Cash Flow, Currency Risk, and the Cost of Capital

Transcription

1 Cash Flow, Currency Rsk, and he Cos of Capal Workng Paper Seres Ocober 2011 Dng Du Norhern Arzona Unversy The W. A. Franke College of Busness PO Box Flagsaff, AZ (928) Fax: (928) Ou Hu Deparmen of Economcs Youngsown Sae Unversy Youngsown, OH (330)

2 1. Inroducon Cash Flow, Currency Rsk, and he Cos of Capal Frms are mpaced by currency flucuaons. Hung (1992) fnds ha he loss due o currency flucuaons for U.S. manufacurng frms s abou 10% per year of gross profs n he 1980s. A Phladelpha Federal Reserve Bank survey ced by Francs, Hasan, and Huner (2008) fnds ha over 45% of U.S. frms repored ha hey are affeced by currency movemens (p. 177). Nucc and Pozzolo (2010) documen a sascally sgnfcan effec of exchange rae varaons on employmen, hours worked and wages n a represenave panel of Ialan manufacurng frms. (p.121) A volumnous leraure has developed assessng currency rsk. os heorecal models of currency rsk emphasze he mpac of currency flucuaons on frms cash flows (e.g. Sulz, 1984; Smh and Sulz, 1985; and Froo, Sharfsen, and Sen, 1993). However, he common emprcal approach s o focus on sock reurns nsead (e.g., Adler and Dumas, 1983). Ths approach, n general, fnds ha currency rsk s no prced (e.g., Joron, 1990, 1991). 1 Despe he weak resuls from hs common approach, few sudes have aemped o examne currency rsk from a cash flow perspecve. Ths paper nends o fll hs gap. Our focus on cash flows s also movaed by Hou, Karoly and Kho (2011) who recenly fnd robus evdence ha cash flow s relaed o a covarance rsk facor n global sock markes. Specfcally, Hou, Karoly and Kho (2011) fnd ha he cash flow mmckng porfolo always carres an economcally and sascally sgnfcan rsk premum even afer conrollng for varaon n he cash flow characersc. In erms of emprcal mplemenaon, we use he mmckng porfolo approach ha s conssen wh domesc asse-prcng models of Fama and French (1992, 1993) and ulzed by Kolar, oorman, and Sorescu (2008) o examne currency rsk. 2 Conssen wh Kolar, oorman, and Sorescu (2008), we frs esmae he sensvy of each frm s operang cash flows o currency movemens n a rollng fashon (o allow for me-varaon n currency exposure); hen we consruc a zero-nvesmen porfolo ha akes long posons n socks whose operang cash flows have posve sensvy o currency movemens and shor posons n socks whose operang cash flows have negave currency sensvy. We refer o as PN (or posve mnus negave). Snce he reurn of he PN porfolo s drven by currency movemens, PN s he facormmckng porfolo of currency rsk and s average reurn represens he rsk premum on currency rsk. PN s based on frms fundamenals (.e. operang cash flows). Therefore, f s sgnfcan n he sandard asse-prcng ess, s unlkely due o possble spurous correlaon dscussed by Lewellen, Nagel and Shanken (2010). 3 Ths s one major advanage of our approach. Our approach has wo more advanages. Frs, prevous sudes (see Foonoe 1) ypcally use raw exchange rae changes, whch, as macroeconomc varables (no reurns), conan nformaon ha s rrelevan o asse prcng and may also have measuremen errors. In conras, our mmckng facor porfolo capures only he nformaon n currency movemens ha s pernen o sock reurns, and herefore should reduce he nose n esmaons. See Chan, Karcesk and Lakonshok (1998, 1999), Kolar, oorman, and Sorescu (2008), and Hou, Karoly and Kho (2011) for more dscusson and applcaons of he mmckng porfolo approach. Second, prevous sudes usually do no allow for me varaon n currency exposure, whch as Francs, Hasan, and Huner (2008), among ohers, pon ou s a 1 See also Khoo (1994), Barov and Bodnar (1994), Allayanns (1997), Chow, Lee and Sol (1997), Vassalou (2000), Grffn (2002), Bodnar and Wong (2003), Barram (2004), Barram and Bodnar (2005), and Barram (2007). Alhough Barram (2008) and Barram, Brown, and non (2010) argue ha frms use hedges o grealy reduce currency exposures, her argumens do no seem o be conssen wh he frm-level evdence n Hung (1992), Francs, Hasan, and Huner (2008) and Nucc and Pozzolo (2010). 2 The major dfference beween hs paper and Kolar, oorman, and Sorescu (2008) s ha we focus on cash flows whch s well movaed by heores, where Kolar, oorman, and Sorescu (2008) cener on sock reurns. 3 Ths observaon moves Da (2009) o focus on cash flows o es consumpon rsk. 1

