20th. anniversary. The Tax Benefits of Relaxing the Long- Only Constraint: Do They Come from Character or Deferral?

Transcription

1 Volume 21 No. 4 jwm.prjournals.com Sprng 2019 The real wsdom s smply to recall that one cannot solve a problem wthout recognzng that t exsts and that the ultmate form of sllness s to keep dong the same thng all the whle expectng dfferent outcomes. Jean Brunel 20th annversary PART FOUR Nathan Sosner, Stanley Krasner, and Ted Pyne The Tax Benefts of Relaxng the Long- Only Constrant: Do They Come from Character or Deferral?

2

3 Nathan Sosner s a managng drector wth AQR Captal Management n Greenwch, CT. nathan.sosner@aqr.com Stanley Krasner s an assocate wth AQR Captal Management n Greenwch, CT. stanley.krasner@aqr.com Ted Pyne s a managng drector wth AQR Captal Management n Greenwch, CT. ted.pyne@aqr.com Sprng 2019 The Tax Benefts of Relaxng the Long-Only Constrant: Do They Come from Character or Deferral? Nathan Sosner, Stanley Krasner, and Ted Pyne Pror lterature shows that relaxng the long-only constrant sgnfcantly mproves pretax mplementaton effcency. 1 More recently, Berkn and Luck [2010] and Salm and Sosner [2018] found that relaxng the longonly constrant also enhances tax effcency, n partcular for tax-aware strateges. In ths study, we propose a decomposton of the current year s total tax beneft (or lablty) of a strategy nto what we defne as character and deferral components. Our decomposton s mathematcally straghtforward, ntutve, and n our vew helpful to taxable nvestors and ther advsors seekng to understand and mprove the after-tax performance of ther nvestment portfolos. We use ths decomposton to dentfy the source of tax benefts resultng from relaxaton of the long-only constrant. Our methodology clearly shows whch taxable nvestors would beneft the most from combnng a lmted amount of shortng wth tax-aware rebalancng. Numerous books and artcles have advsed taxable nvestors to optmze ther 1 The argument n favor of mplementaton effcency of relaxed-constrant strateges was made by Clarke, de Slva, and Sapra (2004); Jacobs and Levy (2006); Berger (2008); and Ang, Mchalka, and Ross (2017). In addton, Jacobs and Levy (2007) dspelled a number of wdespread msconceptons related to relaxed-constrant portfolo constructon and the rsks and costs assocated wth managng relaxed-constrant strateges. nvestment portfolos for taxes. 2 However, to our knowledge, our study s the frst n ths expansve lterature to defne the decomposton of tax benefts nto character and deferral and explan ts relevance for assessng the tax benefts of several wellknown nvestment strateges. There are two ways of achevng a tax beneft at the level of an overall nvestment portfolo held n a taxable account. Frst, an nvestor can favorably affect the character of realzed captal gans and ncome at the overall portfolo level by tltng the balance of net realzed gans n a gven year from short-term to long-term and from ordnary ncome to qualfed dvdends and tax-exempt ncome. 3 The beneft results 2 A short and ncomplete lst of examples of such advce nclude Sten and Narasmhan (1999); Arnott, Berkn, and Ye (2001b); Brunel (2001, 2006); Rogers (2001, 2006); Sten (2001); Berkn and Ye (2003); Horvtz and Wlcox (2003); Qusenberry (2003); Sten and McIntre (2003); Paulson and Tavel (2005); Wlcox, Horvtz, and dbartolomeo (2006); Horan and Adler (2009); Jennngs et al. (2011); Km, Dougherty, and Klen (2011); Israel and Moskowtz (2012); Bouchey, Santodomngo, and Sreklove (2015); Bouchey, Brunel, and L (2016), Lucas and Sanz (2016); and Bouchey and Prtaman (2017). 3 Techncally speakng, captal gans (and losses) and ordnary ncome (and deductons) are the only two true categores of character. Long-term and shortterm captal gans are subcategores wthn the captal gans character. However, they are often referred to as havng a dfferent character because under the current US tax law they are subject to dfferent tax rates. The Journal of Wealth Management 1

