An Analysis of Nonprofit Organization Federal Income Tax Avoidance Using Publicly Available Income Tax Return Data

This study uses publicly available U.S. federal income tax return data to examine tax avoidance by Internal Revenue Code (I.R.C.) § 501(c)(3) nonprofit organizations. Nonprofits are subject to the Unrelated Business Income Tax (UBIT) under I.R.C. § 511, which imposes income taxes on profits from activities that are unrelated to a nonprofit's tax exempt purpose. Over 26 percent of nonprofits with assets of $1 million or more conduct taxable business activities that are subject to the UBIT. The UBIT is the newest and smallest of the four federal income taxes in the U.S., and is one of only two entity-level business income taxes in the U.S.¹ Although I.R.C. sections 6103 and 6104 have historically prevented public disclosure of federal income tax returns, the Pension Protection Act of 2006 modified I.R.C. § 6104 by requiring nonprofit organizations, beginning with tax year 2007, to disclose their previously confidential income tax returns, the IRS form 990-T. This marks the first time in the history of the U.S. federal income tax that one of the four previously confidential federal income tax returns have been subject to public disclosure (Lenter, Slemrod, and Shackelford 2003).² Unlike the corporate and individual income taxes, revenue generation was never envisioned as a goal of the UBIT. Rather, the success of the UBIT can be best measured by how well it discourages nonprofits' pursuit of ancillary revenues by reducing the financial returns to unrelated activities. Because tax avoidance diminishes the UBIT's ability to accomplish its regulatory goals it is important to understand how nonprofits avoid the UBIT and what factors are associated with UBIT avoidance.

I focus my analysis on 501(c)(3) nonprofits which, in addition to being tax exempt on their income (other than UBIT), can also receive tax deductible contributions from donors.³ Nonprofits include a wide variety of well known organizations such as the American Accounting Association, the American Economic Association, Harvard and Stanford Universities, the Professional Golfers’ Association, the National Collegiate Athletic Association, Wikipedia, the Smithsonian Institution, the San Diego Zoo, Boy Scouts of America, the National Rifle Association, and the American Association of Retired Persons.

UBIT avoidance is believed to be widespread. Aggregate IRS data show that over the past three decades fewer than 40 percent of nonprofits with taxable activities reported net profits on those activities, and more than 40 percent report taxable losses. Given that the only purpose of unrelated activities is to generate profits that can be used to subsidize nonprofits’ charitable missions the lack of reported taxable profits is suspected to be the result of tax avoidance (Cordes and Weisbrod 1998).

Yetman (2001) and Hofmann (2007) used small samples of voluntarily supplied nonprofit income tax returns from the mid-1990s and documented UBIT avoidance by expense shifting for some types of nonprofits. Both of these studies used data from time periods before public disclosure of UBIT data and thus are subject to selection bias. Furthermore, prior research has not examined a robust set of factors that could be associated with UBIT avoidance. The goal of my analysis is to fill these gaps in prior research using publicly available UBIT data.

My empirical tests are designed to determine the factors that encourage or deter UBIT avoidance in the post public disclosure period. I use regression analysis where my dependent variables are two plausible measures of UBIT avoidance and provide separate results for each of the three primary nonprofit categories (educational, medical, and charitable). My first dependent variable is the amount of reported taxable income which I presume that, over time in the cross section, will be lower for nonprofits that engage in more aggressive UBIT avoidance. My second dependent variable is equal to 1 if the nonprofit reported taxable losses, and 0 otherwise which I presume is an indicator of UBIT avoidance.

My first hypothesis tests the extent to which taxable loss activities are used to avoid the UBIT. Prior IRS audits suggest that one way nonprofits could avoid the UBIT is to intentionally create and operate taxable loss activities, and use those losses to offset the profits of other taxable activities (IRS 2013; Reichert 2013). I find that nonprofits that operated more than one taxable activity (and thus could be using taxable loss activities to offset the profits of other taxable activities) report lower amounts of taxable income and are more likely to report taxable losses. In terms of magnitude, I find that on average educational and charitable nonprofits that operate multiple taxable activities reported over 100 percent less taxable income than nonprofits that operate only a single taxable activity, suggesting that prior to the Tax Cuts and Jobs Act (TCJA) of 2018 nonprofits were using taxable loss activities to offset the profits of other taxable activities. This is the first empirical evidence that nonprofits were using taxable loss activities to avoid the UBIT.

My second hypothesis tests the deterring effect of the silo provisions of the TCJA of 2018 on UBIT avoidance. The silo provisions were specifically targeted at the use of loss activities discussed above, and beginning in tax year 2018 disallow taxable losses from a taxable activity to offset the taxable profits of other taxable activities. I find that after the silo provisions became effective in 2018 the amount of taxable income reported by nonprofits that operated more than one taxable activity increased by more than the taxable income reported by nonprofits that operated a single taxable activity, suggesting that the silo provisions were effective in reducing UBIT avoidance. In terms of magnitude, I find that nonprofits that operated multiple taxable activities reported average taxable losses of approximately $200 thousand in 2017, but after the TCJA reported average taxable profits of roughly $200 thousand in 2018, a swing of almost $400 thousand in reported taxable income in a single year. This is the first empirical evidence that the silo provisions of the TCJA were effective in mitigating UBIT avoidance.

