The Budget Lab’s Model for Tax Depreciation

Introduction

Businesses are subject to tax rules governing how and when their capital costs can be deducted from revenues to arrive at taxable income. This set of rules, called tax depreciation or cost recovery policy, is an important dimension of how businesses are taxed, impacting the timing of government receipts and effective tax rates on capital investment.

The Budget Lab has developed an open-source model called Cost-Recovery-Simulator that allows users to simulate the budgetary and incentive effects of changes to depreciation policy. At a high level, the model has four main components:

1. Configuration. The model takes as input several user-defined scenario configuration files which specify parameters for tax law, policy take-up, and more.
2. Input data. The model relies on an exogenously specified projection of investment by asset class and industry.
3. Tax calculation. Given a scenario and investment projections, the model calculates gross depreciation deductions and accounts for net operating loss carryforwards when deductions exceed taxable income.
4. Post-processing. The model calculates the government revenue effects of policy changes as well as tax burden impacts by industry, asset class, entity type, and more.

This report describes these features in more detail.

Configuration

Tax-Depreciation is a tool for scenario analysis: it projects depreciation deductions under several scenarios, then compares the results. Scenarios are characterized by the parameters described in this section, the implementation of which can be found in the /config folder.

Tax law

Tax law parameters represent the underlying features of cost recovery schedules for each kind of asset. We characterize tax depreciation policy by four key parameters that vary by asset class and industry as defined in CBO’s CapTax model (our source for these parameter values):

• Recovery period (“L”): the number of years over which property must be depreciated. Also known as “useful life”.
• Acceleration factor (“B”): per the Modified Accelerated Cost Recovery System (MACRS), the geometricdecay rate for the declining balance portion of a cost recovery schedule.¹ MACRS works by offering declining balance depreciation at a rate of B until a straight-line alternative is more generous. When B=1, MACRS generalizes to straight-line depreciation at a rate of 1/L per year.
• Statutory bonus depreciation rate (“bonus”): theadditional amount of depreciation allowed in the first year of investment allowed under section 168(k), commonly known as “bonus” depreciation. Expressed as a fraction of investment.
• Effective section 179 rate (“s179”): the fraction of investment that is eligible for section 179 expensing. While the parameters listed above are statutory, this parameter is calculated by CBO and reflects firm size composition by asset class and industry.

The user must supply a tabular CSV file specifying the value of each of these parameters for all asset class-industry pairs. Each CSV will define policy for a generic year. These CSVs are named descriptively and saved in parameter-specific subfolders.

But tax law can vary over time. Therefore, the user must create a “master” tax law CSV which links these parameter files to years of simulation and legal entity type (either C corporation or pass-through). This file contains columns for two additional tax law parameters which are assumed not to vary by asset class or industry:

• Indexation. This parameter governs whether depreciation deductions are indexed. There are three possible parameter values: “inflation” indexes deductions to CPI; “timevalue” indexes deductions to the projected 10-year Treasury yield; “none” indicates no indexation.
• Tax rate. This is the investment-weighted average effective marginal tax rate on positive business income by legal entity type. For C corporations, this is simply the statutory corporate tax rate. For pass-through entities, we estimate this parameter by calculating the income-weighted marginal rate on partnership income using our tax microsimulation model.

Other assumptions

There are four additional configuration files which allow users to specify assumptions related to tax policy and firm behavior:

• Expensing take-up. Not all firms who are eligible for bonus depreciation or section 179 expensing elect to use these provisions, instead opting for normal depreciation schedules. This file allows users to specify the take-up rates—the fraction of eligible investment claimed as additional first-year deductions under these provisions. Baseline parameter values are based on averages from Kitchen and Knittel (2016).
• Deduction usage. Some firms who accrue depreciation deductions do not have sufficient taxable income to offset. This file allows users to specify the fraction of gross depreciation deductions that cannot be immediately used for this reason, by year and legal entity type. This is a reduced form parameter that captures the joint distribution of gross depreciation deductions and taxable income (excluding depreciation deductions). Our baseline parameter value of 40% is a calibrated value such that, all else equal, our revenue estimates match those of recent JCT revenue estimates of bonus depreciation changes.
• NOL carryfoward usage. Depreciation deductions that are unused (due to insufficient taxable income) can be carried forward to future years. This file allows users to specify the fraction of unused deductions that get used in each subsequent year, by year and legal entity type. This is a reduced form parameter that reflects unmodeled micro-level factors like the evolution of taxable income at the firm level and firm deaths/acquisitions. Baseline parameter values are based on Lim et al. (2018) and Cooper and Knittel (2006).
• Investment shifting. Reforms to cost recovery rules may lead firms to change investment plans—either on a permanent basis or as a temporary, intertemporal shifting of investment from higher-taxed years to lower-taxed years. This file allows users to specify the ratio of total investment under a reform scenario to total investment under the baseline, by year. By definition, this parameter is equal to 1 for all years under the baseline.