4 from a lower amount of hghly taxed short-term captal gans and ncome as a fracton of all realzed gans and ncome. Second, at the overall portfolo level, an nvestor can defer the realzaton of captal gans to future years and beneft from a reducton n the current year s taxable gans. In ths case, the beneft arses from a lower amount of realzed gans as a fracton of the total nvestment profts realzed and unrealzed. Whereas the character beneft s permanent (.e., tax labltes are reduced permanently by payng tax at a lower rate), the deferral beneft or lablty s temporal: Barrng a taxexempt portfolo lqudaton resultng from a donaton to charty or step-up n cost bass at death, an ncrease n current unrealzed gans (current deferral beneft) leads to hgher lqudaton taxes, whereas a decrease n current unrealzed gans (current deferral lablty) leads to lower lqudaton taxes. Despte ts temporal nature, current deferral beneft adds real value because t allows the nvestor s wealth to apprecate (compound) at a faster rate, even after adjustng for a hgher future lqudaton tax resultng from gan deferral. 4 At the level of a sngle strategy wthn an overall nvestment portfolo, a character beneft occurs when the strategy allocates gans and ncome n low taxed characters, such as long-term captal gans and qualfed dvdends, and a matchng amount of losses and deductons n hghly taxed characters, such as shortterm captal losses and ordnary deductons. Ths s because short-term losses offset short-term gans before offsettng any long-term gans. Thus, a strategy realzng a smlar amount of long-term gans and shortterm losses tlts the balance of net realzed gans n a gven year from short term to long term at the overall portfolo level. Ths s consstent wth the defnton of character beneft at the overall portfolo level defned n the prevous paragraph. (Note that here and throughout the artcle we assume that a strategy s managed n a separately managed account or n a lmted partnershp both vehcles allow pass-through of strategy postons realzed losses and deductons to the nvestor.) Smlarly, at the sngle strategy level, a deferral beneft occurs when the strategy allocates losses and deductons n excess of gans and ncome. A strategy realzng deferral beneft s reducng the net 4 As wll be seen later, our after-tax return calculaton methodology explctly accounts for the lqudaton tax costs (benefts) of an ncrease n unrealzed gans (losses). realzed gans and ncome of the overall nvestment portfolo n the current year, whch s agan consstent wth the defnton of deferral beneft n the prevous paragraph. 5 Contnung wth ths defnton, a strategy that adds realzed gans and/or ncome to the nvestment portfolo realzes a deferral lablty. The flp sde of what we defne as deferral beneft (lablty) s an ncrease (decrease) n unrealzed gans and thus an expected future tax lablty (beneft). 6 Relaxed-constrant strateges are partcularly nterestng for ths type of character-deferral analyss because they combne features of tradtonal actve long-only asset management and alternatve long short hedge fund nvestng. Smlar to actve long-only, they seek to provde benchmark exposure and a hgherthan-benchmark return and thus can be vewed as a substtute for tradtonal actve management. Smlar to long short hedge fund strateges, they use leverage and shortng to acheve actve return. Through the lens of the character-deferral decomposton, we explore whch types of tax benefts (and labltes) relaxedconstrant strateges nhert from ther long-only and long short counterparts. Our emprcal evdence shows that for tax-aware strateges, relaxng the longonly constrant results n a large ncrease n tax benefts, n partcular character beneft. We thus conclude that tax-aware relaxed-constrant strateges are more attractve than tax-aware long-only strateges to taxable nvestors. In the fnal secton, we dscuss caveats that mght affect ths concluson. Before we present our character-deferral tax beneft decomposton, n the next secton, we ntroduce a tax-aware relaxed-constrant equty strategy and compare t to other types of tax-aware equty strateges 5 An mportant caveat s that the tax benefts of a strategy arse when ts losses offset gans from other nvestments n the nvestor s portfolo. Ths assumpton wll be employed throughout the artcle. 6 A comparson between, for example, an equty exchangetraded fund (ETF) and a long-dated equty exchange-traded note (ETN) mght help clarfy the concept of deferral lablty. Even a tax-effcent equty ETF would stll dstrbute dvdend ncome pad out by the underlyng stocks and thus would have a deferral lablty n any gven year. A long-dated ETN only has one payment at a dstant future maturty date and, as a result, has no deferral lablty n the current year but a larger expected tax lablty n a future year when t matures. Under our defnton, the ETN does not have a deferral beneft, though, because t does not offset current gans or ncome of other strateges n the nvestment portfolo. 2 The Tax Benefts of Relaxng the Long-Only Constrant: Do They Come from Character or Deferral? Sprng 2019

5 E x h b t 1 Comparson of Relaxed-Constrant Strategy to Other Well-known Strateges, descrbed n pror lterature. 7 Readers famlar wth the topc may skp the next secton. A BRIEF OVERVIEW OF TAX-AWARE STRATEGIES Relaxng the long-only constrant mproves the mplementaton effcency of actvely managed portfolos: Although long-only strateges are lmted n ther ablty to express negatve vews on stocks, relaxed-constrant strateges can mplement negatve vews va short sellng. 8 Importantly, for a taxable nvestor, relaxed-constrant strateges are even more attractve: Berkn and Luck (2010) and Salm and Sosner (2018) found that combnng relaxaton of the long-only constrant wth tax-aware rebalancng enhances the tax effcency of actvely managed strateges. Exhbt 1 compares an actvely managed tax-aware relaxed-constrant strategy based on a value momentum alpha model to three other tax-aware strateges (see Appendx D for further detals on smulatons of all the strateges). The strategy n the column to the mmedate rght of the relaxed-constrant s smlar to the one orgnally modeled by Sten and Narasmhan [1999]. The strategy harvests losses whle targetng a lmted annual trackng error of 1% to the Russell 1000 benchmark. Borrowng from Sten and Narasmhan, hereafter we refer to ths strategy as tax-managed passvendexed (TMPI). The two strateges further to the rght are actvely managed tax-aware strateges based on the same alpha model as the relaxed-constrant strategy and closely follow the portfolo constructon methods of Salm and Sosner (2018). 9 7 We assume that all the tradng nstruments employed by the strateges are physcal (or cash) equtes. Although the dscusson of taxaton of dervatves s outsde of the scope of ths artcle, we would lke to pont out that f dervatves (e.g., swaps or forwards) were used, the tax results would be very dfferent from the ones we descrbe. 8 See Clarke, de Slva, and Sapra (2004); Jacobs and Levy (2006); Berger (2008); and Ang, Mchalka, and Ross (2017). Sprng All strateges are rebalanced monthly, and ther returns are smulated over a 30-year perod from January 1988 to December The actvely managed strateges target 4% actve rsk and use a value momentum alpha model wth equal rsk allocatons to value and momentum factors. Factor portfolo constructon methodology follows Salm and Sosner (2018), who n turn rely on an earler work by Asness et al. (2015). The Journal of Wealth Management 3