My third hypothesis tests the governing effect of a nonprofit's organizational form on UBIT avoidance (Petrovits and Yetman 2023). Nonprofits must choose one of three organizational forms: the charitable corporation, the charitable trust, and the unincorporated association. Officers and trustees of charitable trusts are subject to enhanced legal obligations and personal liability as compared to officers and directors of corporations or associations. I find that medical and charitable nonprofits organized as charitable trusts report higher amounts of taxable income and are less likely to report taxable losses, suggesting that the enhanced governance features of the charitable trust form mitigate UBIT avoidance. This is the first evidence that organizational form plays a role in mitigating UBIT avoidance. However, given that less than 2 percent of nonprofits are organized as charitable trusts the mitigating effect of organizational form is limited to very few nonprofits.

My fourth and final hypothesis tests the effect of public donors on a nonprofit's decision to avoid the UBIT. Public donors are often considered to be effective external governing agents over nonprofit behaviors (Khumawala and Shroff 2023; Petrovits and Yetman 2023). I find that educational and medical nonprofits that receive higher proportions of their total donations from the public report higher amounts of taxable income and are less likely to report taxable losses. This is the first evidence that public donors play a governance role in mitigating UBIT avoidance.

Although not formal hypotheses as they have been examined by prior research, I also examine the effects of common cost complementarities and income tax rates on UBIT avoidance.

Consistent with Yetman (2003), I find that the presence of common cost complementarities between taxable and tax exempt activities increases nonprofits' ability to avoid the UBIT. Consistent with Omer and Yetman (2007) I find that higher income tax rates encourage UBIT avoidance.

I.R.C. § 513(a) defines a taxable activity as “any trade or business the conduct of which is not substantially related (aside from the need of such organization for income or funds or the use it makes of the profits derived) to the exercise or performance by such organization of its charitable, educational, or other purpose or function constituting the basis for its exemption.” For example, consider a nonprofit aquarium the operates a gift shop to generate extra funds to support the exempt mission. The gift shop sells sea otter dolls and coffee cups. Sea otter doll sales are not taxable if there is any information attached to the doll informing the purchaser why sea otters should be protected, or if a live sea otter was floating somewhere in a tank in the aquarium. However, the revenues from the sale of coffee cups are taxable as coffee cups are not related to the tax exempt mission of informing the public of the need to protect the oceans (IRS Private Letter Ruling 201429029). Not all unrelated activities are taxable as there are a variety of exclusions. The most common exclusion is for passive income from interest, dividends, and capital gains. Some exclusions are very narrow, such as the exclusion for profits earned from games of chance (gambling) in the state of North Dakota (no other state has this special legislative dispensation).

Congress had several motivations for enacting the UBIT including preventing unfair competition, tax base protection, preventing mission drift, and ensuring economic efficiency (Kaplan 1980; Hansmann 1989; Sansing 1998; Weisbrod 1998; Stone 2005). Absent the UBIT nonprofits could operate commercial businesses completely tax free under the umbrella of their tax exemption (Kaplan 1980; Stone 2005). For example, prior to the UBIT New York University owned and operated several businesses under its tax-exempt umbrella including Mueller Macaroni Company (at the time the largest selling brand of pasta in the U.S.), Ramsey Corporation (a manufacturer of automobile piston rings), the American Limoges China Company, and the Howes Leather Company (Kaplan 1980; Stone 2005).

The most significant change to the UBIT since its enactment in 1950 was the addition of I.R.C. § 512(a)(6) in the TCJA of 2018. Prior to 2018 Treasury Regulation 1.512(a)-1(a) allowed nonprofits to aggregate profits and losses across all of their taxable activities such that losses in some taxable activities could offset profits in other taxable activities. As previously discussed IRS audits uncovered that this rule was possibly being exploited by nonprofits that intentionally created taxable loss activities and used those losses to offset the profits of other taxable activities. I.R.C. § 512(a)(6) was specifically targeted to mitigate this type of UBIT avoidance by requiring nonprofits to place each separate taxable activity into its own “silo” such that losses from one activity can no longer offset the profits of another activity. Losses from a taxable activity may be carried forward and used to offset future profits of the same activity that created the losses.

The UBIT had two distinct public disclosure events. The first was when UBIT returns, the form 990-T, became subject to public disclosure in 2007. The second was in 2008 when taxable income started to be disclosed on nonprofits’ publicly available financial statements, the form 990. There are two ways to reliably get copies of a nonprofit’s 990-T, and both involve a written request. First, nonprofits must mail copies of their the most recent three years of 990-Ts (not including years prior to 2007) to anyone upon written request, but are entitled to compensation for any copying and mailing costs. Second, the IRS will supply copies of form 990-T for any year it has available (not including years prior to 2007) in response to a properly filed IRS form 4506-A and the receipt of a fee based on the number of pages in total. The IRS does not maintain a publicly available electronic database of 990-Ts.

