Although most types of revenue earned by nonprofit

Transcription

1 Forum on Tax Research in Economics and Accounting Why Do Nonprofi ts Have Taxable Subsidiaries? Abstract - Nonprofit organizations operate taxable activities in two general ways: as unrelated businesses operated by the nonprofit or through controlled subsidiaries. Prior research and regulatory attention has focused on unrelated business activities, although taxable subsidiaries generate at least as much taxable revenue. We find that nonprofits place their taxable activities into subsidiaries when those taxable activities are relatively large, and when the taxable activities are relatively more risky. Nonprofits trade off possible benefits of the subsidiary form with the costs of reduced tax planning ability. Michelle H. Yetman & Robert J. Yetman Graduate School of Management, The University of California at Davis, Davis, CA National Tax Journal Vol. LXI, No. 4, Part 1 December 28 INTRODUCTION Although most types of revenue earned by nonprofit organizations are tax exempt, nonprofits engage in two types of taxable activities (Plunkett and Christianson, 24). First, a nonprofit can conduct taxable activities without the use of a separate legal entity. These activities are referred to as unrelated business activities and are taxable if they are unrelated to the organization s primary exempt mission (Internal Revenue Code section 512). Unrelated business activities are reported on the Internal Revenue Service (IRS) form 99 T. Second, a nonprofit can place taxable activities into taxable subsidiaries organized as either corporations or partnerships. The taxable subsidiaries report their results on IRS Form 112s (for corporations) or IRS Form 165s (for partnerships). 1 The first of these two organizational choices, unrelated businesses, has been the subject of significant descriptive analyses, several academic studies, and a recent spate of 1 As an example, consider the National Geographic Society, which has both unrelated businesses and taxable subsidiaries. The Society s charitable mission is to increase and diffuse geographic knowledge in its broadest sense (National Geographic Society 24 IRS Form 99, page 2, part III). The society reported total revenues in 24 of $5 million of which $1 million was from taxable unrelated businesses, primarily from advertising in National Geographic Magazine. Advertising of this sort is typically taxable to a nonprofit as unrelated business income. The Society s wholly owned taxable subsidiary, NGHT, Inc., operates a cable TV channel and earned taxable revenues of approximately $1 million in 24. It is clear that the Society s taxable activities are significant, comprising roughly 29 percent of total revenues. 675

2 NATIONAL TAX JOURNAL regulatory debate and legislation. The IRS annually produces a large amount of descriptive information about nonprofits unrelated business activities. Cordes and Weisbrod (1998) and Hines (1999) examine why nonprofit organizations would engage in unrelated business activities. Sansing (1998), Cordes and Weisbrod (1998), Yetman (21), Omer and Yetman (23), Yetman (23), Hofmann (27), and Omer and Yetman (27) provide evidence that nonprofits aggressively (but not necessarily illegally) avoid the tax on unrelated businesses primarily by shifting expenses from tax exempt to taxable activities. Congressional concerns over the increase in and aggressive tax planning by nonprofits unrelated businesses led to a provision in the Pension Protection Act of 26 that requires nonprofits to publicly disclose their unrelated business income tax returns (i.e., their IRS 99 Ts). However, that same law does not require that nonprofits disclose their taxable subsidiaries tax returns (i.e., their subsidiaries 112s or 165s), which is an issue we discuss further in the paper. Although the attention given to nonprofits unrelated business activities by researchers and regulators is warranted, we suggest that it is incomplete as it ignores taxable subsidiaries. Our paper attempts to fill this gap. First, we seek to describe, as much as our data permit, the taxable subsidiaries of nonprofit organizations. Nonprofits are allowed to set up and operate controlled taxable subsidiaries with few restrictions, and many have done so, yet there is little by way of descriptive information on taxable subsidiaries. Second, we empirically examine some factors associated with the choice of conducting taxable activities via a subsidiary versus conducting them directly as an unrelated business. Finally, we frame our results within their policy implications. Given the existing empirical and descriptive analyses of nonprofits unrelated business activities, it is perhaps surprising that so little is known about their taxable subsidiaries. To our knowledge no empirical analyses of taxable subsidiaries have been conducted. Existing research on taxable subsidiaries is limited to one brief descriptive analysis (Steuerle, 21) and one research paper by Sansing (2), which indirectly examines taxable subsidiaries by modeling the social welfare effects of joint ventures between nonprofit and for profit organizations. We are unaware of any IRS descriptive analyses of nonprofits taxable subsidiaries, although it has plans to collate information on them (Internal Revenue Service, 2). To summarize our results, we find that one of every nine nonprofit organizations in the IRS Statistics of Income sample has a taxable subsidiary. 2 Of large nonprofits with total assets above $1 million (about 15 percent of the IRS Statistics of Income sample), over one in three has a taxable subsidiary. The taxable revenues earned by nonprofits taxable subsidiaries are at least as large as the taxable revenues earned by their unrelated businesses. The average (median) taxable revenue generated by a subsidiary is approximately $1 million ($3,). Our cross sectional empirical tests of organizational form choice suggest that nonprofits place their taxable operations into subsidiaries rather than conduct them as unrelated businesses when the relative amounts of taxable revenue are high, consistent with attempts to mitigate 2 The IRS database does not include information on all 2, plus nonprofits, but rather is a sample of approximately 15, organizations each year. Although the IRS database contains only about eight percent of the population by number, because of size weighted sampling methods (i.e., all large nonprofits are included along with a stratified random sample of smaller organizations), the sample accounts for over 8 percent of total population assets and revenues. 676

3 Forum on Tax Research in Economics and Accounting loss of tax exempt status due to excessive unrelated business revenues (a nonprofit cannot lose its exempt status for large amounts of taxable subsidiary revenues). We also find that nonprofits are more likely to place relatively more risky taxable activities into subsidiaries, consistent with the attempt to provide some degree of legal liability protection for the exempt parent (separate subsidiaries provide some degree of legal distance whereas unrelated businesses do not). We also find that prior to the Taxpayer Relief Act of 1997, some nonprofits used a subsidiary as a tax planning tool (the Tax Act largely removed subsidiary tax planning ability). Nonprofits appear to trade off these benefits with costs of using a subsidiary. For some nonprofits, tax planning is more difficult with a subsidiary as compared with an unrelated business. Also, stricter state governance makes use of the subsidiary form more costly relative to an unrelated business. We also examine the time trend of unrelated business revenues around taxable subsidiary creations and dissolutions. A taxable subsidiary can be formed from a completely new business, or from the movement of an existing unrelated business or previously exempt activity from the parent nonprofit. Likewise, a taxable subsidiary can be dissolved as a result of the parent nonprofit abandoning the activity, or by the parent nonprofit transferring the activity to an unrelated business or exempt activity. Our evidence suggests that when nonprofits create new taxable subsidiaries, a substantial portion of the new subsidiaries revenues were previously earned by unrelated businesses, consistent with many taxable activities beginning life as unrelated businesses but at some point migrating to taxable subsidiaries. In addition, our evidence suggests that the parent nonprofit abandons taxable activities when taxable subsidiaries are dissolved. The paper proceeds as follows. In the next section, we provide a descriptive analysis of taxable subsidiaries. The following sections present and empirically test our cross sectional model of subsidiary choice. Then we examine taxable subsidiary creation and dissolution. Finally, we discuss the policy implications of our results, and conclude. DESCRIPTION OF TAXABLE SUBSIDIARIES Data To provide our descriptive and empirical analyses, we collect information about nonprofit organizations and their taxable activities. 3 The most direct source of taxable subsidiary information is the IRS form 112 if it is a corporation or its 165 if it is a partnership, neither of which is publicly available. Unrelated business activities are likewise reported on an income tax return (i.e., the IRS 99 T), which is not publicly available during our sample time frame (although the Pension Protection Act of 26 requires nonprofits to make 99 Ts filed after August 17, 26 publicly available). Although the primary data sources are not publicly available, nonprofits do provide some information about both their unrelated businesses and their taxable subsidiaries on their publicly available IRS form 99. Specifically, on part VII of the IRS 99, a nonprofit reports the amount of total revenue earned from unrelated 3 Our analysis focuses on nonprofit organizations whose tax exempt status was granted under Internal Revenue Code section 51(c)(3). With the exception of religious organizations, all nonprofit organizations operating under Internal Revenue Code section 51(c)(3) with revenues in excess of $25, must file an IRS form 99 annually. Page format and computer readable IRS 99 data is available from the National Center for Charitable Statistics (NCCS) at a project of the Center on Nonprofits and Philanthropy at the Urban Institute. 677

4 NATIONAL TAX JOURNAL businesses, and on part IX it reports the amount of total revenue earned by controlled (5 percent or more owned) taxable subsidiaries, as well as the percentage owned and the end of year total assets. 4 Many nonprofits report owning multiple taxable subsidiaries. For purposes of our analysis, we sum the taxable revenues of subsidiaries for each nonprofit. 5 The IRS began to collect taxable activity data in 1991, and the most recent year of available data is 22, providing us with 12 years of data. After removing observations that did not earn any type of taxable revenue, we are left with 28,123 panel observations across 12 years. Descriptive Analysis Table 1 contains a description of the taxable revenue earned from unrelated businesses and taxable subsidiaries for the most recent year of the sample. In the 22 sample year, there are 4,184 observations that earn some amount of taxable revenue, either directly through unrelated business activities or indirectly through controlled subsidiaries, or both. Because of the size weighted sampling methods used by the IRS to create this database, the sample contains a larger proportion of larger nonprofits, and since larger nonprofits are more likely to conduct taxable activities, all statistics in the paper should be interpreted in light of this sample. We present three columns in the table. The first contains the 3,284 observations that operate an unrelated business, regardless of whether they also had a taxable subsidiary. The second column contains the 1,845 observations that operate a taxable subsidiary, regardless of whether they also had an unrelated business. The third column contains the 945 observations that have both unrelated businesses and taxable subsidiaries. The total number of unique observations is 4,184 (3,284 plus 1,845 less 945). Nearly 2 percent of the 22 SOI population report operating an unrelated business. The average (median) amount of unrelated business revenues earned by these observations is $1.1 million ($127 thousand). Roughly ten percent of the unrelated businesses report negative revenues. 6 Almost 11 percent of the 22 SOI population report taxable subsidiaries. The average (median) number of taxable subsidiaries per nonprofit is 1.8 (1.), and ranges from 1 to 12. These subsidiaries are organized as either corporations or partnerships. The average (median) amount of taxable revenues earned by these observations subsidiaries is $8.8 million ($28 thousand). These results show that the amounts of taxable revenues earned by nonprofits subsidiaries are at least as large as the amounts of unrelated business revenues earned. About ten percent of the subsidiaries report negative revenues. Approximately six percent of the 22 SOI population report operating both unrelated businesses and taxable sub- 4 Prior research using a confidential set of 99 Ts finds that the amount of taxable revenues reported on matched sets of IRS 99s and IRS 99 Ts have a correlation of 8 percent (Yetman, Yetman, and Badertscher, forthcoming). 5 Total income as reported on the parent s IRS 99 Part VII is taken directly from line 11 of the IRS form 112. As such, total income represents total revenues with two exceptions. First, revenues from product sales are netted against their respective costs of goods sold, which can include a multitude of costs per IRC 263A. Second, sales of assets such as property or equipment are reported on a net of depreciated cost basis. The same is true for the amount of total unrelated business income reported in Part X. In our analyses we will refer to these taxable components as revenues, and avoid using the IRS terminology of income, which could be confused with net income. 6 Total revenues can be negative if gross profit on sales is negative (i.e., cost of goods sold is higher than sales revenues) or there were net losses on the sales of capital or business assets. This applies to both unrelated business revenues and taxable subsidiary revenues. 678

5 Forum on Tax Research in Economics and Accounting TABLE 1 TYPES AND MAGNITUDES OF TAXABLE REVENUES EARNED BY NONPROFIT IN Sample Year Observations Percentage of 22 SOI population Observations with Unrelated Businesses 3,284 2% Observations with Taxable Subsidiaries 1,845 11% Observations with Both Unrelated Businesses and Taxable Subsidiaries d 945 6% Taxable Unrelated Businesses a Mean revenues ($) Mean revenues as percent of parent s exempt revenues Median revenues ($) 1,1,35 1.4% 127,119 2,388,11 2.8% 383,789 Taxable Subsidiaries Mean revenues ($) b Mean revenues as percent of parent s exempt revenues Median revenues ($) Mean ownership percentage Percentage 1% owned 8,798, % 27,946 88% 68% 12,169,2 19.% 437,165 86% 65% Mean assets ($) c Mean assets as percent of parent s assets Median assets ($) 21,66, % 1,694,91 27,914,3 8.4% 2,519,693 Notes: All data taken from the publicly available Internal Revenue Service Form 99 for 22 (the most recent year of available data) from the IRS Statistics of Income data files. The full SOI sample includes 16,782 observations. a Taxable unrelated business revenues are from taxable activities carried on directly by the nonprofit organization. Nonprofits are required to disclose the amounts of gross taxable unrelated revenues on page five of the publicly available Form 99 (more detailed results of unrelated business activities are reported on Internal Revenue Service form 99 Ts, which are not publicly available). b Taxable subsidiary revenues are from taxable activities carried on subsidiaries at least 5 percent owned by the nonprofit. Nonprofits are required to disclose the amounts of revenues earned by their taxable subsidiaries on page five of the publicly available Form 99 (more detailed results of the subsidiaries activities are reported on the subsidiaries tax returns, which are not publicly available). c Taxable subsidiary assets exclude observations that report negative assets. d These 945 observations operate unrelated businesses and have taxable subsidiaries. These observations are included in each of the first two columns of the table. sidiaries. The average (median) amount of taxable revenues earned by these observations subsidiaries is $12.2 million ($437 thousand), while the average (median) amount of unrelated business revenues is $2.4 million ($384 thousand). This result again shows that the amounts of taxable revenues earned by nonprofits taxable subsidiaries are at least as large as the average amounts of unrelated business revenues earned. In terms of relative assets, Table 1 shows that the assets of nonprofits taxable subsidiaries are roughly one seventh the size of the parents total assets on average, suggesting that the subsidiaries are not trivial in size. Turning to ownership percentages, we find that average ownership is approximately 88 percent, while approximately 68 percent of the observations are 1 percent owned. Steuerle (21) suggests that one reason a nonprofit would establish a separate taxable subsidiary is to raise outside equity capital and/or to pay equity based compensation to its officers. Our statistics suggest that while some taxable subsidiaries possibly raise outside equity capital and could pay equity based compensation, at least 68 percent do not. The National Taxonomy of Exempt Entities (NTEE) as established by the Internal Revenue Service classifies nonprofits into 26 primary industry codes 679

6 NATIONAL TAX JOURNAL using the letters A through Z. Figure 1 shows the proportionate use of taxable subsidiaries across nonprofit industries. We omit industries Q (international) and X (membership organizations) as there are fewer than 1 observations in the panel sample, industry Y (religious organizations) as they are not required to file an IRS form 99, and industry Z (organizations with unknown industries). The sample for the graph is limited to observations that earn some type of taxable revenue, either via an unrelated business or a taxable subsidiary. By percentage, housing and shelter nonprofits (L) are the largest users of taxable subsidiaries followed by health care organizations (E). Several industries, such as public safety (M) and animal related (D), use relatively fewer taxable subsidiaries when conducting taxable activities. It is difficult to say ex ante why subsidiary use varies considerably by industry, although several factors likely play a role, some of which we empirically investigate later in the analysis. In raw terms, our sample is comprised of roughly one third each of Educational (B) and Health Care (E), with the remaining Figure 1. Percentage of Observations with Taxable Subsidiaries by Industry (for observations with some type of taxable revenues) Percentage A B C D E F G H I L J K M Industries N O P R S T U V W Notes: Sample is limited to observations that earn either taxable unrelated business revenues or have a taxable subsidiary. Industry Codes are as follows. A Arts, culture, and humanities L Housing and shelter B Education M Public safety C Environment N Recreation and sports D Animals O Youth development E Health care P Human services F Mental health R Civil rights G Diseases S Community improvement H Medical research T Philanthropy I Crime and legal U Science and technology J Employment V Social science K Food and nutrition W Public benefit 68

7 Forum on Tax Research in Economics and Accounting industries accounting somewhat evenly for the remaining third. With respect to taxable subsidiary concentration by industry, Figure 2 shows that health care accounts for over 6 percent of all taxable subsidiaries, whereas this industry comprises of only approximately 3 percent of the total sample of nonprofits. We conjecture that the large concentration in the health care industry is due to several factors, such as the ability to provide an array of medical related outputs that tend to be taxable (such as ambulance for hire services and blood and tissue sales). The next industry with the most subsidiaries is educational institutions, which account for approximately 2 percent of the total sample and have approximately ten percent of all taxable subsidiaries. To summarize our descriptive analysis, approximately one out of nine nonprofits has a taxable subsidiary, and on average these subsidiaries earn millions of dollars of revenue. The taxable activities conducted by these subsidiaries are at least as large as nonprofits unrelated businesses. In the next section, we examine the factors Figure 2. Distribution of Taxable Subsidiaries across Nonprofit Industries (for observations with some type of taxable revenues) Percentage A B C D E L F G H I J K N O Industries P R S T U V W Notes: Sample is limited to observations that earn either taxable unrelated business revenues or have a taxable subsidiary. Industry Codes are as follows. A Arts, culture, and humanities L Housing and shelter B Education M Public safety C Environment N Recreation and sports D Animals O Youth development E Health care P Human services F Mental health R Civil rights G Diseases S Community improvement H Medical research T Philanthropy I Crime and legal U Science and technology J Employment V Social science K Food and nutrition W Public benefit 681

8 NATIONAL TAX JOURNAL associated with operating a taxable activity as either an unrelated business or via a taxable subsidiary. CROSS SECTIONAL MODEL OF SUBSIDIARY CHOICE We begin our informal model development with a set of general facts based on nonprofit tax law as contained in the Internal Revenue Code and based on discussions on the use of taxable subsidiaries (Tenenbaum, 1996; Steuerle, 21; Plunkett and Christianson, 24). Nonprofit organizations are permitted to operate taxable activities and can choose to operate those activities as either unrelated businesses or taxable subsidiaries. 7 A nonprofit can have multiple taxable activities, some of which are organized as unrelated businesses and others as taxable subsidiaries. To the extent that the costs and benefits associated with unrelated businesses and taxable subsidiaries vary, nonprofits will prefer one form to the other. A nonprofit can choose to move its existing taxable activities between unrelated businesses and taxable subsidiaries, although there are certainly costs of doing so. We do not find many examples of taxable activities migrating between unrelated businesses and taxable subsidiaries, although a longer time series, which we do not have, might identify more instances of this. We model an organization s choice of how to conduct its taxable activities as a function of various costs and benefits of each form. To conduct our analysis we need to identify various costs and benefits that affect the choice of how to organize a taxable activity. To the extent that our measures of costs and benefits apply exclusively to either taxable subsidiaries or unrelated businesses, we must exercise care when interpreting those results for observations that simultaneously have unrelated businesses and taxable subsidiaries. One alternative is to simply exclude observations that have both unrelated businesses and taxable subsidiaries, but this would preclude us from presenting some potentially interesting findings. Rather, we will include all three potential cases in our empirical analyses (i.e., unrelated businesses only, taxable subsidiaries only, and both unrelated businesses and taxable subsidiaries), but we will focus most of our attention on the results related to operating either unrelated businesses or taxable subsidiaries. EMPIRICAL ANALYSIS OF SUBSIDIARY CHOICE Research Design To estimate the effects of various costs and benefits on the choice of conducting taxable activities as either unrelated businesses or as taxable subsidiaries, we estimate the following multinomial logit model: [1] Subsidiary Choice it = α + β 1 Taxable Revenue Ratio it + β 2 Insurance Ratio it + β 3 Rental Revenue it + β 4 Rental Revenue it * Post 1997 it + β 5 State Tax Rate it + β 6 State Governance it + β 7 Size it + ε. The dependent variable Subsidiary Choice is equal to zero if the parent nonprofit operates only an unrelated business, one if the parent has only a wholly owned taxable subsidiary, and two if the parent nonprofit has both an unrelated business and a taxable subsidiary. As discussed, our primary interest is in comparing a Subsidiary Choice of zero (unrelated business only) to a Subsidiary Choice of one (taxable subsidiary only). We include the other 7 We do not address the decision as to whether a taxable activity should be engaged in, but rather take that choice as given. See Cordes and Weisbrod (1998) and Hines (1999) for empirical analyses of this decision. 682

9 Forum on Tax Research in Economics and Accounting two possible comparisons (i.e., comparing unrelated business only to using both, and comparing a subsidiary only to using both) for completeness purposes. We limit our analysis sample to taxable subsidiaries that are wholly (i.e., 1 percent) owned by their nonprofit parent, and do so to avoid measurement error problems. Subsidiaries that are 1 percent owned are prima facie corporations, as partnerships are always less than 1 percent owned by a single entity. It is certainly true that by eliminating all less than 1 percent owned subsidiaries, we are eliminating at least some corporations. Although we would like to include all corporations, 1 percent owned as well as less than 1 percent owned, we are not able to determine from our available data whether these less than 1 percent owned subsidiaries are corporations or partnerships. Because it is quite likely that these less than 1 percent owned subsidiaries include at least some partnerships, including them in our sample would introduce measurement error as our empirical measures of costs and benefits apply strictly to corporations and do not apply to partnerships. 8 Our first independent variable, Taxable Revenue Ratio, is a measure of the relative size of a nonprofit s taxable activities to their total activities, and is measured as the ratio of taxable unrelated business revenue (line 14b) plus taxable subsidiary revenue (part IX) to the sum of total nonprofit revenue (including exempt activities, unrelated businesses, and taxable subsidiaries). Taxable Revenue Ratio captures how large the taxable activities would be as if they all were operated as an unrelated business. 9 The larger is the size of taxable activities relative to total activities, the more likely it is that a nonprofit will use a subsidiary for a variety of reasons. First, nonprofits will lose their exempt status if the scope of their unrelated businesses becomes too large [Treasury Regulation 1.51(c)(3) 1(c)]. A nonprofit can avoid jeopardizing its tax exempt status by placing relatively larger taxable activities into taxable subsidiaries rather than operating them as an unrelated businesses as there are no limits as to the scope of taxable subsidiaries (Steuerle, 21). The point at which unrelated business revenues becomes excessive is not clear, but IRS guidance provides some insight. In Technical Advice Memoranda the IRS summarizes the position of the courts as [g]enerally courts have denied exemption to organizations that conducted nonexempt activities which generated revenue in excess of approximately twenty five percent of the organization s total annual revenue. However, because the courts apply a facts and circumstances test, this ratio is not a hard cutoff. Nonetheless, it does provide us with some justification for conjecturing an inflection point in Taxable Revenue Ratio variable at a value of 25 percent. While the loss of tax exempt status would seem to provide a powerful motive for placing relatively large taxable activities into subsidiaries, there are other plausible reasons for doing so. For example, reporting revenues from unrelated business activities tends to reduce charitable donations to nonprofits (Yetman and Yetman, 23). To the extent that a nonprofit can effectively hide the extent of its taxable activities from donors by placing them 8 It is possible that some of the 1 percent owned subsidiaries are organized as limited liability corporations (LLC), which are treated as partnerships for tax purposes. The presence of LLCs in our sample will result in measurement error. Unfortunately we do not know the extent to which 1 percent owned LLCs are in our sample, but we suspect it is not large as the use of LLCs in the exempt arena is relatively recent (Borden, 28). 9 Observations with ratios of less than zero (which can happen if the reported amounts of taxable revenues are negative) are removed from the analysis, as prior research suggests that a negative value for revenues is indicative of an error in reporting. (Yetman, Yetman, and Badertscher, forthcoming). 683

10 NATIONAL TAX JOURNAL into taxable subsidiaries, there is a clear incentive to place relatively larger taxable activities into subsidiaries. In addition, as a taxable activity becomes relatively large, a nonprofit might find it more cost effective (in terms of record keeping or organizational structure) to operate the activity as a subsidiary. In the end we are not able to rule out these competing explanations. The next independent variable is a measure of the legal liability risk that the taxable activity imposes on the nonprofit parent. Because taxable activities organized as subsidiaries provide some degree of legal separation (often referred to as the corporate veil), it seems plausible that a nonprofit would use a taxable subsidiary to operate its relatively risky taxable activities, whereas unrelated businesses would not provide such legal distance (Steuerle, 21). To construct this variable, we presume that insurance companies appropriately price risk and that liability insurance expense at the nonprofit parent level is a reasonable proxy for the relative risk of the organizations taxable activity operations. We collect liability insurance expenses paid by a random sample of ten unique observations for each of the 46 two digit NTEE industry codes that have at least 3 observations per industry. 1 These liability expenses were hand collected from the detailed schedules of the organizations IRS form 99s for the most recent year of data used in our analysis (i.e., 22). 11 Using these 46 samples of ten observations each, we calculate the ratio of liability insurance expense to total assets for each observation. We use total assets as the scalar as insurance contracts are typically based on asset values, although our results are robust to scaling by total expenses as well. We then take the industry average of these ratios (ten observations per industry), providing us with 46 industry level ratios, which are then applied to all observations in our sample. Thus, our legal liability measure (Insurance Ratio) varies across industries but not across observations within an industry. 12 Our next independent variable is intended to measure the extent to which taxable subsidiaries are used for tax planning purposes. Prior to 1997, nonprofits could effectively reduce overall taxes by having their controlled taxable subsidiaries make payments of interest, rents, or royalties up to the nonprofit parents. To the extent the subsidiary was properly structured (i.e., as a second tier), these payments were tax deductions to the subsidiary (i.e., they reduced the subsidiary s taxable revenue) yet were not taxable at the parent nonprofit level. 13 The Taxpayer Relief Act of 1997 eliminated this ability by making such payments taxable to the parent nonprofit as unrelated business revenue. Using this event, we examine 1 We excluded observations with taxable subsidiaries from our random sample because it is possible that some insurance expense was shifted to the subsidiary. Furthermore, we conditioned the random sample on observations from the same size categories as the observations that do have taxable subsidiaries because larger organizations are more likely to have subsidiaries, and we wish to know how much insurance expense is paid by those same sized organizations. 11 Data contained in detailed schedules of the IRS 99 (which is where liability insurance is reported) are not included in the computer readable IRS database, requiring us to hand collect these data. 12 Other studies on organizational choice in the for profit setting have used various measures of firm risk such as employee injury rates or default rates. These and other similar measures are not available in the nonprofit setting. 13 In order for the passive payments to avoid taxation at the nonprofit parent level, the subsidiary making the payments must have organized as a second tier, or a subsidiary of a subsidiary owned by the nonprofit parent. After the 1997 tax act, payments of passive income from second tier subsidiaries to nonprofit parents are taxable to the nonprofit parent. Interestingly, this rule was again changed in 26, allowing deductible payments from subsidiaries and no taxation to parent nonprofits if the payments are at arms length. 684

11 Forum on Tax Research in Economics and Accounting whether nonprofits took advantage of this tax planning ability by testing whether the gross amount of rental revenue received by the nonprofit parent (line 6a of the IRS 99 scaled by total nonprofit revenues) decreased after 1997 for nonprofits with taxable subsidiaries by more than it decreased for nonprofits without taxable subsidiaries. We limit our measure to rental revenue (and exclude interest or royalties), as rental revenue is the only specifically identified line item on the Our primary test variable is the interaction of Rental Revenue with the Post 1997 indicator (which is equal to one in the post 1997 period and zero otherwise). If nonprofits reduced the amounts of rental payments from their taxable subsidiaries after 1997, the coefficient on the interaction will be negative. The Rental Revenue variable will likely be positive, as observations with subsidiaries are likely to receive more rental revenue by virtue of having a subsidiary. We exclude the Post 1997 indicator as our model includes year indicators. Turning to measures of the costs of using a taxable subsidiary, we start with a measure of tax costs, which is the highest marginal tax rate in a state for the given year (State Tax Rate). Operating a taxable activity as a subsidiary can make tax planning more difficult in certain circumstances. Nonprofits commonly reduce the taxes on their unrelated businesses by shifting costs from their tax exempt activities (Cordes and Weisbrod, 1998; Yetman, 21; Omer and Yetman, 23; Yetman, 23; Hofmann, 27; and Omer and Yetman, 27). Sansing (1998) analytically demonstrates that the ability to shift costs from tax exempt to taxable activities is a function of the separability of the taxable activity s cost function. He shows that, if the cost function is inseparable (meaning that the taxable and exempt activities use common production inputs), it is easier to shift costs than if the cost function were separable. Yetman (23) empirically documents this effect. Based on this, we presume that unrelated businesses (which are most frequently operated by and within the nonprofit s overall operations) use common inputs more than do taxable subsidiaries. If true, it would be more difficult to reduce taxes by shifting costs between a nonprofit parent and its taxable subsidiary than it would be to shift costs between a nonprofit s taxable unrelated business and non taxable activities. Thus, the benefits of avoiding relatively higher state income tax rates favor the unrelated business form, and impose a cost on the subsidiary form. Our next measure of the cost of using a subsidiary is a state level governance variable, State Governance, which measures the extent to which a state regulates transactions between nonprofits and their for profit subsidiaries. Prior research shows that the states are the primary governance mechanisms affecting the behavior of nonprofit organizations (Fisman and Hubbard, 25; Desai and Yetman, 28; Krishnan, Yetman, and Yetman, 26), and some states have specific laws that apply to for profit subsidiaries. The general intent of these laws is to prevent nonprofits from reincorporating as for profit firms and subsequently paying out their assets to officers and directors. These laws impose additional costs only on the subsidiary form, making it more costly relative to the unrelated business form. We include total assets as a control, as well as yearly indicator variables. Standard errors are cluster corrected at the organization level (Peterson, forthcoming). Our analysis implicitly presumes that the economic bulk of activities 14 We attempted to use other possible line items on the IRS 99 where intercompany transfer payments could be recorded, such as other expenses. We were unable to document a statistically different relationship between these proxies for intercompany transfer payments and the use of a taxable subsidiary across the year

12 NATIONAL TAX JOURNAL conducted by taxable subsidiaries would have been taxable had they been directly conducted by the nonprofit parent. 15 Descriptive Statistics for Empirical Analyses Table 2 contains the descriptive statistics of our sample of 28,123 panel observations divided into our dependent variable bins (zero if only an unrelated business equal 18,976 observations, one if only a taxable subsidiary equal 4,24 observations, and two if both equal 4,943 observations). Many of the variables are ratios, and due to significant heterogeneity in nonprofits there are large outlying observations. To mitigate the effects of these outlying observations, we winsorize several of the continuous variables (i.e., Taxable Revenue Ratio, Rental Revenue, and Size) at the 1 st and 99 th percentiles, and those winsorized values are reported in Table 2. As is obvious from Table 2, even after winsorizing, several of our variables are extremely skewed. In particular, the Taxable Revenue Ratio and Size variables exhibit extreme left skewness. Kernel density plots confirm the skewness in these variables. To mitigate skewness, we employ a method suggested by Box and Cox (1964) that transforms the variable using a power function with a result of near zero skewness. We present the untransformed variables in Table 2 because the transformed variables are difficult to interpret, but the regression models use the transformed variables (later robustness tests will employ the original untransformed variables). In addition to presenting the descriptive statistics, we also conduct t tests for differences in means across the three bins. These t tests are in the first columns of Table 2 and show that Taxable Revenue Ratio is significantly larger for taxable activities operated in subsidiaries. Subsidiary use is also more frequent when insurance ratios are higher. Higher state tax rates and stronger state governance appear to favor the use of unrelated businesses. The correlation matrix in Table 3 shows that none of the independent variables (other than the partitioned Taxable Revenue Ratio variables, which is expected) has particularly high correlations, mitigating concerns over multicollinearity. Empirical Results Our regression results are contained in Table 4. To get a sense of overall model fit, we calculate the percentage of correct predictions (i.e., the predicted probability of an observation being in the proper category is greater than 5 percent). Of the observations that have only an unrelated business, our model correctly predicts 37 percent. Of the observations that have both an unrelated business and a subsidiary, our model correctly predicts 7 percent. Of the observations that have both an unrelated business and a subsidiary, our model correctly predicts 64 percent. These predicted probabilities suggest that our model does a good job in predicting which nonprofits will operate a taxable subsidiary, either alone or in conjunction with an unrelated business, but only a fair job predicting which nonprofits will operate an unrelated business. The first column compares the choice between the use of a subsidiary or an unrelated business (but not both). The second column 15 Steuerle (21) suggests that nonprofits might place an otherwise tax exempt activity into a wholly owned taxable subsidiary in order to provide liability protection. This is not likely to occur, as other non taxable alternatives exist, such as a nonprofit subsidiary. Another reason suggested is to avoid voluntary or legal salary ceilings at the nonprofit level that would limit the amount a nonprofit can pay its employees. However, this practice is not likely to avoid excessive compensation issues as the parent nonprofit must report all amounts of compensation paid to its officers, directors, and key employees and also separately state all compensation received from related taxable or tax exempt organizations if that amount exceeds $1,. 686

13 Forum on Tax Research in Economics and Accounting Observations with Unrelated Businesses Only: Taxable Revenue Ratio < = 25% Taxable Revenue Ratio > 25% Insurance Ratio Rental Revenue Rental Revenue * Post 1997 State Tax Rate State Governance Size (in millions) TABLE 2 DESCRIPTIVE STATISTICS FOR ORGANIZATIONAL CHOICE TEST F Test t values Mean % Median Comparing UB only to SUB only % Std. Dev Observations with Taxable Subsidiaries Only: Taxable Revenue Ratio < = 25% Taxable Revenue Ratio > 25% Insurance Ratio Rental Revenue Rental Revenue * Post 1997 State Tax Rate State Governance Size (in millions) Comparing SUB to UB and SUB Observations with both Unrelated Businesses and Taxable Subsidiaries: Taxable Revenue Ratio < = 25% Taxable Revenue Ratio > 25% Insurance Ratio Rental Revenue Rental Revenue * Post 1997 State Tax Rate State Governance Size (in millions) Comparing UB and SUB to UB only Notes: Contains 28,123 observations from the publicly available Internal Revenue Service Form 99 as collected and distributed by the Statistics of Income division of the Internal Revenue Service, unless otherwise noted. The sample is a panel dataset from 1991 to 22 and is restricted to observations that have an unrelated business activity and/or a taxable subsidiary. Observations with less than 1 percent owned taxable subsidiaries are excluded to remove possible partnerships from the sample is the first year in which data on taxable subsidiaries were collected, and 22 is the most recent year of data available. There are approximately 3,3 unique observations per year. Variable Definitions: UB refers to unrelated businesses and SUB refers to subsidiaries. In the regression models, the dependent variable is Subsidiary Choice, which is equal to zero if the parent nonprofit operates an unrelated business only (18,976 observations), one if the parent has a wholly owned taxable subsidiary only (4,24 observations), and two if the parent nonprofit has both an unrelated business and a taxable subsidiary (the remaining 4,943 observations). Taxable Revenue Ratio is equal to the ratio of taxable unrelated business revenues (line 14b) plus taxable subsidiary revenues (part IX) to the sum of total nonprofit revenues (including exempt activities, unrelated businesses, and taxable subsidiaries). This variable is split at the value of.25. Insurance Ratio is the average ratio of liability insurance expense to total assets for a random sample of ten observations in the 46 two digit industries that have at least 3 observations. Rental Revenue is equal to the amounts of rental revenue received by the nonprofit parent (line 6a) scaled by the total revenues of the nonprofit parent. State Tax Rate is the highest marginal state corporate rate in effect for the given year (from Commerce Clearinghouse Tax Guides). State Governance is equal to one if the state where the nonprofit is located has a law limiting transfers of assets between nonprofit and for profit organizations, and zero otherwise. Size is total assets (line 59b). Several of the continuous variables (i.e., Taxable Revenue Ratio, Rental Revenue, and Size) contained large outliers and, thus, those variables were winsorized at the 1st and 99th percentiles, and those winsorized values are shown in this table. For the regression models that follow, the Taxable Revenue Ratio variables and the size variable were transformed using the method of Box and Cox (1964) to mitigate severe skewness. The untransformed variables are shown in this table for expositional purposes. 687

14 NATIONAL TAX JOURNAL Taxable Revenue Ratio < 25% Insurance Ratio Rental Revenue Rental Revenue * Post 1997 Indicator State Tax Rate State Governance Size TABLE 3 CORRELATION STATISTICS FOR ORGANIZATIONAL CHOICE TEST Taxable Revenue Ratio >= 25%.693*.55*.186*.144*.35*.35*.67* Taxable Revenue Ratio < 25%.48*.152*.127*.38*.31*.43* Insurance Ratio.28*.19*.1..98* Rental Revenue.74*.25*..37* Rental Revenue * Post 1997 Indicator.23*.6.22* State Tax Rate.47*.23* State Governance.9 Notes: Variable definitions are in Table 2. Pearson correlations indicated by * are significant at the five percent level. Constant TABLE 4 TAXABLE ACTIVITY ORGANIZATIONAL FORM CHOICE Unrelated Business Compared to Subsidiary (14.8) Unrelated Business Compared to Subsidiary and Unrelated Business (23.51) Subsidiary Compared to Subsidiary and Unrelated Business (9.48) Taxable Revenue Ratio <= 25% (17.41) (23.26) (4.89) Taxable Revenue Ratio > 25%.871 (.41) (6.76) (7.92) Insurance Ratio (6.42) (5.54) (.57) Rental Revenue (5.33) (4.11).832 (1.21) Rental Revenue*Post (5.1) (3.38) (1.67) State Tax Rate (3.67) (3.93).233 (.14) State Governance.17 (1.6).276 (2.53).16 (.88) Size.5 (9.93).125 (21.93).75 (12.73) Observations Pseudo R 2 28, , , Notes: The dependent variable is equal to zero if the observation had an unrelated business only, one if the observation had a taxable subsidiary only, and two if the observation had both an unrelated business and a taxable subsidiary. Independent variables are described in Table 2. The model used is a three way multinomial logit, and results for all three pairwise comparisons are reported. Coefficient estimates are shown with robust and cluster corrected (by nonprofit) two tailed z statistics in parentheses. All models include year indicator variables. 688

15 Forum on Tax Research in Economics and Accounting compares the use of a taxable subsidiary and an unrelated business to an unrelated business only. The final column compares the use of a subsidiary to the use of a subsidiary and an unrelated business. As discussed earlier, our emphasis is on the first column. All models employ yearly effects and report logistic coefficients and robust z statistics clustered at the organization level. We do not use organization fixed effects as there is little within organization variation (i.e., approximately ten percent of the observations created or dissolved a subsidiary in the sample periods and, thus, there is no within firm variation for approximately 9 percent of the sample). The logistic results show that the use of a taxable subsidiary is statistically positively associated with Taxable Revenue Ratio values equal to or less than 25 percent, but not statistically correlated with values greater than 25 percent. Untabulated F tests show that these two coefficients are statistically different. Thus, we conclude that when the ratio of taxable to total revenues rises above 25 percent, the amount of unrelated business revenues earned has little affect on the choice of whether to use a subsidiary. One possible interpretation is that nonprofits with very high values of Taxable Revenue Ratio that are conducting their taxable activities as unrelated businesses are not in any particular danger of losing their exempt status, perhaps because if their exempt status were susceptible to an IRS challenge in court, it likely would have already occurred. 16 To gain insight into economic magnitudes, we calculate the change in the probability of using a taxable subsidiary only as the independent variables move from their 1 th to their 9 th percentiles. We acknowledge that this is quite a large swing in percentiles, but given the underlying distribution of many of our variables, it provides a useful calibration. Moving from the 1 th to the 9 th percentile of Taxable Revenue Ratio (i.e., a movement from less than one percent to approximately ten percent) increases the probability of using a taxable subsidiary from ten percent to 27 percent, a 17 percent increase. Results show that higher ratios of insurance to total expenses are associated with a higher probability of using a taxable subsidiary, suggesting that nonprofits place more risky taxable activities into subsidiaries. In terms of magnitude, moving from the 1 th to the 9 th percentile of Insurance Ratio (i.e., a movement from.3 percent to 1.5 percent) increases the probability of using a taxable subsidiary from 11 percent to 18 percent, a 64 percent increase. The amount of rental revenue received by a parent nonprofit is higher for observations with taxable subsidiaries, but this effect is smaller after the Tax Reform Act of This result suggests that subsidiaries were used for tax planning purposes prior to the Act of However, the economic magnitude of this effect is comparatively small. Moving from the 1 th to the 9 th percentile of Post97*Rental Revenues decreases the probability of using a taxable subsidiary from 15 percent to 13 percent, a modest change of 15 percent. Interestingly, this latter result does not hold for observations that have both a taxable subsidiary and an unrelated business. One explanation for this result is that nonprofits reporting both subsidiary and unrelated business revenue did not structure their subsidiaries properly and, thus, the rental payments were taxable 16 Consider a nonprofit that rents out unused facilities. Although the amounts of revenues might grow quite large, generation of passive rental income could well require very little employee time and effort, and thus not compromise focus on core charitable missions. Unfortunately we are not able to discern the types of unrelated business revenue earned using publicly available information. 689

