Skip to content

Arkansas Retail Sales — A New Data Set from AEDI

Tracking the economies of states and regions is often limited by the availability of data.  In the recent context of the COVID-19 pandemic and the associated economic dislocations, economists have found the development of new sources of information to be increasingly helpful in tracking the rapidly evolving economic situation.

One glaring omission in the data-measurement tools of regional economists is the lack of data on consumer spending.  The Census Bureau calculates Retail Sales on a national level, but not for smaller geographic subdivisions.[1] The Bureau of Economic Analysis publishes annual estimates of Personal Consumption Expenditures by state, but annual data is not terribly helpful in tracking situation during the pandemic.

At the Arkansas Economic Development Institute, we have been tracking one set of proxies for consumer spending since 2015—Arkansas Taxable Sales (ATS).  ATS is a simple extrapolation from state sales tax receipts: Dividing total receipts by the tax rate yields an estimate of the original expenditures.[2]  Using state data on gallons of taxable motor fuel sold, along with data on average gasoline prices in the state, we have added a component to create Arkansas Taxable Sales Including Gasoline (ATSIG).

Figure 1 shows compares ATSIG to the Census measure of U.S. Retail Sales and Food Services.

Figure 1:

Sources: Arkansas Department of Finance and Administration, U.S. Census Bureau, Oil Price Information Service, Arkansas Economic Development Institute

This comparison strongly suggests that Arkansas was spared the experience of a severe contraction in spending during March and April.  Indeed, by April, ATSIG was already recovering from a downturn in March, and was only lower than the previous April because April 2019 was a relatively strong month.

Although ATSIG is intended to capture trends in spending, it is not an ideal measure of consumer spending on the retail level. Sales taxes are levied on many transactions that are not consumer retail transactions, and the underlying data on net sales tax receipts is distorted by rebates for some non-retail sales and other transactions exempt from sales tax.

Constructing Arkansas Retail Sales: The Basics
An alternative source of sales tax data comes from the sales tax remittances reported to counties and municipalities.  Sales taxes are collected at the state level by the Arkansas Department of Finance and administration, then allocated to local governments, with accompanying reports of sales tax remittances by industry sector. These reports are published for the most recent 36 months at Local Tax Distribution by NAICS.  Our strategy for constructing a data set for Arkansas Retail Sales is to extract and aggregate information on sales tax receipts for all of Arkansas 75 counties, specifically for the sectors that correspond to the components of the U.S. Census Bureau’s report on Monthly Retail Trade and Food Services.

Appendix 1 shows an example of the Local Tax Distribution reports—specifically for Jackson County for October 2019.  The report lists net sales tax, net use tax and their sum by industry, delineated by four-digit NAICS codes.[3]  It also notes rebates and audits that were used to adjust the net figures.  The report is dated October 2019, corresponding to the month in which the sales tax revenue was credited to the County.  It was collected the previous month, from vendor remittances based on sales during the month before that.  So the figures reported in the October 2019 report roughly correspond to sales in August 2019.

Conceptually, the statistics on total net sales and use tax collections provide the basis for a useful measure of final spending, after rebates and audits are subtracted out.  However, the rebates and audits are recorded when they are received, rather than in the month in which the original sale occurred.  Consequently, the adjustments add noise to the monthly time series.  In constructing our measure of Arkansas Retail Sales, we start by calculating the net total less rebates and audits.[4]

To adjust for changes in local tax rates, the revenue figures are divided by the county sales tax rate in effect during the month in which the underlying sale took place, yielding a measure of taxable sales.

To construct measures comparable to the Census’ Retail Trade we select the same subset of 3-digit NAICS subsectors designated as retail, 441 through 454, plus the subsector 722 – Food Services and Drinking Places.  Table 1 lists the specific subsectors included.  The data reported in the Local Tax Distribution reports are reported for 4-digit industry groups, where the classification is self-identified by the tax remitter.

One shortcoming of the data is the practice of suppressing data for industries with fewer than three individual businesses reporting (for confidentiality reasons).  In the report for Jackson County shown in Appendix 1, 13.4% of the net total and 17.7% of the gross total are in the category “NAICS with Less Than 3 Businesses.”  Our supposition is that few 4-digit industries within the retail sector will fall into that category, and we maintain the assumption that changes in the composition of unreported industries do not bias our measurements.  Aggregating from 4-digit to 3-digit industry groups, and from county to statewide measures, should help smooth out some of the idiosyncratic fluctuations from this factor, as well as from other potential measurement problems that may be embedded in the raw data.