In order to make some limited UBIT information more easily available to the public the IRS redesigned nonprofits’ publicly available financial statement form 990 in 2008 to include (at the top of the first page) the amount of taxable income as reported on the organizations' federal tax return 990-Ts. The form 990 is nonprofits' primary publicly available financial statement and is available from a variety of sources including many nonprofit websites as well as IRS public use databases. Prior to 2008, 990s disclosed the amounts of taxable revenues nonprofits earned, but did not disclose taxable income. The taxable income data I use for this analysis are from the form 990, which in turn is taken directly from the underlying 990-Ts.

Prior research has often discussed but never measured the effects of public disclosure on U.S. federal income tax avoidance (Lenter et al. 2003). A few prior studies have used non-U.S. data to examine individual and corporate taxpayer responses to public disclosure (Hasegawa, Hoopes, Ishida, and Slemrod 2013; Bø, Slemrod, and Thoresen 2015; Hoopes, Robinson, and Slemrod 2018). In general, those studies find little evidence that public disclosure increased reported amounts of taxable income or the percentage of firms reporting positive taxable profits.

Although UBIT information became publicly available in 2007 and 2008 the lack of nonprofit level data prior to 2007 prevents a nonprofit level comparison of reported taxable income across pre and post disclosure periods. However, the IRS Statistics of Income (SOI) produces an annual report that contains population average taxable income data for nonprofits over time.⁴Figure 1 shows the population average distribution of unrelated business taxable income for all nonprofits that engaged in taxable activities for the 20-year period beginning 1991 (when the IRS began to collect the data) up to 2010 (three years after disclosure). Fewer than 40 percent of nonprofits (the solid line) have ever reported positive profits on their taxable activities over the sample period. On average roughly 45 percent of nonprofits (dotted line) reported taxable losses over the sample period.

The vertical line in Figure 1 at 2006 denotes the period after which UBIT information first became publicly available. Figure 1 reveals no obvious effect of public disclosure on nonprofits' reported taxable income distribution. The percentage of observations reporting positive taxable income appears to fall slightly after disclosure, the exact opposite of what one would expect if disclosure mitigated avoidance. Because these data are aggregated averages and are not nonprofit level it is not possible for me to control for other factors that could have affected reported taxable income and thus Figure 1 does not provide direct evidence of the effect of UBIT disclosure on avoidance. Nonetheless, Figure 1 shows that public disclosure of UBIT information had no obvious effect on the aggregate distribution of reported nonprofit taxable income.

As previously discussed prior research on UBIT avoidance has focused on documenting expense shifting in periods prior to public disclosure. The first study to consider the effects of UBIT expense shifting was Sansing (1998) who analytically shows that an effective UBIT will increase economic efficiency by preventing nonprofits from producing excessive quantities of taxable output, but that expense shifting will cause excessive production reducing efficiency. The first study to examine expense shifting as a UBIT avoidance technique was Cordes and Weisbrod (1998) who did not have access to 990-T tax return data but instead used taxable revenue data from nonprofits’ 990s and find that the set of nonprofits that earn taxable revenues report roughly the same amount of total labor costs as the set of nonprofits that do not earn taxable revenues. However, when reporting labor costs on the form 990 the set of nonprofits that earn taxable revenues include a larger proportion those labor costs as cost of goods sold, which reduces net profits on inventory sales. If nonprofits with inventory sales are more likely to earn taxable revenues (an untested but plausible assumption) this result suggests that nonprofits are attempting to reduce their taxable income by shifting labor costs into inventory.

Yetman (2001) and Hofmann (2007) are the first studies to examine UBIT avoidance via expenses shifting. Because UBIT returns were not subject to public disclosure at the time both Yetman (2001) and Hofmann (2007) obtained their samples by contacting nonprofits via mail and asking them to voluntarily supply the most recent three years of their IRS form 990-T. Yetman (2001) contacted 2,316 public charities and received 1,824 tax returns from 703 unique organizations across the years 1995 to 1997. Hofmann (2007) contacted 1,152 membership associations and received 399 tax returns from 126 unique organizations across the years 1994 to 1997. Using a taxable-expense prediction model both studies found that organizations shift expenses from their tax-exempt to their taxable activities. One notable exception was that Yetman (2001) failed to find shifting by any type of public charity other than medical and educational organizations. Using the same data as Yetman (2001), Omer and Yetman (2007) show that expense shifting is higher for nonprofits located in higher tax rate states.

