Does Gold Love Bad News? Hedging and Safe Haven of Gold against Stocks and Bonds

Transcription

1 Does Gold Love Bad News? Hedging and Safe Haven of Gold agains Socks and Bonds Samar Ashour* Universiy of Texas a Arlingon samar.ashour@mavs.ua.edu (682) January *Corresponding auhor: Samar Ashour docoral suden Universiy of Texas a Arlingon Box Deparmen of Finance and eal Esae Arlingon Texas

2 Does Gold Love Bad News? Hedging and Safe Haven of Gold Agains socks and Bonds SAMA ASHOU ABSTACT This paper examines he conemporaneous and dynamic relaion beween gold excess reurns and he socks and corporae bonds excess reurns. The resuls show ha gold can be used as a hedge and safe haven agains bonds and socks. This paper is he firs o use CAPM and Fama French hree facor models o check he relaion beween gold marke and he sock marke as well as corporae bond marke. Over he enire sample period from 1986 o 2012 here is a significan relaion beween gold excess reurns and marke excess reurns as well as he corporae bonds excess reurns even afer conrolling for Fama French hree risk facors. Addiionally during exreme marke condiions such as crashes and crises here is a negaive relaion beween sock reurns and gold reurns in crashes imes and here is a zero correlaion beween sock and bonds reurns and gold reurns in crises imes and consequenly gold can be used as a hedge agains socks and bonds in he ension imes. The dynamic relaion beween hree markes using VA and IF indicaes ha gold can be used as a safe haven agains socks and bonds. I. Inroducion Gold is a well-known as a precious meal and sore of value for hundreds of decades. Gold is sill he mos famous precious meal for invesmen purposes since many people ransfer heir invesmens from he sock and corporae bonds o miigae heir losses in hese risky markes during he recession periods especially during crashes and crises. Gold rading and prices seing are similar o sock rading. The purpose of his sudy is o invesigae he role of gold as a safe haven or hedge agains losses in financial markes hrough he equiy marke and corporae bond marke by using recen daa and by applying CAPM and Fama French hree facor models o examine he saic relaion beween socks reurns and gold reurns and using vecor auoregressive model (VA) and impulse response funcion (IF) o invesigae he dynamic relaion beween he hree 1

3 markes. In paricular I examine he iner-relaionships beween hree differen markes gold marke sock marke and bond marke over he period from 1986 o 2012 and during he bad saes of he economy and exreme marke performance especially in financial crashes such as 1987 Sock Marke crash (Black Monday) Sepember eleven crash in 2001 as well as he mos recen crises (Asian crisis and he global housing crisis ( ). The conribuion of his sudy is wofold. Firs unlike Baur and Lucey (2010) ha examine he relaion beween hree markes using GACH and using daa from 1995 o 2006 and find ha gold canno be hedge agains bonds I used more recen and comprehensive daa and he mos imporan asse pricing models (CAPM and Fama French hree facor models) and I find a significan negaive relaion beween bonds and gold. Second unlike he exising lieraure ha examines eiher he relaion beween gold prices and sock prices (e.g. Baur and Lucey 2010 and Baur and McDermo 2011) or beween gold prices and exchange raes (e.g. Capie Mills and Wood 2005) or even invesigae he riskiness of he gold (McCown e.al. 2006) his sudy aemps o link many sreams of lieraure by examining he co-movemens beween he gold reurns sock reurns and bond yields using differen models. Third his aricle is he firs aemp o use CAPM and Fama French hree facor models o know he saic relaion beween gold socks and bonds. Moreover his is he firs sudy o vecor auoregressive model (VA) and impulse response funcion (IF) o examine he marke inerdependencies and he dynamic relaion beween he hree classes of asses during bad saes of he economy wheher in financial crashes or financial crises. A relaively lile research has been done in his area. My argumen is ha he dependence srucure beween he hree markes is no consan and i may be differen beween normal and calm periods. Mos invesors realized grea losses especially in he las financial crisis because of he significan drop in he sock prices and he experienced subsanial losses 2

4 due o his phenomenon. This paper ries o provide a suggesion for invesors o include gold in heir invesmen porfolios especially if gold commove agains boh socks and bonds. I find ha he negaive relaion beween socks and gold is sronger and more economically and saisically significan han he negaive relaion beween bonds and gold. To verify ha hese resuls are no spurious I used differen models o conrol for Fama French hree facor models; SMB which is he difference beween reurns on he small firms and he reurns on he large firms HML he difference beween he high book o marke reurns and low book o marke reurns. Addiionally o know he dynamic relaion beween hree markes I used he lagged reurns of he hree markes and run vecor auoregressive regression o make sure ha he resuls are no spurious due o auocorrelaion or he endogeniy problem and o ge he linkage beween hree markes assuming every ime ha each of hem is dependen variables and i is lagged variable as well as he oher wo markes in heir lagged forms are explanaory variables o his marke reurn. There is an imporan sudy in his area by Baur and McDermo (2011) bu hey resric his sudy o gold and socks bu my paper will invesigae he inerrelaionship beween hese hree variables. Moreover Baur and Lucy (2010) use index reurns daa only o represen socks and hey did no conrol for any variables and hey did no use subsamples o check he robusness of he safe haven finding. However my sudy uses a novel daa se in his area of lieraure. My sudy is disinguished by using he value weighed porfolio of all NYSE NASDAQ and AMEX from CSP as well as using S&P 500 from CSP afer comparing i wih FED o correc for missing observaion. Addiionally Baur and Lucy (2006) uses MSCI bond index bu I used boh AAA bonds and BAA bonds from Moody s service by using FED websie. The resuls provide convincing evidence ha gold can be used as a hedge and safe haven agains socks and bonds. 3

