The January sales tax collections data came out recently and I want to take a longer view of the data here. This typically means messier graphs because the time scale gets compressed. I will put up a post with a shorter timeline tomorrow. Why the longer timeline? I think it is necessary if we are going to try and interpret or understand the numbers we are currently seeing.
The graphs show two different measures, both in percentage change terms. The first is the percent change (year-over-year) for the monthly collections data. That simply implies a comparison of January 2018 with January 2017. Those who read my blog in the past will recall this is an easy way to remove some of the seasonality in the data. If there is a “January effect” in the sales tax data this approach controls for it.
There is quite a bit of volatility in the data series as well as some significant (and almost regular?) outliers. The Flood of 1997 shows up in the data with the typical increase in sales after a natural disaster, and then the drop off afterwards. Note it appears to be the case that the series has a significant number of observations below zero towards the end of the time horizon. In fact the January 2018 number was over 16% down compared with January 2017. In fact, if we go back the last 25 observations (the last 2 years plus January 2018) 15 of those months were below the year ago level. It is not uncommon for this window to reach double digits for Grand Forks, though it seems to be happening more frequently since around 2010 which likely points to possible negative effects of the oil boom in the Bakken region for Grand Forks.
The City of Grand Forks prefers to use a rolling twelve month window for its measure. For those that do not know what this means, you would look at a period of January to December in year t as one twelve month window. When you get the next observation (January of year t+1) the last January observation leaves the window so you have a February to January data series. Basically every month you generate a new measure of the last year of sales tax collections. The city writes this is preferred because it is a more stable measure. Looking at the data back to 1988 it does have a lower standard deviation.
Just a casual glance demonstrates less volatility in the data. There are a few other features to note including the presence of a “Flood effect.” In addition, the drop off in sales after the increase due to the flood is easier to see. There are fewer significant outliers in this series and the end of the series is again more negative. In fact, using the same widow of twenty-five months there are 18 months where the rolling window moved in a negative fashion (recall this is a month to month percent change). As a result. The only time this was higher was in the window after the Flood of 1997 when there was a clear increase in sales that would naturally lead to a fall off. We do not have that happening currently so an excellent question is why this occurs now?
Is it a loss of retail establishments such as Macy’s? That cannot help but seems unlikely to be the complete cause. Are there population and income effects here? These also seem to be possible, partial, explanations. There are other candidates as well including the U.S.-Canada exchange rate that could weigh on sales. Clearly we need an answer if we are going to better inform policy decisions in the city so send me your favorite candidates as I continue to model possible solutions.