Ideally my dependent variable would be a direct measure of UBIT avoidance, although directly measuring tax avoidance is not possible even with access to actual income tax return data. As previously discussed Yetman (2001) and Hofmann (2007) identified UBIT avoidance by deriving estimates of expense shifting from tax-exempt to taxable activities using a regression-based cost allocation model. Their primary motivation for using this method is that their samples were voluntarily supplied by the nonprofits themselves (their sample periods pre-date required public disclosure) and thus the reported amounts of taxable income were subject to selection bias. Both studies avoided using reported taxable income in favor of their regression-based models in an attempt to extract a measure of pre-avoided taxable income that would plausibly be free of selection bias. Although those estimation models are plausible, they are subject to several estimation issues as discussed by Hofmann (2007).

In contrast to those prior studies, my sample is not voluntarily provided by the nonprofits but rather is taken directly from the IRS database in the post-public disclosure period and is thus free of nonprofit chosen selection bias. With access to a larger sample that is free of nonprofit chosen selection bias I am able to use the reported amounts of taxable income as measures of UBIT avoidance and thus avoid the estimation issues in Yetman (2001) and Hofmann (2007). Federal disclosure law has prohibited any prior study of U.S. federal tax avoidance using publicly available data. However, there is a small body of prior research that used non-U.S. income tax return data furnished by the relevant taxing authorities. All of those studies inferred tax avoidance using the reported amounts of taxable income presuming that, all else equal, lower (or negative) reported taxable income over time and in the cross section is correlated with more aggressive tax avoidance (Hasegawa et al. 2013; Bø et al. 2015; Hoopes et al. 2018). Consistent with these studies I use two measures of UBIT avoidance. My first measure is the amount of reported taxable income and my second measure is an indicator equal to 1 if the nonprofit reported taxable losses, and 0 otherwise.

In order to interpret the amounts of reported taxable income as an indicator of tax avoidance it is important to control for other factors that influence the amount of taxable income reported both inside and outside the nonprofit. For example, the amount of taxable income can be a function of the amount of taxable revenues earned, the relative profitability of a nonprofit's tax-exempt activities, size, industry, operating efficiency, as well as external market factors such as competition and overall economic conditions. It is important to include controls for these factors so that the reported amounts of taxable income are a reliable indicator of UBIT avoidance.

My analysis sample comes from the public use IRS form 990 nonprofit level database from 2008 (the first year when taxable income was included on the form 990) to 2019 (the most recent year available).⁵ I removed observations that did not engage in a taxable activity (zero taxable revenues and zero taxable expenses) as well as observations with negative values for accounts that should properly be positive (i.e., negative assets or negative total revenues), or are located in a foreign jurisdiction, leaving me with a total analysis sample of 47,979 observations. It is important to note that this public use nonprofit level database does not include the entire population of nonprofits that earned taxable revenues. Rather, the IRS SOI division provides a stratified random sample that captures an average of 13 percent of the population of nonprofits that earned taxable revenues in a year, but due to size-weighted sampling techniques captures an average of 73 percent of total taxable revenues earned in each year. The IRS sampling technique is based on asset size, and not the amount of UBI earned, making the samples less subject to bias in reported UBI.⁶ The goal of the publicly available IRS SOI samples is to provide samples that achieve two goals; to capture the majority of economic activity by overweighting larger organizations, and to provide reliable statistical analyses.

Variable definitions are in  Appendix A. I estimate Model (1) using OLS when the dependent variable is the amount of reported taxable income, and use a logistic model when my dependent variable is an indicator of reported taxable losses. The data are an unbalanced panel with nonprofit i in state s for year t. For both my OLS and logistic models I cluster the standard errors at the nonprofit level to control for repeated observations and use robust standard errors.⁷ I remove possibly influential observations that have a studentized residual greater than the 1st and 99th percentiles (Belsley, Kuh, and Welsch 1980). Next I discuss my data and test variables followed by a discussion of my controls.

My first hypothesis is that, prior to the TCJA of 2018, nonprofits avoided the UBIT by intentionally creating taxable loss activities and using those losses to offset the profits of profitable taxable activities (IRS 2013; Reichert 2013). Prior to the TCJA of 2018 nonprofits were allowed to aggregate the profits and losses of all their taxable activities, and pay income taxes on the net taxable income across all taxable activities. The ideal variable to test this hypothesis would be a measure the profits and losses of each of a nonprofit's taxable activities, although neither the 990 nor the 990-T reports taxable income separately for each taxable activity but rather aggregates taxable income across all activities to produce a single number of net taxable income. However, Part VIII of the form 990 breaks out taxable revenues into ten categories allowing me to identify which nonprofits are earning taxable revenues from more than one taxable activity. Only nonprofits with multiple taxable activities could possibly be using taxable loss activities to offset profits in other taxable activities. The variable Multiple Taxable Activities is an indicator equal to 1 if the nonprofit reported earning taxable revenues from more than one of these ten taxable revenue sources and 0 otherwise.⁸ If nonprofits were using taxable loss activities to offset the profits of other profitable taxable activities I expect to find a negative coefficient on Multiple Taxable Activities prior to the TCJA of 2018.

My second hypothesis is that the silo provisions of the TCJA of 2018 reduced the use of taxable loss activities to avoid the UBIT. The silo provisions became effective for tax years 2018 and were specifically designed to prevent the use of taxable losses from some taxable activities to offset the taxable profits of other taxable activities. I.R.C. § 512(a)(7) requires that each taxable activity is to be placed into its own “silo” and income taxes paid for each taxable activity separately. To empirically test the extent to which the silo provisions were effective in mitigating UBIT avoidance I use a difference-in-differences approach which exploits the fact the silo provisions should have no effect on nonprofits that operate only a single taxable activity and should have had no effect on the use of loss activities prior to 2018. The variable TCJA is an indicator equal to 1 for years after 2017 and 0 otherwise.

As discussed above, the main effect Multiple Taxable Activities, will capture the effect of operating multiple taxable activities on reported taxable income (and the propensity to report taxable losses) prior to the TCJA. The interaction of the TCJA indictor with Multiple Taxable Activities will capture the effect of the TCJA on the difference in reported taxable income (or the propensity to report taxable losses) between nonprofits with and without multiple taxable activities. If the silo provisions were effective in reducing UBIT avoidance I expect to find a positive coefficient on the interaction of the TCJA indictor with Multiple Taxable Activities.

The main effect TCJA will capture the incremental effect on reported taxable income due to the TCJA for the set of nonprofits that operate a single taxable activity. Because the silo provisions of the TCJA do not apply to nonprofits that operate only a single taxable activity I would expect to find, all else equal, an insignificant coefficient estimate for TCJA. However, there was a second provision added by the TCJA that could affect the estimated coefficient for the main effect of TCJA in Model (1) even for the set of nonprofits that operate a single taxable activity. The TCJA also included I.R.C. § 512(a)(7) which imposed the UBIT on transportation benefits paid by the nonprofit to its employees (Boris and Cordes 2019). These benefits include free parking or public transportation vouchers. This controversial provision (which was repealed for years after 2019) imposed the UBIT on an expense rather than on income. Because it was an expense item it is not included in the ten measures of nonprofit revenues, and I would not code it as an additional source of taxable revenues. Furthermore, there were no deductions allowed and as a result nonprofits that provided these benefits could report higher taxable income for the years 2018 and 2019. This provision could have the effect of producing a positive coefficient for my TCJA main effect variable. Thus, I am unable to make a prediction for the expected value of the TCJA main effect.

My third hypothesis examines the effect of a nonprofit's organizational form on UBIT avoidance. In order to qualify for tax exempt status under I.R.C. § 501(c)(3) a nonprofit must choose one of three organizational forms (Fremont-Smith 2004; Mehlman and Watts 2007). The most common form is the charitable corporation. In my sample charitable corporations make up 98.2 percent of the observations, and report an average amount of taxable revenues of $1,545,685. The next most common organizational form is the charitable trust, which in my sample makes up 1.2 percent of the observations, but reports over twice as much average taxable revenues of $3,361,780. The least common organizational form is the unincorporated association. In my sample unincorporated associations make up only 0.6 percent of the sample, and report a relatively smaller average amount of taxable revenues of $805,712.

Managers and directors of nonprofit corporations are subject to the same business judgment rule as are managers and directors of for-profit corporations which requires that decisions be made in good faith with diligence, care, and attention. The business judgment rule focuses on the manner in which a decision is made, rather than on the outcome of the decision itself. Officers and directors are not liable for poor outcomes as long as their decisions were made in good faith. Furthermore, in charitable corporations the corporation itself (and not its officers or directors) is considered to be the contracting party for business arrangements, limiting the personal liability of the officers and directors. Officers and directors of unincorporated associations are treated in a similar manner as they are in charitable corporations.

The business judgment rule does not apply to charitable trusts. Instead, officers and trustees of charitable trusts are subject to the significantly higher standards of the duties of care and loyalty (Fremont-Smith 2004; Mehlman and Watts 2007). In a charitable trust the officers and trustees are considered to be the contracting parties for business arrangements (and not the trust itself), exposing them to personal liability for business decisions.

All these factors suggest that the additional duties and liabilities imposed on officers and trustees of charitable trusts will cause them to be more risk averse than managers of either charitable corporations or associations and thus less aggressive in avoiding the UBIT. However, a second issue with trusts is that they have on-average higher tax rates than do corporations. Trusts must use individual rather than corporate tax rates and reach the highest tax rate at relatively low levels of taxable income (i.e., $12,500 in 2018). To the extent that higher tax rates encourage tax avoidance I would expect to find that charitable trusts are more aggressive in avoiding the UBIT than are nonprofits organized as corporations or associations. Thus it is an empirical question as to which factor will have a stronger influence on UBIT avoidance by charitable trusts. In Model (1) the variable Charitable Trust is equal to 1 if the nonprofit is organized as a charitable trust and 0 otherwise. If the additional duties and liabilities imposed on officers and trustees of charitable trusts cause them to be more risk averse than managers of either charitable corporations or associations and thus less aggressive in avoiding the UBIT I will find a positive coefficient for Charitable Trust. If higher trust tax rates cause trusts to be more aggressive in avoiding the UBIT I will find a negative coefficient for Charitable Trust.

My fourth and final hypothesis examines the effect of public donors on a nonprofit's decision to avoid the UBIT. Public donors (individuals and corporations) are often considered to be an influential group of external stakeholders. Khumawala and Shroff (2023) and Petrovits and Yetman (2023) provide excellent discussions of the role of donors as users of nonprofit financial information and as governing mechanisms. Cordes and Weisbrod (1998) hypothesized that that public donors could have a natural aversion to nonprofits' taxable activities whereas other types of donors (corporations and governments) might not be as averse. They find that nonprofits are less likely to engage in taxable activities if they receive larger proportions of their donations from the public, conditional on a total assets size control. M. Yetman and R. Yetman (2003) find that public donors give less to nonprofits that earn larger amounts of taxable revenues. Although neither of these papers examined the effects of public donors on UBIT avoidance, it is plausible to presume that in addition to disfavoring taxable activities in general public donors also disfavor low (or negative) amounts of reported taxable income as an indicator of UBIT avoidance. Based on this prior research, I chose to measure donor aversion to taxable activities as the ratio of public donations to total donations, and then to include total assets as a size control. The variable Public Donations is equal to the ratio of public donations from Part VIII line 1f to total donations from Part VIII line 1h. If public donors react negatively to lower reported amounts of taxable income (or negative taxable income) I expect to find a positive coefficient on Public Donations.

Omer and Yetman (2007) found that nonprofits shift more expenses from their tax-exempt to their taxable activities in states with higher UBIT rates. I attempt to duplicate their results here using different measures of UBIT avoidance. One persistent problem facing tax research is the proper identification of a tax rate. Given that federal tax rates vary little over time and also vary little in the cross-section, researchers frequently exploit the variation in state tax rates to gain econometric identification (Feenberg 1987). Following Feenberg (1987) and Omer and Yetman (2007), the variable Tax Rate in Model (1) is an indicator equal to 1 if the state UBIT tax rate is greater than its full sample median value of 7.1 percent, and 0 otherwise. If a state UBIT rate is based on a schedule (rather than a flat rate) I used the average schedule rate for that state. I use a dichotomous measure of tax rates to avoid issues that arise from tax rate schedules that are proportionate to the amounts of taxable net income. If I attempted to use the applicable tax rate schedules I would first need to estimate pre-avoidance taxable income and then apply that taxable income to the tax rate schedules. Given that I have no direct measure of UBIT avoidance I cannot estimate pre-avoided taxable income.

Cordes and Weisbrod (1998) and Yetman (2003) find that the ability to shift expenses depends on how functionally integrated a taxable activity is with a nonprofit's tax exempt activities. Some taxable activities rely almost exclusively on existing labor and facilities already in place for the tax exempt mission, whereas others use little existing labor and facilities. It is far easier to identify a pool of expenses that could be shifted to a taxable activity if the taxable activity is heavily integrated with the nonprofit’s tax exempt activities. I attempt to duplicate their results here using different measures of UBIT avoidance. Of the ten categories of taxable revenues disclosed in Part VIII of the 990 taxable revenues from programs by definition uses existing labor and facilities. According to the instructions to the form 990, “Program revenues are primarily those that form the basis of an organization’s exemption from tax.” Because taxable program outputs use the same facilities and labor that are used to produce the nonprofit’s tax exempt outputs it is easier to identify a pool of common expenses and shift a disproportionately large amount of them to the taxable activity. The variable Common Costs is equal to the ratio of taxable revenues from programs to total taxable revenues.

I include a variety of control variables that could be associated with variations in UBIT avoidance and also correlated with my test variables. The variable Year is a time trend (0 in 2008, 1 in 2009, and so on) and measures the extent to which reported taxable income (or the propensity to report taxable losses) changes over time. Overall Profit Ratio and is the ratio of total nonprofit net income to total nonprofit revenues. It is possible that variations of a nonprofit’s taxable income over time and in the cross section is an artifact of overall nonprofit profitability. Taxable Revenues is the amount of reported taxable revenues earned. Nonprofits with relatively larger amounts of taxable revenues could have more incentive and opportunity to avoid the UBIT, or alternatively have a difficult time avoiding high levels of taxable income. Accounting Fees are the amount of fees paid to external accounting firms divided by the nonprofit's total assets. Prior research shows that the primary determinant of accounting fees is organizational size measured as total assets (Simunic 1980; Hay, Knechel, and Wong 2006). Accounting firm advice could increase UBIT avoidance if the firm assists the nonprofit in aggressive tax planning, or reduce avoidance if the firm assists the nonprofit in proper tax reporting. The Program Ratio is the ratio of mission-related program expenses to total expenses, and is a common measure of nonprofit operational efficiency (Weisbrod and Dominguez 1986; Tinkelman 2004). Although I include this variable only as a control, it is not universally agreed to be a measure of nonprofit efficiency (Khumawala and Shroff 2023; Lecy, Searing, and Li 2023). Competition is a Herfindahl style variable that measures the relative amount of taxable activity competition a nonprofit faces in its location by industry and year. Competition over time across locations can affect a taxable activity’s relative profitability. I define location by breaking up the U.S. into 100 distinct geographic regions using U.S. postal zip codes. I define industry according to the IRS’s National Taxonomy of Exempt Entities (NTEE) industry classification which categorizes all nonprofits into 25 industries. Assets is a size control, and SGDP is the one year change in state level gross domestic product that controls for changes in the underlying economy across states and time, which could in turn affect a taxable activity’s profitability. Finally, I include industry indicators for the NTEE 25 industries as discussed above.