3 major mehodologcal weakness. In conras, our approach esmaes frms currency sensves n a rollng regresson fashon, whch akes no accoun me varaon n exposure n a non-srucural framework. See Dodge, Grffn, and Wllamson (2006) for more dscusson. ore specfcally, we frs consruc 25 sze and cash-flow-sensvy porfolos as our es asses n lne wh Fama and French (1993) and Kolar, oorman, and Sorescu (2008). Cash-flow sensvy refers o he sensvy of frms operang cash flows o currency movemens. We hen compare he sandard Fama-French hree-facor model (as well as he four-facor model ha employs exchange rae changes as he currency rsk proxy) wh our new four-facor model ha ncludes PN as he currency rsk facor from pole o pole n sandard me-seres and cross-seconal asse prcng ess. We fnd ha our new four-facor model ncludng PN ouperforms boh he Fama-French hree-facor model and he sandard four-facor model. Taken all he evdence ogeher, our fndngs sugges ha currency rsk s relevan for asse prcng f we focus on s mpac on frms cash flows. Our resuls are robus regardless of wheher we use a dfferen se of esng asses or wheher we use real or nomnal exchange raes. Based on our four-facor model ncludng PN, we fnd ha he mpac of currency rsk on he cos of capal s no only sascally bu also economcally sgnfcan. Wh he average currency bea beng around 0.22 n absolue value and he average currency rsk premum beng abou 6% per year, our esmaes sugges ha he mpac of currency rsk on he cos of capal s abou 1.3% per year (or 11.4% of he average rsk premum). Over me, he mpac of currency rsk on he cos of capal does no decrease bu ncreases. The remander of he paper s organzed as follows: Secon 2 dscusses he daa and how we consruc he currency rsk facor PN. Secon 3 presens he emprcal mehodology and resuls. Secon 4 s robusness check. Secon 5 concludes he paper wh a bref summary. 2. ovaon and Currency Rsk Facor 2.1 Daa A frm s value depends on cash flows generaed from s operaons or operang cash flows (OCFs). Therefore, we focus on operang cash flows n hs paper. We use he OCF measure ha s commonly used n he leraure (e.g., Hrshlefer, Hou and Teoh, 2009). ore specfcally, we defne operang cash flow as earnngs afer deprecaon less accruals, where accruals s he change n non-cash curren asses less he change n curren lables excludng he change n shor-erm deb and he change n axes payable, mnus deprecaon and amorzaon expense. Followng Huang (2009), we sandardze OCF by sales. As Huang (2009) pons ou, usng sales as he scalar has he advanage of removng seasonaly from cash flows. The quarerly accounng daa are obaned from he Compusa daabase. To be conssen wh relevan leraure (e.g., Francs, Hasan, and Huner, 2008), we focus on all NYSE, AEX, and NASDAQ socks n he 36 ndusres ha are mos lkely o be affeced by currency rsk. 4 The monhly sock reurns daa are obaned from CRSP. The Fama-French facors daa are from Kenneh French s webse. Our sample covers July 1980 o December Followng relevan leraure (e.g. Kolar, oorman, and Sorescu, 2008), we focus on he Federal Reserve s ajor Currences Index (CI) based on foregn exchange values of he dollar agans currences of major ndusral counres from he Federal Reserve Bank n S. Lous. The ajor Currency Index ncludes he Euro Area, Canada, Japan, Uned Kngdom, Swzerland, Ausrala, and Sweden. Panel A of Fgure 1 shows he raw CI over our sample perod. As we can see, he value of he radeweghed dollar changes subsanally over our sample perod. 5 4 They are made up of 31 raded-goods (manufacurng) ndusres and fve nonraded-goods ndusres (Eneranmen, Consrucon, eals, Real Goods and Bankng). 5 CI s a nomnal exchange rae seres. However, n he robusness check secon, we show ha usng real exchange raes would yeld qualavely smlar resuls. 2

4 2.2 Currency Rsk Facor PN Snce heory emphaszes he mpac of currency movemens on frms cash flows, we consruc he currency facor porfolo based on he sensvy of frms cash flows o currency movemens. Essenally, we consruc a zero-nvesmen porfolo ha akes long posons n socks whose OCFs have posve sensvy o currency movemens and shor posons n socks whose OCFs have negave currency sensvy. Snce he currency facor porfolo s based on frms fundamenals (.e. OCFs), s unlkely o be a spurous facor n he Lewellen, Nagel and Shanken (2010) sense. In lne wh Fama and French (1993), we form our currency facor porfolo n wo seps. The frs sep s o form sx value-weghed sze and cash-flow-sensvy porfolos wh all socks n he 36 ndusres for whch we have approprae accounng daa. The porfolos, whch are consruced a he end of each June, are he nersecons of wo porfolos formed on sze (marke equy) and hree porfolos formed on cash-flow sensvy. The sze breakpon for year s he medan NYSE marke equy a he end of June of year. The cash-flow sensvy for June of year s esmaed wh he pror fve years quarerly daa (.e. from he hrd quarer of year -5 o he second quarer of ) n a regresson wh he quarerly sandardzed OCF as he dependen varable and he quarerly exchange rae change as he ndependen varable. Tha s, OCF Sales a b CI e (1) where OCF s he quarerly operang cash flow defned as Hrshlefer, Hou and Teoh (2009), Sales s he quarerly sales, and CI s he percenage change n quarerly CI. b s he cash-flow sensvy of frm. The cash-flow sensvy breakpons are he 30h and 70h percenles. These porfolos are held for one year (from July of year o June of year 1) and rebalanced a he end of June of year 1. Tha s, a he end of June of year 1, we reclassfy socks based on her sze a he end of June of year 1 and her cash-flow sensvy based on he coeffcen esmae over he pror fve years (.e. from he hrd quarer of year -4 o he second quarer of 1). By rebalancng he porfolos on an annual bass n a condonal fashon, we allow currency exposure of frms OCFs o be me varyng, whch as Francs, Hasan, and Huner (2008) sugges s mporan for sudyng currency rsk. Table 1 shows he annual mean reurns and oher relevan summary sascs of he sx sze and cash-flow-sensvy porfolos. Conssen wh he leraure (e.g. Fama and French, 1993), small frms n our sample have hgher average reurns han large frms. However, wha s new s ha he mean reurn generally ncreases monooncally wh he cash-flow sensvy whn a sze group. For nsance, whn small socks, he mean reurns for he frms wh negave, neural and posve cash-flow sensvy are 15.57% per year, 17.25% per year and 20.62% per year, respecvely. A smlar paern s also found n large socks. These resuls herefore sugges a posve lnear rsk premum on he currency rsk facor, whch s conssen wh general percepon. For nsance, Sarks and We (2005) argue ha currency flucuaons can push a frm no fnancal dsress. Therefore, currency rsk may be a dsress rsk much lke he value facor n he hree-facor model of Fama and French (1993). Gven he rsk premum of he value facor s lnear and posve, he rsk premum on he currency rsk facor should be lnear and posve oo. 3

