Revenue Estimating Methods

Overview

One major function of our Economic and Statistical Research Bureau is to provide revenue estimates for proposed changes to the California Revenue and Taxation Code. In addition to developing revenue estimates for legislative bills, we receive revenue estimate requests from the Department of Finance (DOF), legislators, board members of the Franchise Tax Board, and other elected officials. We also report annually on the potential effects of conforming to the prior year federal tax law changes, as well as, the impacts from special provisions in California’s tax code (i.e. tax expenditures). In total, we produced over 700 revenue estimates in 2012.

Objective

We produce accurate, consistent, and impartial estimates of the revenue effects of alternative tax policies that the Governor, members of the Legislature, and the public can rely upon during the policy making process.

How we construct revenue estimates

The baseline

We start with the most recent tax revenue data available. We then apply growth projections to the data elements of interest to project the amount of tax that will be collected in the future if there are no changes in tax law. These projections rely heavily on forecasts of general economic growth and on many specific economic parameters (such as the growth in capital gains realizations) provided by third parties. Baseline revenue changes from year-to-year in response to projected changes in economic conditions, but it assumes there will be no changes in tax law in future years.

Tax year changes

After establishing the baseline, revenue analysts recalculate expected revenues for three to five tax years after the proposed change would become effective. The new calculations consider any changes in the amount of income subject to taxation, available tax deductions, tax rates, and tax credits resulting from the proposed policy changes. Projected revenues with the policy change are then compared to the baseline estimate to generate the estimated effect for each tax year.

Fiscal years and accruals

Revenue effects are first calculated on a tax year basis. Tax year effects of a policy proposal are then converted to the appropriate fiscal years (FY), July 1 to June 30, used in the state's budget planning process. We then project anticipated cash flows to the state resulting from the policy change. For example, a:

  • $100 tax reduction in tax year 2014 may result in a $20 reduction in withholding and estimated payments made between January and June of 2014 and assigned to FY 2013/14.
  • $40 reduction in withholding and estimated payments made between July 2014 and January 2015 assigned to FY 2014/15.
  • $30 decrease in final payments made with returns filed between January and April 2015 also assigned to FY 2014/15.
  • $10 decrease in payments made with extension returns filed between July and October 2015 assigned to FY 2015/16.

Any increase/decrease in refunds issued is treated identically to a decrease/increase in payments made to the state.

Finally, our estimates are adjusted for accruals. Accruals are accounting rules that apply to certain transactions, such as tax payments made past their due date. The rule used in our estimates is that proposals affecting old tax years are accrued backwards one year. For example, suppose a law providing a new collection tool results in $10 million in payments in January 2014 that should have been made in 2011. In this example, the payments are actually received in FY 2014/15, but because they are attributable to an earlier tax year, they are assigned to FY 2013/14.

Estimation tools

Revenue estimators have a variety of tools at their disposal to estimate the revenue effects of proposed changes to tax law. We provide a brief description of the most commonly used methods.

Microsimulation models

Microsimulation models calculate tax liabilities for alternative policies at the individual taxpayer level. We maintain both a Personal Income Tax (PIT) and a Corporation Tax (Corp) microsimulation model.

For the tax year 2011 PIT microsimulation model (the most recent year for which data are available), we collected about 500 data items from approximately 300,000 tax returns. The PIT sample is a stratified random sample. Taxpayers are divided into 42 groups based on their income and filing status. Enough tax returns from each group are chosen for the sample to produce statistically valid predictions for the tax paid by each group. The model is calibrated so that in the absence of proposed changes to tax law, it will reproduce projected tax liabilities for future years. This model is very good at estimating a variety of policy proposals, such as changes to tax rates or to existing credits and deductions. The model is less useful for proposals concerning exclusions or new deductions because tax returns do not contain any data directly relevant to such proposals.

Because the model recalculates taxes with real data on individual taxpayers, it automatically adjusts for interactions between the proposed policy and other features of the tax code. For example, when estimating the effects of an increase in tax rates, the model will identify which taxpayers have unused credits available to offset the increase in tax from the proposed rate change. Similarly, when estimating a proposal to eliminate a tax deduction, the model will know which tax bracket each taxpayer currently claiming the deduction is in and which taxpayers will be pushed into a higher tax bracket by removing the deduction.

