At best, VAT is a net zero on inequality

The main take-away from Table 2 is that the VAT appears to have led to an increase in inequality when inequality is measured based on disposable income. This finding is very robust across specifications, in particular, across different ways of instrumenting the VAT. Since this is a main finding where a statistically significant effect is found, we have also assessed its sensitivity to violations of the exclusion restriction by using Local-to-zero (LTZ) estimation method developed by Conley, Hansen, and Rossi (2012). Figure A1 in the Appendix, shows the effect of the VAT on income-based Gini that also allow for a direct impact of the instrument (NeighbourV) on inequality. These estimates assume that the direct effect of the instrument to be uniformly distributed between zero and δ. The results show that as long as δ remains smaller than 15 in panel C (13 in panel D), the effect of the VAT on income-based Gini remains significant at 10% risk level. Because the corresponding reduced form effect of the instrument is 17.5 (and 15.8) in panel C (and D), the IV results for income-based Gini are robust to substantial deviations from perfect exogeneity.

However, Table 2 shows no impact on inequality for countries which use consumption-based inequality measurement; in some cases the impact is negative but it is not consistently significant. The difference between income- and consumption-based results must be interpreted with care, as the impact can also differ between other dimensions. In particular, high-income countries are overrepresented in the sample where Gini is income based. Additional analysis (not reported for brevity) suggests that the impact of the VAT on income-based inequality is similar in middle-income and high-income countries, Electronic copy available at: https://ssrn.com/abstract=2924448 21 suggesting that the main driving force for the result may be measuring inequality using income and not consumption rather than differences in the income levels of countries. One reason for why the impact of the VAT differs between income- and consumption-based measurement is that in consumption-based analysis, the consumers’ response to differentiated commodity tax structure can also be taken into account.

A caveat to the result is that when including region*time period fixed effects to an IV specification with inverse distance weights, the impact of VAT adoption on inequality for income-based measurement countries is not significant (results available on request). It remains significant for the fixed effect OLS model with region*time period fixed effects. While controlling for region-by-time-period fixed effects is a useful robustness check, including these additional controls also takes away some of the remaining variation. Since it is unclear how strong a concern the period-specific regional unobservable effects are, we would tend to favour the results with country and period fixed effects only.

The results do not seem to differ in a statistically significant way between income groups or depending on the degree of openness. One might also be concerned that the results using the time span until 2010 can be problematic, since most of the identifying variation in countries adopting the VAT took place earlier (see Figure 2), and the latest observations can be affected by the financial crisis. We therefore also report results for a shorter time span (ending in 2000). These results are reported in the Appendix, Table A4. The results for the IV remain insignificant, whereas for the OLS, the sign is retained but significance drops. The positive impact of VAT adoption when inequality is measured using income remains valid for the shorter sample, as well.