5 Table 1 ean Reurns of Sx Sze and Cash-Flow Sensvy Porfolos Number of frm monhs Cash Flow Sensvy Percen sgnfcan a 10% level Average annual raw reurn Sze CF Sensvy Esmae Percen Posve Sze Small Negave Neural Posve Bg Negave Neural Posve Table 1 shows he annual mean reurns of he sx sze and cash-flow sensvy porfolos. The sensvy o cash flow s esmaed over a fve-year perod. Sze s calculaed as he prce mes he number of shares ousandng n June of year. The able shows porfolo averages for sze, he cash-flow sensvy, and reurns. Frm monhs used n each porfolo are shown also, along wh he percenage of frm monhs for whch cash-flow sensvy s posve or sgnfcan a he 10% level over he formaon perod.] Agan, n lne wh Fama and French (1993), he second sep s o defne he currency rsk facor as he average reurn on he wo posve sensvy porfolos mnus he average reurn on he wo negave sensvy porfolos. Tha s, our currency rsk facor, PN, s BP SP BN SN PN (2) 2 2 where BP, SP, BN and SN are he reurns on large and posve sensvy, small and posve sensvy, large and negave sensvy, and small and negave sensvy porfolos, respecvely. Snce he reurn of he PN porfolo s drven by currency movemens, PN s he mmckng-facor porfolo of currency rsk and s average reurn represens he rsk premum on he currency rsk. Table 2 Correlaon arx of he Relevan Varables R SB HL PN R SB HL PN Panel B of Fgure 1 shows he currency rsk facor (PN) over our sample perod. In Table 2, we repor he correlaons of PN wh he Fama-French hree facors. As we can see, he currency rsk facor PN s orhogonal o he Fama-French hree facors. The correlaons are all very close o zero, rangng from 0.00 wh he marke facor o wh he sze facor. Therefore, f PN s sgnfcan n our emprcal ess, canno be due o s spurous correlaon wh he Fama-French facors. Tha s, PN mus be a relevan rsk facor for asse prcng. 4

6 150 Fgure 1 CI and Currency Rsk Facor Panel A Trade -Weghed Value of he U.S. Dollar Panel B Currency Rsk Facor PN Panel A shows he raw CI over our sample perod, whle Panel B depcs he currency rsk facor PN. If PN s a relevan rsk facor, s mean reurn s an esmae of s rsk premum. The mean reurn of he currency rsk facor PN s 0.40% per monh or abou 5% per year, whch s no only sascally bu also economcally sgnfcan. 3. Emprcal ehodology and Resuls Emprcally, n lne wh Fama and French (1993), we consruc 25 sze and cash-flow-sensvy porfolos as our esng asses. Cash-flow sensvy agan refers o he sensvy of frms operang cash flows o currency movemens. These 25 porfolos are consruced n a smlar way as he sx sze and cash-flow-sensvy porfolos n Secon 2.2. Table 3 shows he annual mean reurns and oher relevan summary sascs of hese porfolos. Agan, he same paerns emerge: he mean reurn ncreases monooncally wh cash flow sensvy (and decreases monooncally wh sze), suggesng a posve lnear rsk premum on currency rsk. We ake no accoun frm sze, because he sze-based analyss may offer some nsgh on he role of hedgng n currency exposure. Large frms may be more exposed o currency rsk due o he fac ha hey end o be operang more globally. As a resul, hey may have more ncenve o engage n hedgng. Clearly, large frms may also have more resources for professonal rsk managemen. Thus, f hedgng plays a crcal role n currency exposure as Barram (2008) and Barram, Brown, and non (2010) sugges, we expec ha large frms have sgnfcanly smaller exposure han do small frms. On he oher hand, f hedgng does no ell he full sory of he exposure puzzle as Francs, Hasan, and Huner (2008) sugges, we expec ha large frms have sgnfcanly larger exposure o currency rsk. 5

7 Table 3 ean Reurns of 25 Sze and Cash-Flow Sensvy Porfolos Number of frm monhs Sensvy Percen sgnfcan a 10% level Average annual raw reurn Sze CF Sensvy Esmae Percen Posve Sze Small Negave Posve Negave Posve Negave Posve Negave Posve Negave Posve We form 25 value-weghed sze and cash-flow sensvy porfolos as our esng asses. Table 4 shows he annual mean reurns of hese porfolos. Cash-flow sensvy s esmaed over a fve-year perod. Sze s calculaed as he prce mes he number of shares ousandng n June of year. The able shows porfolo averages for sze, cash-flow sensvy, and reurns. Frm monhs used n each porfolo are shown also, along wh he percenage of frm monhs for whch cash-flow sensvy s posve or sgnfcan a he 10% level over he formaon perod. Nex, we compare he sandard hree-facor model (as well as he four-facor model ha employs he exchange rae change as he currency rsk proxy) wh our new four-facor model ha ncludes PN as he currency rsk facor from pole o pole n sandard me-seres and cross-seconal asse prcng ess. 3.1 Tme-Invaran Currency Exposure and Rsk Premum Currency Exposure To es our hypohess, we consder hree asse-prcng models. The frs one s he sandard Fama-French hree facor model: odel 1: r α β, β, SB SB β, HL HL ε (3) where r s he excess reurn on asse n perod,, SB and HL are he reurns on he marke, he sze, he book-o-marke facors. The β s are he assocaed facor loadngs, and ε s he dsurbance. The second one s he sandard four-facor model n he exchange rae leraure ha ncludes he Fama-French hree facors and changes n CI. As we have emphaszed, exchange rae changes, as 6