16 NATIONAL TAX JOURNAL to the parent. As previously discussed, a nonprofit could receive tax free rental revenue from a subsidiary prior to the 1997 Tax Act, but only if the subsidiary were structured correctly (i.e., as a second tier subsidiary). If the subsidiary were not structured correctly, the rental revenue would be unrelated business revenue to the exempt parent. Higher state corporate income tax rates are negatively associated with the probability of using a taxable subsidiary, suggesting that loss of tax planning ability is a cost associated with using a taxable subsidiary, although the economic magnitude is again comparatively small. Moving from the 1 th to the 9 th percentile of State Tax Rate decreases the probability of using a taxable subsidiary from 17 percent to 14 percent, a change of 14 percent. Finally, we do not find evidence that stronger state governance aimed at making it difficult to engage in certain abusive transactions with taxable subsidiaries is associated with less subsidiary use, although the coefficient is nearly significant and is in the correct direction. However, even if the coefficient were significant, its economic effect is very minor. Moving from State Governance values of zero to one decreases the probability of using a taxable subsidiary from 15 percent to 14 percent. To summarize the results of our organizational choice model, we provide evidence that the decision to use a taxable subsidiary is a function of several benefits and costs. Our results are consistent with nonprofits using subsidiaries to avoid loss of exempt status due to excessive unrelated business revenues, although it is not possible to completely rule out alternative explanations for placing relatively large taxable operations into subsidiaries. We also find that nonprofits place relatively more risky taxable activities into subsidiaries. These two results have relatively strong economic effects. Taxes also appear to have some effects, favoring the unrelated business overall and particularly so 69 after 1997, but the economic magnitudes of these effects are comparatively small. SENSITIVITY TESTS To assess the sensitivity of our findings, we conduct a series of robustness tests and report the results in Table 5. We report the results on the choice between operating taxable activities as an unrelated business or as a taxable subsidiary, but not both (i.e., similar to the first left most column in Table 4). We begin by examining the effect that transforming our heavily skewed variables has on our results (recall we use Box Cox transformations of the heavily skewed variables in our primary analysis). The first column reports the results, showing no changes in general inferences, although the magnitude of the coefficient estimates for the untransformed variables are smaller. Second, we examine whether the results are being driven by primary care hospitals (with an NTEE code of E), which comprise more than half of the primary sample. Although the magnitudes of several of the coefficient estimates fall, they remain statistically significant. Of note is the fall in the Insurance Ratio coefficient, which falls by nearly 5 percent. This suggests that hospitals are more sensitive to risk (as we measure it) as compared to other types of nonprofit organizations. Next we examine whether results are robust to industry fixed effects (using two digit NTEE code effects) as shown in the third column of Table 5. Note that we exclude the Insurance Ratio variable as it is identified only by industry. Again the general tenor of our results is unchanged. Alternative Functional Forms between Ratio and Subsidiary Choice Our primary tests include a linear relation between Taxable Revenue Ratio and Subsidiary Choice, with an inflection at a value of 25 percent. We examine several

17 Forum on Tax Research in Economics and Accounting Model: Constant TABLE 5 TAXABLE ACTIVITY ORGANIZATIONAL FORM CHOICE ALTERNATIVE TESTS (11.23) (1.46) (14.77) 4.73 (13.73) (14.9) (11.2) Taxable Revenue Ratio <= 25% (17.36) (7.98) (17.68) (18.37) Taxable Revenue Ratio > 25%.868 (2.36) (.68).213 (.1) (2.89) Lagged Taxable Revenue Ratio <= 25% (1.22) Lagged Taxable Revenue Ratio > 25% (11.2) Taxable Revenue Ratio (26.42) (11.95) Taxable Revenue Ratio (5.56) Insurance Ratio (5.7) (2.98) (6.9) (6.7) (5.8) Rental Revenue 3.72 (4.67) 4.99 (3.76) 3.57 (4.59) 4.59 (5.51) (5.8) 4.33 (4.87) Rental Revenue*Post (4.92) (3.43) 3.84 (4.74) (4.94) (5.41) 4.2 (4.21) State Tax Rate (4.95) (3.39) (3.1) 6.17 (3.88) (3.58) (3.57) State Governance.199 (1.92).71 (.44).131 (1.24).179 (1.68).169 (1.58).25 (1.77) Size. (2.98).46 (6.78).48 (8.73).48 (9.69).5 (9.93).45 (7.88) Observations 28,123 13,888 3,511 28,123 28,123 23,3 Pseudo R Notes: The dependent variable is equal to zero if the observation had an unrelated business only and one if the observation had a taxable subsidiary only. Independent variables are described in Table 2. This corresponds to the model in column 1 of Table 4. Coefficient estimates are shown with robust and cluster corrected (by nonprofit) two tailed z statistics in parentheses. All models include year indicator variables. Model Specifications with model in Column 1 of Table 4 representing Base Model: 1 - Base Model with untransformed (i.e., before the Box Cox transformation) Taxable Revenue Ratio and Size variables. 2 - Base Model excluding Hospitals (NTEE code of E, Primary Healthcare Organizations). 3 - Base Model using double digit NTEE industry fixed effects (note: Insurance Ratio variable does not vary within industry and is, therefore, excluded from this model). 4 - Base Model with untransformed Taxable Revenue Ratio set to 25 percent if greater than 25 percent. 5 - Base Model with transformed Taxable Revenue Ratio and its square. 6 - Base Model with lagged Taxable Revenue Ratio variables. additional measures including resetting all values of Taxable Revenue Ratio to 25 percent when the value is above 25 percent rate (fourth column of Table 5), as well as including a squared measure of Taxable Revenue Ratio rather than partitioning the variable at a value of 25 percent (fifth column of Table 5). The results for these models do not change the general conclusions of our results. 691

