Paris-Dauphine Universiy Sin Sock Reurns over he Business Cycle Augus 007 Sin socks are socks of companies involved in producing obacco, alcohol and gaming. This paper ries o lis he sylized facs ha exis on sin socks from previous academic and professional lieraure, and ess some underlying hypoheses on he U.S. marke. Sin socks ypically have a posiive alpha and a bea below one; hey are considered as value socks, and some argue ha hey ouperform he marke during bad imes. I invalidae some of hese feaures, based on a condiional analysis accouning for differen business condiions. One ineresing resul is ha he abnormal reurn on he sin porfolio is higher during recessions han during expansions. However, he expeced reurns on sin socks are lower han on oher socks, especially during good imes. Inroducion Sin socks are he socks of companies involved in producing alcohol, obacco and gambling. Why is i ineresing o sudy he behavior of sin sock reurns over he business cycle? Sin socks are usually discarded from many funds known as socially responsible. More and more invesors avoid his vice based invesing, because of social norms, or because of social, ehical, and environmenal crieria. Bu here is no evidence ha avoiding sin socks leads o higher porfolio performance. I seems ha invesors include non-financial ases in heir invesmen decision. Are socially responsible invesors also socially responsible consumers? Some people neglec sin socks, bu do hey neglec sin producs? Alcohol, obacco and gambling are a paricular class of producs: heir consumpion consiues an addicive behavior, considered as unhealhy, and hey have no close subsiue, which implies demand inelasiciy. Addiced consumers coninue o drink, smoke, or gamble, even if hey don inves in hese secors. Anecdoal evidence highlighs he virues of vice based invesing. A manager of he American Vice Fund argues ha in aggregae, hese (sin) indusries are defensive in naure and have ended o ouperform when he economy was sressed and he broad marke was sruggling 1. Oher evidence highlighs he fac ha people buy cigarees and alcohol regardless of economic condiions and poliical ensions [see for insance Berman (00), Money Managemen (006), Ahrens (004), and Waxler (004)]. To my knowledge, here is very lile research on sin sock reurns, and resuls are boh ineresing and debaable. In his paper, I review some sylized facs on sin socks from boh academic and professional lieraure, and derive and es some hypoheses abou sin sock reurns over he business cycle. Here is a shor lis of previous evidence on sin socks, which I will discuss below in he lieraure secion. 1. Sin socks as a whole are defensive (bea < 1), bu gaming socks seem o be more aggressive (bea > 1);. They are underpriced and can be considered as value socks (because of high average book-omarke raio); 1 In Vice Fund$ Nice, May 1, 007, New York Pos. - 1 -
Paris-Dauphine Universiy 3. They ouperform oher socks, especially during bad imes, because of limied risk sharing; 4. The excess reurn on sin socks is higher when he marke is down han when he marke is up; 5. Sin socks use more privae deb financing han equiy financing; 6. They have superior financial reporing qualiy han oher socks. I will check and es some of hese paerns, using U.S. daa over he period 196-005. I will also es oher hypoheses, based on previous lieraure on business cycles. For insance, is a condiional model beer in explaining ime-series sin sock reurns han a radiional four-facor model? Or are sin socks more sensiive during expansion or recession periods? I will also sudy he behavior of each sin indusry over he business cycle, namely alcohol, obacco and gaming socks. The remainder of he paper is divided ino five secions, which will expose a lieraure review, explain he hypoheses formulaion and mehodology, describe he daa sample, presen he empirical resuls and draw he conclusion, respecively. Lieraure The sudy of sin sock reurns over he business cycle is moivaed by anecdoal evidence on he fac ha sin socks are likely o be less sensiive o business condiions han oher socks. Moreover, academic research by Hong and Kacperczyk (007) documens ha sin socks are underpriced and ha hey exhibi posiive alphas, which is inconsisen wih porfolio heory. They sudy he performance of sin socks on he American marke, over he period 1965-003. Based on an uncondiional four-facor model which conrols for marke premium, size, book-o-marke and pas reurns, hey argue ha sin socks ouperform he marke because hey are no held by insiuions subjec o social norms. While gauging he relaive imporance of liigaion risk versus his negleced sock effec, he auhors find ha liigaion risk canno explain he abnormal reurns on sin socks. This neglec effec implies limied risk sharing, which implies higher expeced reurns. Hong and Kacperczyk (007) also sudy he financing decision of sin companies and find ha hey use more privae deb financing han equiy financing. Kim and Venkaachalam (006) examine wheher his negleced sock effec is aribuable o differenial informaion risk for hese firms; i.e. sin socks may possess greaer informaion risk due o poor financial reporing qualiy. They show ha sin firms financial reporing qualiy is superior o a conrol group of firms, implying ha he neglec by marke paricipans is no aribuable o financial reporing facors. This superior financial reporing qualiy is explained by greaer regulaory scruiny. Hong and Kacperczyk (007) are he firs o sudy sin socks as a porfolio sraegy. They form a sin porfolio and compare is average reurn o an indusry-comparable porfolio. The performance of sin socks over he business cycle has no been examined from an indusry poin of view, excep for gambling socks. According o Goodall (1994), who looks a gaming sock reurns over a 0 year horizon, gaming socks end o be more volaile han he marke as a whole, and some special evens can cause he gaming socks o move in a direcion opposie o he general marke. Boh resuls are - -
Paris-Dauphine Universiy driven by he relaively small capializaion of hese socks. In addiion, he shows ha gaming socks are more sensiive o sock marke declines. Chen and Bin (001) sudy US gaming sock reurns over up and down markes ; using a CAPM regression wih ime-varying alphas and beas, in a GARCH esimaion framework. They show ha invesors earned a negaive excess reurn and bore an abovemarke-average sysemaic risk for holding socks of gaming companies over 1993-1997. Afer changes in marke condiions were considered, however, he gaming socks provided heir invesors wih a relaively normal reurn (alpha no significan) and an abnormal risk level agains he marke average (bea above one). Bu he excess reurn on sin socks was higher during down markes han during up markes. They also invesigae he effecs of legislaion evens on gaming sock reurns, and find ha small casino operaors are more reacive o deregulaion / regulaion acions in comparison wih large casino gaming firms. Hong and Kacperczyk (007), and Kim and Venkaachalam (006) agree on he fac ha sin socks earn abnormal reurns because of negligence. Bu is his risk-adjused excess reurn consisen wih he neglec effec sory, or jus an anomaly ha should disappear? If you hink abou he join hypohesis problem, his anomaly ha canno be explained using he four-facor model suggess eiher marke inefficiency, or he need of anoher asse pricing model. In his paper, I ry o miigae his behavioral explanaion, which assesses ha invesors have non-financial ases, by allowing for ime-varying