feenberg
commented
8 years ago On Thu, 15 Oct 2015, Matt Jensen wrote:
One of the assumptions in the data extrapolation routine that stands out is “what will the wage distribution be in the future.” As John has pointed out, it is very difficult to predict this distribution over a ten-year horizon. Currently we assume that the distribution of wages and salaries is the same throughout the budget year as it was in 2013.
It would be nice to know how sensitive our model’s revenue estimate (and those of other similar models) is to the future wage distribution. Depending on the answer; we may want to add the ability to vary the wage distribution as a feature of the Tax-Calculator API and of TaxBrain.
Just the wage distribution? Capital gains varies a lot from year to year and extrapolations of revenue from good years are therefore particularly optimistic. Also, it is odd to vary wages but not deductions.
@feenberg has suggested that the historical variation in the wage distributions could serve as a range for such a sensitivity analysis. For instance, we could show how our 2015 revenue estimate would change if the wage distribution in 2015 matched the wage distribution from each year for the last couple of decades — we would keep everything else, including aggregate wages, constant.
@amy-xu has suggested that we might implement this via generating new different stage II targets of wage and salary by real AGI. We can get these historical targets from SOI historical wage distribution by AGI class tables (Table 1.4). This is an expensive way to do things though, because it would require re-running the extrapolation routine in tax data.
If expense is an issue, why not reweight the aggregates according to the proportion of taxpayers in each AGI class for various years past years. That would be computationally light.
dan
A couple of questions:
1.
Are there other interesting experiments besides: “How sensitive is the 2015 revenue estimate to the wage and salary distribution within the range of the last couple of decades?”2.
Are there other ways to accomplish this, other than by using stage 2 targets of wage and salary by AGI class?— Reply to this email directly or view it on GitHub.[AHvQVZuHJrk761BzKLNyi_jHG9BRKd_vks5o8FMLgaJpZM4GP8Sw.gif]
MattHJensen
martinholmer
Amy-Xu
MaxGhenis
jdebacker
One of the assumptions in the data extrapolation routine that stands out is “what will the wage distribution be in the future.” As John has pointed out, it is very difficult to predict this distribution over a ten-year horizon. Currently we assume that the distribution of wages and salaries is the same throughout the budget year as it was in 2013.
It would be nice to know how sensitive our model’s revenue estimate (and those of other similar models) is to the future wage distribution. Depending on the answer; we may want to add the ability to vary the wage distribution as a feature of the Tax-Calculator API and of TaxBrain.
@feenberg has suggested that the historical variation in the wage distributions could serve as a range for such a sensitivity analysis. For instance, we could show how our 2015 revenue estimate would change if the wage distribution in 2015 matched the wage distribution from each year for the last couple of decades — we would keep everything else, including aggregate wages, constant.
@amy-xu has suggested that we might implement this via generating new different stage II targets of wage and salary by real AGI. We can get these historical targets from SOI historical wage distribution by AGI class tables (Table 1.4). This is an expensive way to do things though, because it would require re-running the extrapolation routine in the taxdata repository.
A couple of questions: