Over the past few weeks the conversation about sales tax has been pretty consistent and fairly intense. That is probably as it should be given the overall importance of revenue generation across the state right now. What have I done so far? I think the best thing to date may be the discussion surrounding the rolling twelve month series used by the city. Rather than accuracy it is more about stability. Eleven of twelve months in the series are the same from one month to the next so of course the series is stable. The flaws of this approach can be seen when you look at the recent decline. A good month is totally outweighed by the totality of the previous eleven. It also begs a statistical question: should all these months be given equal weight while in the series and then just disappear? That seems highly suspect.
So it is all well and good to poke holes in the approach followed by others. There is a place for that certainly and it can and does inform policy in important ways. However, better is to offer up something different or new. My model is still in its infancy but I offer up a forecast of sales tax collections. I do this mostly because we will soon get data about the new revenues collected from the sales tax increase approved in the fall. Those are not built into this forecast yet but later versions will have it. Frankly, it will be interesting to see if it even reverses the trends demonstrated in my earlier posts (here, here, and here).
The data come from the City of Grand Forks sales tax release and it is of the monthly collections. The black line represents that actual data series released by the city while the red line are model predictions. Like I said, I am working on the model still but this is far enough along I thought I would start the discussion. The blue line and the blue shaded areas are the forecast prediction and the confidence intervals around the prediction.
The starting point for the forecast is a bit difficult to determine precisely. I should say the proper starting point is a question. I know when this series started (January 2000). However was there still a “hangover” from the flood at this time? There was the air base realignment in this interval too. These are factors I continue to investigate so expect there to be updates to this model along with the demographic and other predictions.