Journal of Public Economics 118 (2014) 120–127

Contents lists available at ScienceDirect

Journal of Public Economics

j ourna l homepage: www.e lsev ie r .com/ locate / jpube

To own or not to own your life insurance policy?

David Joulfaian ⁎US Department of the Treasury, 1500 Pennsylvania Avenue, NW, Washington, DC 20220, USA

⁎ Tel.: +1 202 622 0234.E-mail address: david.joulfaian@treasury.gov.

http://dx.doi.org/10.1016/j.jpubeco.2014.07.0040047-2727/Published by Elsevier B.V.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 30 October 2012Received in revised form 7 July 2014Accepted 7 July 2014Available online 14 July 2014

JEL classification:D64G22H24H31

Keywords:Life insuranceBequestsEstate tax

Life insurance proceeds are generally subject to the estate tax. An exception is when the policy is owned by thebeneficiaries and the insured gives up ownership and control, including the ability to change beneficiaries.Should the insured strategically own the policy contract and potentially subject proceeds to estate and inheri-tance taxes, or relinquish control, with the beneficiaries owning the policy, and escape such transfer taxes?This paper addresses how the estate tax influences the choice of life insurance ownership. Using samples of estatetax returns, the empirical evidence suggests that those facing high estate tax rates are more likely to forgo own-ership and have proceeds excluded from their estates, and provides further evidence on the incentive effect oftaxes and in support of the strategic bequest motive.

Published by Elsevier B.V.

1. Introduction

The adequacy of life insurance and the factors that shape thedemand for it over the life cycle are often the key questions of interestin the literature (e.g., Auerbach and Kotlikoff, 1991; Bernheim et al.,2003). This demand may be motivated by the desire to leave bequests(Yaari, 1965; Campbell, 1980). It may as well be shaped by income taxconsiderations in the case of whole life contracts, as the inside buildup is not taxable, or by liquidity constraints faced in disposing of one'sestate at death and succession planning (Holtz-Eakin et al., 2001; Hau,2000; Verdon, 2008).

Life insurance coverage is also employed in gauging the strength ofbequest motives (Bernheim, 1991). Little attention, however, is paidto the nature of these motives (e.g., altruistic vs. strategic), and howthis may shape the form of ownership of life insurance contracts. Aparent, or the insured in general, may give up ownership of a policycontract to designated beneficiaries and with it the ability to changesuch beneficiaries in the future. Alternatively, the insured may retainthe ability to change beneficiaries at any point in time. In the latter,and because the insured maintains ownership, proceeds pass throughthe estate and are potentially subject to the US estate tax. In the former,

and because the policy is owned by the beneficiaries, proceeds bypassthe estate and go directly to the heirs free of tax.

To be more specific, an individual may acquire an insurance policy,pay the premiums every year (or lump sum in a single premium), andthen have the proceeds included in their estate and potentially subjectto estate taxation. Alternatively, the individual may transfer ownershipto the recipients (preferably in the form of a life insurance trust), andthereby have the proceeds bypass the estate tax. For a given contract,if bequests are altruistic and the objective is to maximize the net oftax bequests, then the beneficiaries (trusts) should own the contracts.On the other hand, if bequests are strategic as in Bernheim et al.(1985), then the insured (e.g. parent) may wish to maintain control,and with it the ability to change beneficiaries. But this ownership sub-jects the proceeds to estate taxation. With the top estate tax rate wellover 50% during much of the past seven decades, the implications ofthe form of ownership are not trivial.

In many ways the ownership question is reminiscent of the findingsin the literature on the timing of transfers and their allocation betweenlifetime or inter vivos gifts and bequests (Bernheim et al., 2004;Joulfaian, 2005; McGarry, 1999; Page, 2003). The key distinction hereis in the timing of the designation of the recipients of life insuranceproceeds. The various bequest motives have different implications forthe form of ownership as the estate tax can be avoided if the insuredis willing to part away with ownership and control in designating theultimate beneficiaries.Which form is pursued is ultimately an empiricalquestion. In this paper, I explore the determinants of the form of life

121D. Joulfaian / Journal of Public Economics 118 (2014) 120–127

insurance ownership with particular attention to the role of the estatetax. To my knowledge, I believe this paper is the first to examine thissubject.1

Section 2 of this paper provides a brief description of the estate taxtreatment of life insurance policies. Section 3 addresses issues relatedto modeling the ownership decision, as well as data sources. Adminis-trative records consisting of estate tax returns of decedents in 1989through 2003 are employed, which provide information on life insur-ance proceeds included as well as amounts excluded from the estate,in addition to the size and composition of wealth and demographics.Empirical findings are reported in Section 4. The results suggest thattaxes are an important consideration in shaping the form of ownership.Section 5 concludes.

2. A brief review of the estate tax treatment of life insurance proceeds

The estate tax was enacted in 1916, and life insurance proceeds firstbecame taxable under the Act of 1918. Under the Act, proceeds frompolicies owned by the decedent, plus proceeds in excess of $40,000from life insurance policies owned by others (on the life of the insured),were included in the gross estate and potentially subject to the estatetax. This remained the law until the Act of 1942, when all proceedsfrom policies where the decedent paid the premiums or had an inci-dence of ownership (“premium paid” test) were also made taxable.2

With the introduction of the Internal Revenue Code in 1954 and theconsolidation of the various acts, the “premium paid test”was droppedand only proceeds from policies directly owned by the insured, orwhere the insured had any control, remained taxable and includablein the estate.

Current law continues the tax treatment of life insurance introducedin 1954. More specifically, insurance proceeds are subject to the estatetax only when the insured owns or in any way maintains any controlover the life insurance policy contract. The latter may include the rightto change the listed beneficiaries; the right to borrow against the policyor pledge it as a collateral for a loan; the right to assign policy owner-ship; and the right to cancel the policy or, in the case of whole life insur-ance, surrender it for its cash value.