5 II. Lieraure eview Gold has many advanages over invesmen in socks and bonds or even foreign exchange rae. Gold is a angible good so ha when he invesor purchases gold yields an objec of angible value. Firs from he consumpion side he gold invesor no only purchases gold for hedging bu also he invesor benefis from using i as jewelry so ha he uiliy of gold is higher because i is considered consumpion good. Second from he producion side he producer can hold gold invenory o hedge himself form he sock ou risk. Third he gold is very liquid especially in ension imes so ha he invesor can pay higher prices and realizes lower reurns compared o oher risky asses However people fligh o less risky invesmens during bad saes of economy and consequenly he price of gold afer 2009 is subsanially increase. This fac agains he famous invesmen rule: buy low sell high. This sriking high prices lead o an expecaion ha i will coninue o rise forever bu according o he pas gold hisory his is no always he scenario.(davision 2013) My work is relaed o wo srands of lieraure. One sream focuses on examining he gold prices Fama and French (1988) sudy he relaion of meal pricing wih business cycles and hey find ha posiive demand chocks during peak periods lower he meal invenories and causes severe price conversions according o he heory of sorage while he oher srand invesigaes he relaionship beween gold prices and sock prices. Baur and Lucey (2010) invesigae wheher gold can serve as a safe haven and hedge agains sock and/or bonds during bad imes. The resuls show ha here is a fligh o qualiy from socks o gold during periods of ension bu no from bonds o gold. Baur and McDermo (2010) provide inernaional evidence o he role of gold as a hedge and a safe haven agains sock prices. In paricular hey run muli-counry analysis using daa from 53 inernaional sock markes (which include developed and 4

6 developing markes) from March 1997 o March Alernaively Capie Mills and Wood (2005) provide evidence on he role of gold as an exchange rae hedge and hey find ha here is a negaive relaion beween gold prices and Serling-Dollar and Yen-Dollar exchange raes. This paper is relaed o he fligh-o-qualiy lieraure. Gulko (2002) examines wheher he invesors ransfer heir invesmens from socks o bonds in he exreme marke condiions (Gonzalo and Olmo (2005)) saed ha invesors fligh o qualiy (from socks o bonds). from he above discussion i is clear ha here is one srand of lieraure ha finds ha Gold is a sock marke hedge while he oher sream provides evidence ha Gold is no hedge agains corporae bonds. Given he scarciy of lieraure on he role of gold as a hedging ool in he financial markes in general he conribuion of his sems from linking he above wo srands of lieraure by examining he relaion beween gold prices bond prices and sock prices over he enire sample period and during he bad saes of he economy wheher in recessionary and he financial crises sub periods. Barro and Misra (2013) find ha here is a saisically insignifican covariance of change in he gold s real rae wih consumpion and GDP growh rae and hey documen ha he prediced reurns on gold is very close o he risk free rae. Jaffe (1989) jusifies why ha he gold reurn is lower han socks and bonds by referring o is non-pecuniary benefis. The gold invesor no only purchase gold for hedging bu also he invesor benefis from using i as jewelry so ha he uiliy of gold is higher because i is considered consumpion good. Moreover from he producion side he producer can hold gold invenory o hedge himself form he sock ou risk. Finally he gold is very liquid especially in ension ime so ha he invesor can pay higher prices and realizes lower reurns compared o oher risky asses. McCown (2006) sudy he degree of he riskiness of gold by using CAPM model and Arbirage pricing model and find ha 5