6 For calculatng tax costs and benefts, we assume that the tax rate applcable to long-term captal gans (losses) and qualfed dvdend ncome s 20% and the tax rate applcable to short-term captal gans (losses), nterest ncome, and deductons resultng from n-leu dvdends on short postons s 35%. In our tax-aware portfolo rebalancng, the tax cost functon assumes that all the realzed gans are taxed mmedately at ther respectve character tax rates and that all the losses mmedately offset gans of the same character realzed by other unrelated strateges. As a result, short-term captal gans are vewed as more puntve than long-term captal gans, and short-term captal losses are vewed as more attractve than long-term captal losses. In addton, usng the methodology from Poterba [1999], we calculate the tax rate applcable to unrealzed gans to be approxmately 10% (see Appendx A for detals). 10 The latter tax rate allows us to estmate the present value of tax costs resultng from future lqudaton of unrealzed gans. Although we track and report these lqudaton tax costs, they are not ncluded n the tax-aware portfolo constructon. Smlarly, dvdend taxes and n-leu short dvdend deductons are not explctly ncorporated nto the tax-aware portfolo constructon, but they are ncluded n after-tax return calculaton. 11 The Taxes panel n Exhbt 1 shows both the level of the actve tax (defned as the strategy s total tax n excess of the tax of the relevant benchmark) and the total tax. Appendx B shows the components comprsng the actve and total taxes reported n Exhbt 1. Actve tax s used n calculaton of actve after-tax net return. Estmaton of actve taxes, however, requres estmaton of taxes on a benchmark ndex, whch mght be a challengng task (for more on ths see Sosner, Sullvan, and Urruta [2018]). Because the purpose of our study s to llustrate the character-deferral decomposton rather than to resolve the complexty of an approprate benchmark 10 In hs example, Poterba (1999, 31) assumed an effectve tax rate on unrealzed gans of 10%. Ths assumpton s conservatve; earler lterature estmates the effectve tax rate applcable to unrealzed gans to be approxmately a quarter of the statutory captal gans tax rate (see Baley 1969 and Protopapadaks 1983). Our man conclusons are robust to reducton of the unrealzed gans tax rate below 10%. If anythng, such reducton would make the relaxedconstrant tax-aware strategy even more attractve. 11 In analyss not reported here, for the type of strateges modeled n ths study, we could not fnd convncng evdence that penalzng dvdend exposure s benefcal for after-tax strategy returns. tax calculaton, after settng the stage wth Exhbt 1, we wll focus on total, not benchmark-relatve, taxes. Relatve to the Russell 1000 benchmark, the TMPI strategy realzes an annual tax beneft of 30 bps (see the Actve Tax lne) approxmately half of the 58 bps actve tax beneft of the tax-aware relaxed-constrant strategy. Ths estmate s n lne wth a number of prevous studes smulatng a smlar TMPI-type strategy. 12 The long-only strategy realzes 26 bps of annual tax lablty compared to the relaxed-constrant strategy s 58 bps of annual tax beneft n excess of the Russell 1000 benchmark. The long short strategy realzes a very meanngful tax beneft of 2.29% annually relatve to ts three-month T-bll benchmark. The results for the three actvely managed strateges long-only, relaxed-constrant, and long short are qualtatvely consstent wth fndngs reported by Salm and Sosner [2018]. Importantly, out of the three beta-one strateges benchmarked to Russell 1000, relaxed-constrant s the most tax effcent. Because benchmarks themselves tend to generate tax costs, the total tax lne n Exhbt 1, whch does not subtract the benchmark tax, shows ether a hgher tax lablty or a lower tax beneft. 13 The long short strategy s the most tax effcent wth a tax beneft of 98 bps (see the Total Tax lne). Of the three beta-one strateges, relaxed-constrant shows the lowest tax cost at 97 bps, followed by TMPI wth a tax cost of 1.25% and longonly wth a tax cost of 1.81%. 14 To be clear, the tax costs of tax-aware strateges need to be compared to tax costs of other nvestment alternatves. For example, f a taxable nvestor s lookng 12 See, for example, Sten and Narasmhan (1999); Arnott, Berkn, and Ye (2001a); Berkn and Ye (2003); and Sten, Vadlamund, and Bouchey (2008). In addton, n our smulaton TMPI realzes a small postve alpha, whch we fnd to be fully explaned by a momentum exposure resultng from tax awareness. Israel and Moskowtz (2012) and Salm and Sosner (2018) ponted out that tax-aware optmzaton tends to ntroduce a small momentum tlt. 13 Tax costs of the Russell 1000 benchmark result mostly from qualfed dvdend ncome and the present value of the future lqudaton tax on unrealzed gans. To a smaller extent, there are longterm captal gan tax costs assocated wth perodc reconsttuton of the ndex portfolo. Tax costs of the three-month T-bll benchmark result from nterest rate treated as ordnary ncome. 14 Note that the addtonal turnover of long-only compared to TMPI causes an addtonal annual tax cost of 55 bps but also generates 1.8% of the pretax returns, leadng to a hgher expected after-tax return of the long-only strategy. 4 The Tax Benefts of Relaxng the Long-Only Constrant: Do They Come from Character or Deferral? Sprng 2019

7 E x h b t 2 Strategy Annual Tax Benefts n Down- and Up-Market Years, for a passve equty allocaton, TMPI mght be a more attractve alternatve than a passve ndex: Exhbt 1 shows that the tax cost of TMPI s lower and ts aftertax return s hgher than those of the Russell 1000 benchmark. Smlarly, f a taxable nvestor s lookng for an allocaton to equtes, actve or passve, accordng to our smulatons (and subject to the caveats dscussed n the last secton), tax-aware relaxed-constrant presents a more tax-effcent alternatve than ether TMPI or tax-aware long-only: Exhbt 1 shows that taxes on relaxed-constrant are lower, and ts after-tax returns are hgher than those of TMPI and long-only. Fnally, we would lke to pont out that tax costs and benefts of tax-aware strateges can vary sgnfcantly between rsng and fallng market perods. Exhbt 2 shows how the tax benefts are dstrbuted across upand down-market years. The relaxed-constrant strategy nherts the features of both the long-only and long short strateges. Smlar to long-only, t realzes sgnfcantly hgher tax benefts n declnng markets than n rsng markets. At the same tme, because of shortng, t realzes meanngfully lower tax costs than long-only n rsng markets when the long-only strategy tends to realze sgnfcant tax costs. Smlarly, n comparson to TMPI, relaxed-constrant realzes hgher tax benefts n down markets and lower tax costs n up markets. Importantly, separatng the sample nto up- and down-market years does not change our concluson: Relaxng the long-only constrant results n tax benefts. The relaxed-constrant strategy s more tax effcent than the other two beta-one strateges n both rsng and fallng markets. Does ths relatve tax effcency arse from character or deferral? TAX EFFICIENCY OF A TAX-AWARE RELAXED- CONSTRAINT STRATEGY THROUGH THE LENS OF A CHARACTER-DEFERRAL DECOMPOSITION Defnng Character and Deferral Benefts Exhbt 3 helps to llustrate the concept of current perod character and deferral tax benefts. Suppose a portfolo of strateges realzes a total of $100 of captal Sprng 2019 The Journal of Wealth Management 5

8 E x h b t 3 Illustratve Example of Character and Deferral Amounts 6 The Tax Benefts of Relaxng the Long-Only Constrant: Do They Come from Character or Deferral? Sprng 2019

9 gans, wth $50 realzed as a long-term gan and $50 as a short-term gan. In Panel A, we add a tax-aware strategy to ths portfolo of strateges (to smplfy the example, we add the tax-aware strategy rather than allocatng to t whle at the same tme reducng the allocaton to other strateges). The tax-aware strategy realzes a long-term gan of $20 and a short-term loss of $30, wth a net total captal loss of $10. After combnng all the gan realzatons, at the overall portfolo level, the amount of long-term realzed gans s $70 and the amount of short-term realzed gans s $20. Thck borders of the bars n Exhbt 3 mark the long-term and short-term gans before and after addng the tax-aware strategy to the mx. Two thngs changed because of the addton of the tax-aware strategy. Frst, the overall level of taxable realzed gans decreased from $100 ($50 long term plus $50 short term) to $90 ($70 long term plus $20 short term). Second, the character of gans shfted favorably from equal amounts of long term and short term ($50 and $50) to mostly long term ($70 versus $20). We defne the matchng amount of ncrease n long-term gans and decrease n short-term losses, whch n ths example equals $20, as the character beneft amount of the tax-aware strategy (shown by the lght gray bars). We defne the decrease n total realzed gans, whch n ths example s $10, as the deferral beneft amount (shown by the dark gray bar). It s possble for a tax-aware strategy, rather than allocatng a net total loss, to allocate a net total gan. We llustrate such a scenaro n Exhbt 3, Panel B. In ths scenaro, the character beneft amount s $20 ths s the matchng ncrease n realzed long-term gans and decrease n realzed short-term losses at the portfolo level. However, the deferral beneft s negatve (.e., a deferral lablty) because the total amount of realzed gans at the portfolo level ncreases from $100 ($50 long term plus $50 short term) to $110 ($80 long term plus $30 short term). In ths example, the tax-aware strategy stll realzes a character beneft by tltng the balance of the overall portfolo realzed gans from short term to long term, but t realzes a deferral lablty because the total amount of realzed gans ncreases. To sum up, the character beneft of a tax-aware strategy results from matchng amounts of realzed longterm gans and short-term losses, and the deferral beneft (lablty) results from a net realzed total loss (gan). The algorthm of the decomposton s derved mathematcally n Appendx C. The algorthm formalzes and generalzes the deas llustrated n the prevous two examples and captures all possble permutatons of gans, losses, ncome, and deductons. The product of the character beneft amount and the dfference between the short-term and long-term tax rates s defned as the character beneft. The product of the deferral beneft (lablty) amount and the applcable tax rate s defned as the deferral beneft. The applcable tax rate s long term f long-term gan exceeds short-term loss or long-term loss exceeds short-term gan; t s short term f short-term gan exceeds long-term loss or short-term loss exceeds long-term gan. If long-term and short-term results are both gans or are both losses, then the approprate long-term and short-term tax rates are appled to the respectve amounts of long-term and short-term gans or losses. Importantly, n our calculatons we make the assumpton that when the strategy allocates losses, there are always suffcent gans of the same character from other nvestments that can be offset by those losses. Ths assumpton helps assess the tax-savng opportuntes afforded by the strategy rather than a specfc tax stuaton of any possble nvestor. These opportuntes mght materalze dfferently for dfferent nvestors, and ths s exactly why the character-deferral decomposton s valuable. It s partcularly pertnent n the context of tax-aware strateges that seek to delver tax benefts by realzng short-term captal losses and allocatng them to nvestors. A tax-aware strategy s character beneft s ncreased when short-term gans and ordnary ncome are taxed at a sgnfcantly hgher rate than long-term gans and qualfed dvdend ncome, as n the case of US federal taxes (n partcular n the hgher tax rate brackets), and when nvestors have substantal shortterm gans from other strateges n ther portfolos. The character beneft s reduced when nvestors have only a small amount of short-term gans or when the dfference between short-term and long-term captal gans tax rates s ether small (under the alternatve mnmum tax) or nonexstent (under most state and local taxes). 15 A deferral beneft can be acheved by a tax-aware strategy that realzes a net captal loss as long as the 15 Strateges that only realze character benefts mght not yeld tax benefts for US corporatons. The same s true n countres where there s no dfference between long-term and short-term captal gans rates (e.g., the Unted Kngdom, Japan, and many countres n contnental Europe). Sprng 2019 The Journal of Wealth Management 7

10 E x h b t 4 Annual Character and Deferral Tax Benefts Assumng Average Year Realzaton of Gans and Losses, nvestor has any type of captal gans from other strateges n the portfolo, long term or short term. Ths beneft s temporal though, because barrng a tax-exempt lqudaton of the strategy (e.g., chartable gftng or bequeathal at death), nvestors wll have to pay the deferred taxes upon lqudaton. The temporal nature of the deferral beneft s captured by the lqudaton tax lablty, whch ncreases wth the current perod s deferral beneft. Because of the dfference between tax rates applcable to current realzed gans and the effectve tax rate estmated for future realzed gans, deferral beneft yelds a very real value for a taxable nvestor. Character-Deferral Decomposton of the Tax Benefts of Tax-Aware Strateges Exhbt 4 apples the character-deferral decomposton to a representatve year for the relaxed-constrant strategy and compares t to the three other strateges. The relaxed-constrant strategy allocates long-term gans and qualfed dvdends taxed at the lower longterm captal gans rate and short-term losses and deductons, provdng an offset aganst gans and ncome taxed at the hgher short-term captal gans rate. The overlappng annual amount, whch we defne as the character beneft amount, s 5.6%. There are excess long-term gans and qualfed dvdend ncome n the annual amount of 3.5%. We defne ths amount of excess gans and ncome as the deferral lablty amount. Because ths s a lablty, n Exhbt 4 we show ths amount wth a negatve sgn. Exhbt 4 shows that n a typcal year, relaxedconstrant, and all the other tax-aware strateges, realzes a character beneft. Actvely managed strateges tend to realze a larger amount of character beneft than TMPI, and ther character benefts ncrease wth leverage: The character beneft amount of relaxedconstrant s an ntermedate case between long-only and long short. As for the deferral beneft, all the beta-one strateges realze a deferral lablty, whereas the long short strategy realzes a small amount of deferral beneft. The short postons of relaxed-constrant help t acheve a substantally smaller deferral lablty amount than long-only. Smlar to the character beneft, wth respect to the deferral lablty, relaxed-constrant acheves an ntermedate result between long-only and long short. The relaxed-constrant strategy s character beneft amount s hgher than that of long-only and lower than that of long short. Its deferral lablty amount s lower than long-only and hgher than long short. 8 The Tax Benefts of Relaxng the Long-Only Constrant: Do They Come from Character or Deferral? Sprng 2019