expeced reurns, consisen wih he raional pricing heory. Asse pricing heory describes prices and reurns in erms of condiional momens. If he price of any asse p gives righs o a payoff x + 1, he basic asse pricing equaion, in a SDF (sochasic discoun facor) represenaion, is E ( m x ) p = +1 +1, where + 1 m is he SDF, and E means expecaion condiional on ime- informaion. In erms of reurn, he equaion becomes ( m R Z ) 1 = E + 1 + 1, where R+ 1 is a gross reurn, and Z denoes he informaion se. Condiional asse pricing models use predeermined variables o capure he ime-series variaion of risk premiums. Assuming he exisence of some reurn predicabiliy, hey use a se of insrumens z Є Z, for which E(R +1 Z ) varies over ime. Condiioning variables are ofen chosen from variables ha can predic business cycle or forecas he marke premium 3. Evidence has been shown empirically ha sock reurns are predicable from variables ha A bull (or up) marke is defined as a period when he monhly reurn on he value-weighed CRSP sock index exceeds he monhly reurn on a 90-day US Treasury Bill. Conversely, if he sock marke does no perform beer han he risk-free ineres rae, such a period is defined as a bear (or down) marke. 3 Chen, Roll and Ross (1986), Ferson and Harvey (1991) and Campbell (1993) find ha reurns are predicable from macroeconomic variables such as he erm spread, he defaul spread, he indusrial producion and he unexpeced inflaion. Jagannahan and Wang (1996) use he defaul premium as a proxy for condiional marke risk premium. Over a long horizon, expeced reurns are also predicable from dividend yields and erm spreads [Fama and French (1988)] and earnings-o-price raio [Campbell and Shiller (1988)]. Oher condiioning variables include he invesmen growh [Cochrane (1996)], he consumpion-wealh raio [Leau and Ludvigson (001a)], and he labor income o consumpion raio [Sanos and Veronesi (006)]. Leau and Ludvigson (001b) documen ha HML has a ime-varying bea on boh he marke reurn and on consumpion. Though here is very lile uncondiional correlaion beween HML and recession measures, Leau and Ludvigson - 3 -
Paris-Dauphine Universiy are informaive abou he business cycle. Lamon (001) consrucs economic-racking porfolios desined o rack economic news, raher han conemporaneous macroeconomic variables 4, and finds significan risk premiums associaed wih hese porfolios. Chordia and Shivakumar (00b) consruc an earnings based zero-invesmen porfolio ha is relaed o fuure macroeconomic condiions. This porfolio is long on socks wih high pas earnings changes and shor on socks wih low pas earnings changes. The auhors show ha he reurn on his long-shor porfolio is correlaed wih fuure growh in GDP, indusrial producion, consumpion, labor income, inflaion and T-bill reurns. The evidence suggess ha he equiy premium is ime-varying, and his variaion in risk can cause uncondiional models o fail. Many empirical ess of condiional models use only one predeermined variable a a ime. Bu in a muli-facor framework, he use of only one informaion variable may no be appropriae, because differen risk facors may need differen condiioning variables. Wang (006) runs a horse race among eigh risk facors and eigh condiioning variables in order o find which risk facor, condiioning on which variable, beer explains he cross-secional variaion of average reurns. He finds ha a consumpion growh facor, condiioning on lagged business income growh, is he mos successful in explaining cross-secional variaion of average reurns. Anoher branch of lieraure examines he behaviour of sock reurns for differen marke condiions, referred o as expansion and recession periods. Fama and French (1989), and Chordia and Shivakumar (00a, 00b) show ha sock reurns are predicable from macroeconomic variables, based on Naional Bureau of Economic Research (NBER) business cycles. 5 They show ha expeced reurns are higher a business cycle roughs han a peaks. In oher words, here is a negaive relaionship beween expeced reurns and business cycles. Moreover, earnings [Chordia and Shivakumar (00b)] and profis o momenum sraegies [Chordia and Shivakumar (00a)] are relaed o business condiions. Using a dividend discoun model, DeSefano (004) finds ha expeced sock reurns vary inversely wih economic condiions. Reurns decrease hroughou expansions and become negaive during he firs half of recessions; and reurns are larges during he second half of recessions. Finally, Kosowski (006) shows ha risk-adjused muual fund performance (alpha) is higher in recession han expansion periods. This srand of lieraure underlines he imporance of business cycles for sock reurn predicabiliy. The ime-varying risk premium, condiional on he sae of he economy, can be due o consumpion smoohing or a variaion in macroeconomic risks [Fama and French (1989)]. Consumpion smoohing means ha consumers/invesors save more when imes are good and save less (consume more) during bad ime. This has a direc impac on he expeced documen ha HML is sensiive o bad news, in bad imes. See also Fama and Schwer (1977), Campbell (1987), Fama and French (1989), Ferson (1989), and Evans (1994) who use he same se of informaion variables. 4 Chen, Roll and Ross (1986) use conemporaneous macroeconomic facors. However, Chan, Karceski and Lakonishok (1998) conclude ha porfolios consruced o mimic he conemporaneous macroeconomic facors are basically noise. 5 Periods of expansion begin a he rough dae and end a he peak dae, and periods of recession begin a he peak dae and end a he rough dae. - 4 -
Paris-Dauphine Universiy reurn, i.e., expeced reurns are high during bad imes and low during good imes. Is he consumpion of alcohol, obacco and gambling also smoohed over good imes and bad imes? The consumpion of hese producs is considered as an addicive behaviour. Becker and Murphy (1988) define an addiced person as follows: a person is poenially addiced o c if an increase in his curren consumpion of c increases his fuure consumpion of c. The consumpion of sin producs should no (or less) be dependen on he business cycle, since consumers are addiced. Bu does his addicion have an impac on he ime-variaion of sin sock reurns? Braun and Larrain (005) sudy he imporance of finance in he propagaion of business cycles (under he hypohesis of imperfec financial markes) hrough he noion of exernal finance dependence of an indusry. They show ha recessions have a higher impac on indusries which are more dependen on exernal funds. According o heir daa, he exernal finance dependence is very low for he obacco indusry. Moreover, Hong and Kacperczyk (007) show ha sin companies use relaively more privae deb financing han oher companies, implying ha sin socks should be less sensiive o recessions ha oher socks. This lieraure review makes i ineresing o sudy sin sock reurns over business cycles from wo poins of view: sin socks as a whole compared o oher socks, and each sin indusry compared o one anoher. Tesable hypoheses and mehodology Before esing my hypoheses, I use a radiional uncondiional four-facor model. The uncondiional four-facor model is usually wrien in a bea represenaion as R R f, = α + β 1 EXRm + β SMB + β 3 HML + β 4 MOM + ε, =1,,T (1) where R R f is he reurn on various porfolios, ne of he risk-free rae; EXRm is he marke premium (marke reurn in excess of he risk-free rae); SMB is he size premium (small capializaion minus big capializaion porfolios); HML is he value premium (high book-o-marke minus low book-o-marke porfolios); and MOM is he momenum facor (pas winner porfolio minus pas looser porfolio). I consruc several porfolios; he firs one conains all sin socks, he second one conains all oher socks, and he hird one is a long-shor porfolio, long on sin socks and shor on oher socks. In order o evaluae he accuracy of my porfolios, I can compare my regression esimaes o hose from Hong and Kacperczyk (007). I should find a significan alpha on he sin porfolio and/or on he long-shor porfolio. I hen invesigae hree differen issues. Firs, I examine wheher a condiional model can explain he ime-series variaion of expeced reurns on sin socks no explained by he four-facor model. Second, I sudy wheher expeced reurns on sin socks vary negaively wih he business cycle, consisen wih previous lieraure on business cycles, or wheher he addicive behaviour of sin - 5 -