7 he gold has a grea hedging abiliy because i has very similar expeced reurn of Treasury bill. They provide convincing evidence ha gold is zero bea asse and gold have no marke risk. However my paper ries o provide convincing evidence ha gold has negaive bea (i.e. comovemens agains marke porfolio) and o exend his o see he relaion beween gold marke sock marke porfolio and corporae bond marke afer aking ino accoun all Fama French hree risk facors. III. Mehodology 1. Hedging Hypohesis: My firs hypohesis herefore is ha gold may be used as a hedge insrumen agains socks and bonds if here is a zero or negaive relaion beween gold reurns and socks and bonds reurns during he enire sample period from 1986 o According o Baur and Lucey (2010) gold can be considered as a hedge agains bond and socks when here is a zero or negaive relaion beween gold and bonds and/or socks on average during periods of calm and ensions. 1.1.CAPM Since I am ineresed in he response of gold prices o flucuaions in sock reurns I run he following CAPM regression: (1) f 0 1 Sock f Where ) he daily reurn on gold prices is ) is he daily reurn on S&P index and ( f ( gold ( S& P ) is he risk-free rae. In order o examine he inerrelaions beween bonds sock and gold markes I incorporae he daily gold reurn ) o he above CAPM equaion as follows: ( bond (2) f 1 Sock f 2 bond f 6

8 Equaion (2) is he cenral because i examines he conemporaneous relaionship beween gold socks and bond in one equaion so ha as a robusness I used many measuremen for marke excess reurns. I used S&P500 Value weighed reurns wihou excluding dividends which is he same as marke excess reurn from Fama French Library VW value weighed reurns including dividends VWD equal weighed reurns EW and equal weighed reurns including dividends EWD. In empirical resul secion I will sar by displaying he resuls of his equaion and hen I will presen he resuls of CAPM and Fama French models. 1.2 Fama French Model: In order o accoun for he size and value premium I regress he reurn on gold on he hree Fama-French facors as follows: HML SMB (3) f 1 Sockj f 2 3 In addiion I also incorporae he bond reurns o equaion (3) as follows: gold f 3 1 Sock f 2 HML f SMB (4) f 1.2. Vecor Auoregression Model (VA) The VA sysem reas all of he variables in he model as endogenous variables. Therefore he dynamic relaion beween gold marke sock marke and bond marke can be examined by esimaing he following hree-vecor auoregressive (VA) sysem: I J 4 bond gold - f =l 0 + å a i gold-1 - f + å b j Sock- j - f + åh k bond-k - f + e i (5) i=1 j=1 k=1 K f I J K Sock f 0 i Sock 1 f j gold j f i 1 j 1 k 1 k bond k f i (6) 7

9 I J K bond f 0 i bond 1 f j gold j f i 1 j 1 k 1 k Sock k f i (7) 1.3. Impulse esponse Funcion (IF) The problem wih he variance-covariance esimaed from he VA model is ha errors are unlikely o be diagonal. This means ha i is difficul o shock one variable while holding oher variables consan. Therefore I use he impulse response funcion (IF) o measure he response of gold reurns o a lagged uni impulse in sock reurns while he bond reurns consan. Conversely I focus on examining he impac of he impulse bond reurns on gold reurns while holding he S&P index reurns consan. 2. Safe Haven Hypohesis: My second hypohesis is ha here should be fligh o qualiy from corporae bonds and sock marke o gold during exreme marke condiions. In oher words he safe haven hypohesis saes ha US invesors in he sock marke and bond marke view he gold as a hedge and safe haven o compensae hem for losses during sock marke crashes. Gold can be considered as a safe haven raher han hedge agains socks and bonds if here is a zero or negaive relaion beween gold reurns and bonds and socks during he ension imes only. In order o examine he safe haven hypohesis I esimae he following regression models: gold_ Tension f 0 1 (7) Sock _ Tension f gold_ Tension f gold_ Tension gold 0 f f 1 (8) 3 Sock _ Tension 8 f 2 bond_ Tension f 1 Sock _ Tension f 2 HML _ Tension 3 SMB _ Tension (9) 1 Sock 2 _ f HML Tension Tension SMB (10) _ Tension 4 bond_ Tension f

10 I will also used he VA and IF o es he second Hypohesis as follows: I J K gold _ Tension f 0 i gold _ Tension 1 f j Sock _ Tension j f i 1 j 1 k 1 I J K Sock _ Tension f 0 i Sock _ Tension 1 f j gold_ Tension j f i 1 j 1 k 1 I J K bond_ Tension f 0 i bond_ Tension 1 f j gold_ Tension j f i 1 j 1 k 1 k bond_ Tension k k k f bond_ Tension k Sock_ Tension k i f f (11) i i (12) (13) Where ) ) and ) are ineracion variables of reurns and ( gold _ Tension ( Sock _ Tension ( bond_ Tension dummy variable (Tension) ha akes value of 1 during he ension ime and zero oherwise. These ineracion variables represen daily reurn on gold Marke index and bonds during he crisis period respecively. IV. Daa and Variable Measuremen My sample period is exended from April ill Ocober The sample period will be divided furher ino differen subsamples o cover he periods of four exreme marke condiions (Ocober 1987 Crash Asian Crisis Sepember eleven 2001 crash and he recen housing crisis ). This sudy uses daily price daa from hree differen markes - gold sock and corporae bonds. The gold prices are he gold fixing price 3:0 P.M. (London ime) in London Bullion Marke based on U.S. dollars obained from London Bullion Marke Associaion hrough FED. Gold prices are expressed in U.S. dollars per roy ounce from April 3rd 1986 o December 31s I use he log daily reurns of he value-weighed (VW) and equal-weighed (EW) porfolios obained from CSP. The log ineres rae (TBL) is compued from he one-monh Treasury bill which is also obained from CSP. Addiionally I obain he daily daa of Corporae Aaa and Baa bonds yields in a percenage form from he Board of Governors of he Federal eserve Sysem which is originaed in Moody s invesor service ha includes bonds wih remaining mauriies as close as possible o 30 years 9

