Is past prologue? Prospects for state and local sales tax bases

Abstract

General sales taxes provide substantial fractions of state and local revenues in the US. However, state and local sales tax bases have been eroding steadily during the past 50 years. Base erosion contributes to fiscal stress in the states; therefore, prospects for continued sales tax base erosion are important to state tax administrators, policymakers and public finance economists. This article offers a quantitative assessment of base erosion. We construct time series of a representative state sales tax base and its price index, and estimate a structural demand system for ‘taxed’ and ‘untaxed’ commodities. We use the estimates to forecast the sales tax base over coming years. Time-series forecasts and a weighted-average forecast are constructed, to reduce the likelihood of forecast error. The results suggest slow, but relentless, base erosion and possible recurring fiscal stress, in states where sales tax revenues make up sizable fractions of total tax revenues.

Acknowledgements

For many very helpful comments and suggestions I wish to thank Donald Bruce, Stanislav Radchenko, E. Frank Stephenson, Jennifer Troyer, Rick Zuber, participants at sessions of the Southern Economic Association annual meetings and an anonymous referee. Chris Irwin, in particular, and Shannon Wei provided very timely and able research assistance.

Notes

¹ In each sales tax state, citizens are required to remit use tax on commodities purchased from remote vendors, if the vendors do not collect the state's sales tax.

² For example, the Federation of Tax Administrators (FTA) (1997, p. 1) states ‘long-term viability of state and local retail sales taxes continues to be threatened…. One of the primary threats is the increasing proportion of economic activity related to the provision of services as opposed to goods.’ Also see Duncombe (1992) and Mazerov (2003).

³ Dye and McGuire's representative narrow base is personal consumption expenditure minus food for home consumption, most services and motor vehicle fuels. Dye and McGuire carefully note that national consumption data do not account for sales taxes on business inputs. Ring (1999) estimates that on average in the US, 41% of sales tax collections result from tax on sales of business inputs. He points out, however, that states exempting specific commodities from sales taxes on household spending tend to exempt the same commodities from tax on business inputs. Business spending on services is likely to follow a trend similar to personal consumption of services, so this is unlikely to affect the forecasts in a significant way.

⁴ They report that the representative sales tax base grew about 2.2%, annually, between 1968 and 1987.

⁵ The categories are, food consumed at home, food consumed away from home, apparel services, medical and health spending, entertainment, personal care, household services, apparel, utilities, alcohol, tobacco, gas and oil.

⁶ And it will become clear, below, that the ratio of services spending to income tends to overstate base erosion.

⁷ Detailed discussion of taxed commodities follows later.

⁸ Liviatan (1961) describes the problem.

⁹ The ratio could decline, if the decline in saving causes demand for tangible goods to decline. But there is no reason to believe this would occur, even if saving causes a change in the composition. For example, if the increase in saving causes goods spending to increase, but by less than the increase in services spending, the ratio increases.

¹⁰ This raises an important econometric issue. Generally, measurement error in the left-hand side variable of a regression equation does not cause Ordinary Least Squares (OLS) estimates to be inconsistent. It is standard in demand analysis to use a spending ratio as the left-hand side variable, and total expenditure on the right-hand side. For the reasons just discussed, if saving changes, total expenditure may not be exogenous with respect to the spending/income ratio, and so OLS estimates may be inconsistent.

¹¹ See the data Appendix for more detail.

¹² Alaska, Delaware, Montana, New Hampshire and Oregon do not levy state sales taxes. The numerator of the taxed commodities ratio includes spending not taxed in these states: the denominator includes spending in all states. Adding a dollar to numerator and denominator increases the ratio, so Fig. 2 overstates the tax base. But population in these five states is only 3% of the US population. Local sales taxes are levied in Alaska, so more than 97% of US residents live in jurisdictions with sales taxes. And there is no a priori reason to think spending in these five states has evolved differently than in other states, so measurement error should be small.

¹³ The trend lines were constructed by splitting the sample into two parts, 1947–1961 and 1961–2002, and running separate regressions of taxed commodities on constants and linear trends.

¹⁴ Stone's index is . According to Deaton and Muellbauer (1980b), the Stone index is a good approximation if prices are collinear. The correlation between log prices of taxed and untaxed commodities is 0.99.

¹⁵ For the average marginal income tax rate, we extended Stephenson's (1998) time series on the federal individual income tax. See the data Appendix for details.

¹⁶ Stock and Watson (2003) describe the procedure.

¹⁷ We used Augmented Dickey–Fuller tests to check for stochastic trends. Results are available on request.

¹⁸ The estimation results using differenced data are available on request.

¹⁹ The uncompensated own-price elasticity of demand for taxed commodities is −0.73. Table 1 reports estimates of the ordinary, Marshallian, demand curves. These are appropriate for forecasting. Hicksian, compensated, demands, would be appropriate in a welfare analysis.

²⁰ The uncompensated own-price elasticity of demand for untaxed commodities is −0.88.

²¹ The income elasticity of demand for taxed commodities is 1.39.

²² Graphs for the time-series forecasts are not shown, and are available on request.

²³ The order in which variables enter VARs affect results. The rule is to place last the variable that is most likely to be affected by other variables in the model. The order we used is: (1) personal expenditure, (2) price of taxed commodities, (3) price of untaxed commodities, (4) average weekly hours, (5) labour force participation rate, (6) fraction spent on taxed commodities. Sims’ (1980) chi-square test suggests including two lagged values of each variable.

²⁴ Evolution of economic relationships provides the rationale for dropping the oldest data points. As relationships evolve, the oldest data become less representative of current reality.

²⁵ Recall this forecast is constructed from aggregate annual National Income and Product Account data, 1961–2002, produced by the BEA. Because policymakers are concerned about the way individual households in their jurisdictions are affected, the referee suggests using individual household data. We collected Consumer Expenditure Survey (CES) data reported by the Inter-University Consortium for Political and Social Research, and constructed a dataset for taxed and untaxed commodities purchased by individual households. There are some difficulties matching categories of taxed and untaxed commodities in the two datasets. Also the CES data begins in 1984, but is quarterly, in comparison with our aggregate annual sample, which extends back to 1961. Nevertheless, the individual household data generates forecasts that are very close to the ones reported here: in this case, the forecast for 2003 is 0.410 and declines to 0.337 in 2017.

²⁶ Baumol argues that labour input is an indicator of quality in many services, making it difficult to reduce the amount of labour used to produce them. In contrast, quality in most manufactured products does not depend on the amount of labour input: consumers do not measure the quality of refrigerators on the basis of the amount of labour hours it takes to produce them. Thus, it is relatively easy to increase labour productivity in tangible goods, and so the relative price of tangible goods tends to decline. Note that Bosworth and Triplett (2003) offer a counter-argument.