PUF asset imputation using SCF data

[codykallen]

I've been thinking that it may be useful to attempt to impute asset information for PUF and/or CPS records., using the Survey of Consumer Finances. They have a public use file, but I'm not particularly familiar with it or its quality. However, if the SCF public use file is not usable, results could perhaps be imputed using estimates from their published results, which can be downloaded here. In particular, for each year, Table 6 [year] sheets provide the percent of families holding different types of assets, and the Table 6 [year] means sheets provide the mean values of families' asset holdings of each type conditional on holding the asset.

These tables include breakdowns by income groups, by age of the head of household, by family type (children information), and work status (perhaps connected to income types). Given this information, and combining it with information in the PUF on income from some of these asset types, I would think it would be possible to impute tax units' asset holdings for at least some asset types.

In terms of applying this, I believe it would be applicable to BRC, in that it would make it significantly easier to distribute the static burden of the corporate income tax to tax units in the PUF, and it would make the distribution of the corporate tax burden significantly more accurate, in the BRC is not currently capable of distributing the burden to unrealized capital gain/loss, only to dividends and realized capital gains/losses.

If it helps, Table 10 in the linked Excel file provides information on unrealized capital gains for some asset types. Also, the Table 13 [year] sheets provide information on home mortgages with the same distributional breakdowns, which could be useful for improving the imputation of mortgage interest paid.

@andersonfrailey @martinholmer @MattHJensen

[martinholmer]

@codykallen, Thanks for the SCF information in taxdata issue #285. One quick question: are the published tabulations for each unit type (like income class and head age range) individually or are there tabulations by (for example, income class and head age range)? In other words, can we see asset holdings for older heads with high income in comparison to asset holdings for older heads with low income?

@andersonfrailey @MattHJensen

[codykallen]

@martinholmer asked:

are the published tabulations for each unit type (like income class and head age range) individually or are there tabulations by (for example, income class and head age range)? In other words, can we see asset holdings for older heads with high income in comparison to asset holdings for older heads with low income?

Unfortunately, the published tabulations do not include cross-tabulations. However, we should be able to use their public use dataset, either to produce our own cross-tabulations or to build a predictive model that can be applied to the PUF.

The full SCF page is available here. If also recommend looking at their Net worth flowchart, which simplifies the process of identifying relevant wealth variables and/or measures.

[martinholmer]

@codykallen said in taxdata issue #285:

Unfortunately, the published tabulations [of the SCF] do not include cross-tabulations. However, we should be able to use their public use [SCF] dataset, either to produce our own cross-tabulations or to build a predictive model that can be applied to the PUF.

The full SCF page is available here. If also recommend looking at their Net worth flowchart, which simplifies the process of identifying relevant wealth variables and/or measures.

@codykallen, thanks for the links and the idea. I agree: if we could impute some wealth holdings and unrealized capital gains variables onto the puf.csv file, it would (as you say) make it much easier to integrate the corporate income tax incidence logic into Tax-Calculator. My schedule is such that I will not be able to look into the feasibility of doing this until early October. I'm sure I'll have more questions then, particularly about the exact nature of the data needs of the corporate income tax logic.

@MattHJensen @andersonfrailey

[martinholmer]

Assigned working on taxdata issue #285 to @codykallen because he said during the online developers meeting that he was working on this.

[codykallen]

I've been working the last few days on doing a statistical matching between the PUF and the SCF. The work for this is over in a separate repo (https://github.com/codykallen/puf-scf-match), with several different versions available.

[MattHJensen]

Tagging #34 as related.