11 and drops he bonds if he remaining life falls beween 20 years if he bond is suscepible o redempion or if he raings change. Baur uses GACH o esimae he volailiy of socks included in his sample. This sudy also calculaes he condiional volailiy of gold and socks using GACH (1 1). I use he excess log daily reurns of gold socks and bonds. The marke excess reurns he Fama French hree risk facors and he reurns on he one monh Treasury bill are obained from Kenneh French websie. In order o es he safe haven hypohesis I need o focus on periods of bad saes of economy specifically speaking during sock marke crashes which are he Black Monday crash in 1987 Asian crises in 1997 he Sepember aack and he recen Housing crisis in Table (1) presens he saring dae for each marke crash or financial crisis. Table (1): Marke Crashes and Crisis Periods Crises Saring dae Marke crash (Black Monday) Ocober 17 h 1987 Sepember eleven Crash Sepember Asian Crisis July 2 nd 1997 Subprime crisis (Housing Bubble) July 1s Preliminary esuls: V. Empirical esuls Figure (1) shows he prices of gold from 1986 o Alhough he price of gold was no changed considerably from 1986 o 2002 i sars o increase dramaically afer he recen financial crisis. In addiion figure (2) show ha he gold volailiy and sock volailiy is very high during he seleced subsamples ha is why I concenraed on hese four criical periods o invesigae if he gold can be used as a safe haven agains socks and bonds or no. 10

12 Figure (1) Gold prices and sock prices from 1986 o 2012 The gold prices in his paper are he gold fixing 1 price 3:0 P.M. (London ime) in London Bullion Marke based in U.S. dollars obained from London Bullion Marke Associaion hrough FED. Gold prices are expressed in U.S. dollars per roy ounce from April 3rd 1986 o December 31 s S& P500 daily prices and reurns from CSP (U.S. Dollars per Troy once) GOLDPICES 1600 S&p500 Prices in U.S. Dollars LEVELOFTHES&P500INDEX LEVELOFTHES&P500INDEX Gold Prices 1 The fixings is an open process a which marke paricipans can ransac business on he basis o single quoed price. Orders can be changed hroughou he proceedings as he price move higher and lower unil he price is moved higher and lower unil such ime as buyers' and sellers' orders are saisfied and he price is said o be 'fixed'. Orders execued a he fixings are conduced as principal-o-principal ransacions beween he clien and he dealer hrough whom he order is placed. 11

13 Figure (2) Condiional Volailiies (GACH (1 1)) of Socks and Gold (Percen) SP500VOLATILITY.0010 (Percen).0008 (Percen) GOLDVOLATILITY S&P Volailiy Gold Volailiy 12

14 Table (2) shows he summary saisics of gold prices S&P prices and he bond prices in addiion o he excess reurn on gold bond and sock. Table (3) presens he correlaion marix beween gold reurn bond reurn and sock reurn. Table (2): Summary Saisics 1986:04: :10:31(6510obs) Series Mean Sandard Deviaion Minimum Maximum GOLDPICES S&PPICES f VWD-TBL VW-TBL EWD-TBL EW-TBL SP-TBL AAA-TBL BAA-TBL Table (3): The Correlaion Marix Gold eurn Bond Sock eurns Series f VW-TBL VWD-TBL EW-TBL EWD-TBL SP-TBL AAA-TBL BAA-TBL f 1.00 (0.00) VW- TBL (0.05) 1.00 (0.00) 1986:04: :10:31(6510 Obs.) VWD- TBL EW-TBL EWED-TBL SP-TBL (0.05) (0.74) (0.75) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) 0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) 1.00 (0.00) AAA- TBL (0.00) (0.58) (0.66) (-0.76) (0.72) (0.66) 1.00 (0.00) BAA- TBL (0.00) (0.47) (0.56) (0.77) (0.75) (0.66) 0.96 (0.00) 1.00 (0.00) From he correlaion marix i is obvious ha here is a negaive correlaion beween gold excess reurns and socks and bonds excess reurns. Addiionally here is insignifican correlaion beween socks and bonds excess reurns and consequenly boh of hem can be included in he same equaion wihou he presence of he problem of mulicollineariy. 13

15 2. The Empirical esuls of esing he Hedging Hypohesis: 2.1. The Conemporaneous relaion beween gold socks and bonds. In his secion I examine he firs sudy hypoheses. I invesigae wheher gold can be used as a hedge agains socks and bonds or no. If I find significan negaive coefficien of he socks and bonds over he enire sample period his means ha invesors can depend on gold invesmen o hedge hemselves from he losses in he sock and bond markes. Table (4) repors he resuls of regressing he gold excess reurns on he sock reurns (S&P500 excess reurns mean excess value weighed reurns VW and he mean excess equal weighed reurns) and he corporae bond yields (Aaa and Baa) during he enire sample period ( ). These resuls provide convincing evidence ha here is a negaive relaion beween he sock reurns and he gold prices reurns. In oher words if he value weighed reurns of all NYSE AMEX NASDAQ porfolio drops on average by 1% he gold reurns move o he opposie direcion by 0.02% and can be used as a hedge agains any surprising falls in he sock prices. This relaion is economically and saisically significan. The negaive relaion beween S&P500 and gold reurns are more dramaic and saisically significan because any drop in S&P500 reurns by 1% can be parially compensaed by increase in he gold reurns by 0.4% he saisics and he P-values show he probabiliies of acceping he null hypohesis. The negaive relaion beween corporae bonds yields and socks are more saisically bu less economically significan. We can imply from hese resuls ha gold can be used as a hedge agains drop in he sock reurns. This resul is Consisen wih Baur and Lucey (2006)) and Baur and McDermo (2009). Addiionally here is a saisically evidence ha gold can be used as hedge agains he drop in he corporae bond yields (Inconsisen wih Baur and Lucey (2006)). 14

16 My resuls can add o he exising lieraure by finding significan negaive relaion beween gold and bonds by using more comprehensive and curren daase for Aaa and Baa corporae bonds. Table (4) Hedging predicabiliy of gold over he period from 1986 o 2012 Table (4) repors he relaion beween gold reurns wih socks reurns and bonds yields. The dependen variables is he gold excess reurn f mean excess value-weighed and equal-weighed porfolios VW-TBL EW- TBL excluding dividends and he mean excess value weighed porfolios including dividends VWD-TBL EWD- TBL; and S&P 500 mean excess rerun are all proxy o he sock reurns. AAA-TBL and BAA-TBL yields are he bonds reurns. All raes are represened in annualized percenage poins.. The firs number in he hird column refers o he -saisic of he es and he second number refers o p value under he null hypohesis ha he gold canno be used as a hedge agains drop in he sock and bond reurns. 1986:04: :10:31 (6510 Obs.) f -saisic VW-TBL ** (0.054) VWD-TBL ** (0.054) EW-TBL (0.76) EWD-TBL (0.76) SP500-TBL -0.04*** (0.00) AAA-TBL *** BAA-TBL *** The nex par use many models o es wheher his relaion is robus across many models especially afer conrolling for Fama-French hree facor model. 2.2.CAPM and Fama-French esuls: Table (5) repors he relaion beween he gold reurns and socks and bonds reurns. By using CAPM model o deermine if gold has zero bea or negaive bea wih he marke porfolio. Gold can be a srong hedge because is reurns commove agains he marke reurn. Afer adding he bonds o he CAPM model o deermine he relaion beween bonds reurns and gold reurns we find very significan negaive relaion beween bond reurns and gold reurns. Even afer conrolling for Fama French risk facors he negaive relaion beween marke excess reurns and gold remains robus. However i is obvious ha here is a posiive relaion (0.00) (0.00)

17 beween he beween small companies reurns and he large companies reurns. This implies ha small socks and gold may move wih he same direcion. However even afer conrolling for his posiive relaion he marke excess reurns as well as bonds excess reurns commove agains gold reurns. Table (5) CAPM and Fama French esuls [he enire sample period] Table (5) documens he relaion beween he gold reurns wih he oher wo markes for he enire sample period using four models. CAPM model CAPM model including excess bonds reurns in he righ-hand side Fama French hree facor models and Fama French Model including bonds excess reurns in he righ hand side. The GOLD- TBL is he gold excess reurn. VW-TBL is he excess value-weighed reurn and S&P 500 is he excess rerun on S&P index reurn. AAA-TBL and BAA-TBL yields are he bonds reurns. All raes are represened in annualized percenage poins. -saisics are in (). 1986:04: :10:31(6510 Obs.) Series MKTF 1-VW-TBL 2-S&P500-TBL CAPM * (-1.91) *** (-3.37) CAPM+Bonds ** (-1.93) -0.04** (-3.38) FF * (-1.65) ** (-2.45) FF+Bonds * (-1.65) * (-2.74) SMB HML BONDS AAA-TBL BAA-TBL 0.07*** (3.83) 0.08*** (3.33).08*** (3.8) -0.07*** (3.27) (-0.04) (0.15) (0.16) (-0.03) *** (-3.06) *** (-2.72) *** (-2.97) *** (-2.97) 3- The Empirical esuls of esing he Safe Haven Hypohesis: 3.1. The Conemporaneous relaion beween gold socks and bonds ( During ension imes) Alhough he gold proves ha i can used as a hedge ool agains socks and bonds risks This does no mean ha i can be used as a safe have. Tha is why I inerac he socks and bonds reurns wih he ension imes- dummy variable o es wheher ha gold can be used as a safe haven in he crises and crash imes whenever he socks and bonds prices decreases dramaically. 16