Paris-Dauphine Universiy consumpion has an impac on he ime-variaion of sock reurns. I finally invesigae he behaviour of each sin indusry: alcoholic beverages, obacco, and gambling, respecively. Predicabiliy of sin sock reurns from macroeconomic variables: Condiional asse pricing presumes he exisence of some reurn predicabiliy. There should be insrumens Z for which E(R +1 Z ) varies over ime. As shown previously in he lieraure, many informaion variables can empirically predic sock reurns. These variables are he erm spread, he defaul spread, he indusrial producion, he unexpeced inflaion, he dividend yield, he consumpion-wealh raio, he labor income o consumpion raio, and ohers. If hese macroeconomic variables can explain he variaion of risk premiums, any excess reurn (or alpha) should disappear. The firs hypohesis I es is wheher he abnormal risk-adjused reurn on sin socks survives a condiional four-facor model ha accouns for ime-varying premiums. In order o es his hypohesis, I urn o a condiional four-facor model. Asse pricing generally describes sock reurns in erms of condiional momens: E -1 (R ) R f, = E -1 (EXRm ) + E -1 (SMB ) + E -1 (HML ) + E -1 (MOM ). () The condiional expecaion E -1 is based on he informaion se a ime -1, denoed as I -1. In order o ransform he uncondiional model ino a condiional model, he facors mus depend on he informaion se a ime -1. In ha case, he parameers are ime-varying and equaion () is wrien as R R f, = α -1 + β 1, -1 EXRm + β, -1 SMB + β 3, -1 HML + β 4, -1 MOM + ε. (3) Unforunaely, he informaion se I -1 is no fully observable and no all variables in I -1 can be condiioning variables. There are hree requiremens for macroeconomic variables o be legiimae insrumens [Hodrick and Zhang (001)]. Firs, hey should be in he informaion se a ime -1. Second, hey should capure he ime variaion of risk premiums. Third, he number of condiioning variables canno be oo large; oherwise he number of parameers will be oo large o obain reliable esimaion. Assuming ha some variables Z -1 are par of I -1 and are available for all invesors, he expecaion E -1 is condiioned on hese informaion variables Z -1. Cochrane (1996) assumes ha he coefficiens in equaion (3) are linear funcions of condiioning variables. If z -1 Є Z -1 is a predeermined variable, he parameers in equaion (3) are defined as α -1 = α 0 + α 1 z -1, β 1, -1 = δ 0 + δ 1 z -1, β, -1 = φ 0 + φ 1 z -1, β 3, -1 = γ 0 + γ 1 z -1, β 4, -1 = η 0 + η 1 z -1. The condiional model in equaion (3) can be rewrien as an uncondiional facor model wih consan coefficiens: R R f, = α 0 +α 1 z -1 + δ 0 EXRm + φ 0 SMB + γ 0 HML + η 0 MOM - 6 -
Paris-Dauphine Universiy + z -1 (δ 1 EXRm + φ 1 SMB + γ 1 HML + η 1 MOM ) + ε. In his equaion here is only one condiioning variable z -1. Many empirical ess of condiional models use only one predeermined variable a a ime. Since we esimae a muli-facor model, he use of only one informaion variable may no be appropriae. The previous specificaion implicily assumes ha z -1 capures he ime variaion of risk premiums for all four facors. Bu differen risk facors may need differen condiioning variables. If we consider one differen condiioning variable for each facor, we have he following specificaion: R R f, = α 0 + α 1 z 0, -1 + δ 0 EXRm + φ 0 SMB + γ 0 HML + η 0 MOM + δ 1 z 1, -1 EXRm + φ 1 z, -1 SMB + γ 1 z 3, -1 HML + η 1 z 4, -1 MOM + ε. As Wang (006) poins i ou, I need o ry several predeermined variables and es wheher I can find one (or more) condiioning variable(s) for which he unexplained reurn on he sin sock porfolio will vanish. If no, his will mean ha he abnormal reurn canno be explained by a dynamic sraegy replicaing for he risk exposure, or for he business condiions. H1: There exis some predeermined macroeconomic variables ha accoun for imevarying risk premiums on sin socks, and make he abnormal risk-adjused reurn on hese socks disappear. Sin sock reurns over he business cycle: We have seen ha consumpion smoohing implies expeced reurns o be higher during recessions han during expansions. Wha abou sin socks? Sin producs are characerized by an addicive consumpion, meaning ha people coninue o smoke, drink and gamble, whaever he economic condiion. Thus, he consumpion of alcohol and obacco does no vary much wih he business cycle, bu can even vary couner-cyclically. The second hypohesis I es is wheher he expeced reurn on sin socks is higher during recession han during expansion, jus like oher socks. H: Expeced reurns on sin socks are higher during recession periods han during expansion periods. Moreover, i has been shown in he professional lieraure ha sin socks are defensive and end o ouperform he marke during bad imes. Sin indusries seem o be less sensiive o recessions. Bu his assessmen in based on reurns no adjused for he risk facors. I hus wan o es wheher sin socks are defensive agains he marke and compared o oher socks, and wheher hey ouperform boh he marke and oher socks during bad imes, even afer conrolling for he four-facor risk premiums. H3: The bea on sin socks is less han one and/or he bea on he long-shor porfolio is negaive. H4: The abnormal reurn on sin socks is higher during bad imes han during good imes. - 7 -
Paris-Dauphine Universiy In order o es hese hree hypoheses, I look a recession and expansion periods. These periods can be defined using a specific macroeconomic variable, like he growh in GDP, or using a se of economic variables, such as he NBER classificaion. The NBER defines periods of recession and expansion by publishing peak and rough daes in economic aciviy. Periods of expansion begin a he rough dae and end a he peak dae, and periods of recession begin a he peak dae and end a he rough dae. Oher sudies on ime-varying risk premiums alk abou marke condiions, referring o he sock marke movemen beween up and down [see for insance Chen and Bin (001)]. Each monh, he sock marke is up (or bull) if he marke premium is posiive, i.e. if he reurn on he marke porfolio exceeds he reurn on he risk-free asse. Conversely, he marke is down (or bear) if he marke premium is negaive. I hus es my hypoheses using boh business cycle classificaions: expansion vs. recession periods, and up vs. down markes. I follow Chen and Bin (001) who examine he US gaming sock performance across various marke condiions. They es a CAPM model, conrolling only for he marke risk premium, wih a dummy variable for up markes (when ExRm is posiive). I wan o es for ime-varying parameers in a four-facor specificaion: R R f, = α* + δ*exrm + φ*smb + γ*hml + η*mom + α r D r + δ r D r EXRm + φ r D r SMB + γ r D r HML + η r D r MOM + ε. where R is he reurn on equally-weighed porfolios