Example: Estimating the effect of Proposition 30

Proposition 30, passed by voters in November of 2012, raised PIT rates for the years 2012 to 2017. The PIT microsimulation model was used to calculate tax liabilities under both the old and new tax rates for each taxpayer in the PIT sample. After scaling the results up to the entire population, the model estimated the difference to be an income tax increase of about $5 billion per tax year.

The sample used for the Corp microsimulation model is also a random stratified sample. For the 2011 Corp sample, we collected about 300 elements from approximately 17,000 tax returns pulled from 200 groups of taxpayers based on income and industry. This model is also calibrated so that in the absence of proposed changes to tax law, it will reproduce projected tax liabilities for future years. Like the PIT model, this model is also used for estimates on changes in tax rates as well as changes to existing credits and deductions. In addition, the Corp model can be used for analyses of issues specific to corporations such as changes in apportionment factors (which are used to determine how much of a corporation’s income is attributable to California). The Corp sample also contains industry codes, enabling estimation of policies affecting specific industries.

We also maintain several special datasets used to model specific aspects of tax policy. These include a sample of capital assets sales transactions, a database on enterprise zone credits, and a file that tracks net operating losses (NOLs) over time at the taxpayer level.

Estimates based on federal data

There are many estimates for which no relevant data is available from the California tax return, but for which estimates are available for federal versions of the same policy. This situation arises most frequently in the case of exclusions – transfers of money that taxpayers are not required to report as income – such as the exclusion of employer provided educational benefits. Most of the federal estimates we use are provided by Congress's Joint Committee on Taxation. Others may come from the Treasury Department. Federal estimates are often prorated (scaled down) to a California estimate by applying two adjustments. One adjustment is an estimate of the percentage of relevant California economic activity, and the second adjustment is for the ratio of the appropriate state and federal tax rates.

Example: Estimating the tax expenditure for miscellaneous fringe benefits

California conforms to federal tax provisions, such as the exclusion for employer contributions for certain employee benefits (collectively known as miscellaneous fringe benefits) from taxation. Our estimate of the annual cost to California for this exclusion is derived from federal estimates. According to the Joint Committee on Taxation, this exclusion reduced federal tax revenues by about $7 billion in federal FY 2011/12. We estimate that California taxpayers are about 13 percent of the population affected by the federal provision, and the average California tax rate for these taxpayers is about 29 percent of their federal tax rate. Thus, we estimate the impact of this exclusion on California revenues is just under 4 percent of the impact on federal revenues, or about $260 million for FY 2011/12.

Use of nontax data sources

We are asked to analyze many proposals that provide special tax treatment for specific taxpayer activities or taxpayers in specific circumstances. To estimate these proposals, our analysts use the best available data from reputable sources.

Example: Qualified medical services credit

AB 248 of 2011 would have created a new credit of up to $5,000 per taxpayer for certain volunteer medical services. For this estimate, we combined data on the number of medical personnel from a Rand California Health and Socioeconomic Statistics report with data from the Corporation for National and Community Service on the percentage of medical personnel who volunteer to estimate that about 15,000 taxpayers would qualify for the credit each year. After estimating the average amount of credit each taxpayer would be able to use, a cost to the state of approximately $50 million per year was estimated.

Modeling taxpayer responses to changes in tax law

Many changes in tax law will induce taxpayers to change their behavior. These behavioral changes may, in turn, change the revenue impact of the tax law change. Revenue estimates that ignore behavioral responses are often referred to as 'static', while those incorporating behavioral responses are called 'dynamic'. Our estimates incorporate dynamic elements by taking into account changes in microeconomic behavior and some changes in the relative size of different sectors of the economy. We do not, however, estimate changes in the overall size of the California economy.

Constant gross state product assumption

Our analysis assumes there is no change in the overall size of the economy. The primary justification for this assumption is that as long as California has a balanced budget requirement, the macroeconomic effects of any particular change in tax law are likely to be very small relative to the effects that are estimated. With a balanced budget requirement, any reduction in California taxes must be offset by either a reduction in state spending or an increase in some other part of the California tax system. Any beneficial effects on economic growth from the initial reduction in taxes will be roughly offset by the drag on the economy created by the corresponding reduction in state spending or other tax increase. In more technical terms, the balanced budget requirement prevents the state from engaging in expansionary fiscal policy. Our estimates also assume the effect of changes in the tax code that alter the overall efficiency and long run performance of the economy will not become significant until after the first few years of revenue changes that we estimate.