18 Table (6) epors he resuls of regressing he gold reurns in he ime of crisis on all ineracion erms ha represen ha sock and bonds reurns in he exreme marke condiions. esuls show ha gold can be used as a safe haven agains socks. Negaive relaion beween wo variables ha ranges from (-0.063) using value weighed reurn porfolio and using S&P 500 as a proxy for sock reurns. There is no significan relaion beween bond reunes and he gold reurns in he ension periods. These resuls indicae ha gold can be used as a safe haven agains sock bu no agains bonds. Table (6) The conemporaneous relaion beween gold sock and bonds in he ension imes (Safe haven hypohesis) Table (6) repors he relaion beween gold reurns wih socks reurns and bonds yields. The dependen variables is he gold excess reurns such as crises or crashes GOLD_Tension. Mean excess value-weighed and equal-weighed porfolios VW-TBL EW-TBL excluding dividends and he mean excess value weighed porfolios including dividends VWD-TBL EWD-TBL; and S&P 500 mean excess rerun are all proxy o he sock reurns during ension imes. AAA-TBL and BAA-TBL yields are he bonds reurns during ension imes. All raes are represened in annualized percenage poins.. The firs number in he hird column refers o he -saisic of he es and he second number refers o p value under he null hypohesis ha he gold canno be used as a safe haven agains drop in he sock and bond reurns. VW-TBL *** VWD-TBL *** EW-TBL ** EWD-TBL ** SP500-TBL *** AAA-TBL BAA-TBL :04: :10:31 Gold_Tension -saisic f

19 3.2. CAPM and Fama French resuls: Table (7) The relaionship beween gold reurns and he sock and bonds reurns during exreme marke condiions for he whole sample period from 1986:04:1-2012:10:31 using CAPM and Fama French models (Safe haven hypohesis) Table (7) documens he relaion beween he gold reurns wih he oher wo markes during exreme marke condiions using four models. CAPM model CAPM model including excess bonds reurns in he righ-hand side Fama French hree facor models and Fama French Model including bonds excess reurns in he righ hand side. The dependen variables is he gold excess reurn GOLD_Tension.. The firs row of each model documens mean excess value-weighed and equal-weighed porfolios VW-TBL which is he same as (MKTF) mean excess value weighed porfolios excluding; and he second row of each model represens S&P 500 mean excess rerun are all proxy o he sock reurns during ension imes. AAA-TBL and BAA-TBL yields are he bonds reurns during ension imes. All raes are represened in annualized percenage poins.. -saisic are in (). Series CAPM *** (-3.22) -0.08*** (-4.14) CAPM+Bonds *** (3.27) *** (04.16) 1986:04: :10:31(1440bs.) MKTF 1-VW-TBL 2-S&P-TBL FF *** (-3.43) *** (-3.88) SMB HML BONDS AAA BAA 0.098** (2.58) * (1.75) -0.09** (-2.27) ** (-2.41) (-1.3) (-1.1) FF+Bonds *** (-3.44) *** (-3.89) (2.57) 0.07* (1.73) -0.09** (-2.38) ** (-2.25) (-1.19) (-1.01) 4- obusness Tess: 4.1.Subsamples As a check for robusness I will analyze he relaion beween he change in he gold prices wih he sock and bond reurns in each crisis separaely o make sure if he gold can be used as a safe haven agains socks and corporae bonds reurns. I will analyze he relaion beween gold prices wih he oher risky asses in he 20 days afer he crash or crisis sar. 18

