Total value of Child Tax Credit differs between CPS, PUF, and other sources

[MaxGhenis]

This issue summarizes discrepancies between total values of the Child Tax Credit between the CPS, PUF, and other sources. In #157, @andersonfrailey said he has ideas on improving n24 in the CPS which should help, so this just provides some context using tax year 2018:

• Using CPS data and Tax-Calculator (notebook), the estimated spend is $123.6B. This is the difference in after-tax income between a 2018 baseline and a reform that sets _CTC_c and _DependentCredit_Child_c to zero.
• Using PUF via TaxBrain to set the same parameters to zero (https://www.ospc.org/taxbrain/32874/), the total spend is $112.4B.
• Combining external reports suggests it could be about $120B: this 2017 TPC report estimated $52B, and JCT estimated that TCJA's non-SSN-related reforms would cost $68B starting in 2019 (the 2018 estimate of $29B does not capture the refundable portions). TaxBrain estimates only $0.3B difference between 2018 and 2019, so between these two it's likely $120B within a rounding error.

Based on this I think the CPS data is too high. The simplest explanation is that n24 currently includes 17-year-olds when it should not (#164).

I'm also curious whether there might be an explanation for the $8B difference between TaxBrain PUF and TPC+JCT, or if there's another external benchmark (or if it's considered a reasonable tolerance).

[MaxGhenis]

I think I mistakenly filed this in the wrong repo because I was thinking about n24 including 17yos. Closing in lieu of PSLmodels/Tax-Calculator#1929.