In order to avoid the tax, and at itsmost basic level, the policy shouldnot name the estate of the insured as the beneficiary and avoid any in-cidence of ownership. An insuredmay transfer ownership of an existinginsurance policy to the beneficiaries. Alternatively, the insured mayform an irrevocable life insurance trust to benefit the heirs and transferownership to it. The same process can be extended to transfers at thetime a new policy is acquired. Insurance trusts typically cost about$1000–2000 to draft and implement. And so it is not too difficult orvery costly for thewell off to exclude life insurance proceeds from an es-tate and avoid taxation provided that the insured is willing to give upcontrol.3

The insured need not assign ownership to beneficiaries at the timeofthe signing of the life insurance contracts, as they are able to transferownership at any point in time in the future. In the case of whole life in-surance policy, the cost of transferring an existing policymay increase ifthe owner (insured) decides to change ownership a few years into thecontract as gift taxes may apply on the cash value of the policy (insidebuild-up). Gift taxes may apply if this value exceeds the annual exclu-sion ($10,000 or $20,000 per couple, indexed) and the size of exemptedestate ($600,000 for most of the period studied in the paper).4

1 Also, it is not clear from the literature whether previous studies of the demand for lifeinsurance address insurance contracts not owned by the insured.

2 The Act repealed the $40,000 life insurance exemption from policies owned by bene-ficiaries. At the same time it expanded the general exemption of $40,000 that applied to allestates to $60,000; this was mostly a revenue neutral change.

3 Dynastic trusts and other complex arrangements can be costlier to set up.4 The annual exclusion in 2013 is $14,000 or $28,000 per couple, and the size of the

exempted estate is $5,120,000.

Gift taxes may also apply to the transfers made to trusts to pay forthe premiums. Again, these premiums have to exceed the annual exclu-sion and the size of the exempted estate before triggering the gift tax.Insurance premiums vary by the age and health of the insured, as wellas the type of insurance (term vs permanent). Consider a 70 year oldmale entrepreneur in excellent health. The premium on a $5 million10-year term policy contract is about $45,000 per year. If the insuredtransfers $45,000 per year to the insurance trust, the annual gift taxexclusion will shelter part of the transfer from the gift tax. In time, thecumulative transfers in excess of the annual exclusion will eat into theestate tax exemption. Nevertheless, there are elaborate schemes thatcan be devised to avoid potential gift taxes on transfersmade to a life in-surance trust to pay the insurance premiums (Johnston, 2002). Moreimportantly, and at a tax rate of 55% for a taxable estate of$3 million, the $5 million insurance proceeds will trigger an additionalestate tax of $2.75 million when the insured has direct ownership.

3. Modeling the form of life insurance ownership

3.1. The role of preferences

Consider a bequestmotivated individualwhowishes to leave behindlife insurance policy proceeds L to benefit his daughters. If the insuredholds or owns a fraction α of the policies directly, then his estate willpay an estate tax of ταL, where τ is the estate tax rate. The daughtersreceive (1− τ)αL from policies held by the insured owner, and receivethe remaining (1− α)L from policies they own, or which are held by anirrevocable trust set up on their behalf and out of the control of theinsured, free of tax.

The choices are clear: hold the insurance contract directly and havethe estate pay a tax on proceeds, or forgo ownership and control andhave the proceedsflow free of tax to the daughters. The insured choosesfrom these two forms of ownership so as to maximize his utility:

max U O;B½ � ð1Þ

where O= (1− τ)αL represents ownership by the insured, and B=(1− α)L represents ownership by the beneficiaries.Whendifferentiatedwith respect to α this yields:

1−τð ÞUO−UB ¼ 0 ð2Þ

or,

UB

UO¼ 1−τ: ð3Þ

The insured allocates contracts between the two forms of ownershipat the pointwhere themarginal rate of substitution is equal to the net oftax rate. The heirs receive only (1 − τ) of every dollar in life insuranceproceeds from contacts controlled by the insured, but this loss inbequests is offset by the greater utility derived from controlling thepolicy when bequests are, say, strategic (UO N UB).

The above framework can be modified to focus on the allocation ofpremiums rather than proceeds. But because the underlying contractsand terms are identical except for the nature of the owner, their alloca-tion should be identical. Another possible extension is to introducefinancial education where those acquiring large insurance contractsL may receive better and most likely free advice related to ownershipforms, thereby rendering α endogenous to L.5 This is not dissimilarfrom the effect of financial education on retirement savings (Bayeret al., 2009; Bernheim and Garrett, 2003; Choi et al., 2002).

5 The size of contracts (L) may also mitigate the effects of the fixed cost of setting up atrust for those whowish to forgo ownership but prefer not to hand over the policy to ben-eficiaries directly.

122 D. Joulfaian / Journal of Public Economics 118 (2014) 120–127

In addition, the age of the insured may critically shape preferencesand timing of transfers of ownership. Assume, as in Bernheim et al(1985), that the parent (insured) induces his children to provide hima stream of services st…st + n through his n years in retirement in ex-change for bequests. Then, as the insured gets older and consumesmore of the services provided by the children, the uncertainty aboutthe ultimate beneficiaries is reduced and he is more likely to transferownership. Direct ownership should become rarer as individuals age.Alternatively, the bequest motives can be altruistic and the parent,reflecting on the uncertainty of the future well being of each of hisdaughters, may postpone the transfer of ownership. Here, again owner-ship will decline with the age of the insured.

The business share of estate may also influence the life insuranceownership share depending on whether the parent plans to pass thebusiness on to his children (heirs) to manage and run. If the parenthas no such plans, then the business share will likely lead to a reductionin the parent's share of direct insurance ownership. Under buy/sellagreements, for instance, the parent and his business partners buyinsurance policies on each other's lives. At death, the survivingpartner(s) buys out the share of business owned by the deceased. Thisbuy/sell option is identical to having the heirs own the life insurancepolicy but with the added benefit of protecting the value of the businesswith active and uninterruptedmanagement (with no concerns over anypossible gift tax on premiums payments).

A similar outcome is expected if, alternatively, life insurance cover-age is acquired to insure against future income shocks (e.g. death ofbusiness founder), and the parent cares (dynastically or otherwise)about the survival of the business as an ongoing enterprise owned andmanaged by the heirs; transferring life insurance ownership insuresthe availability of greater (tax-free) resources and liquidity to theheirs to finance the enterprise. In addition, directly owning the life in-surance policy and keeping the proceeds within the estate not onlymakes less cash available from life insurance (after-tax), it may alsomake the estate less liquid in meeting its tax obligations.6

3.2. Data sources and variable construction

I employ data from estate tax returns prepared by the Statistics ofIncome Division of the Internal Revenue Service. Estate tax returnsprovide information on wealth and its disposition for the wealthiestindividuals. Information is available on asset categories, debts, andexpenses of settling the estate such as attorney fees and executor com-missions. They also provide information on jointly held assets and lifeinsurance proceeds not included in the estate.7 Demographic informa-tion is available on age, marital status, and gender.