11 From these observatons t s clear that relaxed-constrant wll be more tax effcent than long-only but less tax effcent than long short. A comparson to TMPI s harder because relaxed-constrant has a much larger character beneft amount than TMPI but also a larger deferral lablty amount. To wegh these conflctng results, we must use the respectve tax rates applcable to the character and deferral amounts. The next panel n Exhbt 4 shows those rates. The character beneft effectve tax rate s the dfference between the shortterm and long-term captal gans tax rates. The tax rates applcable to the deferral beneft or lablty correspond to the category of the excess gan or loss amount ether long term or short term. For example, relaxed-constrant realzes excess long-term gans and qualfed dvdend ncome that are taxed at the long-term rate, whereas long short realzes excess short-term losses and ordnary deductons that are credted wth the short-term tax rate. The total tax benefts and labltes n Exhbt 4 correspond exactly to those shown n Exhbt 1. However, the character-deferral decomposton n Exhbt 4 allows us to measure the sources of the tax effcency and tax neffcency of the four smulated strateges. For example, relaxed-constrant realzes a deferral lablty that s 23 bps hgher than TMPI 69 versus 46 bps. If nvestors can use the short-term losses and deductons realzed by the strateges effcently to offset short-term gans and ordnary nvestment ncome elsewhere n ther portfolos, the character beneft of relaxed-constrant s 62 bps hgher than TMPI 85 versus 23 bps. If the nvestor assgns no value to the character beneft (as, for example, would be the case when the nvestor has no short-term gans to offset), then the value of the character beneft s zero, and the total prelqudaton tax s just the deferral lablty. Note that for the sake of brevty we do not decompose the character and deferral benefts (and labltes) nto those due to captal gans and losses and those due to qualfed dvdends and ordnary ncome and deductons. However, ths can be easly done usng the same logc as we apply here. Exhbt 5 shows the character-deferral decomposton for an average year durng up- and downmarket years and durng the full strategy smulaton E x h b t 5 Character-Deferral Tax Beneft Decomposton n Down- and Up-Market Years, (contnued) Sprng 2019 The Journal of Wealth Management 9

12 E x h b t 5 (contnued) Character-Deferral Tax Beneft Decomposton n Down- and Up-Market Years, The Tax Benefts of Relaxng the Long-Only Constrant: Do They Come from Character or Deferral? Sprng 2019

13 perod. 16 In Exhbt 2, we saw that relaxed-constrant s more tax-effcent than TMPI n both up and down markets. The character-deferral decomposton n Exhbt 5, Panel A, makes t clear that the source of ths relatve tax effcency of relaxed-constrant s the character beneft, whch remans hgh and stable across up- and down-market years 70 and 90 bps, respectvely. The TMPI s character beneft s 40 bps n down markets but s reduced to just 10 bps n up markets, resultng n an average character beneft of 20 bps compared to 80 bps for relaxed-constrant. Exhbt 5, Panel B, shows why the relaxedconstrant actve approach domnates the long-only actve approach from the perspectve of tax effcency. Frst, shortng allows relaxed-constrant to acheve a sgnfcantly hgher character beneft n both rsng and fallng markets especally n up-market years, when relaxed-constrant realzes 70 bps of character beneft compared to only 20 bps of character beneft for longonly. Second, n rsng markets, the long-only strategy realzes a deferral lablty much larger than that of the relaxed-constrant strategy 1.9% versus 0.9%, respectvely. 17 Although the tax effcency of relaxed-constrant results n a larger expected lqudaton tax cost as compared to long-only, ths hgher lqudaton tax cost s more than compensated by a sgnfcantly hgher character beneft and a sgnfcantly lower deferral lablty of relaxed-constrant. Salm and Sosner [2018] ponted out that tax benefts of long short exhbt a postve correlaton wth market return. Exhbt 5, Panel C, shows that both the character and deferral benefts of long short are hgher n rsng than n fallng markets. Interestngly, n down markets relaxed-constrant outperforms long short on all metrcs of tax effcency, whereas n up markets t underperforms. Because the underperformance of relaxed-constrant n up markets s larger than outperformance n down mar- 16 The decomposton nto character and deferral s nonlnear (the dervaton n Appendx C makes ths clear). As a result of ths nonlnearty, the character-deferral decomposton, on average, as we show n Exhbt 5, wll be dfferent from the character-deferral decomposton usng gans and losses of an average year, as we show n Exhbt 4. However, despte ths nonlnearty, the results of character-deferral decomposton are qualtatvely and quanttatvely smlar between the two computaton methods. 17 Ths s consstent wth the fndngs of Salm and Sosner [2018], who ponted out that n up markets an average short poston creates opportuntes for a tax-aware strategy to realze losses that can partally offset the gans realzed on an average long poston. Sprng 2019 kets and because up markets are more frequent than down markets (25 up years versus only 5 down years durng our sample perod), the average tax beneft of long short s hgher. In fact, all categores character, deferral, and unrealzed contrbute to the hgher tax beneft of long short. 18 PRACTICAL SOLUTIONS FACILITATED BY CHARACTER-DEFERRAL DECOMPOSITION Relaxed-constrant strateges straddle the world of tradtonal long-only asset management and alternatve nvestng. As a result, these strateges can be thought of as ether a replacement of a porton of a tradtonal equty allocaton or an allocaton to alternatves. From the pretax perspectve, these strateges yeld an attractve actve return. From the tax perspectve, f managed n a separately managed account or n a lmted partnershp, tax-aware relaxed-constrant strateges can play the role of a tax-effcent core n a core and satellte structure orgnally proposed by Brunel [2001]. 19 In such a structure, a beta-one tax-effcent core strategy helps offset captal gans realzed by tax-neffcent satellte managers. Other authors who advocated the benefts of a core satellte structure for taxable nvestors consdered TMPI-type strateges as the tax-effcent core strategy. 20 Our study makes a number of contrbutons to ths lterature on core satellte portfolo structure. Frst, we show that barrng prolonged market downturns, the tax benefts of both the tradtonal TMPI strategy and our 18 It s mportant to note that the long short strategy n our study s smulated at a low average gross notonal exposure of about 230% of the net asset value (NAV) 115% long and 115% short. If the gross notonal exposure were to ncrease, all the tax costs and benefts would ncrease proportonately. For example, f the leverage of our smulated strategy were to double to 230% long and 230% short, n our calculatons, the character beneft of long short would also approxmately double. A word of cauton s approprate here: Wth an ncrease n leverage, the rsks and costs of managng a levered portfolo also ncrease. Because of these rsks and costs, dependng on the stuaton, leverng up the strategy portfolo beyond a partcular level of leverage mght become economcally untenable. 19 Separately managed account and lmted partnershp vehcles ensure that losses realzed by the strategy can be passed through to the nvestor. Ths would not be the case for a regulated nvestment company, such as a mutual fund or an ETF, whch cannot dstrbute losses to nvestors. 20 See, for example, Rogers (2001), Sten (2001), Qusenberry (2003), and Bouchey and Prtaman (2017). The Journal of Wealth Management 11