Types of behavioral adjustments incorporated in our revenue estimates

Our estimates include a variety of behavioral responses to tax law changes without violating the assumption there is no change in the overall size of the state economy. Many tax law changes will induce taxpayers to shift from nontax-favored to tax-favored activities.

Examples of behavioral adjustments into tax-favored activities

  • A revenue estimate of a proposal to create a new jobs credit for employers of disabled veterans begins with data on the number of disabled veterans currently being hired; we then augment the data with an estimate of the additional number of jobs that will be filled with disabled veterans instead of another type of employee.
  • A revenue estimate for a proposal to eliminate the deduction for interest paid on mortgages on second homes begins with data on interest currently paid on second mortgages, which we then adjust because a portion of affected taxpayers would respond by borrowing less when purchasing second homes.

Our estimates also consider changes in the timing of income recognition in response to tax law changes. For example, when estimating proposals to change depreciation rules, we will consider whether some investments will be accelerated or postponed in order to receive more favorable tax treatment. Similarly, estimates on proposals to reduce the tax rate on capital gains will include an adjustment for taxpayers who delay recognition of some gains until they can receive more favorable treatment (see example below).

Uncertainty about whether or not we would go over the federal "fiscal cliff" at the end of 2012 led to another interesting application of behavioral principles. Because taxpayers were expected to respond to scheduled increases in 2013 federal tax law (i.e., paying some bonuses and dividends in 2012 rather than in 2013 and recognizing some capital gains in 2012 instead of 2013), many recently produced estimates have shown relatively large revenue impacts for tax year 2012 and more muted impacts for tax year 2013.

Additional considerations in estimating

Vintaging

Some tax law changes may produce revenue effects that are spread out over a period of years even for a single transaction. In these cases, the effects of transactions in different years may need to be calculated separately, and then added together at the end. For example, consider a policy proposal to allow expensing for certain investments. Many types of investments are depreciated. That is to say taxpayers deduct a portion of their investment cost each year for several years. When expensing is allowed, a taxpayer can deduct the entire value of their investment in the year of the investment. This leaves them with no value to deduct in future years. Thus, the revenue effect for an item expensed this year will include both a revenue loss this year, and revenue increases in any future year during which the taxpayer would have had a depreciation deduction. The revenue effects of each year’s investments (vintages) must then be added together. For example, the total revenue loss for the third year after the policy change would be equal to the revenue loss from increased deductions for investments made in the third year minus the revenue gain from the depreciation deductions that taxpayers can no longer claim for investments they expensed in year one and year two.

Carryovers and grandfathering

Estimates of proposals to remove current features of the tax code incorporate any lingering effects of those provisions.

Example: Modeling credits used after a tax program ends

An estimate of a proposal to eliminate enterprise zones will consider several institutional details about how the zones would be eliminated. One of these questions concerns the treatment of employees hired before the elimination of the zones. Under current law, employers get a credit for the first five years that a qualified employee works for them. If the proposal being analyzed allows credits for any employee hired before the elimination data, we employ a vintaging analysis to model how much credit will be generated in the five years after the policy change by employees that were hired before the policy change.

The analysis also includes estimates of the use of carryover credits. Some taxpayers generate more tax credits than the amount of tax liability they have against which to use their credits. They may then be able to use the excess credits in a subsequent tax year. Because of this, an analysis of eliminating the enterprise zone program, but allowing the use of carryover credits, would show a small revenue gain at the beginning, increasing over a period of years as more and more taxpayers exhaust their supply of unused credits.