More specifically, the data in this paper are drawn from the estatetax returns of decedents in 1989 through 2003, before the large increasein the filing threshold. The sample is limited to estates over the legal

6 The estate tax attributable to the business can be deferred and paid in installmentsover 14 years (interest only during first 4) at preferential rates provided that the businessmakes up at least 35% of the adjusted estate (gross estate less debts and estate expenses).The gross estate becomes largerwhen insurance is held by the insured and can potentiallyjeopardize the installment payment method. This can be illustrated with a simple exam-ple. Consider a hypothetical estate in 1995 valued at $5 million, consisting of a houseworth $2 million, $2 million business, and $1 million in liquid investments, with no debtor estate expenses; there is an insurance policy with face value of $2 million owned bythe beneficiaries and excluded from estate. The tax liability of this estate is $2,198,000.The business makes up 40% of the estate, and accordingly 40% of the tax liability is de-ferred; only $1,320,800 needs to be paid with the tax return. With the heirs having re-ceived $2 million in insurance proceeds, plus the $1 million in liquid assets, there are noliquidity concerns in funding the business or paying the tax. In contrast, if the insuredhad owned the life insurance policy, the reported estatewould have been larger by $2mil-lion, and the tax liability growing to $3,298,000.With the business share of the estate nowat less than 35%, its share of the ensuing tax liability can no longer be deferred; the estatewill have to exhaust the insurance proceeds, the liquid assets, and liquidate either thehouse or the business to pay the tax.

7 Proceeds from contracts on the life of the insured but owned by others is reported onForm 712 by insurance companies which is attached to the estate tax return, Form 706.

filing threshold with total assets over $600,000 in 1989 through 1997;$625,000 in 1998, $650,000 in 1999, $675,000 in 2000–2001, and$1 million in 2002–2003. I exclude observations where the individualresided outside the US, as well as those under the age of 30. Observa-tions with negative wealth are also excluded.8

3.2.1. Variable constructionThe share of insurance proceeds frompolicies owned by the insured,

the key variable of interest α, is defined as the ratio of life insuranceproceeds included in the estate (taxable) to the sum of all life insuranceproceeds included and excluded (tax exempt) from the estate. The keyexplanatory variable, the tax rate, is computed as the average tax rateassuming all insurance proceeds are taxable regardless of whetherthey are included or not, and is thus exogenous to the form of owner-ship. More specifically, the tax on insurance proceeds is computed asthe difference in tax liability on wealth defined to include all proceedsand that defined to exclude all proceeds, regardless of form ofownership.9

Reflecting on the estate planning process, and following Kopczukand Slemrod (2001), tax rates are computed using the tax law in effect10 years before the date of death as the date of insurance contract ortransfer of policy is not known. Estate tax rates varied considerably dur-ing the sample period. Prior to 1982, the maximum tax rate was 70%which was reduced in steps to 50% by the Economic Recovery Tax Actof 1981. Similarly, the size of the exempted estate was increased from$120,000 to $600,000 in steps. In addition, the deduction for spousal be-quests was increased from the greater of $500,000 and 50% of the estateto an unlimited deduction. Furthermore, the special use valuation ex-emption for businesses was increased from $500,000 to $750,000. Abubble marginal tax rate of 60% was created in 1987 as the benefit ofthe graduated tax rates and the unified tax credit were phased out.Combined, these changes introduced considerable variations in taxrates independent of the size of the estate.

In the presence of phased-in legislated changes, the fully phased-intax rates and exemptions are employed. Thus for decedents in 1992,the fully phased-in tax changes anticipated in 1982 are used. Becausethe maximum tax rate declined from 70 to 65% in 1982, and wasplanned to decline to 50% by 1985, the relevant tax rate is assumed tobe 50%. Similarly, the size of the exempted estate was expanded insteps to $600,000. The maximum tax rate was temporarily frozen at55% in 1984, again in 1987, and permanently in 1993. Thus, for these pe-riods, the maximum tax rate is set at 55%.

A number of control variables that may further shape preferencesare considered. These include demographic variables such as age, sex,and marital status, as well as business share of estate, the size of insur-anceproceeds, andwealth. The value of businesses is defined as the sumof farms, closely held corporations, real estate partnership, and othernon-corporate businesses, and the business share variable is set equalto the value of businesses divided by the gross estate (plus excludedinsurance); the value of businesses and the gross estate are expandedby adding back the special use valuation discount.10

Table 1 provides descriptive summary statistics. There are 178,305observations with mean wealth of $5.8 million (sd = 31), and meanage of 71.4 years at death (sd = 15.4). The mean value of proceeds(face value) from insurance policies is $290,882; conditional on having

8 Wealth is defined as bequeath-ablewealth and set equal to the gross estate less debts,plus excluded insurance proceeds. Post 2003, the estate tax filing population shrank con-siderably following the sizable increase in the size of exempted estates (Joulfaian, 2012,Table 2.4 and Table 4.1). Note that because thefiling threshold is based on the size of grossassets, low net worth or nontaxable estates are common in the data.

9 The tax liability is computed before applying the credit for state death taxes so that,say in 1995, the maximum statutory tax rate is 0.55 and not 0.39 (i.e. 0.55 less the maxi-mum credit rate of 0.16) for the wealthiest of estates; credit is set to zero.10 Business and farm owners are able to value their land based on the capitalized incomeand not the market value; the difference, up to $750,000 (indexed) in the sample periodstudied, is excluded from the estate.

Table 1Sample descriptive statistics.

