Who would win from a multi-rate GST in New Zealand: evidence from a QUAIDS model*

Abstract

The merits of New Zealand moving away from its broad-based single-rate GST structure – particularly by removing GST on food – are often raised in public discourse and political campaigns. This paper investigates who would benefit from the introduction of a multi-rate GST structure in New Zealand and, in particular, whether reduced GST rates would be a more effective way of providing support to poorer households than New Zealand’s current income-tested tax credit approach. Behavioural simulation results from a QUAIDS model confirm previous findings that applying reduced GST rates to food and beverages would have a small progressive effect, but that richer households would benefit more than poorer households in aggregate terms. Meanwhile, reduced GST rates applied to recreational and cultural expenditure would have a regressive effect. Additional simulation results clearly show that the family tax credit is a far superior mechanism for providing support to poorer households than reduced GST rates. New Zealand should therefore maintain its current approach of a broad-based single-rate GST and income-tested tax credits.

KEYWORDS:

• GST
• VAT
• QUAIDS
• reduced rates
• cash transfers
• tax credits
• distributional effects

JEL CODES:

• D12
• H23
• H24
• H31

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 See, for example, Cnossen (2002).

2 Most OECD countries use their GST/VAT systems to achieve distributional goals by applying reduced rates to consumption items that typically make up a greater proportion of the expenditure of poorer as compared to richer households. The pursuit of cultural and social objectives, amongst others, have also led to the application of reduced GST/VAT rates on an even broader range of consumption in many OECD countries (see, e.g., OECD/KIPF, 2014).

3 The New Zealand First party proposed zero-rating ‘basic food’ in its 2017 election campaign. The Labour party proposed zero-rating ‘fresh fruit and vegetables’ in its 2011 election campaign, though it subsequently dropped the proposal.

4 Kaplow (2006) and Laroque (2005) both show that this result still holds when the income tax is not optimal. Limited empirical evidence (see, e.g., Browning & Meghir, 1991) suggests that weak separability is unlikely to hold in practice. However, the empirical difficulty in determining appropriate tax rates based on complementarities with leisure suggests that uniform indirect taxation remains preferable from an efficiency perspective (Crawford et al. 2010).

5 Microsimulation modelling is typically the preferred tool for distributional analysis of indirect taxes. This is because it is based on household level data (typically household budget surveys) which enables highly detailed distributional breakdowns to be obtained. The large number of expenditure categories (typically at least 200) provided in household budget surveys also enables highly differentiated GST/VAT systems to be accurately modelled. A weakness of microsimulation modelling is that it does not take account of general equilibrium effects. A CGE modelling approach would incorporate general equilibrium effects. However, CGE models are highly limited in terms of the distributional analysis they can provide as they are based on aggregated household types.

6 As specified in Banks et al. (1997). The original specification in Deaton and Muellbauer (1980a) varies slightly. Note that setting $\lambda (p)$ = 0 reduces equation 1 to the indirect utility function of the AI model.

7 Alternatively, one could rearrange for the expenditure function and apply Sheppard’s lemma, as in the original derivation of the AI budget share equations in Deaton and Muellbauer (1980a, 1980b).

8 Note, though, that these price data will not correspond perfectly to the prices facing every consumer within a region due, for example, to transport costs (especially in the large and spatially heterogenious ‘rest of North Island’ and ‘rest of South Island’ regions).

9 Twenty-four different sets of prices (6 years x 4 quarters) vs 80 (4 years x 4 quarters x 5 regions).

10 Modelling was attempted for a range of expenditure groupings and compositions using the larger dataset, with group prices initially calculated as averages of the prices of the constituent expenditure items weighted by their average population within-group expenditure shares. Following IFS (2011), attempts were made to increase price variation by calculating group average prices based on the within-group expenditure shares of each household, so that the average price varied depending on each household’s actual consumption pattern. However, this approach risks conflating quality variation with price variation resulting in spurious relationships. This was indeed the case here, leading, for example, to positive price elasticities for some expenditure groupings.

11 Expenditure groupings can be justified on the basis of weak separability which requires that preferences for goods within a particular group can be described independently of the quantities in other groups (Deaton & Muellbauer, 1980b).

12 Including durables for either of the two datasets led to clearly spurious results, including several positive own-price elasticities.