Descriptive statistics for my variables are in Table 1 whereas their definitions are in  Appendix A. Statistics in Table 1 show that educational nonprofits report average losses on their taxable activities and that 51 percent of the observations report taxable losses. Medical nonprofits report a small amount of average profits on their taxable activities whereas 38 percent of the observations report taxable losses. Finally, charitable nonprofits also report a small amount of average profits on their taxable activities whereas 45 percent of the observations report taxable losses. Over 40 percent of educational nonprofits earn taxable revenues from multiple sources, whereas only 27 percent of charitable nonprofits earn taxable revenues from multiple sources. Nearly half of medical nonprofits' taxable revenues are earned from taxable program activities where common costs are most likely to be present. Almost 3 percent of charitable nonprofits are organized as trusts, whereas only 0.3 percent of medical nonprofits are organized as trusts. Charitable nonprofits receive the largest proportion of donations from the public (32 percent), whereas medical nonprofits receive the smallest (3 percent).

Table 2 reports the results of my OLS model where the dependent variable is a continuous measure of logged reported taxable income. Table 3 reports the results of my Logistic model where the dependent variable is equal to 1 if the nonprofit reported a taxable loss, and 0 otherwise. Results in Table 2 for Multiple Taxable Activities, which captures the effects of operating more than one taxable activity prior to the TCJA, is significantly negative for educational and charitable nonprofits, but not for medical nonprofits. In terms of magnitude the set of educational nonprofits with multiple taxable activities prior to the TCJA reported taxable income that was 181 percent lower than the set of educational nonprofits with only a single taxable activity. Charitable nonprofits with multiple taxable activities prior to the TCJA reported taxable income that was 116 percent lower than the set of charitable nonprofits with only a single taxable activity. My Logistic results in Table 3 show a statistically negative coefficient for Multiple Taxable Activities for all three types of nonprofits. These results suggest that nonprofits used taxable loss activities to offset the profits of other taxable activities prior to the TCJA.

My null result for medical nonprofits in Table 2 could be due to possible misclassification of some nonprofits that are actually operating multiple taxable activities as only operating a single taxable activity. This could result from nonprofits that report earning only a single type of taxable revenues (and thus I code as only operating a single taxable activity) actually operating multiple taxable activities all of which produce a single source of revenues. Analysis of my data reveals that this possibility is particularly acute for medical nonprofits as they earn the majority of their taxable revenues from a single source (programs).

Figure 2 displays the essence of my difference in differences analysis by showing the amounts of taxable revenues for my sample partitioned into observations with multiple and single taxable activities. In the two years prior to the TCJA (2016 and 2017) nonprofits that operated multiple taxable activities reported average taxable losses of approximately $200 thousand, rising to average profits of roughly $200 thousand in each of the two years after the silo provisions became effective (2018 and 2019). In contrast reported taxable income for the set of nonprofits that operated a single taxable activity remains fairly constant before and after the TCJA. This dramatic swing of an average of $400 thousand in reported taxable income for the set of nonprofits that operated multiple taxable activities suggests that the silo provisions mitigated UBIT avoidance.

Results in Table 2 for the interaction of TCJA with Multiple Taxable Activities show that reported taxable income for the set of educational nonprofits that operated multiple taxable activities is an average of 429 percent higher than for the set of nonprofits that operated a single taxable activity after the TCJA. Taxable income for the set of medical nonprofits that operated multiple taxable activities is 318 percent higher, and 309 percent higher for charitable nonprofits. The relative magnitudes of these effects are large, consistent with the results in Figure 2. My Logistic results in Table 3 are consistent in terms of statistical significance. These results suggest that the silo provisions of the TCJA were effective in significantly (statistically and economically) increasing reported taxable income and reducing the frequency of reported taxable losses, which in turn suggests that the silo provisions were effective in reducing UBIT avoidance by the use of loss activities.