Runscripts

The last step in model configuration is creating a “runscript”—a list of user-defined scenarios to run. A runscript is a CSV where rows represent scenarios and columns represent attributes defining those scenarios, corresponding to the parameters described above (tax law, expensing take-up, deduction usage, NOL carryforward usage, and investment shifting). Users must also specify the start and end years for the simulation for each scenario. Examples can be found here.

Input data

Tax depreciation rules vary by asset class, industry, and legal entity type. To support full parameterization of the tax code, Cost-Recovery-Simulator requires a projection of investment across these three dimensions.

First, Cost-Recovery-Simulator reads as input an exogenous projection of investment by asset class. This data is constructed outside of the model using the projections for nonresidential and residential investment in CBO’s economic projections. We then hold the subcomponent shares of investment (i.e., by industry and by asset class) fixed at their averages over the last five years. Data on investment in nonresidential assets by industry and by type are from the BEA’s Detailed Fixed Assets Tables. Since our model occasionally uses finer grain industries, we impute some values using the detailed industry weights in CBO’s CapTax model. Data on investment in residential assets is from the BEA’s Fixed Assets Accounts Table 2.7. These values are adjusted to exclude owner-occupied housing using the data in Fixed Assets Accounts Tables 5.7.

The final step is to impute the breakdown of investment by legal entity type (C corporation versus pass-through business). Since data on the joint distribution of investment in a given asset class by industry by legal form of business is not available, we use the shares of capital stock from CBO’s CapTax as a proxy. These proportions of investment are held fixed from 2021 onward. For historical years, we index the C corporation share of investment to changes in the C corporation share of business receipts. The C corporation proportion of business receipts is computed as the ratio of business receipts of active corporations, other than Forms 1120S, 1120-REIT, and 1120-RIC² to the sum of business receipts of all active corporations,³ nonfarm sole proprietorships, and partnerships. Data for business receipts by industry by legal form of organization are from SOI Tax Stats. Our model, however, uses finer-grained industry definitions than does the SOI Tax Stats data. Since the industry definitions for the all corporations data best matches our industry definitions, we compute the values for missing industries in the all corporations data using their respective shares of historical investment, and then use the now complete industry shares of all corporations business receipts to compute the values for missing industries in the data for other legal forms.³ Once matched to the correct industry, we use the data on business receipts by industry by legal form to compute the C corporation share of business receipts in each industry. Finally, we compute the C corporation share of investment by industry based on the changes in the C corporation share of business receipts by industry.

Other exogenous economic variables used in Cost-Recovery-Simulator, such as inflation and interest rates, come from the Budget Lab’s Macro-Projections model, which is largely an aggregation of CBO’s long-term economic projections.

Tax calculation

For a given user-defined scenario, the model generates cost recovery schedules for each asset class-industry-legal entity type combination based on the configuration files. MACRS schedules are assumed to operate under the half-year convention. Schedules are adjusted for bonus/section 179 expensing take-up rates and indexation where specified.

These schedules are combined with the investment projections to produce a projection of gross depreciation deductions by investment year, deduction year, asset class, industry, and entity type. These projections are then aggregated to the deduction year level and the usage assumptions are applied. This step generates two buckets of deductions: those that are used immediately, and those that represent NOLs available to be carried forward to future years. The latter bucket is transformed into future-year deductions model based on the assumed NOL carryforward usage rates. The result is a projection of total deductions by year.

Deductions are converted to tax liability by applying the tax rates specified for the scenario. The final step is to convert calendar-year liability to fiscal year receipts based on historical shares of revenues by quarter.

Post-processing

For each non-baseline scenario, Cost-Recovery-Simulator calculates revenue estimates—the change in receipts due to tax depreciation policy changes. The model also generates a series of files summarizing the “cost recovery ratio”, a measure of how burdensome the tax code treats investment through depreciation policy. For a given investment, the cost recovery ratio is calculated as the present value of depreciation deductions divided by the initial investment outlay. The model calculates cost recovery ratios by asset class, major industry, and entity type, and assumes a discount rate of the projected 10-year Treasury rate plus 2 percentage points.