8 macroeconomc varables (no reurns), conan nformaon ha s rrelevan o asse prcng and may also have measuremen errors. As a resul, usng exchange rae changes may nroduce nose and mask he rue underlyng relaonshp. odel 2: r α β, β, SB SB β, HL HL β, CI CI ε (4) The hrd model s our new four-facor model ha ncludes he Fama-French hree facors and he currency rsk facor PN (whch reflecs he mpac of currency movemens on frms cash flows). odel 3: r α β, β, SB SB β, HL HL β, PN PN ε (5) We run me-seres regressons porfolo by porfolo o esmae currency exposure. If as we conjecure currency rsk s relevan bu he change n exchange rae s a nosy proxy of currency rsk, we expec ha odel 2 wll no ouperform odel 1, bu our odel 3 wll ouperform boh odel 1 and odel 2. The currency exposure resuls for he 25 sze and cash-flow-sensvy porfolos based on Eqs. (3), (4) and (5) are repored n Table 4 for he whole sample perod from 1980:7 o 2008:12. To save space, we only repor Alphas, currency exposure esmaes and adjused-r 2 s. The -raos are based on Newey-Wes HAC sandard errors wh he lag parameer se equal o 12, and he sgnfcan esmaes (a he 10% level for wo-sded ess) are n bold. As we can see, odel 2 does no ouperform odel 1 or he Fama-French hree-facor model: he average of he absolue values of Alphas of odel 2 s 0.35 whch s equal o ha of odel 1; he average of he adjused-r 2 s of 0.54 s also equal o ha of odel 1; furhermore, here are only hree porfolos ha have sgnfcan exposure o he change n CI and he loadngs do no change accordngly wh he cash flow sensvy. Noe ha 25 sze and cash-flowsensvy porfolos are consruced based on he sensvy of frms operang cash flows o currency movemens. In conras, odel 3 seems o margnally ouperform boh odel 1 and odel 2. The average of he absolue values of Alphas of odel 3 s 0.33 whch s slghly smaller han hose of odel 1 and odel 2. The average of he adjused-r 2 s of 0.55 s also slghly hgher han hose of odel 1 and odel 2. ore mporanly, 11 ou of 25 or 44% of he sze and cash-flow-sensvy porfolos have sascally sgnfcan exposure o he currency rsk PN, and negave/posve cash-flow sensvy porfolos generally have negave/posve loadngs on he PN facor. Therefore, our resuls ndcae ha socks are exposed o currency rsk f we focus on he mpac of currency flucuaons on cash flows. As we can also see from Table 4, large frms do no have smaller exposure o currency rsk han do small frms. For nsance, four ou of fve porfolos wh he larges sze have sgnfcan exposure o he currency rsk PN, where none of fve porfolos wh he smalles sze has sgnfcan exposure o PN. The resuls herefore sugges ha hedgng may no ell he full sory of he exposure puzzle. Large frms have more ncenve o engage n hedgng because hey end o be operang more globally. Large frms may also have more resources for professonal rsk managemen. Thus, f hedgng plays a crcal role n currency exposure as Barram (2008) and Barram, Brown, and non (2010) sugges, we would expec ha large frms have smaller exposure han do small frms (no he oppose as we fnd n Table 4). Currency Rsk Premum To esmae he rsk premum on he currency rsk facor, we use he Black, Jensen, and Scholes (1972) and Fama and acbeh (1973) wo-pass mehodology esmang facor sensves n he frs pass, and usng hose o oban rsk premum n he second pass wh sandard refnemens: he Shanken (1992) correcon o oban errors-n-varables (EIV) robus sandard errors, accounng for he fac ha facor sensves are esmaed, and he Shanken and Zhou (2007) correcon o generae msspecfcaon (IS) robus sandard errors. 6 6 See also Km (1995) and Jagannahan and Wang (1998). 7

9 Table 4 Currency Exposure based on Nomnal CI over OCF odel 1 odel 2 odel 3 Sze Sensvy α Adj-R 2 α β, CI Adj-R 2 α β, PN Adj-R 2 Small Negave ( 1.26 ) ( 1.25 ) ( ) ( 1.52 ) ( ) ( 0.94 ) ( 0.98 ) ( 1.44 ) ( 0.96 ) ( 0.38 ) ( 3.11 ) ( 3.20 ) ( 2.21 ) ( 3.09 ) ( 0.54 ) ( 1.92 ) ( 1.97 ) ( 1.19 ) ( 1.92 ) ( ) Posve ( 2.46 ) ( 2.46 ) ( ) ( 2.22 ) ( 1.32 ) 2 Negave ( ) ( ) ( 0.93 ) ( ) ( ) ( 2.44 ) ( 2.48 ) ( ) ( 2.51 ) ( ) ( 0.79 ) ( 0.79 ) ( 0.62 ) ( 0.78 ) ( ) ( 2.15 ) ( 2.17 ) ( ) ( 2.15 ) ( ) Posve ( 2.04 ) ( 2.06 ) ( 0.69 ) ( 1.74 ) ( 2.61 ) 3 Negave ( 0.01 ) ( 0.00 ) ( ) ( 0.32 ) ( ) ( ) ( ) ( ) ( 0.07 ) ( ) ( 0.01 ) ( 0.00 ) ( ) ( 0.02 ) ( ) ( ) ( ) ( ) ( ) ( 1.19 ) Posve ( 0.21 ) ( 0.21 ) ( 0.41 ) ( ) ( 4.99 ) 4 Negave ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( 1.04 ) ( ) ( ) ( ) ( ) ( 1.76 ) ( ) ( ) ( 0.32 ) ( 0.32 ) ( 0.16 ) ( 0.43 ) ( ) Posve ( 0.18 ) ( 0.19 ) ( 0.68 ) ( ) ( 1.72 ) Bg Negave ( ) ( ) ( 3.57 ) ( 0.15 ) ( ) ( 2.14 ) ( 2.16 ) ( ) ( 2.79 ) ( ) ( 1.09 ) ( 1.08 ) ( 0.27 ) ( 1.43 ) ( ) ( 1.09 ) ( 1.12 ) ( 0.69 ) ( 0.67 ) ( 2.33 ) Posve ( 3.94 ) ( 3.90 ) ( ) ( 3.33 ) ( 3.27 ) odel 1: r α β β SB β HL ε,, SB, HL odel 2: r odel 3: r α β β SB β HL β CI ε,, SB, HL, CI β, β, SB SB β, HL HL β, PN α PN ε where r s he excess reurn on asse n perod,, SB, HL, PN and CI are he marke, he sze, he book-o-marke, he currency rsk, and he conemporaneous change n exchange rae. 8