14 new tax-aware relaxed-constrant strategy come from character and not deferral. Thus, from the tax perspectve, f ether strategy s used as a beta-one tax-effcent core, the core satellte structure works best when the satellte managers tend to realze a substantal amount of short-term captal gans. Second, the tax-aware relaxedconstrant strategy realzes a sgnfcantly hgher character beneft and thus s expected to yeld a hgher tax beneft than TMPI n the presence of satellte managers wth short-term gans. Fnally, an actvely managed tax-aware long-only strategy s unlkely to functon as a beta-one tax-effcent core nearly as effectvely as tax-aware relaxed-constrant for two reasons: Its character beneft s a fracton of the character benefts of relaxed-constrant, whereas ts negatve deferral acceleraton of gans realzaton s twce as hgh as that of relaxed-constrant. 21 To sum up, the character-deferral decomposton helps nvestors and ther advsors understand the sources of the tax benefts of a gven strategy. Relaxed-constrant, because t realzes substantal and persstent character benefts, offers sgnfcant tax benefts to nvestors wth large short-term captal gans from other nvestments n a core satellte-lke portfolo structure. To such nvestors, relaxed-constrant s more attractve than TMPI. On the other hand, the decomposton shows that both relaxed-constrant and TMPI partally lose ther attractveness as a tax-management tool at the overall nvestment portfolo level for nvestors wth long-term captal gans only. Ths s because n an average year both strateges realze a deferral lablty rather than a deferral beneft. 22 As a result, our character-deferral decomposton allows for a quck and nformatve assessment of tax benefts of dfferent tax-aware strateges wthout modelng varous nvestor-specfc stuatons. 21 Note that tax-aware rebalancng hghly benefts longonly strateges by substantally ncreasng ther after-tax returns. Nonetheless, ther after-tax returns and nformaton ratos reman less attractve than those of tax-aware relaxed-constrant strateges, whch further beneft from shortng. 22 Sten, Valdamund, and Bouchey (2008) showed that the tax benefts of TMPI can be enhanced through strategc realzaton of long-term captal gans. However, those addtonal benefts agan wll be character benefts as realzed long-term gans are compensated by the ablty to realze more short-term losses. It s mportant to note that n TMPI strateges such long-term captal gans realzatons need to be trggered for a tax reason, whereas n actvely managed strateges gan realzatons are a natural result of the strateges turnover. FURTHER CONSIDERATIONS RELATED TO RELAXED-CONSTRAINT STRATEGIES As always, dfferent approaches have ther respectve advantages and dsadvantages. In our smulatons we fnd that tax-aware relaxed-constrant strateges outperform beta-one long-only strateges as a result of a hgher net of costs alpha, lower tax costs of negatve deferral, and sgnfcantly hgher tax character benefts. However, these hgher pretax and tax benefts come wth a number of caveats. Frst, n contrast to TMPI, the relaxed-constrant strategy reles on the performance of a manager s alpha sgnals. There s always uncertanty n how well these sgnals wll perform n the future. Therefore, the selecton of an actve manager mght have a szable mpact on after-tax returns. Moreover, even f the actve strategy on average contnues to perform as expected, t exposes the nvestor to addtonal volatlty above and beyond the volatlty of the passve ndex. For example, durng our sample perod value momentum strateges exhbted strong performance n an average year, yet our smulated relaxed-constrant strategy underperformed the benchmark by close to 9% n 2009 and by approxmately 5% n 1991 and Compared to that, our smulated TMPI strategy underperformed the benchmark by a maxmum of 1.7%. Second, relaxed-constrant strateges mght be sgnfcantly more costly to manage. In our study we assumed transacton and fnancng costs on par wth those of professonal actve managers. If nvestors were to attempt to manage the strategy on ther own, ther tradng and fnancng costs mght be sgnfcantly hgher than those assumed here. Ths brngs us to our fnal pont. Managng leveraged strateges n a separately managed account mght be costly and neffcent, especally for smaller nvestors. As a result, t s possble and even lkely that such a strategy would be managed by a professonal manager n a comngled fund. 23 Investment through a comngled fund, however, makes other tax management technques, such as gftng hghly apprecated ndvdual stock postons to charty or strategcally realzng unrealzed captal gans, vrtually mpossble. As a result, TMPI or longonly strateges, whch are easer to manage n a separate 23 The fund should be organzed as a partnershp for the nvestor to beneft from allocaton of realzed losses. 12 The Tax Benefts of Relaxng the Long-Only Constrant: Do They Come from Character or Deferral? Sprng 2019