Interactions

Our revenue estimates consider the interaction between various parts of the tax code. Consider, for example, a proposal to extend the recent suspension of the use of NOLs by California corporate taxpayers. NOLs are losses incurred by taxpayers in earlier years that are allowed to reduce current year income. Suspending NOLs, therefore, increases taxpayer's taxable income and should increase their tax paid. In estimating an NOL suspension proposal, our analysts consider the extent to which taxpayers have excess tax credits that could not be used under current law, but can be used to offset the increase in taxes caused by the proposed NOL suspension. The microsimulation models described earlier are designed to calculate interactions with credit limitations, the alternative minimum tax, and the corporate minimum tax. Other interactions are modeled by our staff as appropriate.

Putting it all together: A capital gains example

SBX6-10 (Dutton) as amended May 5, 2010, illustrates several techniques we use for estimating revenue. This bill proposed, for taxable years beginning on or after January 1, 2013, and before January 1, 2016, a 50 percent exclusion for capital gains on assets acquired after the effective date and held for more than three years.

We began with the microsimulation model. For each taxpayer in the sample, their tax was first calculated using the actual data on their tax returns. The model then subtracted half of their reported capital gains from their income and recalculated their tax. Because this calculation is done on a sample, the difference between the two calculations for each taxpayer was multiplied by the number of taxpayers they represent. Based on an analysis of data on holding periods in FTB’s Capital Assets sample, this result was reduced by 30 percent to account for the requirement in the proposal that assets be held for at least three years to qualify for the special treatment. Since the most recent available data at the time of the estimate was from the 2007 tax year, the results of the model run were adjusted based on DOF projections of growth in taxes on capital gains. The results were next reduced to reflect the requirement in the proposal that only assets purchased after enactment of the bill would qualify for favorable treatment. Based on an analysis of data on holding periods in our Capital Assets sample, it was estimated that 12 percent of asset sales in 2013, 22 percent of asset sales in 2014, 30 percent of asset sales in 2015, and 37 percent of asset sales in 2016 would qualify. The resulting estimate was a revenue loss of about $2.4 billion for the three affected tax years if we ignored taxpayer behavior.

The estimate for this bill then considered three behavioral reactions by taxpayers. One behavioral adjustment was for the increased willingness of taxpayers to recognize capital gains when tax rates are lower. To make this adjustment we first calculated the proposed tax changes for taxpayers. For example: taxpayers paying a 15 percent federal tax rate and a 9.3 percent state tax rate on their capital gains under then current law would be paying 15 percent federal and 4.65 percent state tax under the proposal. Thus, their total tax would have been reduced by 19 percent (from 24.3 percent to 19.65 percent) and their after tax income increased by 6 percent (from 75.7 percent to 80.35 percent) of the gain. The estimate then applied reaction rates (elasticities) found in the economics literature to these changes in taxes on capital gains for taxpayers of various incomes. The increase in asset sales from this behavioral response reduced the estimated revenue loss by about $250 million over three years. The second behavioral adjustment accounted for taxpayers who would have postponed some asset sales from 2012 into 2013 to take advantage of the lower tax rate. This behavior was estimated to reduce revenues in tax year 2012 by about $15 million and increase revenues in tax year 2013 by about half as much. Similarly, taxpayers would have accelerated some asset sales from 2016 into 2015 to take advantage of the lower tax rate. This behavioral response was estimated to reduce revenues in tax year 2016 by about $100 million and increase revenues in tax year 2015 by about half as much. The third behavioral adjustment is bigger than the second because in the later year more assets will meet the requirement of having been held for at least three years after the passage of the bill (this is an example of vintaging as described above). Adding the behavioral adjustments to the base estimate produced a total revenue loss of $2.2 billion over tax years 2012 to 2016.

Summary

We provide estimates of the revenue effects of proposed changes in tax law. These estimates are used by policymakers in their evaluation of the desirability of the proposed changes. Our estimates compare the revenue expected under proposed policy to revenue expected if there are no changes in tax law. When possible, we generate our estimates from detailed tax return data. We also use information on federal estimates of tax policy impacts and other reputable sources of information on activities or types of taxpayers affected by a tax proposal. Our estimates do include assessments of likely behavioral responses of taxpayers to tax law changes, but assume that those responses will not change the overall size of California’s economy. Our estimates consider institutional features of the tax code such as carryovers and interactions between various tax provisions. We believe that by providing objective estimates of the revenue impacts of tax policy proposals it enables policymakers to make sound tax policy decisions.