13 Imputed rental expenditure of homeowners is estimated by a number of national statistics agencies in their household expenditure surveys, following a range of different methodologies. One possibility would be to base estimation of imputed rental on factors such as region, house size and property rates paid to local councils and reported in the HES (such an approach was undertaken, for example, for the United Kingdom by Brewer & O’Dea, 2012).

14 The three excluded categories are ‘audio-visual and computing equipment’, ‘major recreational and cultural equipment’ and ‘other recreational equipment and supplies’.

15 The aidsills program was chosen because, unlike alternative options such as Poi’s (2012) quaids program, aidsills enables potential endogeneity in total expenditure to be instrumented for. It does have some limitations. First, it does not allow for clustered standard errors. As such, if errors are correlated within groups – as they could be within region, quarter and survey year – then this may result in underestimation of standard errors (although point estimates will not be affected). Second, aidsills arguably provides a less flexible approach to incorporating demographic variables than quaids (see note 21).

16 These are the most commonly applied initial values used in the literature. The Stone price index is calculated as the average price weighted by the mean budget shares: $\sum {\overline{w}}_i{p}_i$.

17 As Lecocq and Robin (2015) note, OLS and SUR would produce identical parameter estimates as the right hand variables in each budget share equation are identical.

18 Following estimation, absolute price effects are then recovered from the relative price effects.

19 Symmetry is only imposed following the iterative process. Lecocq and Robin (2015) note that imposing symmetry during each iteration produces almost identical results but increases the number of estimations that do not converge. 

20 This approach has been adopted in a range of recent studies, including: Cseres-Gergely et al. (2017), Abramovsky et al. (2015), Jansky (2014) and IFS (2011), and in the aidsills program of Lecocq and Robin (2015). There are however a range of ways to include demographic effects (Pollak & Wales, 1981). For example, Poi (2012) applies the demographic scaling approach of Ray (1983) in his quaids Stata program. The scaling approach is arguably more flexible than the translating approach, but the translating approach maintains the conditional linearity of the demand system thereby increasing computational ease and speed. For comparison, I apply both Lecocq and Robin’s (2015) aidsills program and Poi’s (2012) quaids program to model the demand system with the same 11 demographic variables (but without instrumenting total expenditure in either case) and find very similar results. I prefer the aidsills program as it enables potential endogeneity in total expenditure to be instrumented for. It is also computationally faster.

21 The adult equivalent size is calculated using the parametric equivalence scale presented in Section 5.

22 The reference person is normally determined by who takes responsibility for answering the questionnaire.

23 The instrumental variable (two-stage least squares) procedure has two stages: in the first stage, log total expenditure is regressed on the instrumental variable (log disposable income) and the price and demographic variables; in the second stage, the demand equations are estimated with the error term from the first stage regression added as an additional explanatory variable. As two-stage least squares is combined with SUR, the process becomes equivalent to three-stage least squares regression.

24 The case study for the United Kingdom in IFS (2011) found an expenditure elasticity of 0.25 for food subject to the zero VAT rate, but an expenditure elasticity of 1.15 for their standard-rated food and drink category which included restaurant food, takeaways and alcohol.

25 Food own-price elasticity estimates vary considerably across the case studies included in IFS (2011), with estimates of: −0.11 for the United Kingdom; −0.23 for Belgium; −0.43 for Germany; −0.74 for France; and −0.92 for Spain. The latter is a surprisingly large result, particularly in light of the more recent analysis of Bover et al. (2017).

26 New Zealand had the seventh highest rate of passenger vehicle ownership out of 171 countries considered in World Bank (2011)

27 The exceptions being pharmaceuticals, water supply and passenger transport.

28 In 2015–16, the family tax credit provided an amount of NZD 5,303 for the eldest child if aged 16–18 or NZD 4,822 if younger than 16. A further amount was paid per additional child as follows: NZD 4,745 if aged 16–18; NZD 3,822 if aged 13–15; and NZD 3,351 if younger than 13. The total credit amount (including other amounts provided under the WFF package) was withdrawn at a rate of 21.25 cents for every dollar of family income above NZD 36,350. The credit was refundable, so that if it exceeded tax due the unutilised amount was paid out to the family. Note that subsequent reforms have simplified the payment structure as of 2017. The modelling effectively assumes the first child in the family is aged under 16, and the second and subsequent children are aged under 13.