10 Conssen wh prevous analyss, we employ he followng hree cross-seconal regresson models n he second pass. The frs one s he Fama-French hree facor model: odel 1: γ γ ˆ β γ ˆ β γ ˆ β η (6) r 0, SB, SB HL, HL The second one s he sandard four-facor model ha ncludes he Fama-French hree facors and he change n CI. odel 2: γ γ ˆ β γ ˆ β γ ˆ β γ ˆ β η (7) r 0, SB, SB HL, HL CI, CI The hrd model s our new four-facor model ha ncludes he Fama-French hree facors and he currency rsk facor PN (whch reflecs he mpac of currency movemens on frms cash flows). odel 3: γ γ ˆ β γ ˆ β γ ˆ β γ ˆ β η (8) r 0, SB, SB HL, HL PN, PN where he βˆ s are he facor loadngs (jonly esmaed for all he facors) from he frs-pass me-seres regresson, and γ s are he assocaed rsk premums. Agan, f as we conjecure currency rsk s relevan bu he change n exchange rae s a nosy proxy of currency rsk, we expec ha odel 2 wll no ouperform odel 1, bu our odel 3 wll ouperform boh odel 1 and odel 2. The resuls are repored n Table 5. As we can see, odel 2 does no ouperform odel 1 n erms of explanng he cross-seconal dfferences n he reurns of he 25 sze and cash-flow-sensvy porfolos. Frs, he change n CI s no prced regardless of wheher we use OLS or GLS. Ths s conssen wh prevous fndngs (.e. Joron, 1991). Second, addng he change n CI decreases adjused R 2. Noe agan ha 25 sze and cash-flow-sensvy porfolos are consruced based on he sensvy of frms operang cash flows o currency movemens. In conras, odel 3 sgnfcanly ouperforms boh odel 1 and odel 2. Regardless of wheher we use OLS or GLS, he adjused-r 2 mproves subsanally: based on OLS ncreases from 0.55 n odel 1 o 0.70, and based on GLS ncreases from 0.48 n odel 1 o 0.62! Thus, he currency rsk facor PN helps explan cross-seconal reurns. ore mporanly, he currency rsk facor s prced. Based on OLS, he premum assocaed wh hs facor s 0.61 percen per monh wh an EIV-robus - sasc of 2.25 and a IS-robus -sasc of The esmae s que close o he mean reurn of he currency rsk facor seres PN (0.40 percen per monh), whch s wha we expec for a facor porfolo. The GLS esmae s smlar and also sgnfcan even afer we ake no accoun he EIV problem and poenal model msspecfcaons. Therefore, our resuls furher confrm ha currency rsk maers f we focus on he mpac of currency flucuaons on cash flows. 3.3Tme Varyng Currency Exposure and Rsk Premum Currency Exposure Currency exposure of esng asses can be me varyng. To allow currency exposure o change over me, we repea he prevous exercse wh rollng samples. Currency exposure s esmaed wh 10 years of daa o oban meanngful esmaes. Consequenly, he es perod sars n 1990:7. We updae esmaes monhly by droppng he earles observaon and addng he laes observaon. We repor he average currency exposure n Table 6. Snce we use overlappng daa, he -raos are based on Newey- Wes HAC sandard errors wh he lag parameer se equal o Evdence suggess ha 19 ou of 25 (or 76%) sze and cash-flow-sensvy porfolos have sgnfcan currency exposure f we allow me varaon n currency exposure, whch furher confrms ha socks are exposed o currency rsk f we focus on he mpac of currency flucuaons on cash flows. We repor he average currency bea n absolue value for all 25 porfolos over me n Panel A of Fgure 2. The average currency bea s around 0.22 n absolue value and never below 0.18, suggesng ha currency rsk s relevan. 7 The resuls are robus for dfferen lag parameers. 9

11 Table 5 Rsk Premum Esmaes based on 25 Sze and Cash-Flow Sensvy Porfolos over 1980 o 2008 odel 1 OLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL Adj-R GLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL Adj-R odel 2 OLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL CI Adj-R GLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL CI Adj-R odel 3 OLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL PN Adj-R GLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL PN Adj-R odel 1: odel 2: odel 3: r r γ ˆ ˆ ˆ η 0 γ β, γ SBβ, SB γ HLβ, HL r γ γ ˆ β γ ˆ β γ ˆ β γ ˆ β η 0, SB, SB HL, HL CI, CI γ ˆ ˆ ˆ ˆ η 0 γ β, γ SBβ, SB γ HLβ, HL γ PN β, PN where he βˆ s are he facor loadngs (jonly esmaed for all he facors) from he frspass me-seres regresson, and γ s are he assocaed rsk premums. EIV s he Shanken (1992) s errors-n-varables robus -raos, and IS s he Shanken and Zhou (2007) s msspecfcaon robus -raos. 10

12 0.28 Fgure 2 Tme varyng Currency Exposure and Rsk Premum Panel A Tme-Varyng Average Currency Exposure n Absolue Vaue Panel B Tme-Varyng Rsk Premum Panel C Tme-Varyng Two-Pass Regresson Adj-R Panel D Tme-Varyng Impac of Currency rsk