My results for the main effect TCJA, which captures the effect of the TCJA for the set of nonprofits I code as operating only a single taxable activity (because they reported earning their taxable revenues from only one of the ten sources in Part VIII of the 990) are mixed. I find null results for both educational and charitable nonprofits, but a statistically positive coefficient for medical nonprofits. My Logistic results in Table 3 report a statistically positive coefficient for all three types of nonprofits. The most plausible explanation for these positive coefficients is the taxation of transportation benefits discussed earlier which would have raised reported taxable income after the TCJA for all nonprofits that provided transportation benefits to their employees, including nonprofits that operated only a single taxable activity.

Results in Tables 2 and 3 show that educational and charitable nonprofits organized as charitable trusts report higher taxable income and are less likely to report taxable losses, consistent with the additional fiduciary obligations and exposure to personal liability faced by officers and trustees of charitable trusts mitigating UBIT avoidance. In terms of magnitude results in Table 2 show that educational nonprofits organized as trusts reported 577 percent higher taxable income than educational nonprofits organized as either corporations or associations. Charitable nonprofits organized as trusts reported 281 percent higher taxable income than charitable nonprofits organized as either corporations or associations. I do not find that medical nonprofits organized as charitable trusts report more taxable income or are less likely to report taxable losses. One plausible explanation for these null results is that there are so very few medical nonprofits organized as charitable trusts causing a very small size effect in my models. In my sample there are only seven medical nonprofits organized as charitable trusts, which is quite likely too few for my regression models to identify an effect.

Results in Tables 2 and 3 show that reported taxable income is higher and the probability of reporting taxable losses is lower for both educational and medical nonprofits when the organization receives a higher proportion of its total donations from the public. These results are consistent with public donors playing a governing role that mitigates UBIT avoidance. However, I do not find that charitable nonprofits taxable income responds to donations in either Table 2 or 3. One plausible explanation for this null result is that all charitable nonprofits are concerned about upsetting their donors, regardless of where the donations come from. Alternative definition of donations (total donations as a percent of total revenues, governmental grants as a percent of total donations) also produced null results in my models.

Results in Table 2 for my OLS results show that both educational and medical nonprofits report lower taxable income in higher tax rate states, consistent with nonprofits responding to higher tax rates by avoiding the UBIT and with the findings of Omer and Yetman (2007). I do not find that charitable nonprofits respond to higher tax rates. My Logistic results in Table 3 are consistent with those in Table 2. As previously discussed models of tax avoidance show that the sensitivity of tax avoidance to tax rates is not only a function of tax rates, but is also a function of the taxpayer's risk aversion (Feinstein 1991; Slemrod and Yitzhaki 2002). It is possible that charitable nonprofits are inherently more risk averse than are educational or medical nonprofits, and that this risk aversion offsets the financial benefits of tax avoidance in response to tax rates.

My results for the effects of common costs on nonprofits' ability to avoid the UBIT are consistent across all nonprofit types for both my OLS results in Table 2 and Logistic results in Table 3 and show that reported taxable income is significantly lower (and more likely to be negative) when a nonprofit earns relatively more of its taxable revenues from programs. These results are consistent with the findings of Cordes and Weisbrod (1998) and Yetman (2003) who find that the ability to exploit common expenses costs enables UBIT avoidance. My results show that expense shifting remains a common UBIT avoidance technique even after public disclosure.

Nonprofit organizations are subject to the UBIT on the net profits from business activities are not closely related to their tax exempt purpose. UBIT avoidance is suspected to be widespread with fewer than 40 percent of nonprofits reporting positive profits on activities whose only purpose is to produce additional profits to support the charitable mission. To the extent that nonprofit organizations are able to avoid income taxes the UBIT loses its ability to fully achieve its regulatory goals of preventing unfair competition, protecting the tax base, and preventing mission drift.

An historic change to income tax return confidentiality laws in 2007 requires that all 501(c)(3) public charities publicly disclose their previously confidential federal income tax returns, form 990-T. Using this tax return data, I hypothesize that UBIT avoidance will be a function of ability, incentives, and external governance. I find evidence that nonprofits avoid the UBIT by creating taxable loss activities and using those losses to offset the profits of other activities. A provision in the TCJA of 2018 intended to mitigate the use of taxable loss activities appears to have been effective in reducing UBIT avoidance. I find that nonprofits avoid the UBIT by shifting expenses from their tax exempt to their taxable activities, and that higher tax rates encourage UBIT avoidance. The extra liability exposure for managers and trustees of charitable trusts (as opposed to nonprofits organized as charitable corporations) mitigates UBIT avoidance, and that public donors also play a governing role in mitigating UBIT avoidance.

My results suggest that UBIT avoidance has continued in the periods after public disclosure of tax return information, encouraged by the ability and incentive to avoid the UBIT. I find that specifically targeted legislation can mitigate UBIT avoidance, and that organizational form and external donors play a role as well. On the whole these mitigating factors appear weaker than the incentives to avoid the UBIT, and absent further legislative action we can expect UBIT avoidance to remain frequent in occurrence and large in magnitude.