described below, and D r equals one if he monh is in an recession period, zero oherwise. In order o circumven he poenial condiional heeroskedasiciy bias in he usual ime-series process, and hus provide relaively efficien esimaes, his hypohesis is esed in a GARCH esimaion framework in which he error erm is modelled as follows: ( 0, ) ε ~ N σ, where σ γ + γ σ γ ε. = 0 1 1 + 1 I firs classify he economic condiions beween expansion and recession periods, using he NBER daes and he dummy variable D r. I also run he regression using a dummy variable accouning for down sock markes. D d equals one if ExRm is negaive, zero oherwise. Indusry porfolios: Sudying sin indusries as a whole may be inadequae, since hey probably do no behave he same over he business cycle. One can invesigae separaely each vice indusry and see wheher hey are homogeneous in erms of ime-varying risk premiums. For he gaming indusry, previous sudies sugges ha gaming socks have, on average, beas higher han one. Bu i is no clear from hese sudies wheher marke changes affec gaming sock reurns. On he one hand, Goodall (1994) shows ha gaming sock reurns are more sensiive o sock marke declines han o marke upurns. On he oher hand, Chen and Bin (001) conclude ha marke urnings have no effec on gaming sock reurns. They find ha (1) he sysemaic risk for gaming socks is high, and does no change significanly across up and down markes, and () he gaming porfolio excess reurn is - 8 -
Paris-Dauphine Universiy affeced by marke urnings, implying ha he excess reurn is significanly smaller when he sock marke is up han when he sock marke is down. Since alcohol and obacco socks have no been sudied in he lieraure, and since here are close complemens, one can invesigae wheher hey have he same behaviour over good and bad imes. We know ha he sin indusries have a low exernal dependence, meaning ha i does no depend largely on exernal funds. Braun and Larrain (005) show ha recessions have a lower impac on indusries ha are less dependen on exernal finance. Given he above findings, I es he hypoheses H-H4 for each sin indusry. I use he same mehodology as earlier for my hree sin porfolios: gambling, alcoholic beverages, and obacco. H5: Expeced reurns on each sin sock porfolio are higher during recession periods han during expansion periods. H6: The bea on gaming socks is more han one and he bea on alcohol and obacco socks is less han one. H7: The abnormal reurn on each sin sock porfolio is higher during bad imes han during good imes. The Daa I use daa on non-financial U.S. socks over he period 196-005. My daa on monhly reurns and monhly shares ousanding come from CRSP and cover all NYSE, AMEX and NASDAQ socks. I include in my daabase only ordinary common socks, wih share code equal o 10 or 11. Socks wih less han 1 monhs of daa are discarded from he sample, as well as ouliers. Ouliers are firs defined as reurn observaions above 350%. I hen deermined for each monh an abnormal level of reurn, given he monhly disribuion of reurns. For insance, ouliers are reurns above 00% during he period 1941-1974, and reurns above 300% over 1999-000. To proxy for he marke porfolio, I use monhly reurns of he CRSP value-weighed porfolio. SMB, HML and MOM are porfolio reurn series downloaded from Ken French s websie. Sin socks are socks in he following indusries: alcohol, obacco and gaming. I use he Fama and French (1997) classificaion ino 48 indusries o assign socks in he beer and obacco groups 6. This classificaion does no differeniae gaming socks from oher enerainmen socks. I hus use he NAICS classificaion o selec gaming socks 7. There are a oal of 183 sin socks over he period 196-005, wih well-known companies such as Bacardi, Bally, Alria, Anheuser Busch Co., Mandalay, and Trump Hoels. My lis of sin socks is smaller han he lis used by Hong and Kacperczyk (007), since I do no include holdings 8. I consruc several porfolios of sin socks; he firs one conains he lis of all sin socks described above, and he hree ohers are indusry porfolios, including alcohol socks, obacco socks and gaming socks respecively. I also form a non-sin sock porfolio, which conains all oher socks presen in he daabase, and a 6 Group 4 for beer or alcohol, and group 5 for smoke or obacco. 7 Socks wih he codes: 713, 7131, 71310, 7139, 71390, 711, and 7110. 8 Hong and Kacperczyk (007) include 193 disinc sin companies over he period 196-004. - 9 -
Paris-Dauphine Universiy zero-invesmen porfolio, long on sin socks and shor on oher socks. All hese porfolios are equally-weighed every monh 9. I invesigae some macroeconomic indicaors ha could proxy for omied variables in he ime-series of sock reurns, by looking a heir auocorrelaion, and heir correlaion wih porfolio reurns. I choose predeermined variables (wih a one-monh lag) ha are mos used in condiional models: he dividend yield, he risk-free rae, he erm srucure, he defaul spread, he unanicipaed inflaion, and he indusrial producion. The dividend yield is calculaed as he difference beween he value-weighed reurn wih dividends and he value-weighed reurn wihou dividends from he CRSP value-weighed index. The risk-free rae is he 3-monh T-bill yield. The erm spread is he difference beween 10-year Treasury bond yield and 3-monh T-bill yield. The defaul premium is defined as he difference beween BAA-raed bond yield and AAA-raed bond yield. The daa on ineres raes come from he Federal Reserve. The unexpeced inflaion is he difference beween acual and forecased inflaion. Jagannahan and Wang (1996) use he change in inflaion rae, i.e., he difference beween inflaion rae a and inflaion rae a -1. The daa for inflaion rae, no seasonally adjused, are from Daasream. The change in indusrial producion is he growh rae over he previous monh. The daa on indusrial producion are from he Board of Governors of he Federal Reserve Sysem. I also use hree oher condiioning variables which are he growh in GDP, he consumpion o wealh raio, moivaed by Leau and Ludvigson (001), and he labor income o consumpion raio, following Sanos and Veronesi (006). I downloaded he daa from Marin Leau s webpage, who only provides quarerly daa. The business cycle classificaion ino expansion and recession periods is obained from he NBER websie. Finally, up and down markes are defined according o he sign of he monhly marke risk premium. The sock marke is up when he marke premium is posiive or null, whereas he sock marke is down when he risk premium is negaive. Empirical Resuls Panel A of Table 1 gives he coefficien esimaes of he uncondiional regression in equaion (1) for he sin, non-sin and long-shor porfolios. Consisen wih Hong and Kacperczyk (007), he significan alpha of boh he sin porfolio and he long-shor porfolio indicaes reurns no capured by he four-facor model. Moreover, he negaive and significan coefficien for he marke risk premium suggess ha he long-shor porfolio is no a zero-bea porfolio, bu varies negaively wih he marke. One can inerpre in he same way he negaive coefficien for SMB. The coefficien on MOM is significanly posiive, poining o a relaion beween he zero-invesmen porfolio and he momenum effec. Panel B of Table 1 shows esimaes of he same regression by disinguishing all hree sin indusries. Beas are significanly lower han one for boh alcohol and obacco indusries, whereas he gaming indusry has, on average, a bea greaer han one. The alcohol sock porfolio does no exhibi 9 I also consruc value-weighed porfolios, bu do no repor resuls, qualiaively similar o hose wih equallyweighed porfolios. - 10 -