15 account, could n practce demonstrate hgher tax effcency compared to what we report here. A p p e n d x A TAX RATE APPLICABLE TO UNREALIZED GAINS In our dervatons we used expected tax rate t E to estmate the present value of the future tax labltes (benefts) resultng from decrease (ncrease) n the current year s taxable gans and ncome. Poterba [1999] derved the followng formula for parameter t E n the case of a sngle asset: t E = t cg p(1 λ )(1 + r) r + p+ q pq where t cg s the statutory tax rate applcable to realzed captal gans at a future date when the lqudaton occurs, p s the chance of lqudatng the asset n any gven year, l s the probablty that the lqudaton of the asset does not result n a captal gan tax lablty, q s the probablty of step up n the cost bass at death, and r s the after-tax nomnal dscount rate. What s the approprate level of the tax rate t cg n ths formula? Poterba set t cg equal to t L. Under the assumpton that future statutory captal gans tax rate remans constant at ts current level, settng t cg equal to t L s precsely accurate n the case of lqudaton of a sngle asset (whch s exactly the case modeled by Poterba). For portfolo lqudaton, there s no guarantee that all the assets n the portfolo wll be lqudated at a long-term gan some of the assets mght be held for a perod of tme shorter than 12 months. Nonetheless, the assumpton that for portfolo lqudaton the expected level of t cg s smlar to t L s plausble under realstc crcumstances. p(1 λ )(1 + r) The multpler of the tax rate can be r + p+ q pq shown to be between 0 and 1. As the value of ths multpler ncreases, the future lqudaton tax becomes more puntve from the nvestor s pont of vew. An ncrease n the probablty of sellng the asset, p, or n the probablty of realzng gan when sellng the asset, 1 l, leads to an ncrease n the rate of gan realzaton and, assumng a postve after-tax nomnal dscount rate, r, affects the multpler postvely, thus ncreasng the expected tax rate. In other words, the hgher the chance of realzng the deferred gan, the hgher the present value of the tax lablty resultng from realzaton of deferred gans. Increasng the probablty of death, q, effectvely ncreases the nvestment horzon: There s a correspondence between death and nfnte nvestment horzon both elmnate the lqudaton tax, the former because of the step-up n cost bass, the latter because the lqudaton never occurs. Consstent wth ths, an ncrease n the probablty of death reduces the multpler and thus reduces the expected tax rate. In ths study, we make the followng assumptons: p = 10%, 1 - l = 75%, q = 2%, r = 3%, and t cg = t L = 20%, resultng n t E = 10.44%. We round ths rate to 10%. In hs numercal example, Poterba [1999, p. 31] also used an effectve unrealzed gans tax rate of 10%. A p p e n d x B CHARACTER-DEFERRAL DECOMPOSITION OF TAX BENEFITS Exhbt B1 shows the components composng actve and total taxes. We assume that the tax rate applcable to long-term captal gans (losses) and qualfed dvdend ncome s 20% and the tax rate applcable to short-term captal gans (losses), nterest ncome, and deductons resultng from n-leu dvdends on short postons s 35%. We calculate the tax rate applcable to unrealzed gans to be approxmately 10% (see Appendx A for detals of ths calculaton). Postve values represent tax benefts, whereas negatve values represent tax costs. E x h b t B 1 Annualzed Components of Actve and Total Taxes, Sprng 2019 The Journal of Wealth Management 13

16 A p p e n d x C CHARACTER-DEFERRAL DECOMPOSITION OF TAX BENEFITS Let r PT and r AT denote the pretax and postlqudaton after-tax returns of a strategy, respectvely. The strategy realzes net long-term captal gans (losses) n the amount g L and net short-term captal gans (losses) n the amount g S. The strategy also receves qualfed dvdend ncome n the amount q and net ordnary ncome (loss) n the amount. Consstent wth returns, all the gan, loss, ncome, and deducton amounts are expressed as a fracton of the strategy s NAV. Let t L denote the lower tax rate applcable to long-term captal gans and qualfed dvdends and t H denote the hgher tax rate applcable to short-term captal gans and ordnary ncome. In addton, let t E be the expected tax rate on unrealzed gans, as explaned n Appendx A. Usng the defnton from Poterba [1999], the postlqudaton after-tax return of the strategy can be defned as r = r ( g + q) t ( g + t ) ( r g q g t ) AT PT L L S H PT L S E where rpt gl q gs s the amount of unrealzed gan. The frst term on the rght-hand sde of the equaton represents pretax return n a gven perod, and the remanng three terms represent tax costs (n the case of gans or ncome) or benefts (n the case of losses or deductons) n the same perod. The second term thus measures the tax costs or benefts resultng from the realzaton of low-taxed characters such as long-term captal gans (losses) and qualfed dvdend ncome. The thrd term measures the tax costs or benefts resultng from the realzaton of hghly taxed characters short-term captal gans (losses) and ordnary ncome (deductons). Fnally, the last term estmates the tax costs (or benefts) of future lqudaton gans (or losses) resultng from the deferral of unrealzed captal gans of the current perod. Our man nsght s that tax effcency can be attrbuted to character and deferral benefts realzed n the current perod. Character benefts result from realzng losses n hghly taxed characters, such as short-term captal losses and ordnary deductons, whle at the same tme realzng gans and ncome n low-taxed characters, such as long-term captal gans and qualfed dvdend ncome. Deferral benefts arse from postponng gan realzaton to future perods such benefts arse from the dfference between the reducton n the current tax lablty and the ncrease n the present value of the future tax lablty. We recognze that tax effcency mght have an ndrect adverse effect on the level of pretax returns of a strategy. However, such an effect s rarely measureable under realstc crcumstances. As a result, our proposed character-deferral decomposton focuses on observable tax outcomes resultng from tradng or holdng of the strategy postons and gnores the potental effects on the pretax return. There are three scenaros of nterest for the purpose of character-deferral decomposton: 1. g L + q and g S + do not have a dfferent sgn (ths ncludes 0 for ether one or both sums) In ths case, all the tax benefts (labltes) come from deferral (acceleraton) of gans: ( 1)[( gl + q) tl + ( gs + t ) H] deferralbeneft( lablty) 2. g L + q and g S + have dfferent sgn and gs + gl + q In ths case, the character beneft s appled to the amount of long-term gans and qualfed dvdends, and the excess short-term captal and ordnary losses yeld the deferral beneft: ( 1)[( gl + q) tl + ( gs + t ) H] = ( gl + q)( th tl ) + ( 1)( gl + q+ gs + ) th characterbeneft( lablty) deferralbeneft( lablty) For example, f short-term captal losses and ordnary deductons add up to 15% of the NAV and long-term captal gans and qualfed dvdends add up to 10% of the NAV, the decomposton wll dentfy 10% multpled by the dfference between hgh and low tax rates as the character beneft and 5% of excess short-term gans and deductons multpled by the hgh tax rate as the deferral beneft. Ths deferral beneft s temporal, as reflected by the ncrease n the expected tax costs of future lqudaton measured separately by ( rpt gl q gs t ) E. 3. g L + q and g S + have dfferent sgn and gl + q > gs + In ths case, the character beneft s appled to the shortterm captal and ordnary losses and the excess longterm gan and qualfed dvdends result n a negatve deferral beneft, or acceleraton of tax labltes to the current perod: ( 1)[( g + q) t + ( g + t ) ] L L S H = ( 1)( gs + )( th tl ) + ( 1)( gl + q+ gs + ) tl characterbeneft( lablty) deferralbeneft( lablty) For example, f short-term captal losses and ordnary deductons add up to 10% of the NAV and long-term captal gans and qualfed dvdends add up to 15% of the NAV, the decomposton wll dentfy 10% multpled by the dfference between hgh and low tax rates as the character beneft and 5% of excess long-term 14 The Tax Benefts of Relaxng the Long-Only Constrant: Do They Come from Character or Deferral? Sprng 2019