13 Table 6 Tme-Varyng Currency Exposure ean Exposure Cash Flow Sensvy Negave Posve Small Sze Bg T-sasc H 0 : ean exposure 0 Cash Flow Sensvy Negave Posve Small Sze Bg To allow currency exposure o change over me, we repea he currency exposure regressons wh a rollng sample. Currency exposure s esmaed wh 10 years of daa o oban meanngful esmaes. Consequenly, he es perod sars n 1990:7. We updae esmaes monhly by droppng he earles observaon and addng he laes observaon. We repor he average currency exposure n Table 7. Snce we use overlappng daa, he - raos are based on Newey-Wes HAC sandard errors wh he lag parameer se equal o 12. Currency Rsk Premum The currency rsk premum may also change over me. To oban more nformaon abou he pah of he currency rsk premum over me, we repea he prevous exercse wh rollng samples. The rsk premum a each me s esmaed wh 10 years of daa o oban meanngful esmaes. Consequenly he es perod sars n July We updae esmaes monhly by droppng he earles observaon and addng he laes observaon. The resuls are dsplayed n Panel B of Fgure 2. The average rsk premum s 0.49% per monh or abou 6% per year, whch s no only sascally bu also economcally sgnfcan. The evdence furher confrms ha he currency rsk facor s relevan for asse prcng. We also repor he assocaed adjused-r 2 s from he above regressons n Panel C of Fgure 2. As we can see, currency rsk becomes more relevan n recen years for explanng cross-seconal reurns. Ths rend s plausble gven he remendous growh n nernaonal rade. Accordng o our calculaons, U.S. mpors and expors have ncreased from abou 20% n 1980 o abou 30% n 2008 as a percenage of U.S. GDP. I also suggess ha hedgng does no ell he full sory of he exposure puzzle, because f dd we would expec ha currency rsk would become less relevan n explanng cross-seconal reurns over me. Impac of Currency Rsk on he Cos of Capal The mpac of currency rsk on he cos of capal can be measured by he produc of rsk premum and currency exposure n absolue value. We focus on he average esmaes from he rollng regressons n hs secon snce hey allow me varaon n currency exposure and currency rsk premum. The resuls are repored n Panel D of Fgure 2. As we can see, he mpac of currency rsk on he cos of capal vares subsanally over me. The average mpac s abou 1.32% per year (or 11.4% of he average rsk premum). ore mporanly, he mpac of currency rsk ncreases over me. Ths agan suggess ha hedgng does no explan he full sory of he exposure puzzle. 4 Robusness Check 4.1 Usng 25 Cash-Flow Sensvy Porfolos as Tesng Asses Lewellen, Nagel and Shanken (2010) pon ou ha a spurous facor can be sascally sgnfcan n sandard asse prcng ess f es asses are he porfolos formed on sze and book-o- 12

14 marke rao. We emphasze ha our currency rsk facor PN s based on frms fundamenals (.e. OCFs), and herefore s unlkely o be a spurous facor. However, for a robusness check, we use a dfferen se of es asses o repea he prevous ess. ore specfcally, n lne wh Kolar, oorman, and Sorescu (2008), we consruc 25 porfolos based solely on cash-flow sensvy as our es asses. Cash-flow sensvy agan refers o he sensvy of frms operang cash flows o currency movemens. These 25 porfolos are consruced n a smlar fashon as he 25 sze and cash-flow-sensvy porfolos n Secon 3. The currency exposure resuls are repored n Table 7 for he enre sample perod from 1980 o 2008 n he same fashon as Table 4. As we can see, odel 2 (he sandard four-facor model usng he change n exchange raes as he currency rsk proxy) does no ouperform odel 1 (he Fama-French hree-facor model): he average of he absolue values of Alphas of odel 2 s 0.36 whch s equal o ha of odel 1; he average of he adjused-r 2 s of 0.43 s also equal o ha of odel 1; furhermore, here are egh porfolos ha have sgnfcan exposure o he change n CI, bu her loadngs do no change accordngly wh he cash-flow sensvy. Noe agan ha 25 cash-flow-sensvy porfolos are consruced based on he sensvy of frms operang cash flows o currency movemens. In conras, odel 3 ncludng PN margnally ouperforms boh odel 1 and odel 2. The average of he absolue values of Alphas of odel 3 s 0.35 whch s slghly smaller han hose of odel 1 and odel 2. The average of he adjused-r 2 s of 0.44 s also slghly hgher han hose of odel 1 and odel 2. Furhermore, 13 ou of 25 cash-flow-sensvy porfolos have sascally sgnfcan exposure o he currency rsk PN, and negave/posve cash-flow-sensvy porfolos generally have negave/posve loadngs on he PN facor. The rsk premum resuls are repored n Table 8 for he enre sample perod from 1980 o As we can see, odel 2 does no ouperform odel 1 n erms of explanng he cross-seconal dfferences n he reurns of he 25 cash-flow sensvy porfolos. Frs, he change n CI s no prced. Second, addng he change n CI decreases adjused R 2. In conras, odel 3 sgnfcanly ouperforms boh odel 1 and odel 2. Regardless of wheher we use OLS or GLS, he adjused-r 2 mproves subsanally. ore mporanly, he currency rsk facor s prced, and he esmaed premum of 0.72 percen per monh (based on OLS) s very close o he esmae based on he 25 sze and cash-flow sensvy porfolos. Therefore, our resuls furher confrm ha currency rsk maers for asse prcng f we focus on he mpac of currency movemens on frms cash flows. 4.2 Usng Real Exchange Raes Prevous sudes usually fnd ha usng eher nomnal or real exchange raes s no crucal for currency rsk research (.e. Joron, 1990; Sarks and We, 2005). For robusness, we repea he prevous ess wh real CI seres from he Board of Governors of he Federal Reserve Sysem. Tha s, we consruc he PN facor and he 25 sze and cash-flow-sensvy porfolos n he same way as n Secons 2 and 3 excep ha we use real CI. We hen redo all he ess n Secon 3 wh he new PN and 25 sze and cash-flow-sensvy porfolos. When we rees odel 2 (he sandard four-facor model), we also use he real CI change. 8 In general, conssen wh prevous sudes, we fnd ha he resuls based on real exchange raes are qualavely smlar as hose based on nomnal exchange raes. In our case, he four-facor model ncludng PN agan ouperforms boh he hree-facor model and he fourfacor model ncludng he change n real CI, whch agan suggess ha currency rsk maers for asse prcng f we focus on he mpac of currency movemens on frms cash flows. 8 To save space, we do no repor he resuls. Bu hey are avalable upon reques. 13