Paris-Dauphine Universiy any abnormal reurn. On he conrary, he obacco and gambling porfolios have a significan excess reurn of around 5.7 percen and 7.4 percen (annualized) respecively. These findings sugges ha all hree sin indusries do no behave he same. Predicabiliy of sin sock reurns from macroeconomic variables: In order o sudy sin sock reurns over he business cycle, I use macroeconomic variables ha proxy for he business condiions. Table gives he correlaion marix of he monhly informaion variables defined above. None of hese variables are highly correlaed, i.e., hey all capure differen risks and are no redundan. One can consider all of hem as poenial condiioning variables. I also have access o quarerly daa which could proxy for he business cycle: he growh in GDP, he consumpion growh and he growh in labor income. Table 3 gives he correlaion coefficiens of hese hree condiioning variables (conemporaneous and wih one-quarer lag) wih he sin sock porfolio, he non-sin sock porfolio and he long-shor porfolio. The sin sock porfolio excess reurns seem o vary negaively wih he lagged growh in GDP and he lagged consumpion growh. Figure 1 shows he reurns on he longshor porfolio versus he growh in GDP over he period 1948-005, which highlighs he negaive correlaion beween boh variables. Figures, 3, and 4 draw he consumpion o wealh raio versus he alcohol, obacco, and gambling porfolios respecively. Again one can see a negaive correlaion beween he variables, meaning ha he acual reurns on he sin porfolios are high during good imes (when he invesmen is high relaive o wealh) and low during bad imes. Tables 4 and 5 give evidence for he firs hypohesis. Table 4 shows coefficien esimaes from regressions wih monhly variables. The abnormal reurn on he long-shor porfolio disappears for 5 macroeconomic variables ou of 6. All informaion variables have differen effecs on he risk premiums. For insance, he marke premium and he value premium on he long-shor porfolio are negaively correlaed wih he erm spread; and he size premium is posiively relaed o he defaul spread, he dividend yield and he erm spread. If one look a he quarerly variable regressions (Table 5), hey all help explain he excess reurns on he sin and he long-shor porfolios. I appears ha he consumpion growh has more impac on he momenum facor, he growh in GDP on he size facor, and he growh in labor income on boh facors. These resuls imply ha a condiional four-facor model is beer in explaining he ime-series reurn on he sin porfolio. Sin sock reurns over he business cycle: In he second se of hypoheses, I look a expeced reurns, beas, and abnormal reurns on sin socks over good and bad imes. Before sudying he expeced and abnormal reurns, le s look a acual reurns on sin socks vs. oher socks. Table 6 gives he reurns o he zero-invesmen porfolio for differen periods, based on wheher he period is a recession or an expansion. During expansion periods, he reurns on he long-shor porfolio are on average significan and negaive, i.e., here is a negaive premium on sin socks compared o oher socks when he business condiions are good. During conracion periods, however, he reurns on he - 11 -
Paris-Dauphine Universiy long-shor porfolio are posiive (excep for hree subperiods) and significan on average (equal o 0.83% per monh), implying a posiive premium on sin socks. When business condiions are bad, i seems ha sin socks earn higher reurns han oher socks. Moreover, he difference in reurns on he long-shor porfolio beween he wo periods is significan a he one percen level, which is consisen wih he idea ha sin socks behave differenly han oher socks, on average. In order o es he second hypohesis Expeced reurns on sin socks are higher during recession periods han during expansion periods I regress he reurns on my sin and long-shor porfolios on a four facor model, using a dummy variable allowing for recession periods. Regression esimaes are given in Table 7. The parameers of he GARCH equaion are highly significan, meaning ha here is condiional heeroskedasiciy in he ime-series process. 10 The excess reurn on he sin porfolio has a posiive coefficien on γ r, implying ha he value premium is higher during bad imes han during good imes, which is consisen wih he hypohesis. On can look a he long-shor porfolio regression in order o compare he ime-varying expeced reurns on boh sin and non-sin socks. During expansion periods, he long-shor porfolio has negaive loadings on he marke, he size and he value facors, and a posiive loading on he momenum facor. On average, he expeced reurn on sin socks is lower han on oher socks. During recession periods, however, he premiums on sin socks increase relaive o oher socks, which again is consisen wih he hypohesis. The hird hypohesis The bea on sin socks is less han one and/or he bea on he long-shor porfolio is negaive is direcly acceped from Panel A of Table 1, and again in Table 7. The bea on sin socks is 0.75, which is significanly differen from 1, and he long-shor porfolio has a negaive bea of -0.8, implying ha sin socks are defensive agains he marke. However, hese beas don vary over expansion and recession periods (δ r no significan in Table 7). Table 7 also gives evidence for he fourh hypohesis The abnormal reurn on sin socks is higher during bad imes han during good imes since α r is posiive and significan a a 1 percen level for boh he sin and he long-shor porfolios. This hypohesis is also esed over up and down markes, and resuls are given in Table 8. There is no abnormal reurn on sin socks when he dummy variable accouns for down markes insead of recessions. However, he hypoheses H and H3 are no rejeced in Table 8. Indusry porfolios: In his secion I perform he same GARCH esimaion wih a four-facor model and a dummy variable allowing for conracion periods (or down markes) for each sin indusry: alcohol, obacco, and gaming. Tables 9 and 10 are nearly he same as Tables 7 and 8, excep ha he dependen variables here are he excess reurns on alcohol socks, obacco socks, and gaming socks, respecively. The resuls can be summarized as follows: 10 In all he analysis, I omi o repor he GARCH esimaes (all saisically significan) for ransparency preoccupaions. - 1 -
Paris-Dauphine Universiy H5: In Table 9, he loadings for he recession periods are eiher posiive or no significan, implying ha he expeced reurns on each sin porfolio are higher during conracion han during expansion periods. H6: The bea on alcohol and obacco socks is significanly lower han one, and he bea on gaming socks is no significanly differen from one (hese resuls are consisen wih Panel B of Table 1). Moreover, hese beas do no vary over he business cycle. H7: For he alcohol and obacco indusries, α r is significanly posiive; implying ha he average excess reurns on boh obacco and alcohol socks are higher during recessions han during expansions. Conversely, α r for he gaming indusry is no saisically differen from zero, implying ha he excess reurn for he gaming porfolio does no vary wih he business cycle. These findings sill hold when he dummy variable allows for up and down sock markes (Table 10). Comparing he hree sin indusries, I can conclude ha alcohol socks and obacco socks behave very similarly, whereas gaming socks have a specific behavior over he business cycle. I seems ha previous findings on he sin sock porfolio are driven by alcohol and obacco socks, even if gaming socks accoun for around 30 percen of he sin