20 Moreover I will analyze he naure of his relaion over he enire Asian crisis and subprime crisis period. I included here he mos imporan crashes and crises ha experienced dramaic decrease in he sock and bond reurns. For he black Monday crash and Sepember eleven crash I esed he relaion in he weny days afer he sar of crashes. I esed he model for Asian and subprime crises using 20 days afer he crisis sar and on overall crises period. The flowing four ables show all he coefficiens of sock reurns and corporae bonds yields in each specific subsample. Table (8) The conemporaneous relaion beween gold sock and bonds in he Black Monday crash Table (8) repors he relaion beween gold reurns wih socks reurns and bonds yields during he black Monday crash. The dependen variables is he gold excess reurn f mean excess value-weighed and equalweighed porfolios VW-TBL EW-TBL excluding dividends and he mean excess value weighed porfolios including dividends VWD-TBL EWD-TBL; and S&P 500 mean excess rerun are all proxy o he sock reurns. AAA-TBL and BAA-TBL yields are he bonds reurns. All raes are represened in annualized percenage poins.. The firs number in he hird column refers o he -saisic of he es and he second number refers o p value under he null hypohesis ha he gold canno be used as a safe haven agains drop in he sock and bond reurns.the displays he number of observaions and :04: :5:01 (20 Obs.) 2=42% f VW-TBL *** VWD-TBL *** EW-TBL EWD-TBL SP500-TBL *** AAA-TBL BAA-TBL saisic (0.02) (0.64) (0.02) (0.00) (0.74) (0.63) Table (8) repors he relaion of he gold wih he sock and bonds reurns o es wheher he gold can be used as a safe haven in he exreme marke condiion like Black Monday crash. The coefficien of he value weighed reurn in he 20 rading days afer he beginning of he 19

21 crash is and his esimae is saisically and economically significan. There is no significan relaion beween he equal weighed reurn and he gold reurns. The coefficien of S&p500 is very close o he VW ( ) and here is negaive relaion beween he corporae bond yields and he gold reurns bu his relaion is insignifican. Table (9) shows ha gold canno be used as a safe haven agains socks and bonds for Asian crisis. Table (9) The conemporaneous relaion beween gold sock and bonds In he Asian crisis Table (9) repors he relaion beween gold reurns wih socks reurns and bonds yields during he Asian crisis. The dependen variables is he gold excess reurn f mean excess value-weighed and equal-weighed porfolios VW-TBL EW-TBL excluding dividends and he mean excess value weighed porfolios including dividends VWD-TBL EWD-TBL; and S&P 500 mean excess rerun are all proxy o he sock reurns. AAA-TBL and BAA-TBL yields are he bonds reurns. All raes are represened in annualized percenage poins.. The firs number in he hird column refers o he -saisic of he es and he second number refers o p value under he null hypohesis ha he gold canno be used as a safe haven agains drop in he sock and bond reurns.the displays he number of observaions and :07:02-o1997:08:02 (20 obs.) 20days rading afer he saring of Asian crisis 2 =0 saisic f VW-TBL (0.63) VWD-TBL (0.63) EW-TBL (0.64) EWD-TBL (0.39) SP500-TBL (0.57) AAA-TBL (0.44) BAA-TBL (0.48) 1997: :09(405obs.) The Asian crisis period 2 =0 -saisic f (0.11) (0.11) (0.013) (0.015) (0.135) (0.33) (0.301) 20

22 Table (10) The relaion conemporaneous beween gold sock and bonds in he Sepember 11 Crash Table (10) repors he relaion beween gold reurns wih socks reurns and bonds yields during Sepember 11 crash. The dependen variables is he gold excess reurn f mean excess value-weighed and equal-weighed porfolios VW-TBL EW-TBL excluding dividends and he mean excess value weighed porfolios including dividends VWD-TBL EWD-TBL; and S&P 500 mean excess rerun are all proxy o he sock reurns. AAA-TBL and BAA-TBL yields are he bonds reurns. All raes are represened in annualized percenage poins.. The firs number in he hird column refers o he -saisic of he es and he second number refers o p value under he null hypohesis ha he gold canno be used as a hedge agains drop in he sock and bond reurns.the displays he number of observaions and :09: :09:30 (20 Obs.) 2 =50.9 -saisic f (P value) VW-TBL -0.73*** (0.02) VWD-TBL -0.73*** (0.03) EW-TBL -0.62** (0.09) EWD-TBL -0.63** (0.09) SP500-TBL -0.73*** (0.02) AAA-TBL (0.34) BAA-TBL (0.33) Table (10) documens he esimaed coefficien from regressing he gold reurns on he socks reurns and corporae bonds yields for weny days afer he saring of he Sepember eleven crash from. The esimaed coefficiens indicae ha here is a highly significan negaive relaion beween socks reurns and gold reurns in his crash indicaing ha he gold can be used as a safe haven in his criical period. However he bonds reurns coefficien implies no relaion wih he gold reurns. Moreover Table (11) repors he relaion beween he gold prices change and he changes in he sock and bond prices in 20 days afer he crisis sar during he peak of he crisis and during he overall subprime crisis period. There is no significan relaion beween he gold and socks in he early sage of he subprime crisis excep beween S&P500 excess reurns and he gold excess reurns. However here is a significan relaion beween gold prices 21