15 Table 7 Currency Exposure based on he 25 Cash-Flow Sensvy Porfolos over odel 1 odel 2 odel 3 Cash Flow Sensvy α Adj-R 2 α β, CI Adj-R 2 α β, PN Adj-R 2 Negave ( 0.56 ) ( 0.53 ) ( ) ( 0.73 ) ( ) ( 0.13 ) ( 0.12 ) ( ) ( 0.60 ) ( ) ( ) ( ) ( 1.66 ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( 0.84 ) ( 0.85 ) ( 2.52 ) ( 1.35 ) ( ) ( ) ( ) ( 1.83 ) ( 0.18 ) ( ) ( 0.52 ) ( 0.52 ) ( 0.31 ) ( 0.85 ) ( ) ( 1.74 ) ( 1.74 ) ( ) ( 2.43 ) ( ) ( 0.18 ) ( 0.18 ) ( 0.35 ) ( 0.41 ) ( ) ( 0.88 ) ( 0.88 ) ( 0.19 ) ( 1.23 ) ( ) ( 0.53 ) ( 0.52 ) ( ) ( 0.77 ) ( ) ( ) ( ) ( ) ( ) ( ) ( 1.62 ) ( 1.62 ) ( 2.82 ) ( 2.10 ) ( ) ( 1.91 ) ( 1.92 ) ( 0.60 ) ( 1.92 ) ( 0.65 ) ( ) ( ) ( ) ( ) ( 0.18 ) ( 1.88 ) ( 1.90 ) ( 0.55 ) ( 1.83 ) ( 1.04 ) ( 2.02 ) ( 2.09 ) ( 2.11 ) ( 1.98 ) ( 0.64 ) ( ) ( ) ( 2.02 ) ( ) ( 1.35 ) ( 1.52 ) ( 1.59 ) ( 1.70 ) ( 1.27 ) ( 2.23 ) ( 1.47 ) ( 1.46 ) ( ) ( 1.33 ) ( 1.05 ) ( 1.83 ) ( 1.83 ) ( ) ( 1.60 ) ( 1.15 ) ( 1.51 ) ( 1.58 ) ( 1.67 ) ( 1.40 ) ( 0.73 ) ( 2.92 ) ( 2.91 ) ( 0.68 ) ( 2.85 ) ( 1.13 ) ( 2.25 ) ( 2.30 ) ( ) ( 1.86 ) ( 2.28 ) Posve ( 1.19 ) ( 1.17 ) ( ) ( 0.50 ) ( 3.79 ) odel 1: r odel 2: r odel 3: r α β β SB β HL ε,, SB, HL α β β SB β HL β CI ε,, SB, HL, CI β, β, SB SB β, HL HL β, PN α PN ε where r s he excess reurn on asse n perod,, HL, PN and CI are he marke, he sze, he book-o-marke, he currency rsk, and he conemporaneous change n exchange rae. 14, SB

16 Table 8 Rsk Premum Esmaes based on he 25 Cash-Flow Sensvy Porfolos over odel 1: odel 2: odel 3: odel 1 OLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL Adj-R GLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL Adj-R odel 2 OLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL CI Adj-R GLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL CI Adj-R odel 3 OLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL PN Adj-R GLS Coeffcen EIV-robus -rao IS-robus -rao Alpha SB HL PN Adj-R r r γ γ ˆ β γ ˆ β γ ˆ β η 0, SB, SB HL, HL r γ γ ˆ β γ ˆ β γ ˆ β γ ˆ β η 0, SB, SB HL, HL CI, CI γ ˆ ˆ ˆ ˆ η 0 γ β, γ SBβ, SB γ HLβ, HL γ PN β, PN where he βˆ s are he facor loadngs (jonly esmaed for all he facors) from he frs-pass me-seres regresson, and γ s are he assocaed rsk premums. EIV s he Shanken (1992) s errors-n-varables robus -raos, and IS s he Shanken and Zhou (2007) s msspecfcaon robus -raos. 15

17 5. Concluson Alhough mos heorecal models of currency rsk emphasze he mpac of currency flucuaons on frms cash flows, he sandard emprcal approach s o focus on sock reurns nsead. Ths approach, n general, fnds ha currency rsk s no prced. Despe he weak resuls from hs sandard approach, few sudes have aemped o examne currency rsk from a cash flow perspecve. Ths paper nends o fll hs gap. In erms of emprcal mplemenaon, we use he mmckng porfolo approach ha s n lne wh domesc asse-prcng models of Fama and French (1992, 1993) and ulzed by Kolar, oorman, and Sorescu (2008) o examne currency rsk. Essenally, we consruc a currency rsk facor porfolo (PN) ha reflecs he mpac of currency movemens on frms cash flows. Ths facor s shown o be orhogonal o he Fama-French hree facors. To es our conjecure, we focus on he comparson beween he sandard hree-facor model (as well as he sandard four-facor model ha employs he exchange rae change as he currency rsk proxy) and our new four-facor model ha ncludes PN. We show ha, n boh me seres exposure regressons and cross-seconal premum regressons, he sandard four-facor model does no ouperform he Fama-French hree-facor model, bu our new four-facor model ha akes no accoun he mpac of currency movemens on frms cash flows ouperforms boh he sandard fourfacor model and he Fama-French hree-facor model. Our resuls are robus regardless of wheher we use a dfferen se of esng asses or wheher we use real or nomnal exchange raes, whch suggess ha currency rsk maers for asse prcng f we focus on he mpac of currency movemens on frms cash flows. 16

