TAXSIM is the NBER's FORTRAN program for calculating liabilities under US Federal and State income tax laws from individual data. Scholars interested in using TAXSIM to calculate tax liabilities and marginal tax rates should follow the these instructions for submitting data. The program itself is not released.
Version 3.0Beta introduces:
A crosswalk of Census state codes and SOI codes is available, but you have to do the conversion yourself. If you don't want state tax calculations, just set the state id to zero.
This is a simplified list of input variables selected to encompass what might be available in a non-tax oriented survey such as the CPS, CEX, PSID, etc. The items are rich enough to cover the ordinary and supertax brackets, the earned income credit, the child care credit, the partial taxation of Social Security, and UI, the secondary earner deduction, the maximum tax on earned income and other important features of the tax code. At the state level, low income allowances, property and rent credits are calculated if data is available. If you don't know the wage split between husbands and wives, you can put it all in one category. The computer won't get the perfect answer, but only you can judge the suitability of the simplification. Obviously ignored are the capital gains deduction, limits, floors and ceilings on deductions, adjustments, tax preferences, minimum taxes, separate filing etc. These are available in the full TAXSIM model used internally at the NBER, but have been left out of Internet TAXSIM for the sake of simplicity.
Here is a sample entry for a single parent of one child in Idaho with only labor income, and $2000 in child care expenses. Notice that no missing value indicator is used - zero amounts are coded as zero.:
Sample input and output files are are available to allow users to make a quick test of the functioning of the server. To run the test right click on the sample file link to download the 14 case test file to your PC. Then submit it back to our server using the file selection box above. Compare the values returned with the cannonical output file stored on the server. If they are the same, the server has passed a simple test.
The file may include any number of such lines. While the format is flexible, and values may be of variable size, each value must be separated from the next by one or more spaces or a single comma. If one line is short of values, the next will be read to make up the set of 18, but a new case always begins on a new line. We cannot read .prn files, or any file that uses commas to mark thousands. Spreadsheet files of any kind can not be read. Before uploading your file please check for the following common errors:
The tax calculator stops on any error and returns a range error with the case ID. Some users are concerned about imputations. Internet TAXSIM only does one. Social security taxes on wages and business income are imputed for the states that have a deduction for that tax. Anything not listed above becomes a zero on the Form 1040. Zero is always a valid entry, although it may be inaccurate. Formerly we did an imputation for rent when property taxes were less than one percent of income. This has been dropped.
TAXSIM will calculate the federal and state tax liability, and also the marginal tax rate calculated on an additional $1000 in wage or non-wage income. Since the federal and state tax liabilities may depend upon each other, the calculations are iterated, with the calculated state tax entered on the federal form as a (potential) deduction and the federal tax similarly placed in the state calculation. (Yes, we know that in law it is only lagged taxes that matter, but that is less interesting for our purposes). Refundable credits appear as negative liabilities.
In many cases the marginal rate of tax taken from schedule X, Y or Z of the form 1040 correctly characterizes the tax wedge. But for many taxpayers, their effective marginal rate is affected by a phase-out or other special provision of the tax code.
These effective marginal rates are important to us, because one of the most important uses TAXSIM can be put to is calculating correct after-tax prices in models of economic behavior. A problem has been the presence of decreasing marginal tax rates (notches) and otherwise implausible marginal rates in these calculations. We maintain a list of significant notches that we have noticed. Although notches generally involved very small sums of money, they could have disproportionate effects on the results of a linear regression, which is sensitive to the squared difference of any observation from the mean.
Some of the anomalies arose from defects in our FORTRAN code, these have been corrected where found. Some arose from actual notches in the tax laws. For example, in the federal law the child care credit is an integer percent of child care expenses, depending on AGI. In this case we have used a smooth interpolation between adjacent percentages even though the tax code prescribes a step function. Especially at the state level some credits are just a fixed amount for taxpayers with income less than X. In that case the range for an smoothing isn't obvious and we have not done any. So some notches will remain. We have reduced the maximum magnitude of the notches by calculating the marginal rate over a $1000 range instead of the $100 range used in Version 2 of Internet TAXSIM.
We have also set limits on the range of marginal rates the state tax calculator will report. If any calculated rate exceeds 25 percent (or 40 percent for taxpayers with negative liability generated by refundable credits) we recalculate the marginal rate using a $1000 decrease in business income. Since there probably is no notch on the decrease side (and business income is less subject to notches than wage income) this alternative calculation generally returns a more routine estimate of the marginal rate. Because of the biased selection of taxpayers to receive this treatment, the procedure adds a tiny bias to an estimate of the average marginal tax rate, but it may make for more accurate estimates in regressions using the marginal tax rates as regressors. In a run with the 1995 Current Population Survey as the data source, 21 taxpayers out of 51,000 were truncated in this fashion.
We are very interested in other solutions to this problem, and suggestions are welcome.
The returned eight columns of results include:
If you suspect any case has been calculated incorrectly, please extract that case (or a exemplar, if there are many cases) and email it to me with a statement of what you think is wrong. I will get back to you within a couple of days with an explanation or a fix. It is important that you not send large files or attachments. Just send an email with a few cases, the TAXSIM response and the reason you believe the response to be in error.
We do answer all mail, and we do follow up on queries, but it is important to abide by the guidelines in the previous paragraph. An email message with the statement that "state x is broken for feature y" will be returned with a request for an example in Internet TAXSIM format as described above. Once we get that, we will compare our results with hand generated results done from a tax form and any problem will soon be resolved. Comments from users have led to significant improvements in the program.
Some users have expressed concern about our ability to process large files, but files up to 100,000 cases are not a problem. Larger files should probably be broken up and run in the evening. Anytime you are getting timed out, please wait several hours and try again, as the time-outs are a sign that something is overloaded.
Users of the AHEAD and HRS are reminded that they are not permitted to send state identifiers to us. If such users need state tax calculations they should send each record 51 times, once for each possible state, and select the correct record from among the returned records. Some problem records are retained on our server.
This paper and this URL (http://www.nber.org/taxsim) should be cited if any results from TAXSIM are circulated or published. If you intend to make any serious scholarly (or other) use of Internet TAXSIM you should speak with me about your project on the telephone. I am at 617-588-0343 and can offer advice etc.
Daniel Feenberg
NBER
1050 Mass Ave.
Cambridge MA 02138
617-588-0343
feenberg@nber.org
Date last modified: June 2nd, 2000