Skip to content

README.md

Latest commit

 

History

History
139 lines (105 loc) · 6.24 KB

README.md

File metadata and controls

139 lines (105 loc) · 6.24 KB
Raw

Validation of Tax-Calculator against Internet TAXSIM-35

The general cross-model validation process described here is being executed in this directory using TAXSIM-35.

We are in the process of comparing Tax-Calculator and TAXSIM-35 results generated from several assumption sets in the taxsim_input.py script for years beginning with 2018. Each INPUT file is used to generate a TAXSIM-35 OUTPUT file by uploading it to the TAXSIM-35 website and requesting detailed intermediate calculations. And each INPUT file is translated into a CSV-formatted input file that is read by the Tax-Calculator tc CLI tool to generate output that is then transformed into an OUTPUT file having the TAXSIM-35 format. Finally, these two OUTPUT files are compared using the main_comparison.py script. See the tests_35.py script in this directory for more details.

The following results are for INPUT files containing 100,000 randomly-generated filing units for a given year. The random sampling is such that a different sample is drawn for each year. In each INPUT file three state-related variables are set to zero for every filing unit, one variable specifies the year, and another specifies a filing unit id, which leaves twenty-two input variables that are set to randomly-generated values.

In order to handle known differences in assumptions between the two models, we use the taxsim_emulation.json "reform" file to make Tax-Calculator operate like TAXSIM-35. See the taxsim_emulation.json file for details.

In the following results, when we say "same results" we mean that the federal individual income tax liabilities and payroll tax liabilities being compared have differences of no larger than one cent.

For information on the variable names illustrated in taxsim_input.py, the document that generates data for input into TAXSIM-35, see the TAXSIM-35 website listed above.

Instructions

  1. Navigate to taxcalc/validation/taxsim35 and run the Python script tests_35.py.
  2. If you would like to generate new input files and and get new files from TAXSIM-35, just delete all of the .in.out-taxsim files. On Mac/Linux, this can be done with rm -f *.in.out-taxsim.

Troubleshooting

If the TAXSIM-35 validation code throws errors such as .in files not found, .out files not found or that any parameter within policy_current_law.json does not exist, please try these 2 steps:

  1. Make sure that the taxcalc conda package is installed
  2. If you have Tax-Calculator downloaded locally, navigate to the root directory and run pip install -e . This will install the current source code into the taxcalc CLI.

Description of Randomly Generated Assumption Sets

a18 ASSUMPTION SET:

2018 INPUT file that specifies the first twelve of the TAXSIM-35 input variables, which include demographic variables and labor income, but sets to zero all the TAXSIM-35 input variables numbered from 13 through 27.

b18 ASSUMPTION SET:

2018 INPUT file that specifies the first twenty-one of the TAXSIM-35 input variables, which include demographic variables, labor income, capital income, and federally-taxable benefits, but set to zero all the other six TAXSIM-35 input variables except variables 28-35, which are the variables representing the new QBI-related variables. Two of those six are always set to zero because they specify transfer income that is not taxed under the federal income tax or because they specify rent paid that does not affect federal income tax liability. Three of the remaining four input variables are itemized expense amounts and the fourth is child-care expenses.

c18 ASSUMPTION SET:

2018 INPUT file that specifies all the non-state TAXSIM-35 input variables to be randomly generated values.

a19 ASSUMPTION SET:

2019 INPUT file that specifies the first twelve of the TAXSIM-35 input variables, which include demographic variables and labor income, but sets to zero all the TAXSIM-35 input variables numbered from 13 through 27. (This is the same logic as used to generate the a17 sample except that a different stream of random numbers is used so that the 100,000 filing units are completely different.)

b19 ASSUMPTION SET:

2019 INPUT file that specifies the first twenty-one of the TAXSIM-35 input variables, which include demographic variables, labor income, capital income, and federally-taxable benefits, but set to zero all the other six TAXSIM-35 input variables except variables 28-35, which are the variables representing the new QBI-related variables. Two of those six are always set to zero because they specify transfer income that is not taxed under the federal income tax or because they specify rent paid that does not affect federal income tax liability. Three of the remaining four input variables are itemized expense amounts and the fourth is child-care expenses. (This is the same logic as used to generate the b17 sample except that a different stream of random numbers is used so that the 1,000 filing units are completely different.)

c19 ASSUMPTION SET:

2019 INPUT file that specifies all the non-state TAXSIM-35 input variables to be randomly generated values. (This is the same logic as used to generate the c17 sample except that a different stream of random numbers is used so that the 100,000 filing units are completely different.)

Validation Results

The latest expected differences between Tax-Calculator and TAXSIM-35 can be found in the expected_differences folder in this directory.

These differences are summarized and explained in the Differences_Explained markdown file.

Differences are generally small, although we note larger differences due to how Tax-Calculator and TAXSIM-35 handle the following items:

  • The computation of SECA taxes
  • The 2020 Recovery Rebate Credit and Economic Impact Payments
  • The 2021 Child Tax Credit
  • Itemized deductions