Use TC to do UBI reforms with CPS tax unit dataset (3 months)

[Amy-Xu]

In a meeting last week with @MattHJensen, it seems our three-month-goal is to be able do UBI reforms based on CPS tax unit in TC. If this works, then we would add both current tabular merge and the CPS tax unit micro-simulation for UBI to TB. Otherwise, only tabular merge to TB.

Here're a list of issues I think we need to deal with to achieve the 3-month goal:

• Adjust tax liabilities from CPS tax unit to match PUF results. As Anderson has been working on this in taxdata issue #90, it seems Anderson should continue the good work there and ask reviews from Matt and Dan.

• Add benefit data to CPS tax unit for 2014. I suggest Anderson to work on this and get review from Matt and me.

• Add 10-year extrapolation for benefit data. I will be working on this, and hopefully get this done by the end of August.

• Expand UBI functions for reform based on micro data. Suggesting Anderson to work on this once 1 & 2 are completed.

I imagine 1, 2, 4 each might take approximately 1 month. @andersonfrailey Is that a reasonable estimate? Did I miss anything in the list above? Any thoughts?

[andersonfrailey]

@Amy-Xu, I believe ~1 to 1.5 months is a reasonable estimate.

For number 2, can we use the C-TAM with the CPS tax units to impute benefits or will we try and use the raw CPS and add everything together during the tax-unit creation process?

[Amy-Xu]

@andersonfrailey Because C-TAM only imputes benefit at individual/family/household level, in other words to raw CPS, I would expect we have glue the adjusted welfare data to raw CPS and put everything together during the tax-unit creation, but at least we don't have the income-matching problems this time.

[MattHJensen]

If this works, then we would add both current tabular merge and the CPS tax unit micro-simulation for UBI to TB. Otherwise, only tabular merge to TB.

One caveat: If this works, we'll aim to add the CPS tax unit file as a data input option to TB and forgo the tabular merge. If it doesn't work, we'll focus on the tabular merge.

[Amy-Xu]

It seems from Matt's @MattHJensen comment in taxdata issue on CPS file, task 1 in this issue is complete for current stage, is that correct?

[andersonfrailey]

@Amy-Xu, if we're ok with the results shown in the notebook then yes. I also just finished task 2. Working on checking totals now.

[MattHJensen]

@andersonfrailey, @Amy-Xu, it would be good to know why the itemized deductions changed so dramatically w/ the latest run before checking the box.

[andersonfrailey]

@MattHJensen I believe I found the solution. In early iterations of the notebook, I had included a variable HMIE - home mortgage interest expense - as e19200. It totaled over $414 billion and dramatically increased the number of itemizers. I removed it because I decide it was not actually equivalent to e19200.

The graph in the notebook only shows total itemized deductions for itemizers so by increasing the number of them to 60.7 million from 39.5 million, more was counted. Below are the graphs both with and without HMIE as interest expense.

With

Without

[Amy-Xu]

@andersonfrailey I guess what @MattHJensen means is that in the latest version itemized deduction total needs more scrutinization -- the total number of itemizers is not bad, but the total amount of itemized deduction in CPS is 45% lower than PUF, and 49% lower than SOI stats. But what I'm not sure is whether we should adjust it right now or wait until later? @MattHJensen

[Amy-Xu]

@MattHJensen Do you think it's necessary to tweak itemized deduction total amount more before checking the first box here in this issue?

[MattHJensen]

It sounds to me like "Home Mortgage Interest Expense" is a subset of "Interest Paid", and so it makes sense to include it in e19200. After that, I don't think it makes sense to continue tweaking these until everything is publicly available and documented.

[MattHJensen]

Also, for the purposes of documentation, "state and local taxes" are probably better described as "state and local income taxes" as the vast majority or all of "real estate taxes" are state and local as well.

[Amy-Xu]

If this is the case, @andersonfrailey I would suggest to shrink everyone's HMIE by the ratio of HMIE to SOI home mortgage interest of itemizers, and then include HMIE as e19200. In this way, we're able to have state and local income taxes deduction, and not overblow total number of itemizers.

[andersonfrailey]

@Amy-Xu that sounds good. I'll do that and see what happens with both total itemized deductions and number of itemizers.

[andersonfrailey]

@Amy-Xu @MattHJensen scaling down HMIE did make the itemizers numbers better. They still aren't totally accurate though. You can see the results in the notebook I most recently posted in the tax data issue.

[andersonfrailey]

Here is a notebook where I recreated the UBI reforms using the the CPS file.

functions3 is the same as functions2 was for the original UBI, just with a few variable name changes to accommodate the CPS file. Because Tax-Calculator drops variables it doesn't deem useable, I read in the CPS separately and then created new attributes for the Records class after initialization for each of the benefits programs.

@Amy-Xu @MattHJensen

[Amy-Xu]

@andersonfrailey just to clarify last box in this issue -- we would want to include the functionality of of the UBI notebook in TC if everything goes smoothly.

[Amy-Xu]

It seems like the second task is done based on the notebook posted by @andersonfrailey. Anderson, could you add a cell in the notebook printing out the total participants and benefits for each program? So we can compare how tax unit aggregates look like compare to the raw CPS aggregates. @MattHJensen I'm thinking just compare aggregates, not distribution, at this point since there's no valid distribution tax unit to compare with.

[andersonfrailey]

@Amy-Xu I've added them and posted the new notebook in issue #90

[Amy-Xu]

Thanks Anderson!

In the notebook (cell 120) from taxdata issue #90, Anderson posted summary table for each program.

Since we have made a few modifications for C-TAM, here're the totals CPS participation and benefits after adjustment.

Program	Participants (millions)	Benefit (billions)
Medicare	50.5	576
Medicaid	61.6	396
Veteran's Benefit	5.0	146
SNAP	50.7	70
SSI	9.3	59

There will be some dead weight loss loss after tax-units get created, but it would be the best if we could make the diff under 10%.

[andersonfrailey]

	Total	Participants	Avg. Benefit Per Person
VB	146,835,166,041.32	4,908,776.46	29,912.78
Medicaid	368,572,083,338.75	27,937,147.80	13,192.90
SNAP	105,660,492,407.65	28,539,328.92	3,702.28
SSI	54,186,915,454.49	7,276,536.68	7,446.80

When I modified the SAS files to include some of the information Amy needed to extrapolate the benefits programs aggregate totals for them afterwards. I haven't included Medicare in this because I'm running into an issue where benefit totals dropped significantly. I'm looking into why right now and will post when it's fixed.

@Amy-Xu