting matrices:A symmetric 1/(N 1) weighting matrix, which makes the spatial lag for eachunit equal to the simple average of the Y values in the other units (as in Hays),61
International Taxation Handbookand three weighted averages using, respectively, GDP, FDI, and Fixed CapitalFormation (FCF) as weights. The last three spatial-weighting matrices have cellentries that differ across columns, but the rows are identical. For example, forevery country (row) in the sample, the USA (column) – because of its large GDP,capital stock, and flows of FDI – is weighted more heavily than Finland (anothercolumn), but the effect of American tax rates on other tax rates is the same for allcountries (in every row). American tax rates (column) have the same effect onCanada (row) as they do on Austria (another row), for example. These spatialweights are time varying (because the GDP, FDI, and FCF of each country changesover time). Like Hays, Basinger and Hallerberg (2004) included country-fixedeffects in their models, but, unlike Hays, they did not lag the dependent variabledirectly. They did include the lagged level of the tax rate, though, which makestheir model with changes as the dependent variable essentially the same as apartial-adjustment (lagged-dependent-variable) model in levels like the one Haysestimates. Finally, Basinger and Hallerberg (2004) also estimated their models byOLS with panel-corrected standard errors.Both Hays (2003) and Basinger and Hallerberg (2004) found the coefficient onthe spatial lag to be positive and statistically significant – i.e. both found strongevidence of tax competition. The problem with both analyses, however, is thatneither accounts for the endogeneity of the spatial lag, which renders biased andinconsistent the S-OLS estimator used. As we showed in Franzese and Hays(2004, 2006), the simultaneity bias in these circumstances would be toward exaggerationof the strength of interdependence and would also entail an induceddownward bias in the estimated effects of common conditions. Furthermore,both may have underspecified the common-conditions sorts of arguments theyinclude as well, which, again as we showed in Franzese and Hays (2004, 2006),would tend further to depress those estimated effects and inflate the estimatedstrength of interdependence. We therefore conduct now a reanalysis of Hays’s(2003, p. 99) 16 regressions using a new spatial-weight matrix and two consistentestimators – spatial two-stage least squares and spatial maximum likelihood(Tables 3.1 and 3.2). We also re-estimate the regressions, including a set of perioddummies to control better for common shocks (Table 3.3). 17 The results show thatHays may have overestimated the coefficient on the spatial lag with consequencesfor some of the other estimates. In particular, the original results for themediating effect of the capital endowment on capital-account openness are notvery robust across alternative estimators. However, the consensus democracyresults are robust and tend to be even stronger (i.e. larger coefficients and higherlevels of statistical significance) when the consistent estimators are used.62
- Page 31 and 32: International Taxation HandbookThis
- Page 33 and 34: This page intentionally left blank
- Page 35 and 36: International Taxation HandbookMarg
- Page 37 and 38: International Taxation HandbookrS
- Page 39 and 40: International Taxation Handbooka gr
- Page 41 and 42: International Taxation HandbookIt f
- Page 43 and 44: International Taxation Handbookthat
- Page 45 and 46: International Taxation HandbookProp
- Page 47 and 48: International Taxation HandbookTher
- Page 49 and 50: International Taxation Handbookther
- Page 51 and 52: International Taxation Handbook2.3.
- Page 53 and 54: International Taxation Handbookprod
- Page 55 and 56: International Taxation Handbookinpu
- Page 57 and 58: International Taxation HandbookThe
- Page 59 and 60: International Taxation HandbookRefe
- Page 61 and 62: International Taxation Handbook●
- Page 63 and 64: This page intentionally left blank
- Page 65 and 66: This page intentionally left blank
- Page 67 and 68: International Taxation Handbookconf
- Page 69 and 70: International Taxation Handbookelab
- Page 71 and 72: International Taxation Handbookthei
- Page 73 and 74: International Taxation Handbookτ*
- Page 75 and 76: International Taxation Handbookwher
- Page 77 and 78: International Taxation Handbookcoun
- Page 79 and 80: International Taxation Handbookown
- Page 81: International Taxation Handbookand
- Page 85 and 86: 64Table 3.2Capital tax rates and in
- Page 87 and 88: International Taxation HandbookHays
- Page 89 and 90: International Taxation Handbookpape
- Page 91 and 92: International Taxation Handbookthat
- Page 93 and 94: International Taxation HandbookSwan
- Page 95 and 96: This page intentionally left blank
- Page 97 and 98: International Taxation Handbookequi
- Page 99 and 100: International Taxation Handbook4.2.