sample. Conclusion This paper highlighs some sylized facs abou sin sock reurns on he U.S. marke over he period 196-005. Firs, ime-varying expeced reurns are a plausible explanaion for he ouperformance of sin socks in he long run. Indeed, he abnormal reurn on boh he sin porfolio and he long-shor porfolio vanishes when allowing for reurns o vary wih predeermined macroeconomic variables. The erm spread urns o be he bes informaion variable, since i varies negaively wih boh he marke premium and he value premium, and posiively wih he size premium. The momenum premium, however, varies mosly wih he consumpion growh, he relaionship being negaive. Second, sin socks as a whole are defensive (bea around 0.75), his resul being driven by alcohol and obacco socks; whereas gambling socks are marke neural. All hese marke beas, however, are no ime-varying wih he business cycle. The long-shor porfolio has a negaive bea, implying ha his sraegy is marke conrarian. Third, he risk premiums, and hus he expeced reurns on sin socks are higher during conracion han expansion periods, and his resul is sill valid when looking a each sin indusry. However, when comparing beween sin and non-sin socks, expeced reurns on sin socks are lower han on oher socks, especially during good imes. Finally, he abnormal reurn on he sin porfolio is higher during recessions han during expansions; and again he resul holds only for obacco and alcohol socks, whereas he abnormal reurn on gaming socks does no vary wih he business cycle. Furher research will be o invesigae he reasons for hese findings, using specific models such as he raional addicion model. - 13 -
Paris-Dauphine Universiy Reference: Braun Maias and Borja Larrain (005) Finance and he Business Cycle: Inernaional, Iner-Indusry Evidence. Journal of Finance 60, 3, 1097 118. Breeden Douglas (1979) An Ineremporal Asse Pricing Model wih Sochasic consumpion and Invesmen Opporuniies Journal of Financial Economics 7, 65 96. Campbell John (1987) Sock Reurns and he Term Srucure. Journal of Financial Economics 18, 373-399. Campbell John and Rober Shiller (1988) Sock Prices, Earnings, and Expeced Dividends. Journal of Finance 43, 661 676. Carhar Mark (1997) On Persisence in Muual Fund Performance. Journal of Finance 5, 57 8. Chan Louis K. C., Jason Karceski, and Josef Lakonishok (1998) The Risk and Reurn from Facors. Journal of Financial and Quaniaive Analysis 33, 159-88. Chen Dar-Hsin and Feng-Shun Bin (001) Effecs of Legislaion Evens on US Gaming Sock Reurns and Marke Turnings. Tourism Managemen, 539-549. Chen Nai-Fu, Richard Roll and Sephen Ross (1986) Economic Forces and he Sock Marke Journal of Business 59, 383-404. Chordia Tarun and Lakshmanan Shivakumar (00a) Momenum, Business Cycle and Time- Varying Expeced Reurns. Journal of Finance 57, 985-1019. Chordia Tarun and Lakshmanan Shivakumar (00b) Earnings, Business Cycle and Sock Reurns Working Paper. Cochrane John (1996) A Cross-Secional Tes of an Invesmen-Based Asse Pricing Model. Journal of Poliical Economy 104, 57 61. DeSefano Michael (004) Sock Reurns and he Business Cycle. The Financial Review 39, 4, 57-547. Evans M. (1994) Expeced Reurns, Time Varying Risk, and Risk Premia. Journal of Finance 49, 655-679. Fama Eugene F. and Kenneh R. French (1988) Dividend Yields and Expeced Sock Reurns. Journal of Financial Economics, 3 7. Fama Eugene F. and Kenneh R. French (1989) Business Condiions and Expeced Reurns on Socks and Bonds. Journal of Financial Economics 5, 3 49. Fama Eugene F. and Kenneh R. French (199) The Cross-Secion of Expeced Reurns. Journal of Finance 47,, 47 465. Fama Eugene F. and Kenneh R. French (1993) Common Risk Facors in he Reurns on Socks and Bonds. Journal of Financial Economics 33, 3 56. Fama Eugene F. and Kenneh R. French (1997) Indusry Cos of Equiy. Journal of Financial Economics, 153 193. - 14 -
Paris-Dauphine Universiy Fama Eugene F. and Schwer (1977) Asse Reurns and Inflaion. Journal of Financial Economics 5, 115-146. Ferson Wayne (1989) Changes in Expeced Securiy Reurns, Risk, and he Level of Ineres Raes. Journal of Finance 44, 1191-117. Ferson Wayne and Campbell R. Harvey (1991) Sources of Predicabiliy in Porfolio Reurns. Financial Analyss Journal May/June, 49-56. Goodall Leonard (1994) Marke Behavior of Gaming Socks: an Analysis of he Firs Tweny Years. Journal of Gambling Sudies 10, 4, 33-337. Harvey Campbell R. (1989) Time-Varying Condiional Covariances in Tess of Asse Pricing Models Journal of Financial Economics 4, 89-317. Hong Harrison and Marcin Kacperczyk (007) The Price of Sin: The Effecs of Social Norms on Markes. Working Paper. Hodrick Rober and Xiaoyan Zhang (001) Evaluaing he Specificaion Errors of Asse Pricing Models. Journal of Financial Economics 6, 37-376. Jagannahan Ravi and Zhenyu Wang (1996) The Condiional CAPM and he Cross-Secion of Expeced Reurns Journal of Finance 51, 1, 3-53. Jagannahan Ravi and Yong Wang (005) Consumpion Risk and he Cos of Equiy Capial. NBER Working Paper. Kim Irene and Mohan Venkaachalam (006) Are Sin Socks Paying he Price for heir Accouning Sins? Working Paper. Kosowski Rober (006) Do Muual Funds Perform When I Maers Mos o Invesors? US Muual Fund Performance and Risk in Recessions and Expansions. Working Paper. Lamon Owen (001) Economic Tracking Porfolios. Journal of Economerics 105, 1, 161 184. Leau Marin and Sydney Ludvigson (001a) Consumpion, Aggregae Wealh and Expeced Sock Reurns. Journal of Finance 56, 3, 815 849. Leau Marin and Sydney Ludvigson (001b) Resurrecing he (C)CAPM: A Cross-Secional Tes when Risk Premia are Time-Varying. Journal of Poliical Economy 109, 138 87. Liew Jimmy and Maria Vassalou (000) Can Book-o-Marke, Size and Momenum Be Risk Facors ha Predic Economic Growh? Journal of Financial Economics 57, 1 46. Linner John (1965) The Valuaion of Risk Asses and he Selecion of Risky Invesmens in Sock Porfolios and Capial Budges. Review of Economics and Saisics 47, 13 37. Roll Richard (1977) A Criique of he Asse Pricing Theory s Tess. Par I: On Pas and Poenial Tesabiliy of he Theory. Journal of Financial Economics 4, 19 176. Sanos Tano and Piero Veronesi (006) Labor Income and Predicable Sock Reurns. Review of Financial Sudies 19, 1 44. - 15 -
Paris-Dauphine Universiy Sharpe William F. (1964) Capial Asse Prices: A Theory of Marke Equilibrium under Condiions of Risk. Journal of Finance 19, 3, 45 44. Vassalou Maria (003) News Relaed o Fuure GDP Growh as a Risk Facor in Equiy Reurns. Journal of Financial Economics 68, 47-73. Wang Yong (006) Condiioning Informaion, Facor Risk Premia, and he Cross-Secion of Sock Reurns Working Paper, Hong Kong Polyechnic Universiy. Yogo Moohiro (006) A Consumpion-Based Explanaion of Expeced Sock Reurns. Journal of Finance 61,, 539 580. - 16 -
Paris-Dauphine Universiy Table 1. Resuls from regression of my various porfolios on Fama-French and Momenum facors In Panel A, he reurns on he sin sock porfolio, he non-sin sock porfolio, and he zero-invesmen porfolio (long on sin socks and shor in oher socks) are regressed on he Fama-French facors and he momenum facor. In Panel B, he reurns on he indusry porfolios (alcohol, obacco and gaming) are regressed on he Fama-French facors and he momenum facor. Sandard errors are given in parenheses. Panel A: Sin, non-sin and long-shor porfolios Inercep ExRm SMB HML MOM R-sq Sin Rf 0.0030*** 0.75*** 0.6063*** 0.161*** -0.0418* 0.770 (0.0010) (0.0196) (0.0303) (0.09) (0.033) NoSin Rf 0.001*** 1.03*** 0.8786*** 0.1978*** -0.1315*** 0.969 (0.0005) (0.0090) (0.0140) (0.0135) (0.0108) Sin NoSin 0.0017* -0.8*** -0.74*** -0.0357 0.0898*** 0.38 (0.0010) (0.0188) (0.091) (0.081) (0.04) Panel B: Indusry porfolios Inercep EXRm SMB HML MOM R-sq R(alco) - Rf 0.0005 0.81*** 0.8075*** 0.3661*** -0.0106 0.660 (0.0016) (0.030) (0.0467) (0.0451) (0.0360) R(ob) - Rf 0.0046*** 0.68*** 0.39*** 0.1400*** -0.078** 0.509 (0.0015) (0.094) (0.0454) (0.0438) (0.0350) R(game) - Rf 0.0060** 1.00*** 1.3533*** 0.5086*** -0.1830** 0.436 (0.009) (0.0680) (0.0969) (0.105) (0.0740) - 17 -
