Income data

Results based on HBS income data at low income levels may be misleading due to the presence of households with transitorily low income (Bozio et al., 2012; Decoster et al., 2010).¹⁰ For example, many self-employed workers may have low income levels at certain stages of their businesses’ development, but will continue to have unaltered (high) expenditure. Alternatively, some households may be drawing down savings to fund their consumption. In either case it is likely to be misleading to consider them “low-income” households for distributional analysis.

To mitigate this concern, we exclude households from the analysis where:

• • the household reports negative or zero income; and/or
• • the household has an expenditure-to-income ratio of four or greater.

Consumer durables and house purchases

Modelling consumer durables poses a problem as these are infrequent purchases and the HBS data only provides a snapshot of expenditure. For example, a car is likely to be owned for several years before being replaced, so it would be relatively arbitrary whether or not a car was purchased in the survey period (and therefore was included as expenditure). Ideally, we would want to apportion the cost of durables over their useful life in order to reduce any overstatement of expenditure for households that have undertaken such purchases during the survey period (or any understatement for households that made such purchases outside the survey period). However, this would require accurate information on length of ownership and expenditure on durables (both purchased within and outside the survey period), and is therefore not a feasible option.

On the other hand, not modelling durables would underestimate consumption and tax revenue significantly. we therefore include consumer durables (with the exception of housing) in the modelling. Given that the basis of our analysis is the presentation of averages across decile groupings, we are effectively making the assumption that, within each decile group, the number of households that purchase durables in that period, and the number that do not, will “average out” - thereby reflecting approximately the same expenditure for that decile as would be modelled if we were able to apportion the expenditure across the useful life of the durable.

Housing is excluded from the modelling for two main reasons: first, housing expenditure is not available for all countries; and second, where it is available, it constitutes such an infrequent and extremely large expenditure that it is less likely than smaller durable purchases to “average out” within decile groups. A possible alternative to including the entire housing expenditure amount in the modelling would be to use imputed rental data to estimate the annual consumption value relating to the housing purchase (and the related return on investment). However, this data is not available for all countries and hence, in order to maintain a consistent methodology across all countries in the study, we do not adopt such an approach. A consequence of excluding housing is that VAT revenue will be underestimated in those countries where housing is subject to VAT (though this would always be the case for countries where the data is unavailable). Additionally, by not taking account of the imputed income from home ownership, the effective level of income of home owners compared to renters will be underestimated.