All No insurance With insurance

Variable Mean Std. dev Mean Std. dev Mean Std. dev

Wealth 5,757,936 31,100,000 6,076,522 37,900,000 5,573,026 26,400,000Total insurance 290,882 9,590,716 0 0 459,713 12,100,000Included insurance 151,173 444,690 0 0 238,916 539,965Excluded insurance 139,709 9,575,052 0 0 220,797 12,000,000Owned insurance share 0.55 0.49 0.00 0.00 0.87 0.31Age 71.4 15.4 77.4 13.9 68.0 15.1Male 0.62 0.49 0.37 0.48 0.76 0.42Married 0.56 0.50 0.39 0.49 0.66 0.47Widowed 0.30 0.46 0.45 0.50 0.21 0.41Never married single 0.08 0.27 0.10 0.29 0.07 0.26Divorced/separated 0.06 0.24 0.07 0.25 0.06 0.24Business share 0.07 0.17 0.05 0.15 0.09 0.18Tax rate 0.34 0.19 0.40 0.15 0.30 0.20Observations 178,305 65,483 112,822

123D. Joulfaian / Journal of Public Economics 118 (2014) 120–127

any insurance coverage, the mean is $459,713. The mean value of ex-cluded proceeds is slightly smaller, but with greater variance. About62% of the observations are male, 56% married, 30% widowed, and 8%never married singles; the remaining 6% are divorced or separated.Businesses make up 7% of their estates, and the average tax rate oninsurance proceeds is 34%.11

In contrast, the insured, the population of interest, are slightly lesswealthy and younger by some three years. They are more likely to bemale (76%), and married (66%). Businesses make up a greater share oftheir assets (9%), and they face lower tax rates (30%). The mean insur-ance face value is $459,713 of which $238,916 is from policies ownedby the insured; the average share of insurance owned by the insuredis 87%. As shown in Table 2 and Fig. 1, 83% own all of their contracts,with all proceeds included in the estate (93,289 observations). In 5%of the observations, beneficiaries own all of the contracts (5481 estates),with 12% of the remaining observations reporting both forms of owner-ship (14,052 or 12% of the estates).

Table 3 provides further descriptive statistics for different groups inthe sample. The middle panel provides figures for those with insurancecoverage in excess of $50,000. In this sub sample, the insured own asmaller share of the policies, and are more likely to be married, male,and about 5 years younger. The right panel excludes married individ-uals. This subset of individuals own more of the insurance policies, areless likely to be male, face higher tax rates, and are 5 years older.

4. Results

4.1. Basic statistics

As illustrated in Table 2, the profile of the insured varies with theform of ownership. For those who own their contracts, and as reportedin column (2), themeanwealth is $4.4 million withmean life insuranceproceeds of $236,033. Their mean age is 67.6 years, with 76%male, 64%married, 21% widowed, and 8% never married singles. Businesses makeup 7% of their estate, and they face an average tax rate of 28%.

In contrast, thosewho have the beneficiaries own the policies, and asshown in column (3), are wealthier, have greater insurance coverage,are older, less likely to be male, less likely to be married, less likelyto be single, and businesses make up a larger share of their estates(0.13 vs. 0.07). More to the point, they face higher estate tax rates(0.45 vs. 0.28).

As to those who have both forms of ownership, as in column (4),they are slightly wealthier than those who own their policies. Theyhave greater coverage, the bulk of which is excluded from the estate

11 For the uninsured, an imputation of face value is set equal to 8% of the gross estate,roughly the mean for the insured.

and owned by the beneficiaries. They are more likely to be male andmarried, and have greater business presence. The average tax rate ismuch higher than that faced by those who own all of their policies(0.38 vs. 0.28). Along with column (3), this continues to highlight therole taxes play in shaping the form of life insurance ownership.

4.2. Multivariate estimates

To shed further light on the factors that explain thedifferent forms ofownership, Table 4 beginswith a simple OLS estimate of the effect of taxrates on the allocation of life insurance contracts between the two formsof ownership (α). As reported in column (1), the estimated coefficienton the tax rate is negative and significant with a value of −0.424(se = 0.115), consistent with the FOC in Eq. (3).12 This implies that aten percentage point increase in the tax rate leads to 4.2 percentagepoint reduction in the owned share of insurance proceeds.

Column (2) expands on the estimates in column (1) and controls fora number of variables including the size of insurance proceeds, wealth,demographics, business share of estate, and year fixed effects. The esti-mated coefficient on the tax rate is now slightly larger, in absolute value,with a value of −0.496 (se = 0.119). In addition to the effect of taxrates, the ownership share also declines with the size of insuranceproceeds, business share of estate, and age. Compared to never marriedsingles and divorced individuals,married andwidowed individuals owna greater share of their insurance policies, consistent with the basic sta-tistics in Table 2. In addition, men own a greater share of their policies.

Because the key variable is a fraction over the interval (0, 1), the OLSpredictions cannot be guaranteed to lie within this interval.13 FractionalLogit, due to Papke andWooldridge (1996), is next employed in place ofOLS. More specifically, the key variable α for individual i is estimated asa function of a number of explanatory variables x, which include the taxrate τ, or:

E αi=xið Þ ¼ G xiβð Þ; i ¼ 1;…;N; ð4Þ

where 0≤ α≤ 1 again denotes the dependent variable, and G(·) is a dis-tribution function similar to the logistic function G(z) = exp(z)/[1 +exp(z)] which maps z to the (0, 1) interval.

Fractional Logit marginal effect estimates are provided in column(3), which unlike the OLS estimates in column (2) ensure that the pre-dicted values of α lie in the interval (0, 1). The estimated coefficient onthe tax rate retains the sign and significance of the earlier OLS estimates,

12 The sign on the coefficient is reversed if the net of tax rate 1 − τ is employed in placeof the tax rate.13 Its log-odds ratio variant cannot be true if α takes a value of 0 or 1.

Table 2Descriptive summary statistics for the insured by type of policy.

All Owned by insured Owned by beneficiaries Owned by both

Variable Mean Std. dev Mean Std. dev Mean Std. dev Mean Std. dev

Wealth 5,573,026 26,400,000 4,385,884 20,000,000 11,000,000 26,300,000 11,300,000 51,100,000Total insurance 459,713 12,100,000 236,033 492,238 1,131,679 2,917,242 1,682,588 34,100,000Included insurance 238,916 539,965 236,033 492,238 0 0 351,243 835,139Excluded insurance 220,797 12,000,000 0 0 1,131,679 2,917,242 1,331,345 34,000,000Owned insurance share 0.87 0.31 1 0 0 0 0.36 0.30Age 67.95 15.14 67.64 15.46 73.25 12.72 67.99 13.33Male 0.76 0.42 0.76 0.43 0.59 0.49 0.89 0.32Married 0.66 0.47 0.64 0.48 0.61 0.49 0.79 0.41Widowed 0.21 0.41 0.21 0.41 0.30 0.46 0.14 0.34Never married single 0.07 0.26 0.08 0.28 0.03 0.16 0.02 0.14Divorced/separated 0.06 0.24 0.06 0.24 0.06 0.23 0.05 0.22Business share 0.09 0.18 0.07 0.17 0.13 0.22 0.18 0.24Tax rate 0.30 0.20 0.28 0.20 0.45 0.12 0.38 0.16Observations 112,822 93,289 5,481 14,052

6080

nt

124 D. Joulfaian / Journal of Public Economics 118 (2014) 120–127

but is now much smaller in absolute terms, with a value of −0.309(se = 0.027).14 This highlights the bias in employing OLS.

