In the world of public policy, a common area of disagreement among elected representatives is what the revenue forecast will look like in the coming budget years. For instance, in California, legislative analysts recently projected $20 billion more to spend in 2022 than what Governor Newsom projected. That may not seem like much in a world of trillions in government spending, but even in large states, $20 billion can make a big difference for thousands of state employees and the programs they administer.
Given that there can be considerable differences of opinion about the future trajectory of a state’s economy, one might wonder whether mobility — such as SafeGraph’s cell phone point-of-interest data — may be helpful in predicting movements in revenue. The two most often employed revenue sources for state governments are the individual income tax and the sales and use tax.
First step in the analysis
The first step in the analysis is to gather the data. In this exercise, the revenue data on income tax and sales tax is reported quarterly by the U.S. Census Bureau. The mobility data stems from SafeGraph. For academics, this data is free, simply sign up with your .edu email address.
The raw data
The raw data was download from SafeGraph. For non-academics, the best place to start is with the sales team at SafeGraph. They’ll walk you through the details. For academics, you can join the community for free. The raw, downloaded data will have data files with the .gz file extension. To create a quarterly time-series of visits, you’ll need to combine all the files. The following is a screenshot of the files I placed into one file folder.
The next step, done in R, was to combine the data.
setwd(“D:/SafeGraph/Tax Revenue/Files”)names_of_filestaxprep <- list.files(“D:/SafeGraph/Tax Revenue/Files”,pattern=glob2rx(“*.gz”),full.names = TRUE)library(furrr)taxprep<-furrr::future_map_dfr(names_of_filestaxprep,data.table::fread)write.csv(taxprep,”taxprep.csv”)
Prior to importing into R, I limited the analysis to just visits to industry classification code (NAICS) 541213, which corresponds to tax preparation service POIs.
The raw revenue data for income tax and sales tax was structured as a panel data set, beginning in the first quarter of 1963 and ending in the first quarter of 2021.
Once the two files were structured, I used Excel’s VLOOKUP function to match mobility by state and quarter for the two sources. The movement data only goes back to 2018, meaning the final data set had 728 state-quarter observations.
Connecting movement with income tax
The two components of the analysis are set. The next step of the analysis is to connect the two and see if the two have a positive or negative relationship. An upward sloping line is a positive relationship, and means that when visits go up, revenue goes up. A downward sloping line is a negative relationship, and means that when visits go down, revenue goes up. In theory, the former should be correct.
The first view below is the point-in-time relationship between quarter-over-quarter change in visits to quarter-over-quarter change in income tax revenue. Each line represents a state. Clearly, there is wide variation in the contemporaneous relationship.
Interestingly, New Mexico has the strongest negative relationship at -6.7 (p-value = 0.01), while New Hampshire has the strongest positive relationship at 1.8 (0.39).
Connecting movement with sales tax
The following is the contemporaneous relationship between the quarter-over-quarter change in sales tax with the quarter-over-quarter change in raw visits. Interestingly, only a few of the states had a negative relationship between change in mobility and change in sales tax revenue. These states were Connecticut, South Dakota, Wyoming, Nevada, West Virginia, Vermont, and Washington, D.C.
Comparing mobility’s connection with income tax and sales tax
The following “first look” regression has the change in income tax as the dependent variable and the bottom pane has the change in sales tax as the dependent variable. Fascinatingly, there is a consistent negative relationship overall (at least when using panel data structure) between the change in income tax revenue and the change in mobility on a quarter-over-quarter basis. Further refining may find a different result, but interesting nonetheless.
Also interesting, the results on sales tax are generally insignificant. As with income tax, further research could change this result.
The following STATA code was used for the panel analysis. Results, obviously, change when adjusting the parameter assumptions.
clear
import excel “D:\SafeGraph\Tax Revenue\TaxPreparation.xlsx”, sheet(“Sheet1 (2)-Tableau”) firstrow
gen quarterly = yq(Year,Quarter)
format quarterly %tq
xtset PanelID quarterly, quarterly
asdoc xtreg Changeinincometax Changeinvisits L.Changeinin L.Changeinvisits, re
asdoc xtreg Changeinsalestax Changeinvisits L.Changeinsalestax L.Changeinvisits, re
Parting thoughts
Economists use a host of indicators to predict future revenue figures. Although this was a simple introduction to the issue, it offers a potentially fruitful, cutting-edge avenue to address revenue forecasts in the future. The lackluster relationships found in this study might be due to pandemic restrictions in 2020 and 2021. A follow-up study to this will look at this issue.
Questions?
I invite you to ask them in the #safegraphdata channel of the SafeGraph Community, a free Slack community for data enthusiasts. Receive support, share your work, or connect with others in the GIS community. Through the SafeGraph Community, academics have free access to data on over 7 million businesses in the USA, UK, and Canada.