18 References Adler,., and Dumas, B., Inernaonal porfolo choce and corporae fnance: a synhess. Journal of Fnance 38, Allayanns, G., The me-varaon of he exchange rae exposure: An ndusry analyss, Workng paper, Darden Graduae School of Busness, Unversy of Vrgna Barov, E., and G.. Bodnar Frm Valuaon, Earnngs Expecaons, and he Exchange Rae Exposure Effec, Journal of Fnance 44:5: Barram, S.., Lnear and nonlnear foregn exchange rae exposures of German nonfnancal corporaons, Journal of Inernaonal oney and Fnance Barram, Sohnke., 2007, "Corporae Cash Flow and Sock Prce Exposures o Foregn Exchange Rae Rsk," Journal of Corporae Fnance, 13, Barram, Sohnke., 2008, "Wha Les Beneah: Foregn Exchange Rae Exposure, Hedgng and Cash Flows," Journal of Bankng and Fnance, 32, Barram, S.., and Bodnar, G.., The foregn exchange exposure puzzle. Workng paper, John Hopkns Unversy and Lancaser Unversy. Barram, Sohnke., G.W. Brown and B. non, 2010, "Resolvng he Exposure Puzzle: The any Faces of Exchange Rae Exposure," Journal of Fnancal Economcs, 95, Black, F.,. C. Jensen, and. Scholes, 1972, The Capal Asse Prcng odel: Some Emprcal Fndngs; n chael C. Jensen, ed.: (Praeger, New York). Sudes n he Theory of Capal arkes Bodnar, G.., and Wong,. H. F., Esmang exchange rae exposures: ssues n model srucure. Fnancal anagemen 32, Chan, L. K. C., J. Karcesk, and J. Lakonshok, 1998, The rsk and reurn from facors, Journal of Fnancal and Quanave Analyss 33, Chan, L. K. C., J. Karcesk, and J. Lakonshok, 1999, On porfolo opmzaon: Forecasng covarances and choosng he rsk model, The Revew of Fnancal Sudes 12, Chow, E.H., Lee, W.Y., and Sol,.E., The exchange-rae rsk exposure of asse reurns. Journal of Busness 70, Da, Zh, 2009, Cash Flow, Consumpon Rsk, and he Cross-secon of Sock Reurns, Journal of Fnance 64, Fama, E.F., French, K.R., The cross-secon of expeced sock reurns. Journal of Fnance 47, Fama, E.F., and French, K. R., Common rsk facors n he reurns on socks and bonds. Journal of Fnancal Economcs 33, Fama, E. F., and J. D. acbeh, 1973, Rsk, Reurn and Equlbrum: Emprcal Tess, Journal of Polcal Economy 81, Froo, K., Sharfsen, D. and Sen, J., Rsk managemen: Coordnang corporae nvesmen and fnancng decsons. Journal of Fnance 48, Francs, B., I Hasan, and D. Huner, 2008, Can hedgng ell he full sory? Reconclng dfferences n U.S. aggregae- and ndusry-level exchange rae rsk premum, Journal of Fnancal Economcs, 90,

19 Grffn, J.., Are he Fama and French facors global or counry specfc? Revew of Fnancal Sudes 15, Hrshlefer, D., Hou, K., and Teoh, S. H., Accruals, cash flows, and aggregae sock reurns, Journal of Fnancal Economcs 91, Hou, K., A. Karoly and B. Kho, 2011, Wha Facors Drve Global Sock Reurns?, Revew of Fnancal Sudes, forhcomng. Huang, A. G., 2009, The cross secon of cash flow volaly and expeced sock reurns, Journal of Emprcal Fnance 16, Jagannahan, R., and Z. Wang, 1998, An Asympoc Theory for Esmang Bea-Prcng odels Usng Cross-Seconal Regresson, Journal of Fnance 53, Joron, P., The exchange-rae exposure of U.S. mulnaonals. Journal of Busness 63, Joron, P The Prcng of Exchange Rae Rsk n he Sock arke, Journal of Fnancal and Quanave Analyss 26:3: Khoo, A., Esmaon of foregn exchange rae exposure: an applcaon o mnng companes n Ausrala. Journal of Inernaonal oney and Fnance 13, Km, D., 1995, The Errors n he Varables Problem n he Cross-Secon of Expeced Sock Reurns, Journal of Fnance 50, Kolar, J., oorman, T., and Sorescu, S..., Foregn Exchange Rsk and he Cross-Secon of Sock Reurns. Journal of Inernaonal oney and Fnance 27, Lewellen, J. W., Nagel, S., and Shanken, J. A., A Skepcal Apprasal of Asse Prcng Tess, Journal of Fnancal Economcs 96, Nucc, F., and A. F. Pozzolo, 2010, The exchange rae, employmen and hours: Wha frm-level daa say, Journal of Inernaonal Economcs, 82, Shanken, J., 1992, On he Esmaon of Bea-Prcng odels, Revew of Fnancal Sudes 5, Shanken, J., and G. Zhou, 2007, Esmang and Tesng Bea Prcng odels: Alernave ehods and Ther Performance n Smulaons, Journal of Fnancal Economcs 84, Smh, C. W., and Sulz, R., The deermnans of frms hedgng polces, Journal of Fnancal and Quanave Analyss 20, Sarks, L. T., and We, K. D., Foregn exchange exposure and shor-erm cash flow sensvy. Workng paper, Unversy of Texas, Ausn, TX. Sulz, R., Opmal hedgng polces, Journal of Fnancal and Quanave Analyss 19, Vassalou,., Exchange rae and foregn nflaon rsk premums n global equy reurns. Journal of Inernaonal oney and Fnance 19,