The estimated marginal effects on the control variables, with theexception for singles, are also uniformly smaller in absolute valuewhen contrasted with those of the OLS estimates. The estimated mar-ginal effect of the size of insurance proceeds is −0.042 (se = 0.021),and implies that the share of insurance owned or included in an estatedeclines by 4 percentage points when the size of coverage doubles.

When compared to married, widowed, and divorced individuals,never married single individuals own a greater share of their insurancepolicies. Similarly, young individuals hold a larger share of insurancecontracts directly, consistent with our priors; the ownership share isabout 10 percentage points lower for those aged 75 and older.

Equally noteworthy, and again consistent with our priors, is thereduction in ownership with the size of business share of estates. Theestimated coefficient on the business share is −0.059 (se = 0.012); ifthe business share of an estate increases by 20 percentage points, thenthe ownership rate would decline by about one percentage point.

4.2.1. Exclude insured with low coverageIn an alternative set of estimates, column (3) is replicated using the

sub-sample of the insured with coverage in excess of $50,000. Thisthreshold, though somewhat arbitrary, aims at excluding those whomay find the cost of setting up a trust prohibitive as well as those withemployer provided and paid coverage at no cost to the insured; life in-surance coverage may be incidental to employment (Brown, 2001),and not necessarily reflective of a bequest motive.15

As shown in descriptive statistics in themiddle panel of Table 3, thisgroup has greater insurance coverage, more likely to be married, youn-ger by some 5 years, and has a greater share of their estates in busi-nesses. The marginal effects reported in column (1) of Table 5 for thissub-sample are qualitatively in harmonywith the earlier figures report-ed in column (3) of Table 4. The share owned by the insured declineswith coverage, and is greatest for those not married and men. Theshare also declines with age and share of business in the estate. Moreto the point, it declines with the tax rate, with the estimated marginal

14 Note that all standard errors are clustered by the wealth categories in Table 3 as theyotherwise are likely to suffer from bias due to both the Moulton (1990) problem and theBertrand et al. (2004) problem. Federalmarginal tax rates, for instance, are similar for peo-ple at similar wealth levels. If their errors are also correlated, then standard errors are bi-ased. In addition, the federal estate tax rates are correlated over time for people at a givenwealth level, and if errors are also correlated over time for people at thatwealth level, thenstandard errors are likely to be biased.15 Employer paid premiums on coverage up to $50,000may not be taxable under the in-come tax. However, this adjustment to the sample may not be necessary since employeesare able to assign ownership to someone else. Such “assignment” is typically irrevocable,andmaymandate that the insured continue paying premiums for the duration of employ-ment. For an example of the “assignment” treatment in an employee plan, see http://www.opm.gov/insure/life/reference/handbook/assign2.asp.

effect greater in absolute value than that reported earlier for thefull sample of the insured; the estimated marginal effect of the taxrate is −0.657 (se = 0.070) compared to the earlier estimate of−0.309 (se = 0.027).

4.2.2. Exclude married individualsJoint life insurance coverage among married individuals may not be

adequately captured in the above specifications. This is particularly thecase since the underlying data do not provide information on the natureof insurance policy beneficiaries. As such, and as a further robustnesscheck, observations with married couples are excluded and the aboveestimates reproduced.

The revised estimates for these older individuals in this subsample,reported in column (2) of Table 5, are qualitatively similar to those re-ported earlier for the universe of observations in Table 4. The coefficienton the tax rate, however, is smaller in absolute value with an estimateof −0.168 (se = 0.020) compared to the earlier estimate of −0.309(se = 0.027) in column (3) for the full sample.

4.2.3. Control for preference for trusts and similar arrangementsWhile the acquisition of life insurance represents some evidence of

estate and succession planning, individuals may arrange their affairs inthe later years of their lives in ways life insurance coveragemay not ad-equately capture. Indeed, some individuals may not be comfortablewith the formation of trusts and transferring their wealth to them.This may lead to inadequate planning regardless of the tax benefits oftransferring life insurance ownership to a trust.

020

40Perce

0 .2 .4 .6 .8 1Share

Fig. 1. Share of life insurance proceeds from policies owned by insured.

Table 3Descriptive summary statistics for select groups.

All Insurance N $50,000 Exclude married

Variables Mean Std. Dev Mean Std. dev Mean Std. dev

Owned insurance share 0.8717 0.3077 0.8089 0.3581 0.8925 0.2939Tax rate 0.2968 0.1983 0.2626 0.2070 0.4173 0.1418ln Insurance face value 11.3080 2.0340 12.6399 1.1470 10.5442 2.0818Married 0.6600 0.4737 0.7509 0.4325 – –

Widowed 0.2075 0.4055 0.1260 0.3318 0.6103 0.4877Single 0.0724 0.2591 0.0582 0.2342 0.2129 0.4094Male 0.7634 0.4250 0.8423 0.3645 0.5768 0.4941Age 67.9542 15.1383 63.0533 14.6340 72.8326 16.1929Age b 45 0.0772 0.2669 0.1146 0.3185 0.0731 0.260345 ≤ Age b 55 0.1268 0.3327 0.1811 0.3851 0.0881 0.283555 ≤ Age b 65 0.2056 0.4041 0.2556 0.4362 0.1299 0.336265 ≤ Age b 75 0.2309 0.4214 0.2167 0.4120 0.1797 0.383975 ≤ Age b 85 0.2024 0.4018 0.1489 0.3560 0.2382 0.426085 ≤ Age 0.1572 0.3639 0.0831 0.2761 0.2910 0.4542Wealth b $1 M 0.1928 0.3945 0.1608 0.3674 0.2547 0.4357$1 M ≤ Wealth b $2 M 0.3120 0.4633 0.2930 0.4552 0.3079 0.4616$2 M ≤ Wealth b $5 M 0.2684 0.4431 0.2860 0.4519 0.2272 0.4190$5 M ≤ Wealth b $10 M 0.1146 0.3186 0.1291 0.3353 0.1039 0.3052Wealth N $10 M 0.1121 0.3155 0.1311 0.3375 0.1063 0.3082Business share 0.0884 0.1836 0.1084 0.1969 0.0506 0.1422Observations 112,822 69,081 38,359