29 Underestimation of revenue occurs due to a range of factors: the under-reporting of some expenditure by households; survey coverage being limited to private households; the exclusion of expenditure on housing (as new house purchases are subject to GST); and treatment of GST exemptions as zero rates. In addition, fraud is not simulated in the model – which may result in some overestimation of revenue.

30 This is a standard assumption in the empirical literature. In theory, it is possible for the VAT to be less than fully or even more than fully passed on to consumers, depending on the structure of the particular market. Empirical evidence, however, is inconclusive, and so full pass-through is assumed in the absence of clear guidance to the contrary. IHS (2011) present a detailed review of both the theoretical and empirical literature on VAT pass-through. They conclude that full pass-through is more likely to be found in more competitive markets and for broader VAT reforms. More recently, Benzarti, Carloni, Harju, and Kosonen (2020) find evidence for European countries of significantly stronger pass-through of VAT increases than VAT decreases. In contrast, Benedek, De Mooij, Keen, and Wingender (2020) find no significant evidence of asymmetric responses to price changes in European countries. They also find roughly full pass-through of standard VAT rate changes, but only around 30% pass-through for changes in reduced VAT rates. Meanwhile, Gaarder (2018) finds the introduction of a reduced VAT rate on food in Norway to have resulted in full pass-through to prices.

31 Exemptions are treated as zero-rates. This overlooks the likely presence of some tax embedded in the supply chain (due to the inability to claim input tax credits for exempt goods). The model is therefore likely to underestimate the amount of GST collected from exempt goods.

32 The tax revenue calculations for ensuring revenue neutrality are only weighted by the household survey weights.

33 As noted by Creedy and Sleeman (2006), this parametric scale was introduced by Cutler and Katz (1992) and is an extension of the simpler $n_i^{\alpha }$ form used by Buhmann, Rainwater, Schmaus, and Smeeding (1988) and Coulter, Cowell, and Jenkins (1992). A commonly used alternative equivalence scale is the ‘OECD modified’ scale which gives a fixed weighting of 1 to the first adult household member, 0.5 to the second and additional household members aged 14 and over, and 0.3 to each child under 14. While the OECD modified scale adjusts for the relative need of adults and children, it does not continuously adjust for economies of scale as second and subsequent children all receive the same weighting. The equivalence scale parameters chosen in this paper produce a close match with the OECD modified scale, but provide for additional economies of scale at greater household sizes. Sensitivity analysis conducted on the two parameters shows some variation in results for changes in both parameters, but not significant enough to alter the paper’s overall conclusions.

34 Note that some minor variation occurs in the estimated number of individuals within each decile due to the need to allocate unweighted household observations that overlap the boundary between two deciles into one.

35 Technically, a negative compensating variation reflects a welfare gain as it shows the amount of money that would need to be given to a household post-reform in order to maintain their pre-reform utility level. For presentational purposes, however, these welfare gains are presented as positive numbers.

36 The total gains reported in Tables 8 and 9 from reduced rates on food/beverages and recreation/culture are slightly different to those presented for the overall reform in Tables 4 and 5. This is because the simulations for Tables 8 and 9 are undertaken separately for each expenditure group (thereby enabling welfare results to also be calculated). Additionally, the overall reform results in Table 4 and 5 capture some additional revenue generated as a result of a shift in some consumption away from reduced-rated goods towards standard-rated goods in response to the change in relative prices. Minimal difference occurs if results are calculated for each expenditure group under the combined reform, and does not change the conclusions of the analysis in any way.

37 I follow Lambert (1985) and use the negative of the Kakwani index for taxes to measure the progressivity of the tax expenditure from reduced GST rates. As such, a progressive tax expenditure results in a positive Kakwani index, and consequently a positive Reynolds-Smolensky index. The same approach is adopted for the welfare change based calculations.

38 A significant number of households in the bottom expenditure decile feature in the second income decile, resulting in higher losses measured as a percentage of expenditure in the second income decile than in the bottom income decile (though less in aggregate terms).

39 Additionally, a small number of bottom expenditure decile households with children aged under 16 have income levels high enough to restrict their eligibility to the family tax credit.

40 A small number of large households in higher deciles still benefit from a reduced family tax credit amount, but the loss of this reduced amount is outweighed by the gain from the reduced GST rates.