17 gans and qualfed dvdends multpled by the low tax rate as the deferral lablty (.e., acceleraton of net gans and ncome to the current perod). Smlar to the aforementoned deferral beneft, ths deferral lablty s temporal, whch s now reflected by the decrease n the expected tax costs of future lqudaton. These three cases can be convenently summarzed by a sngle formula: where ( 1)[( gl + q) tl + ( gs + t ) H] = [ θ ( gl + q) (1 θ )( gs + )] { th tl} characterbeneft( lablty) + [( 1)( gl + q+ gs + )] { θ th + (1 θ) tl} deferralbeneft( lablty) g + S θ= mn max, ,1. g q g L Scenaros 1, 2, and 3, presented earler, correspond to gs + θ=, q = 1, and q = 0, respectvely. For computatonal purposes, ths formula mght serve as a convenent gl + q+ gs + replacement of the condtonal statements n 1, 2, and 3. A p p e n d x D EMPIRICAL METHODOLOGY Our methodology closely follows Salm and Sosner (2018) wth one addton: We also consdered a strategy that s passve wth respect to alpha but s actvely tax managed. Ths secton descrbes the methodology we used to construct nvestment strateges. Actve Strateges Sten and Narasmhan (1999) made a dstncton between actve alpha and actve tax management. Accordng to Sten and Narasmhan, a manager who s actve wth respect to securty selecton but gnores the tax consequences of tradng s passve wth respect to tax. Actve tax management seeks to mprove after-tax returns va acceleraton of captal losses and deferral of captal gans, a technque otherwse known as loss-harvestng. A manager who s passve wth respect to securty selecton (e.g., seekng only to match an ndex) mght thus stll be actve wth respect to tax. Borrowng from Sten and Narasmhan, we call such strateges TMPI. In ths study, we modeled a S TMPI strategy along wth actvely managed tax-aware value momentum strateges consdered by Salm and Sosner. Value and Momentum Strateges We focused on quanttatve strateges that combne value and momentum style factors. Value nvestng has been researched extensvely n academa (see, among many others, Fama and French 1992) and has been wdely appled n the asset management ndustry. We used the most frequently used academc measure of equty value, the book-to-market rato. Consstent wth Asness and Frazzn [2013], we scaled the book value of a company by ts most recent market captalzaton. Momentum effects n US equtes were frst documented by Jegadeesh and Ttman (1993) and Asness (1994). Specfc defntons of relatve strength n past performance may vary; n our study, we used the smplest and the bestknown measure of momentum: the total return over the precedng 12 months, excludng the most recent month. Combnng value and momentum strateges s partcularly benefcal because these strateges tend to exhbt negatve correlaton. In our alpha model, we combned value and momentum wth equal rsk weghts. Ths approach s justfed because, under realstc portfolo constrants and transacton costs, ntegrated strateges domnate naïve combnatons of ndvdual factor tlts before taxes (see Clarke, de Slva, and Thorley 2016 and Ftzgbbons et al. 2017). From an economc perspectve, these styles have perssted across dfferent asset classes, markets, and tme perods (see Asness, Moskowtz, and Pedersen [2013] and Asness et al. [2015]). From a tax perspectve, value sgnals are negatvely related to past returns and momentum sgnals are postvely related. As a result, n addton to ther establshed pretax economcs, these factors exhbt nterestng tax dynamcs (Israel and Moskowtz [2012]). Alpha Model In our study, we begn the portfolo constructon process wth an alpha model that yelds stock-level alphas. Black and Ltterman (1992) showed that alpha forecasts consstent wth rsk and correlaton forecasts are more effectve n portfolo optmzaton. Jones, Lm, and Zangar (2007) adapted ths nsght to factor nvestng n the context of equty portfolos. We used ther methodology to obtan stock-level alphas because our tax-aware approach reles on optmzaton to acheve a balance between the pretax alpha and the tax costs of portfolo rebalancng. The startng pont for our alpha forecasts s a model portfolo, v, whch s long attractve stocks and short Sprng 2019 The Journal of Wealth Management 15