Table 1:

The data set we assembled runs from July 2017 through November 2020 (sales months), with statewide and county-level time-series calculated for each 3-digit industry group.  At the county level, data for 4-digit industries is sometimes sparse, and aggregation fills in some of the gaps.  However, data limitations—especially for the smaller counties in the state—suggest that estimates of monthly changes in taxable sales for individual 3-digit industry groups may be unreliable, and we report only broader aggregates for individual counties (for now).

As an example, Figure 2 shows the calculated Retail Sales for industry group 448: Clothing and Accessories Stores.  The constructed series has a clear seasonal patterns, with major peaks in December of each year, followed by a January lull and a spring surge round March.  The effect of the COVID-19 pandemic and social distancing restrictions had a clear impact on sales, with sales in April 2020 that were down 54% from the previous April.

Figure 2:

Seasonality is more relevant for some subsectors than for others, but Figure 2 shows that the normal seasonal variation can be large enough to obscure events as significant as the COVID pandemic. Year-over-year growth comparisons can be used to roughly eliminate seasonal effects, but it also induces a 12-month cycle that produces notable feedback a year after large changes like those that occurred in the spring of 2020.

Seasonal Adjustment
In order to account for seasonal variation, and for comparison with seasonally adjusted data on U.S. Retail Trade and Food Services, we implement a simple seasonal adjustment procedure.  The data series are too short to use sophisticated algorithms like the Census X-13 model.  And experimentation with a simple centered-moving-average approach revealed in important complication:  the implementation of Act 822 in 2019, which mandated that out-of-state retailers collect and remit sales and use taxes for online sales.

This change induced a shift in use-tax collections that represented a broadening of the tax base rather than an increase in spending, but it is an effect that turns out to be statistically significant for some subsectors.  Because the change was associated with an abrupt increase in tax collections in July 2019, simple MA seasonal factors were biased by the effect.

Consequently, we implemented a model-based approach to seasonal adjustment.  For each sector a (log) linear regression model was estimated, where the regressors include an implicit constant, a time-trend, and a full set of monthly seasonal dummy variables.  The regression also includes a dummy variable that takes on the value of one for July 2019 forward.  Regressions were estimated using OLSQ for the period July 2017 through February 2020.  Because the model is estimated in natural logs, the coefficient on the July 2019 dummy variable represents a direct estimate of the percent increase in tax-revenues associated with the implementation of Act 822.  The coefficients on the seasonal dummy variables are then used to calculate multiplicative seasonal factors.

This methodology produces, as a by-product, estimates of the magnitude of the out-of-state “internet sales tax effect” that accompanied Act 822.  For Total Retail and Food Service Sales, this produced an estimate of the internet sales tax effect of 4.8%.  For total taxable sales, including non-retail components, the effect was estimated to be 4.2%. These are preliminary estimates and will be explored in more detail in further research.

Figure 3 illustrates how the not-seasonally adjusted series are decomposed into trend and seasonal factors, again using Subsector 448 as an example.  The trend-shift in July 2019 represents an estimated increase of 3.8%, corresponding to the Act 822-effect.  The dashed line shows the trend plus estimates of recurring seasonal factors.  The deviation of the actual data from the dashed line reveals, indirectly, the seasonally-adjusted series.

Figure 3:

Figure 4 displays the seasonally adjusted series, after applying the estimated seasonal factors. With this refinement, the decline of sales to 54% below trend in April is evident.

Figure 4:

Gasoline and Automobiles
The procedures described above were implemented directly for 11 of the 13 subsectors included in Retail Trade and Food Services.  However, two subsectors require special treatment: gasoline and automobiles.  The Local Tax Distribution reports include tax revenue derived from sales at gasoline stations (NAICS 447) and automobile dealerships (NAICS 441), but gasoline itself is not subject to sales tax and automobiles are taxed at the local level on a per-vehicle basis.

Gasoline:  In order to include gasoline in the statewide aggregate, we incorporate the gasoline component from ATSIG:  Namely, statewide gasoline sales are calculated as the product of the monthly average gasoline price for Arkansas (obtained from the Oil Price Information Service) and the number of taxable gallons of gasoline sold (as reported by the Motor Fuel Tax division of the Department of Finance and Administration).

The estimated value for total gasoline sales is added to the estimate of taxable sales at gasoline stations (NAICS Code 441) as calculated using the Local Tax Distribution Reports.  This unweighted sum generates values for gasoline station sales that represents 11% of total Retail Trade and Food Services for Arkansas, compared with an 8% share in the U.S. Census data (2019 averages).

Lacking any information on county-level sales of gasoline, the broadest retail sales measure we are able to generate for individual counties is Retail Trade and Food Services ex Gas.