125D. Joulfaian / Journal of Public Economics 118 (2014) 120–127

The fraction of the agingwealthywhomake lifetime transfers to rev-ocable trusts and engage in similar arrangements that donot necessarilyconvey tax benefits has been growing over time (Joulfaian, 2010). As anindividual becomes incapacitated, the trust steps in tomanage his assetsand estate as his symptoms progress. At death, the trust provides unin-terruptedmanagement of the estate and avoids probate and the associ-ated expenses, and makes assets immediately available for distribution

Table 4Marginal effect estimates of the included share of life insurance proceeds.

OLS OLS Fractional logit

Variables (1) (2) (3)

Tax rate −0.424⁎⁎ −0.496⁎⁎ −0.309⁎⁎⁎

(0.1152) (0.119) (0.0272)ln Insurance face value −0.0725⁎⁎ −0.0418⁎⁎⁎

(0.0172) (0.00210)Business share −0.163⁎⁎⁎ −0.0592⁎⁎⁎

(0.0139) (0.0101)Married −0.0781⁎⁎ −0.0491⁎⁎⁎

(0.0233) (0.00546)Widowed −0.0137 −0.0122⁎⁎⁎

(0.00798) (0.00285)Single −0.00464 0.0136⁎⁎⁎

(0.00416) (0.00310)Male 0.0804⁎⁎ 0.0357⁎⁎⁎

(0.0208) (0.00284)45 ≤ Age b 55 −0.0160⁎ −0.0105⁎⁎⁎

(0.00597) (0.00304)55 ≤ Age b 65 −0.0658⁎⁎ −0.0389⁎⁎⁎

(0.0171) (0.00714)65 ≤ Age b 75 −0.127⁎⁎⁎ −0.0833⁎⁎⁎

(0.0267) (0.00836)75 ≤ Age b 85 −0.148⁎⁎⁎ −0.107⁎⁎⁎

(0.0264) (0.00612)85 ≤ Age −0.141⁎⁎⁎ −0.112⁎⁎⁎

(0.0247) (0.00999)ln Wealth −0.00838 0.00800⁎⁎

(0.0171) (0.00320)Constant 0.998⁎⁎⁎ 2.072⁎⁎⁎

(0.00159) (0.0796)Year fixed effects No Yes YesObservations 112,822 112,822 112,822R-squared 0.075 0.307

Robust, and clustered by wealth category, standard errors in parentheses.⁎⁎⁎ p b 0.01.⁎⁎ p b 0.05.⁎ p b 0.1.

to the heirs. An added benefit is that such trusts reduce the risk of a willcontest, as well as maintain the privacy of the nature of relationship tothe heirs. Best of all, the ownership of these assets remainwith the indi-vidual until his passing; they are part of his estate and are potentiallysubject to the estate tax.

It is very possible that those who form revocable trusts and similarinstruments are more inclined, or at least more comfortable, to alsotransfer the ownership of life insurance policies to trusts regardless ofthe tax benefits. To test for this “taste” or preference, information onsuch transfers is obtained from Schedule G Lifetime Transfers of thefederal estate tax return.16 The earlier estimates of column (3) inTable 4 are replicated in column (3) of Table 5 with the addition of anindicator for the presence of Schedule G. Adding such an indicatordoes not alter any of the earlier estimates on the effect of taxes andother variables. In addition, the estimated marginal effect of the indica-tor is 0.005 (se = 0.004), and is not measured precisely.

4.2.4. Insurance as a tax shelterAnother way to view excluded life insurance is as a form of a tax

avoidance instrument aimed atminimizing taxes on transfers of wealth,and not just insuring against some unforeseen risk. Conceptually, thefindings on the tax effects under this view should be similar to those re-ported above; those facing higher tax rates will engage in greater taxshelter activity. Oneway to test for this is to redefine the dependent var-iable in the above estimated specifications as the ratio of excluded in-surance proceeds to total assets, with the latter defined to include allinsurance proceeds. This excluded life insurance share of the estatehas a mean value of 0.0225, and reflecting on the narrow ownershippattern observed in the right two panels of Table 2, it is highly skewedwith a standard deviation of 0.077 and a maximum value of 0.971.17

Fractional Logit marginal effects are reported in column (4) ofTable 5. The explanatory variables are identical to those employed

16 See Kopczuk (2007) for a recent use of Schedule G. The term “lifetime transfers”should not be confused with outright lifetime gifts as these mostly represent transfers torevocable trusts and where the aging or progressively sick donor exerts control on theuse of the assets and the designation of the ultimate beneficiaries. The gift tax on outrightgifts paid by those fewunfortunate to diewithin three years of the date ofmaking the giftsis also reported in Schedule G.17 There are, for instance, 2329 observations where excluded insurance proceeds arelarger than one half of the net worth (gross estate less debts) reported on the estate taxreturn; 759 observations where proceeds exceed 100% of net worth. These highlight theimportance of using life insurance as means to sheltering wealth from taxation.

Table 5Alternative estimates of fractional logit marginal effects.