18 NATIONAL TAX JOURNAL Endogeneity We recognize the possibility of an endogenous relationship between subsidiary choice and taxable revenues. It is possible that the amount of taxable revenues earned is a function of the subsidiary choice, rather than the other way around, as modeled in our empirical tests. Although it is never possible to completely rule out endogenous relationships, it is possible to address them, and the final column in Table 5 presents one attempt. In this model we add lagged values of the Taxable Revenue Ratio (based on prior year s data). The presence of the lagged variables does not alter the interpretation of our primary analysis (although the Taxable Revenue Ratio > 25 percent becomes statistically different from zero). We also conducted (but do not present) an instrumental variables analysis in which the Taxable Revenue Ratio is instrumented with variables intended to capture the demand for taxable output that are independent of subsidiary choice. The instrumental variables include one period lagged values of Taxable Revenue Ratio, gross state product, state disposable income, and state population. The results of the instrumental variables analysis do not alter the general tenor of our results. TAXABLE SUBSIDIARY CREATION AND DISSOLUTION Because our data is a panel, we can perform targeted analyses around the creation and dissolution of taxable subsidiaries. Switching between unrelated businesses and subsidiaries is likely costly and, thus, firms will consider the long term implications when making this choice. By examining changes in the ratio of unrelated business income to the sum of exempt revenues plus unrelated taxable revenues across the creation or dissolution of a subsidiary, we can shed light on the source or disposition of the taxable activity. When a nonprofit starts up a new taxable subsidiary, the activity in the subsidiary can originate in one of three ways. First, it could be a completely new business. Second, it could be a previously operated unrelated business that was transferred into the subsidiary. Third, it could be a previously operated exempt activity that became taxable. 17 Likewise, when a nonprofit dissolves a taxable subsidiary, the taxable activity it once contained can suffer one of three fates. First, it could be abandoned. Second, it could be transferred to the parent and operated as an unrelated business. Third, it could be transferred to the parent and operated as an exempt activity (assuming the activity in the subsidiary became non taxable). We attempt to distinguish between these multiple alternatives by examining the change in the ratio of unrelated business revenues to the sum of unrelated business revenues plus exempt revenues around the formation and dissolution of a subsidiary. With respect subsidiary formation, a decrease in the ratio suggests that activities from one or more existing unrelated businesses were transferred into the newly formed taxable subsidiary. A constant ratio is consistent with the newly formed subsidiary containing entirely new activities. An increase in the ratio implies that the taxable activities contained in the newly formed subsidiary were previously operated as an exempt activities, but became taxable and were placed into a taxable subsidiary. We develop a similar set of propositions for subsidiary dissolutions. An increase in the ratio suggests that the activities previously contained in the subsidiary 17 Taxable activities can become non taxable (and vice versa) for several reasons including rewriting the corporate charter to specifically include the activity as mission related, charging lower prices for the goods and services, using volunteer rather than paid labor, or offering the products only to specific groups of customers. 692

19 Forum on Tax Research in Economics and Accounting were transferred to unrelated businesses. A constant ratio is consistent with the subsidiary activities being abandoned by the parent nonprofit. A decrease in the ratio implies that the taxable activities previously operated in the subsidiary became non taxable and were transferred to the parent and operated as exempt. Research Design We start the analysis with the data sample used for our previous analyses. We retain observations that have four years of data before and after the event. These restrictions reduce the number of usable observations greatly, but the benefit is that we can conduct more powerful and reliable F tests in the periods around the events. 18 To determine whether the ratio of unrelated business revenue to the sum of exempt and unrelated business revenues changed around the event of starting and dissolving a taxable subsidiary, we estimate the following model: [2] Unrelated Business Revenues Exempt Revenues + Unrelated Business Revenues = β 1 t 4 + β 2 t 3 + β 3 t 2 + β 4 t 1 + β 5 t + β 6 t 1 + β 7 t 2 + β 8 t 3 + β 9 t 4 + ε. In this model, observations are lined up in event time where the indicator t is equal to one in 1995, which is the year the subsidiary was either started or dissolved, and zero otherwise. The variables t 4 through t +4 are similarly coded indicator variables that represent the time period of the observations relative to year in which the subsidiary was started or dissolved. We estimate F tests for differences in various combinations of coefficient estimates at various times around the start of the taxable subsidiary at t. The dependent variable is adjusted for the growth in gross domestic product to allow for the effects of increasing revenue over time due to general inflation and economic growth. The descriptive statistics for the variables are in Table 6. The top section of the table reports on the trends of unrelated business revenues around the time of subsidiary creations and documents that unrelated business revenues (as a percentage of total revenues) appears to fall after a subsidiary is created. The average ratio in the four years prior to a new subsidiary is.13, while the average ratio in the four year period after a new subsidiary is created falls to.65, a decrease of.38, or 37 percent. This suggests that at least some taxable activities operated as unrelated businesses are transferred to the newly created subsidiaries. Finally, the trends suggest that unrelated business revenues are growing sharply over the period just before starting up a new subsidiary, but those revenues on average are still well below the IRS danger threshold of 25 percent. This is consistent with the findings in our previous logit analyses in the sense that nonprofits appear to place their unrelated businesses into taxable subsidiaries when they expect the amounts of taxable revenues to increase, thus increasing the chance of losing their exempt status. The extent to which these differences are statistically significant will be considered in the regression analysis. Turning to the subsidiary dissolution data in the lower half of Table 6, there appears to be very little change in unrelated business revenues across the dissolution of a subsidiary. This suggests that when a subsidiary is dissolved, the taxable activity it contained was abandoned. Interestingly, it appears that the amount of revenues earned in taxable subsidiaries 18 It is possible that some nonprofits created and dissolved a subsidiary in the same year. To the extent that the new subsidiaries have similar ownership percentages, total revenues, and total assets as the new subsidiaries, we cannot detect such changes. 693

20 NATIONAL TAX JOURNAL New Subsidiary Analysis Ratio of SUB to total revenue: t+1 t+2 t+3 t+4 TABLE 6 DESCRIPTIVE STATISTICS FOR TAXABLE SUBSIDIARY LIFE CYCLE TESTS Mean Median Std. Dev Ratio of UB to total revenue: t 4 t 3 t 2 t 1 t t+1 t+2 t+3 t+4 Dissolved Subsidiary Analysis Ratio of SUB to total revenue: t 4 t 3 t 2 t Ratio of UB to total revenue: t 4 t 3 t 2 t 1 t t+1 t+2 t+3 t Notes: UB is unrelated business revenues, SUB is subsidiary revenues, as defined in Table 2. The time indices refer to the number of years relative to the event of starting or dissolving a taxable subsidiary. The sample was restricted to observations that appear in all years from t+4 to t 4. There are 152 new subsidiaries created in the panel sample of 1,729 observations for the new subsidiary analysis, and 61 existing subsidiaries dissolved in the panel sample of 684 observations for the dissolved subsidiary analysis. rises sharply just prior to dissolving the subsidiary. One plausible explanation is that the nonprofit is winding up the activities in the subsidiary just prior to dissolving it and is generating additional revenues from asset dispositions and unwinding accounting reserves. 19 The results of the regression analyses in Table 7 Panel A, as well as a series of F tests reported in Panel B support the univariate observations just discussed. With respect to subsidiary creations, the ratio of unrelated business revenues to the sum of unrelated business revenues and exempt revenues falls after a new subsidiary is created. This finding supports the notion that on average taxable subsidiaries contain activities that were previously operated as unrelated businesses. This finding is consistent with the idea that nonprofits transfer their unrelated businesses into taxable subsidiaries, possibly 19 When a business is dissolved and assets are disposed or sold, any previously deferred gain must be recognized. The sources of these potential gains include gains on asset sales due to low basis caused by accelerated depreciation, legal liability reserves, and potential tax liability reserves. 694

21 Forum on Tax Research in Economics and Accounting Panel A: Regression Analysis Constant t 4 t 3 t 2 t 1 t t +1 t +2 t +3 t +4 Observations Number of Groups Within Group R 2 TABLE 7 TAXABLE SUBSIDIARY LIFE CYCLE TESTS Event Test Around Newly Formed Subsidiaries.6 (8.12).3 (2.96).2 (2.55).3 (3.1).8 (1.57).2 (2.28).1 (.82). (.49). (.2). (.16) 1, Event Test Around Dissolved Subsidiaries.9 (7.79).1 (.46).1 (.66).1 (.78).1 (.8).2 (1.26).2 (1.1).2 (1.8).1 (1.6). (.15) Note: The dependent variable is equal to the amount of unrelated business revenue earned by the parent nonprofit (line 14), scaled by total nonprofit revenues (i.e., line 12, which is the sum of exempt revenues and unrelated business revenues). The independent variables t 4 through t+4 are time period indicators equal to the year relative to the event of starting a new subsidiary or dissolving an existing subsidiary. All models use firm fixed effects. Robust and cluster corrected (by nonprofit) two tailed t statistics are in parentheses. Panel B: F Tests for Changes in Unrelated Business Revenue t 1 = t +1 p value t 2 + t 1 = t +1 + t +2 p value t 3 + t 2 + t 1 = t +1 + t +2 + t +3 p value t 4 + t 3 + t 2 + t 1 = t +1 + t +2 + t +3 + t +4 p value t 4 = t 3 p value t 3 = t 2 p value t 2 = t 1 p value t +3 = t +4 p value t +2 = t +3 p value t +1 = t +2 p value 2.17 (.143) 3.36 (.69) 4.37 (.38) 6.7 (.15).71 (.42) 2.42 (.122) 1.67 (.198).8 (.78) 2.45 (.119) 1.12 (.291). (.982). (.975). (.95). (.992) 1. (.32).3 (.87).1 (.97).75 (.391).4 (.531). (.971) Note: F tests are for differences in coefficient estimates of models in Table 6. Related p values are presented underneath F tests in parentheses. A statistically significant p value suggests that the tested coefficients are not equal. All times are stated relative to the event of starting up a new taxable subsidiary or dissolving an existing subsidiary.