Paris-Dauphine Universiy Table. Condiioning Variable Means, Sandard Deviaions and Correlaion Coefficiens This able gives he mean, sandard error and correlaion coefficiens of six monhly condiioning variables: Dividend Yield (DY), Monhly growh rae in Indusrial Producion (MP), Unexpeced Inflaion rae (UI), 3-monh T-bill yield (TB3M), Term Spread (TERM), and Defaul Spread (DEF). DY MP UI TB3M TERM DEF MEAN 0.64% 0.89% 0.003% 5.56% 1.450% 0.986% STD 0.170%.94% 0.39%.777% 1.40% 0.44% Correlaion Marix DY 1.00 MP 0.15 1.00 UI -0.04 0.0 1.00 TB3M 0.35-0.05-0.01 1.00 TERM -0.03 0.03-0.19-0.40 1.00 DEF 0.9-0.09-0.9 0.54 0.4 1.00 Table 3. Correlaion Coefficiens beween sock porfolio reurns and quarerly condiioning variables This able gives he correlaion coefficiens of he sin sock porfolio, he non sin sock porfolio and he long-shor porfolio wih hree condiioning variables (conemporaneous and wih one-quarer lag): he growh rae of GDP (GGDP), he consumpion growh (Gcons) and he growh in labor income (Gli). GGDP() Gcons() Gli() GGDP(-1) Gcons(-1) Gli(-1) Sin Rf -0.05 0.10 0.0-0.13-0.10-0.05 NoSin Rf 0.01 0.14 0.06-0.08-0.1-0.07 Sin NoSin -0.11-0.11-0.08-0.10 0.05 0.07-18 -
Paris-Dauphine Universiy Table 4. Regression resuls for a condiional model wih a one-monh lagged informaion variable The reurns on he sin sock porfolio and he zero-invesmen porfolio are regressed on he Fama-French facors and he momenum facor. Reurns are condiioned on differen informaion variables: he defaul spread, he dividend yield, he growh in indusrial producion, he 3-monh Treasury bill rae, he erm spread, and he unexpeced rae of inflaion. Sandard errors are given in parenheses. Inercep ExRm SMB HML MOM DEF1(-1) DEF1*ExRm DEF1*SMB DEF1*HML DEF1*MOM R-sq Sin Rf 0.0013 0.83*** 0.37*** 0.56*** -0.8*** -0.01.17 39.4*** -13.71 17.9** 0.740 (0.0034) (0.0807) (0.106) (0.180) (0.0886) (0.355) (7.1645) (10.4191) (11.646) (8.1580) Sin NoSin -0.008-0.11-0.49*** 0.40*** 0.0 0.5 -.01 4.98** -9.0 4.95 0.308 (0.003) (0.0763) (0.1005) (0.111) (0.0838) (0.3079) (6.778) (9.8573) (10.9979) (7.718) Inercep ExRm SMB HML MOM Div(-1) Div*ExRm Div*SMB Div*HML Div*MOM Sin Rf 0.0049** 0.83*** 0.53*** 0.49*** -0.18*** -1.34* 5.37 93.64*** -35.70 30.48 0.737 (0.004) (0.0606) (0.074) (0.0863) (0.054) (0.7883) (1.3768) (9.3309) (31.7576) (0.4177) Sin NoSin 0.0007-0.16*** -0.40*** 0.39*** 0.09* -0.8 14.36 68.74** -47.79-14.63 0.317 (0.00) (0.0566) (0.0693) (0.0807) (0.0507) (0.7370) (19.9869) (7.438) (9.697) (19.0901) Inercep ExRm SMB HML MOM Gip(-1) Gip*ExRm Gip*SMB Gip*HML Gip*MOM Sin Rf 0.0014 0.86*** 0.71*** 0.43*** -0.11*** 0.08 -.6-14.83* -7.35-0.10 0.79 (0.0013) (0.0317) (0.0416) (0.0490) (0.0319) (0.303) (5.873) (7.8417) (9.508) (6.756) Sin NoSin -0.000-0.11*** -0.6*** 0.3*** 0.06** 0.15-6.5-4.89-7.37-8.8 0.98 (0.001) (0.096) (0.0388) (0.0457) (0.097) (0.148) (4.933) (7.3166) (8.883) (6.3003) Inercep ExRm SMB HML MOM TB(-1) TB*ExRm TB*SMB TB*HML TB*MOM Sin Rf 0.0061** 0.71*** 0.59*** 0.59*** -0.18*** -0.07.34**.14 -.41 0.65 0.735 (0.009) (0.0768) (0.1118) (0.1156) (0.0685) (0.0475) (1.1794) (1.8746) (1.8413) (1.051) Sin NoSin 0.0018-0.3*** -0.3*** 0.9*** 0.13** -0.0 3.3*** 0.98 0.49-1.61 0.315 (0.007) (0.0716) (0.1043) (0.1079) (0.0639) (0.0443) (1.1006) (1.7494) (1.7183) (0.9809) - 19 -
Paris-Dauphine Universiy Table 4. (Coninued) Inercep ExRm SMB HML MOM Term(-1) Term*ExRm Term*SMB Term*HML Term*MOM Sin Rf -0.0010 0.95*** 0.54*** 0.51*** -0.16*** 0.18* -5.38** 1.64*** -4.46.54 0.735 (0.000) (0.0468) (0.066) (0.0660) (0.0390) (0.106) (.493) (3.7351) (3.6865) (.3000) Sin NoSin -0.004** -0.01-0.41*** 0.47*** 0.03 0.30*** -7.9*** 10.18*** -11.71***.5 0.330 (0.0018) (0.043) (0.0577) (0.0609) (0.0360) (0.0980) (.997) (3.4465) (3.4017) (.13) Inercep ExRm SMB HML MOM UI(-1) UI*ExRm UI*SMB UI*HML UI*MOM Sin Rf 0.0016 0.86*** 0.7*** 0.44*** -0.13*** -0.83** 13.46-13.71 3.08 13.65 0.730 (0.0013) (0.0314) (0.041) (0.0480) (0.0319) (0.404) (9.600) (13.98) (14.889) (9.6515) Sin NoSin 0.0001-0.1*** -0.8*** 0.33*** 0.06* -0.71* 8.15 8.1.9 7.39 0.300 (0.001) (0.093) (0.0393) (0.0449) (0.098) (0.3777) (8.9705) (1.360) (13.9067) (9.0184) - 0 -
Paris-Dauphine Universiy Table 5. Regression resuls for a condiional model wih a one-quarer lagged informaion variable The reurns on he sin sock porfolio and he zero-invesmen porfolio are regressed on he Fama-French facors and he momenum facor. Reurns are condiioned on differen informaion variables: he consumpion growh, he growh in GDP, and he growh in labor income. Sandard errors are given in parenheses. Inercep ExRm SMB HML MOM Gcons(-1) Cons*ExRm Cons*SMB Cons*HML Cons*MOM R-sq Sin Rf -0.0011 0.96*** 1.09*** 0.68*** 0.1 8.3-17.46-19.6-380.48* -300.1* 0.776 (0.008) (0.1003) (0.1334) (0.1300) (0.1004) (1.49) (150.31) (166.64) (05.43) (160.07) Sin NoSin -0.0040-0.07-0.07 0.53*** 0.** 7.77-37.77-13.14-47.43-370.87** 0.309 (0.0074) (0.0909) (0.109) (0.1178) (0.0910) (11.3) (136.3) (151.03) (186.19) (145.08) Inercep ExRm SMB HML MOM GGDP(-1) GDP*ExRm GDP*SMB GDP*HML GDP*MOM Sin Rf 0.0065 0.80*** 1.18*** 0.55*** -0.01-0.63 9.87 -.65** -9.83-1.74 0.780 (0.0073) (0.0947) (0.130) (0.110) (0.0884) (0.65) (8.83) (10.33) (9.66) (8.5) Sin NoSin 0.004-0.14-0.03 0.48*** 0.09-0.45 6.3-16.98* -1.98-6.36 0.31 (0.0066) (0.0863) (0.1187) (0.10) (0.0806) (0.59) (8.05) (9.4) (8.81) (7.5) Inercep ExRm SMB HML MOM Gli(-1) Li*ExRm Li*SMB Li*HML Li*MOM Sin Rf -0.0035 0.89*** 1.0*** 0.60*** 0.05 6.63-18.6-64.63*** -10.31-89.9 0.780 (0.0066) (0.0769) (0.1104) (0.1148) (0.0859) (5.67) (63.79) (9.58) (109.41) (74.4) Sin NoSin -0.0067-0.07-0.07 0.4*** 0.15* 5.73-46.73-17.68 6.48-147.07** 0.305 (0.0061) (0.0706) (0.1013) (0.1054) (0.0788) (5.0) (58.56) (84.99) (100.44) (68.31) - 1 -
Paris-Dauphine Universiy Table 6. Reurns on he zero-invesmen porfolio in expansionary and recessionary periods This Table gives he average reurn on he porfolio long on sin socks and shor on oher socks in periods classified by business cycle condiions. The cycle periods are obained from NBER classificaion. Long-shor porfolio average reurn during recession during expansion 01/7-11/7 0.86% 1/7-08/9-1.13% 09/9-03/33 1.60% 04/33-05/37-1.% 06/37-06/38 1.07% 07/38-0/45-0.41% 03/45-10/45 0.6% 11/45-11/48-0.41% 1/48-10/49 1.13% 11/49-07/53-0.55% 08/53-05/54-0.83% 06/54-08/57-0.55% 09/57-04/58.0% 05/58-04/60 0.1% 05/60-0/61 0.10% 03/61-1/69-0.34% 01/70-11/70 1.9% 1/70-11/73 0.1% 1/73-03/75-0.50% 04/75-01/80-0.55% 0/80-07/80 0.38% 08/80-07/81-1.9% 08/81-11/8 1.18% 1/8-07/90 0.51% 08/90-03/91-0.90% 04/91-03/01-0.10% 04/01-11/01 1.15% 1/01-1/05 0.56% Mean 0.83% Mean -0.5% -sa.51 -sa -.06 T-sa for difference of means across periods = 3.06 - -