Own share of insurance proceeds Excluded insurance estate share

InsuranceN$50,000

Exclude married Control for trust usea

Variables (1) (2) (3) (4)

Tax rate −0.657⁎⁎⁎ −0.168⁎⁎⁎ −0.311⁎⁎⁎ 0.0662⁎⁎⁎

(0.0704) (0.0198) (0.0265) (0.0125)Tax rate × trust usea 0.00533

(0.00442)ln insurance face value −0.0874⁎⁎⁎ −0.0298⁎⁎⁎ −0.0417⁎⁎⁎

(0.00548) (0.00194) (0.00206)Business share −0.126⁎⁎⁎ −0.0466⁎⁎⁎ −0.0588⁎⁎⁎ 0.0140⁎⁎⁎

(0.0209) (0.0146) (0.0104) (0.00327)Married −0.0940⁎⁎⁎ −0.0489⁎⁎⁎ 0.00939⁎⁎⁎

(0.0118) (0.00545) (0.00231)Widowed −0.0275⁎⁎⁎ −0.00593⁎⁎⁎ −0.0122⁎⁎⁎ 0.00514⁎⁎⁎

(0.00722) (0.00177) (0.00285) (0.00164)Single 0.0286⁎⁎⁎ 0.0105⁎⁎⁎ 0.0137⁎⁎⁎ −0.00853⁎⁎⁎

(0.00493) (0.00224) (0.00319) (0.00134)Male 0.0672⁎⁎⁎ 0.0241⁎⁎⁎ 0.0358⁎⁎⁎ 0.00115

(0.00911) (0.00266) (0.00288) (0.00111)45 ≤ Age b 55 −0.0237⁎⁎⁎ 0.00103 −0.0105⁎⁎⁎ −0.00321⁎⁎⁎

(0.00541) (0.00387) (0.00301) (0.000898)55 ≤ Age b 65 −0.0800⁎⁎⁎ −0.0180⁎⁎⁎ −0.0389⁎⁎⁎ −0.00650⁎⁎⁎

(0.0123) (0.00331) (0.00705) (0.00193)65 ≤ Age b 75 −0.173⁎⁎⁎ −0.0584⁎⁎⁎ −0.0834⁎⁎⁎ −0.0102⁎⁎⁎

(0.0125) (0.00572) (0.00836) (0.00229)75 ≤ Age b 85 −0.221⁎⁎⁎ −0.0723⁎⁎⁎ −0.108⁎⁎⁎ −0.0140⁎⁎⁎

(0.00861) (0.00543) (0.00629) (0.00277)85 ≤ Age −0.223⁎⁎⁎ −0.0716⁎⁎⁎ −0.112⁎⁎⁎ −0.0172⁎⁎⁎

(0.0140) (0.00486) (0.0102) (0.00317)ln Wealth 0.0223⁎⁎⁎ 0.00218 0.00777⁎⁎ 0.00247

(0.00680) (0.00162) (0.00316) (0.00169)Year fixed effects Yes Yes Yes YesObservations 69,081 38,359 112,822 112,822

Robust, and clustered by wealth category, standard errors in parentheses.⁎⁎⁎ p b 0.01.⁎⁎ p b 0.05.

a Schedule G proxy for trusts.

126 D. Joulfaian / Journal of Public Economics 118 (2014) 120–127

thus far except for the exclusion of life insurance proceeds. For this spec-ification, the estimated coefficient on the tax rate variable is 0.066 witha standard error of 0.012. An increase of 20 percentage points in the taxrate leads to an increase of slightly more than one percentage point inthe share of estate that is sheltered from taxation. This continues tohighlight the importance of taxes in shaping the share of insurance pol-icies owned by the beneficiaries.

4.2.5. Other robustness testsIn one additional experiment, liquidity concerns are introduced to

examine whether and to what extent these may shape the observedownership pattern. The question of interest is whether the estate hasadequate amounts of liquid assets such as cash, stocks, and bonds, thatcan be liquidated to payoff the creditors of the estate and the estatetax liability. A dichotomous variable is created that takes the value ofone when liquid assets exceed the obligations of the estate (debts plustaxes).18 When added to the estimated equation in column (3) ofTable 4, the estimated marginal effect of this liquidity indicator is−0.0101 (se = 0.002). While the marginal effect is negative and

18 Because owned insurance proceeds formpart of the estate, the estate tax liability is ad-justed to exclude the share of insurance proceeds in the estate.

statistically significant, it has little economic significance and more im-portantly little effect on the remaining variables.19

In one final robustness test, an alternative measure of the tax rate isemployed. Thus far, the tax rate is computed to reflect law ten yearsprior to the year of death. The attractiveness of this measure, due toKopczuk and Slemrod (2001), is that it is exogenous to the form of lifeinsurance ownership. Nevertheless, it may not be truly representativeof the tax regimes that shape this form. Some individuals may havemade their decisions in increments over time. Others may have doneso in different years. As an alternative, the tax rate measure employedthus far is replaced with a fuzzier measure that averages tax ratesover a number of years. More specifically, the 10 year lagged tax rateis replaced by the average of tax rates lagged 6 to 10 years. As an exam-ple, for an estate in 2000 the relavant tax rate is the average of tax ratesin effect in 1990 through 1994. When the estimates of column (3) ofTable 4 are updated using this alternative measure, the marginal effectof the tax rate variable becomes −0.320 (se = 0.020), slightly larger,in absolute value, than the estimate reported earlier.

19 These estimates,while not reported, are available upon request. In an alternative spec-ification, the indicator is further recoded to take a value of 2 when liquid assets exceedtwice the estate's obligation, and 3 for thrice, and so on. For this specification, the estimat-ed coefficients on the subcategories are positive and turn negative when liquid assets areseveral multiples of the sum of debts and taxes.

127D. Joulfaian / Journal of Public Economics 118 (2014) 120–127

5. Conclusion

This paper addresses the form of life insurance ownership. Morespecifically, it examines the effects of estate taxation on the choice be-tween ownership of insurance policies directly by the insured and thatby the beneficiaries. The former subjects the proceeds to estate taxes,while in the latter proceeds escape taxation but with the insured forgo-ing control. Thefindings suggest that those facing higher estate tax rateshold less of their life insurance directly.

In addition to shedding light on another facet of the distortionaryeffect of taxes, thesefindings have interesting implications for evaluatingbequest motives. For absent taxes, individuals may maintain ownershipand control over life insurance contracts, giving them considerable lever-age over the designation of beneficiaries and providing greater supportfor the exchange motive of bequests.

One noteworthy observation is that, and unlike the existing litera-ture on the demand for life insurance, this paper examines the patternof coverage in effect at the end of life. The observed form of ownershipsummarized above, and coverage in general, may very well divergefrom that prevalent among the living population, whether terminatedat old age or otherwise.