22 NATIONAL TAX JOURNAL in expectation of growing those activities to the point where they might endanger exempt status if conducted as unrelated businesses. The second column of Table 7 Panel B reports results for subsidiary dissolutions and shows that the ratio of unrelated business revenues to the sum of unrelated business revenues and exempt revenues does not change after a subsidiary is dissolved. This finding suggests that the taxable activities once contained in the former subsidiary are abandoned. In summary, the taxable activities contained in subsidiaries appear to flow one way (from unrelated businesses into subsidiaries), but not the other way around. This suggests a specific life cycle for the average taxable activity, starting off as a small unrelated business, moving into a taxable subsidiary as it grows, and eventually abandoned when it becomes no long viable or necessary. DISCUSSION AND POLICY IMPLICATIONS Use of Taxable Subsidiaries In some sense it is not surprising to find that taxable subsidiaries are a common way to carry on taxable activities. One significant benefit of using a taxable subsidiary is that a nonprofit s exempt status is jeopardized when the organization earns modestly large amounts of unrelated business revenues, but that same organization will not lose its exempt status no matter how much taxable revenues it generates via a controlled subsidiary. Perhaps the more interesting question is why so many nonprofits continue to operate their taxable activities as unrelated businesses, and we believe our analysis provides some answers. There are costs of using a subsidiary, including reduction of tax planning ability, as well as the general costs of setting up and maintaining a separate corporate structure. Only when the amounts of taxable revenues 696 grow relatively large do the benefits of the subsidiary form overcome these costs. The Unintended Consequences of the Pension Protection Act of 26 One section of the Pension Protection Act of 26 is devoted to nonprofit organizations, with one sub section requiring that all nonprofits make their IRS 99 Ts (which contain detailed information on unrelated businesses) publicly available. The rationale is that the public deserves to be informed about all of the activities of organizations that receive substantial public support via tax subsidies. However, this new law does not require that nonprofits make their subsidiary tax returns (i.e., 112s or 165s) public. The path paved to increase public awareness of nonprofits taxable activities may have been well intended, but, like many such paths, is likely to lead elsewhere. The cost of conducting an unrelated business has risen due to required public disclosure relative to taxable subsidiaries, and, as a result, many nonprofits will have an additional incentive to shift their existing unrelated business activities into taxable subsidiaries, which have no public disclosure requirement. So the law intended to increase disclosure of nonprofits taxable activities may lead to less disclosure if unrelated business activities are placed into taxable subsidiaries. Future studies can calibrate the extent to which unrelated business activity declines (and taxable subsidiary activity increases) in the coming years. CONCLUSION Taxable subsidiaries constitute an economically important component of nonprofit operations, and these subsidiaries generate as least as much taxable revenue as do unrelated businesses. To date there has been little examination of taxable subsidiaries. Our analysis provides a first look

23 Forum on Tax Research in Economics and Accounting at these organizations and finds several factors associated with their use. Several interesting policy issues arise from our analysis. Is it reasonable that a nonprofit will lose its exempt status if it earns a modest amount of taxable revenues directly via an unrelated business but will not lose its exempt status no matter how much taxable revenue it earns indirectly via a controlled subsidiary? Why is it that nonprofits must begin to publicly disclose their unrelated business income tax return (the IRS 99 T) but do not have to disclose their subsidiaries corporate income tax return (the IRS 112)? These raise a deeper policy issue. How far should we permit our nonprofit organizations to delve into commercial unrelated activities? These activities are a two edged sword in that they raise much needed additional capital but do so at the risk of distracting the organization from its primary tax exempt purpose. Our results suggest that taxable subsidiaries play an important role in how nonprofits structure their taxable activities. The larger issue is how consistent taxable activities are, however structured, with the overall scheme of nonprofit organizations in society. Future research on whether taxable activities conducted in any form have any effect on nonprofit behaviors and whether these effects vary across unrelated businesses and taxable subsidiaries would be valuable. REFERENCES Borden, Bradley T. Limited Liability Companies as Exempt Organizations. Estates, Gifts and Trusts Journal 33 No. 3 (May, 28): 15. Box, George E.P., and David R. Cox. An Analysis of Transformations. Journal of the Royal Statistical Society, Series B 26 No. 2 (1964): Cordes, Joseph J., and Burton A. Weisbrod. Differential Taxation of Nonprofits and the Commercialization of Nonprofit Revenues. 697 Journal of Policy Analysis and Management 17 No. 2 (Spring, 1998): Desai, Mihir and Robert Yetman. Constraining Managers Without Owners: Governance of the Not for Profit Enterprise. Harvard University. Mimeo, 28. Fisman, Raymond, and R. Glenn Hubbard. Precautionary Savings and the Governance of Nonprofit Organizations. Journal of Public Economics 89 No. 11/12 (December, 25): Hines, Jr., James R. Non Profit Business Activity and the Unrelated Business Income Tax. Tax Policy and the Economy 13 No. 1 (June, 1999): Hofmann, Mary Ann. Tax Motivated Expense Shifting by Tax Exempt Associations. Journal of the American Taxation Association 29 No. 1 (Spring, 27): Internal Revenue Service. Compendium of Studies of Tax Exempt Organizations, Statistics of Income 3 No (2). Krishnan, Ranjani, Michelle H. Yetman, and Robert J. Yetman. Expense Misreporting in Nonprofit Organizations. The Accounting Review 81 No. 2 (April, 26): Omer, Thomas C., and Robert J. Yetman. Near Zero Taxable Income Reporting by Nonprofit Organizations. Journal of the American Taxation Association 25 No. 2 (Fall, 23): Omer, Thomas C., and Robert J. Yetman. Tax Misreporting and Avoidance by Nonprofit Organizations. Journal of the American Taxation Association 29 No. 1 (Spring, 27): Peterson, Mitchell. Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches. Review of Financial Studies (forthcoming). Plunkett, J. Patrick, and Heidi Christianson. The Quest for Cash: Exempt Organizations, Joint Ventures, Taxable Subsidiaries, and Unrelated Business Income. William Mitchell Law Review 31 No.1 (October, 24): 1 52.

24 NATIONAL TAX JOURNAL Sansing, Richard. The Unrelated Business Income Tax, Cost Allocation, and Productive Efficiency. National Tax Journal 51 No. 2 (June, 1998): Sansing, Richard. Joint Ventures Between Non Profit and For Profit Organizations. Journal of the American Taxation Association 22 (Supplement, 2): Steuerle, C. Eugene. When Nonprofits Conduct Exempt Activities as Taxable Enterprises. The Urban Institute, Emerging Issues in Philanthropy 4 (May, 21): 1 5. Tenenbaum, Jeffrey. Subsidiaries and Related Foundations: Maximizing the Returns & Minimizing the Risks to your Association. The Exempt Organization Tax Review 14 No. 1 (July, 1996): Yetman, Robert J. Tax Motivated Expense Allocations by Nonprofit Organizations. The Accounting Review 76 No. 3 (July, 21): Yetman, Robert. Nonprofit Taxa ble Activities, Production Complementarities, and Joint Cost Allocations. National Tax Journal 56 No. 4 (December, 23): Yetman, Michelle H., and Robert J. Yetman. The Effect of Nonprofits Taxable Activities on the Supply of Private Donations. National Tax Journal 56 No. 1 (March, 23): Yetman, Michelle H., Robert J. Yetman, and Brad Badertscher. Calibrating the Reliability of Publicly Available Nonprofit Taxable Activity Disclosures: Comparing IRS 99 and IRS 99 T Data. Nonprofit and Voluntary Sector Quarterly (forthcoming). 698