Paris-Dauphine Universiy Table 7. Risk-adjused excess reurn for he sin and he long-shor porfolios, over expansion and recession periods (01/197-1/005) This Table shows he GARCH esimaes of a four-facor model allowing for ime-varying alphas and beas over expansion and recession periods. Sandard errors are given in parenheses. R R f, = α* + δ*exrm + φ*smb + γ*hml + η*mom + α r D r + δ r D r EXRm + φ r D r SMB + γ r D r HML + η r D r MOM + ε. This equaion is esed in a GARCH esimaion framework in which he error erm is modelled as follows: ε ~ N( 0, σ ), where σ = γ 0 + γ 1 σ 1 + γ ε 1. α δ φ γ η αr δr φr γr ηr R-sq Sin Rf 0.0007 0.75*** 0.68*** 0.06** 0.05 0.0078*** 0.0184 0.079 0.193*** -0.039 0.763 (0.0011) (0.01) (0.0316) (0.074) (0.000) (0.0019) (0.0333) (0.0653) (0.0573) (0.0384) Sin NoSin 0.0003-0.8*** -0.0*** -0.14*** 0.070*** 0.0070*** 0.0390 0.057 0.115** 0.067* 0.375 (0.0010) (0.0199) (0.093) (0.074) (0.015) (0.0019) (0.0306) (0.0631) (0.0516) (0.0384) N.B. for -saisics: he porfolio bea is esed agains one, and oher coefficiens (including long-shor porfolio bea) are esed agains zero [for dela, =(d-1)/se(d)] - 3 -
Paris-Dauphine Universiy Table 8. Risk-adjused excess reurn for he sin and he long-shor porfolios, over up and down markes (07/196-1/005) This Table shows he GARCH esimaes of a four-facor model allowing for ime-varying alphas and beas when he marke is up and when he marke is down. Sandard errors are given in parenheses. R R f, = α* + δ*exrm + φ*smb + γ*hml + η*mom + α d D d + δ d D d EXRm + φ d D d SMB + γ d D d HML + η d D d MOM + ε. This equaion is esed in a GARCH esimaion framework in which he error erm is modelled as follows: ε ~ N( 0, σ ), where σ = γ 0 + γ 1 σ 1 + γ ε 1. α δ φ γ η αd δd φd γd ηd R-sq Sin Rf 0.007 0.78*** 0.64*** 0.14*** -0.056** 0.000705 0.014 0.161*** -0.08 0.039 0.766 (0.0018) (0.0308) (0.096) (0.099) (0.046) (0.008) (0.0470) (0.053) (0.0513) (0.0470) Sin NoSin 0.0010-0.7*** -0.6*** -0.09*** 0.055** 0.003414 0.041 0.085* -0.030 0.05 0.376 (0.0017) (0.099) (0.0300) (0.063) (0.017) (0.004) (0.047) (0.0469) (0.0500) (0.040) N.B. for -saisics: he porfolio bea is esed agains one, and oher coefficiens (including long-shor porfolio bea) are esed agains zero [for dela, =(d-1)/se(d)] - 4 -
Paris-Dauphine Universiy Table 9. Risk-adjused excess reurn for indusry porfolios, over expansion and recession periods (01/197-1/005) This Table shows he GARCH esimaes of a four-facor model allowing for ime-varying alphas and beas over expansion and recession periods. Sandard errors are given in parenheses. R R f, = α* + δ*exrm + φ*smb + γ*hml + η*mom + α r D r + δ r D r EXRm + φ r D r SMB + γ r D r HML + η r D r MOM + ε. This equaion is esed in a GARCH esimaion framework in which he error erm is modelled as follows: ε ~ N( 0, σ ), where σ = γ + γ σ γ ε. 0 1 1 + 1 α δ φ γ η αr δr φr γr ηr R-sq R(alco) Rf -0.00006 0.8*** 0.54*** 0.18*** 0.06** 0.0056** -0.059 0.17** 0.13* -0.0 0.641 (0.0014) (0.0316) (0.038) (0.0359) (0.061) (0.006) (0.0494) (0.0853) (0.0699) (0.0581) R(ob) Rf 0.001616 0.64*** 0.45*** 0.06* -0.13*** 0.0074*** 0.018 0.10 0.1* 0.180*** 0.509 (0.0013) (0.078) (0.048) (0.0361) (0.079) (0.004) (0.047) (0.078) (0.0679) (0.056) R(game) Rf 0.00598** 0.98 1.10*** 0.46*** -0.15** -0.0013 0.0089 0.6*** 0.046-0.09 0.438 (0.008) (0.0645) (0.0704) (0.098) (0.0639) (0.0065) (0.1394) (0.370) (0.536) (0.1711) N.B. for -saisics: he porfolio bea is esed agains one, and oher coefficiens are esed agains zero [for dela, =(d-1)/se(d)] - 5 -
Paris-Dauphine Universiy Table 10. Risk-adjused excess reurn for indusry porfolios, over up and down markes (07/196-1/005) This Table shows he GARCH esimaes of a four-facor model allowing for ime-varying alphas and beas when he marke is up and when he marke is down. Sandard errors are given in parenheses. R R f, = α* + δ*exrm + φ*smb + γ*hml + η*mom + α d D d + δ d D d EXRm + φ d D d SMB + γ d D d HML + η d D d MOM + ε. This equaion is esed in a GARCH esimaion framework in which he error erm is modelled as follows: ε ~ N( 0, σ ), where σ = γ + γ σ γ ε. 0 1 1 + 1 α δ φ γ η αd δd φd γd ηd R-sq R(alco) Rf -0.0103*** 1.04 0.74*** 0.3*** 0.137*** 0.019*** -0.0511-0.9*** 0.0698-0.195*** 0.664 (0.001) (0.0385) (0.044) (0.0340) (0.07) (0.0037) (0.0645) (0.0737) (0.0807) (0.0661) R(ob) Rf 0.001 0.66*** 0.43*** 0.03-0.136*** 0.005 0.0059 0.15* 0.0944 0.079 0.504 (0.000) (0.0340) (0.043) (0.0336) (0.030) (0.0034) (0.0558) (0.0805) (0.0750) (0.0598) R(game) Rf 0.0176*** 0.71** 1.1*** 0.50*** -0.357*** -0.0064 0.4513** 0.16-0.189 0.36** 0.440 (0.0050) (0.119) (0.0788) (0.0993) (0.0709) (0.0078) (0.1939) (0.14) (0.00) (0.1450) N.B. for -saisics: he porfolio bea is esed agains one, and oher coefficiens are esed agains zero [for dela, =(d-1)/se(d) - 6 -
Paris-Dauphine Universiy Figure 1. Growh in GDP versus long-shor porfolio reurns (1948-005) These graphs show he growh in GDP versus he reurn on he long-shor porfolio, for wo subperiods: 1948-1976 and 1977-005. In he second graph, he scale is no he same for boh imeseries. Figure 1.a. Growh in GDP (annualized) and EW long-shor porfolio reurns - period 1948-1976 0, 0,1 reurn 0 1948 1953 1958 1963 1968 1973-0,1-0, year GGDP_ann LSew Figure 1.b. Growh in GDP (annualized) and EW long-shor porfolio reurns - period 1977-005 0,0 0,40 0,30 0,10 0,0 Growh in GDP 0,10 LS reurn 0,00 1977 1980 1983 1986 1989 199 1995 1998 001 004 0,00-0,10-0,10 year -0,0 GGDP_ann LSew - 7 -
Paris-Dauphine Universiy Figure. Consumpion o wealh raio versus alcohol porfolio reurns (1960-005) These graphs show he raio consumpion/wealh versus he reurn on he alcohol porfolio, for wo subperiods: 1960-198 and 1983-005. Over he enire period, he scale is no he same for boh imeseries. Figure.a. Consumpion wealh raio and alcohol porfolio excess reurns - period 1960-198 0,4 0,03 0,0 excess reurn on alcohol socks 0, 0,0 1960 1963 1966 1969 197 1975 1978 1981-0, 0,01 0-0,01 consumpion-wealh raio -0,0-0,4 year -0,03 Figure.b. Consumpion wealh raio and alcohol porfolio excess reurns - period 1983-005 0,3 0,04 0, 0,03 0,0 excess reurn on alcohol socks 0,1 0,0 1983 1986 1989 199 1995 1998 001 004-0,1 0,01 0-0,01-0,0 consumpion-wealh raio -0, -0,03-0,3 year -0,04-8 -
Paris-Dauphine Universiy Figure 3. Consumpion o wealh raio versus obacco porfolio reurns (1960-005) These graphs show he raio consumpion/wealh versus he reurn on he obacco porfolio, for wo subperiods: 1960-198 and 1983-005. Over he enire period, he scale is no he same for boh imeseries. Figure 3.a. Consumpion wealh raio and obacco porfolio excess reurns - period 1960-198 0,3 0,03 0, 0,0 excess reurn on obacco socks 0,1 0,0 1960 1963 1966 1969 197 1975 1978 1981-0,1 0,01 0-0,01 consumpion-wealh raio -0, -0,0-0,3 year -0,03 Figure 3.b. Consumpion wealh raio and obacco porfolio excess reurns - period 1983-005 0,4 0,04 0,03 0, 0,0 excess reurn on obacco socks 0,0 1983 1986 1989 199 1995 1998 001 004-0, 0,01 0-0,01-0,0 consumpion-wealh raio -0,03-0,4 year -0,04-9 -
Paris-Dauphine Universiy Figure 4. Consumpion o wealh raio versus gambling porfolio reurns (1960-005) These graphs show he raio consumpion/wealh versus he reurn on he gaming porfolio, for wo subperiods: 1960-198 and 1983-005. Over he enire period, he scale is no he same for boh imeseries. Figure 4.a. Consumpion wealh raio and gaming porfolio excess reurn - period 1960-198 0,6 0,03 0,4 0,0 excess reurn on gaming socks 0, 0,0 1960 1963 1966 1969 197 1975 1978 1981-0, 0,01 0-0,01 consumpion-wealh raio -0,4-0,0-0,6 year -0,03 Figure 4.b. Consumpion wealh raio and gaming porfolio excess reurn - period 1983-005 0,5 0,04 0,4 0,03 0,3 0,0 excess reurn on gaming socks 0, 0,1 0,0 1983 1986 1989 199 1995 1998 001 004-0,1-0, -0,3 0,01 0-0,01-0,0 consumpion-wealh raio -0,4-0,03-0,5 year -0,04-30 -