Automobiles:  Automobiles are subject to county and municipal sales taxes on a per-vehicle basis.  Specifically, the local sales tax for automobiles is $25 per vehicle for each percentage point of the local tax rate.  So, for example, if a county’s tax rate is 1.5%, the tax would be $37.50.  Consequently, the local tax collection data provides a measure of the number of vehicles sold, but not their valuation.

The revenue from this tax—reported on the Local Tax Distribution reports in a separate line from the NAICS code data—is used to create an index of vehicles sold, with the county totals aggregated to yield a statewide index.  The index is combined with data from NAICS code 441, Motor Vehicle and Parts dealers using the following equation:

SALES441 = NAICS441 + α*INDEX,

Where SALES441 is the total retail sales for Subsector 441, NAICS441 is the total retail sales for the non-automobile components, INDEX is the automobile index, and α is a constant that is set so that the ratio of SALES441 to Total Retail Sales and Food Services less Gasoline is equal to the nationwide ratio for the period July-December 2019.  The calculated value of α, 16,043, is assumed to be a constant over the sample period.

The SALES441 variable represents 19.2% of total Retail Sales in Arkansas, compared with 19.9% for the U.S. Retail Trade and Food Services nationwide (2019 averages).

Monroe and Saline Counties
Two counties present an additional complication:  Monroe County has no countywide sales tax, and Saline County established their county-wide tax only in April 2019.  For both counties, we construct county sales data by summing over the municipalities within the county that collect local sales taxes.[5]  These measures are, by construction, incomplete so some scaling is necessary.

For Monroe County, the totals calculated by summing over municipalities represents 0.139% of the total state excluding Monroe.  In terms of personal income, data from the Bureau of Economic Analysis shows that Monroe County’s share of state Personal Income is 0.183%.  Consequently, the raw data from the city aggregates is scaled up, multiplying by a factor of 1.32.

For Saline County, the county sales tax data accumulated since mid-2019 provide a basis for more detailed scaling:  For retail sales in NAICS codes 44 and 45 (excluding 441, Motor Vehicles and Parts Dealers), total county sales tax collections run approximately 20% higher than the sum of the municipalities.  The municipal-tax measures also generate a much smaller total for Motor Vehicles and Parts, with the countywide measure since mid-2019 running almost exactly twice the magnitude of total of municipalities.  On the other hand, total food services sales (NAICS 722) are approximately equal using either the county-wide or sum-of-municipalities approach.  These three scaling factors (1.2, 2.0 and 1.0) are used as multipliers to scale up the components of Retail Sales for Saline County.

It is expected that with the accumulation of additional data over time, our retail sales data for Saline County will be revised to reflect the actual county-wide data, using the municipal data to construct historical values prior to April 2019.  For now, however, the municipal aggregate is maintained as a consistent time series.

Total Retail and Food Services Sales for Arkansas
After constructing the data for each three-digit industry subsector by summing across counties, each subsector is seasonally adjusted and the components are summed to produce seasonally adjusted aggregates.  The final results, series for Arkansas Retail and Food Services Sales, are displayed in Figure 5.

Figure 5:

 

Data for Arkansas Retail and Food Service Sales for July 2017 through December 2020 are available in an Excel Spreadsheet:  Arkansas-Retail-Sales-Dec-2020.

The data set includes statewide aggregates and components, both seasonally adjusted and not-seasonally adjusted.  County-level data for Total Retail and Food Service Sales ex Gas, as well as ex Auto and Gas, are available on a not-seasonally adjusted basis.

A PDF File of this report is available HERE:  Arkansas Retail Sales – A New Data Set from AEDI

 

Endnotes:

[1] The Census Bureau has begun publishing statistics on Monthly State Retail Sales (MSRS), but the data set is still a work in progress.

[2] In implementation, ATS is calculated using the .25% Conservation Tax, which has been invariant to recent changes in the taxable status of groceries and other special exemptions.  More details of its construction is available at https://arkansaseconomist.com/arkansas-taxable-sales/.

[3] NAICS is an acronym for North American Industrial Classification System, https://www.census.gov/naics/.

[4] The two methodologies—the use of total net taxes versus the calculated gross—were compared in alternative data-construction exercises.  For statewide aggregates, the choice made little substantive difference, but comparisons revealed some examples of particular county-sector combinations for which the effects of the rebates and audits on monthly changes was clearly overwhelming the information in the underlying gross collections data.

[5] In Monroe County, the municipalities with sales taxes are Clarendon, Brinkley, Roe and Holly Grove.  In Saline County, taxes are collected by Bryant, Shannon Hills, Benton, Bauxite and Haskell.