An extension to this paper is to explorewhether thefindings reportedabove are sensitive to the relationship to the heirs. If the composition ofthe heirs, say presence of children, influences the ownership choice,then the findings can be biased because of this omission. This may notsound as an unreasonable assumption given the common findings inthe literature that the demand for life insurance is greatest whendependent children are present (Bernheim, 1991; Lewis, 1989). Andreflecting on Hurd (1987), the bequest motive may very well be stron-gest in the presence of children, although Kopczuk and Lupton (2007)provide a contrary view on this. On the other hand the strategic insuredmay not be willing to give up control regardless of the relationship ofthe heirs, and the latter may not be relevant to explaining the choicesmade by the insured. Ultimately this is an empirical question, albeitone hampered by data limitations.20

Acknowledgments

The paper benefited from comments by the editor and two anony-mous referees, as well as the invaluable research assistance of SiobhanO'Keefe. The views expressed are those of the author and do not neces-sarily reflect those of the US Department of the Treasury.

References

Auerbach, A.J., Kotlikoff, L.J., 1991. The adequacy of life insurance purchases. J. Financ.Intermed. 1 (3), 215–241.

20 The relationship of theheirs is not generally captured in the available samples of estatetax returns. Using a subset of the 1992data employed above, andwhere the relationship ofthe heirs is reported (Joulfaian and McGarry, 2004), the findings from this cross sectionshow that the presence of children heirs has little effect on the ownership form.

Bayer, Patrick J., Bernheim, B.D., Scholz, J.K., 2009. The effects of financial education in theworkplace: evidence from a survey of employers. Economic Inquiry, 47(4). WesternEconomic Association International, pp. 605–624.

Bernheim, B.D., 1991. How strong are bequest motives? Evidence based on estimates ofthe demand for life insurance and annuities. J. Polit. Econ. 99 (5), 899–927.

Bernheim, B.D., Garrett, D.M., 2003. The effects of financial education in the workplace:evidence from a survey of households. J. Public Econ. 87 (7–8), 1487–1519.

Bernheim, B.D., Shleifer, A., Summers, L.H., 1985. The strategic bequest motive. J. Polit.Econ. 93, 1045–1076.

Bernheim, B.D., Forni, L., Gokhale, J., Kotlikoff, L.J., 2003. The mismatch between life insur-ance holdings and financial vulnerabilities: evidence from the health and retirementstudy. Am. Econ. Rev. 93 (1), 354–365.

Bernheim, B.D., Lemke, R.J., Scholz, J.K., 2004. Do estate and gift taxes affect the timing ofprivate transfers? J. Public Econ. 88 (12), 2617–2634.

Bertrand, M., Duflo, E., Mullainathan, S., 2004. How much should we trust differences-in-differences estimates. Q. J. Econ. 119, 249–275.

Brown, J.R., 2001. Are the elderly really over-annuitized? New evidence on life insuranceand bequests. In: Wise, David (Ed.), Themes in the Economics of Aging. University OfChicago Press, pp. 91–126.

Campbell, R.A., 1980. The demand for life insurance: an application of the economics ofuncertainty. J. Financ. 35 (5), 1155–1172.

Choi, J.J., Laibson, D., Madrian, B.C., Metrick, A., 2002. Defined contribution pensions:plan rules, participant decisions, and the path of least resistance. In: Poterba, James(Ed.), Tax Policy and the Economy, 16.

Hau, A., 2000. Liquidity, estate liquidation, charitable motives, and life insurance demandby retired singles. J. Risk Insur. 67 (1).

Holtz-Eakin, D., Phillips, J.W., Rosen, H.S., 2001. Estate taxes. Life insurance and smallbusiness. Rev. Econ. Stat. 83 (1), 52–63.

Hurd, M.D., 1987. Savings of the elderly and desired bequests. Am. Econ. Rev. 77 (3),298–312.

Johnston, D.C., I.R.S., July 29, 2002. Loophole Allows Wealthy to Avoid Taxes. New YorkTimes http://www.nytimes.com/2002/07/28/us/death-still-certain-but-taxes-may-be-subject-to-a-loophole.html?pagewanted=all&src=pm.

Joulfaian, D., 2005. Choosing between gifts and bequests: how taxes affect the timing ofwealth transfers. J. Public Econ. 89 (11–12), 2069–2091.

Joulfaian, D. The changing face of estate planning, US Department of the Treasury, 2010,mimeo.

Joulfaian, D., 2012. The Federal Estate Tax: History, Law, and Economicsavailable at http://papers.ssrn.com/sol3/papers.cfm?abstract-id=1579829.

Joulfaian, D., McGarry, K.M., 2004. Estate and gift tax incentives and inter vivos giving.Natl. Tax J. 57 (2, Part 2).

Kopczuk, W., 2007. Bequest and tax planning: evidence from estate tax returns. TheQuarterly Journal of Economics, 122(4). MIT Press, pp. 1801–1854.

Kopczuk,W., Lupton, J., 2007. To leave or not to leave: the distribution of bequest motives.Rev. Econ. Stud. 74 (1), 207–235.

Kopczuk, W., Slemrod, J.B., 2001. The impact of the estate tax on the wealth accumulationand avoidance behavior of donors. In: Gale, William G., Hines Jr., James R., Slemrod,Joel B. (Eds.), Rethinking Estate and Gift Taxation. Brookings Institution Press,Washington, DC, pp. 299–343.

Lewis, F.D., 1989. Dependents and the demand for life insurance. Am. Econ. Rev. 79 (3),452–467.

McGarry, K., 1999. Inter vivos transfers and intended bequests. J. Public Econ. 73 (3),321–351.

Moulton, B., 1990. An illustration of a pitfall in estimating the effects of aggregatevariables on micro units. Rev. Econ. Stat. 334–338.

Page, B.R., 2003. Bequest taxes, inter vivos gifts, and the bequest motive. J. Public Econ. 87(5–6), 1219–1229.

Papke, L.E., Wooldridge, J.M., 1996. Econometricmethods for fractional response variableswith an application to 401(k) plan participation rates. J. Appl. Econ. 11 (6), 619–632.

Verdon, L. L. Why Do Rich People Buy Life Insurance? mimeo, 2008, available at http://papers.ssrn.com/sol3/papers.cfm?abstract-id=1265046.

Yaari, M.E., 1965. Uncertain lifetime, life insurance, and the theory of the consumer. Rev.Econ. Stud. 32 (2), 137–150.