- Page 101 and 102: International Taxation HandbookTo c
- Page 103 and 104: International Taxation Handbookcase
- Page 105 and 106: International Taxation Handbookσ 1
- Page 107 and 108: International Taxation Handbook4.5
- Page 109 and 110: International Taxation HandbookThe
- Page 111 and 112: International Taxation HandbookHube
- Page 113 and 114: International Taxation HandbookTedi
- Page 115 and 116: International Taxation HandbookIt i
- Page 117 and 118: This page intentionally left blank
- Page 119 and 120: This page intentionally left blank
- Page 121 and 122: International Taxation Handbookregu
- Page 123 and 124: International Taxation Handbookterm
- Page 125 and 126: International Taxation Handbooktake
- Page 127 and 128: International Taxation Handbookat l
- Page 129 and 130: International Taxation Handbookothe
- Page 131 and 132: International Taxation HandbookCash
International <strong>Tax</strong>ation Handbookand three weighted averages using, respectively, GDP, FDI, and Fixed CapitalFormation (FCF) as weights. The last three spatial-weighting matrices have cellentries that differ across columns, but the rows are identical. For example, forevery country (row) in the sample, the USA (column) – because of its large GDP,capital stock, and flows of FDI – is weighted more heavily than Finland (anothercolumn), but the effect of American tax rates on other tax rates is the same for allcountries (in every row). American tax rates (column) have the same effect onCanada (row) as they do on Austria (another row), for example. These spatialweights are time varying (because the GDP, FDI, and FCF of each country changesover time). Like Hays, Basinger and Hallerberg (2004) included country-fixedeffects in their models, but, unlike Hays, they did not lag the dependent variabledirectly. They did include the lagged level of the tax rate, though, which makestheir model with changes as the dependent variable essentially the same as apartial-adjustment (lagged-dependent-variable) model in levels like the one Haysestimates. Finally, Basinger and Hallerberg (2004) also estimated their models byOLS with panel-corrected standard errors.Both Hays (2003) and Basinger and Hallerberg (2004) found the coefficient onthe spatial lag to be positive and statistically significant – i.e. both found strongevidence of tax competition. The problem with both analyses, however, is thatneither accounts for the endogeneity of the spatial lag, which renders biased andinconsistent the S-OLS estimator used. As we showed in Franzese and Hays(2004, 2006), the simultaneity bias in these circumstances would be toward exaggerationof the strength of interdependence and would also entail an induceddownward bias in the estimated effects of common conditions. Furthermore,both may have underspecified the common-conditions sorts of arguments theyinclude as well, which, again as we showed in Franzese and Hays (2004, 2006),would tend further to depress those estimated effects and inflate the estimatedstrength of interdependence. We therefore conduct now a reanalysis of Hays’s(2003, p. 99) 16 regressions using a new spatial-weight matrix and two consistentestimators – spatial two-stage least squares and spatial maximum likelihood(Tables 3.1 and 3.2). We also re-estimate the regressions, including a set of perioddummies to control better for common shocks (Table 3.3). 17 The results show thatHays may have overestimated the coefficient on the spatial lag with consequencesfor some of the other estimates. In particular, the original results for themediating effect of the capital endowment on capital-account openness are notvery robust across alternative estimators. However, the consensus democracyresults are robust and tend to be even stronger (i.e. larger coefficients and higherlevels of statistical significance) when the consistent estimators are used.62