23 change and he sock reurns over he peak period of he crisis. The resuls remain robus in respecive of he abiliy of he o play a role of safe haven agains corporae bonds in he exreme marke periods. Table (11) The conemporaneous relaion beween gold sock and bonds in he subprime crisis Table (11) repors he relaion beween gold reurns wih socks reurns and bonds yields during he 20days period afer he saring of he subprime crisis he relaion beween hree markes from he saring of he crisis o he peak of he financial crisis and for he enire crisis period. The dependen variables is he gold excess reurn f mean excess value-weighed and equal-weighed porfolios VW-TBL EW-TBL excluding dividends and he mean excess value weighed porfolios including dividends VWD-TBL EWD-TBL; and S&P 500 mean excess rerun are all proxy o he sock reurns. AAA-TBL and BAA-TBL yields are he bonds reurns. All raes are represened in annualized percenage poins.. The firs number in he hird column refers o he -saisic of he es and he second number refers o p value under he null hypohesis ha he gold canno be used as a safe haven agains drop in he sock and bond reurns.the displays he number of observaions and :07: :08:01(20 Obs.) 20days rading afer he sar of Subprime crisis 2 =11.5% -saisic f (p value) VW TBL (0.18) VWD TBL (0.18) EW TBL (0.16) EWD TBL (0.16) SP * TBL (0.05) AAA-TBL (0.28) BAA-TBL (0.30) 2007:7: :10:10(304 Obs.) during he peak Subprime crisis 2 =1.5% -saisic f (p value) -0.09* (0.17) -0.10* -1.66* (0.17) * (0.134) (0.161) -0.10** -1.93* (0.05) (0.84) (0.34) 2007:7: :04:02(529 Obs.) The enire period of he Subprime crisis 2 =2% -saisic f (p value) (0.91) (0.90) (0.79) (0.79) (0.63) (0.90) (0.28) 3.2.CAPM and Fama French esuls ( Subsamples) This secion displays he resuls of he safe haven hypohesis using four differen models during four differen sock marke crashes. The gold can be used as a safe haven agains socks even afer conrolling for FF hree risk facors. Addiionally gold can be used as a safe haven agains bonds risk because here is zero correlaion beween wo markes. 22

24 Table (12) CAPM and Fama French esuls for Black Monday Crash Black Monday (Marke crash) (20bs.) Series MKTF 1-VW-TBL 2-S&P-TBL CAPM *** (-2.76) (-3.67) CAPM+Bonds *** (-2.67) (-3.47) FF ** (-2.04) (-2.18) FF+Bonds * (-1.65) (-1.69) SMB HML BONDS AAA BAA 0.17* (1.75) (0.91) 0.18* (1.69) 0.19 (0.9) (-1.29) (-1.34) (-1.1) -.34 (1.17) (-0.34) ) (-0.49) (0.32) (0.28) Table (13) CAPM and Fama French esuls: Asian Crisis Series Asian crisis monhs (405)obs.) MKTF 1-VW-TBL 2-S&P-TBL CAPM (1.62) (1.49) CAPM+Bonds FF 0.148** (2.34) 0.152** (2.37) SMB HML BONDS AAA BAA 0.09 (1.3) (1.66) (1.59) (1.65) FF+Bonds 0.164** (2.57) 0.170*8 (2.61) (1.56) 0.166* (1.73) (1.83) (1.9) (-1.4) (-1.45) 23

25 Table (14) CAPM and Fama French esuls: Sepember Aack Series Sepember 11 crash (20bs.) MKTF 1-VW-TBL 2-S&P-TBL CAPM -0.70*** (-2.76) -0.68*** (-2.68) CAPM+Bonds -0.73*** (-2.8) -0.73*** (-2.82) FF *** (-4.46) *** (-4.24) FF+Bonds *** (-4.13) *** (-3.91) SMB HML BONDS AAA BAA 1.15 (1.81) (1.56) 1.15 (1.74) 1.93 (2.69) (0.15) (0.10) (0.14) 0.07 (0.09) (-0.99) (-1.03) (-0.01) (-0.00) Table (15) CAPM and Fama French esuls: Subprime Crisis Panel (A): Subprime crisis firs 20 rading days (20bs.) Series MKTF 1-VW-TBL 2-S&P-TBL SMB HML BONDS AAA BAA CAPM (1.46) (1.39) CAPM+Bonds (1.45) 0.17 (1.37) FF (1.24) 0.16 (1.12) FF+Bonds (1.07) (0.94) (-0.54) (-0.45) (-0.85) ) (-0.01) (-0.01) (-0.26) (-0.26) (0.69) (-0.33) 0.02 (0.95) (-0.16) 24

26 Panel (B): Subprime crisis from July 2007 o 0c 2008 (304bs.) Series MKTF 1-VW-TBL 2-S&P-TBL SMB HML BONDS AAA BAA CAPM (-1.38) -0.10* (-1.93) CAPM+Bonds -0.09* (-1.65) -0.10* (-1.93) FF (-1.02) (-1.58) FF+Bonds (-1.03) (-1.6) (1.53) 0.19 (1.38) 0.21 (1.53) 0.19 (1.38) (-0.01) 0.03 (0.29) -001 (-0.01) (0.12) (-0.19) (-0.34) (-0.25) (-0.25) (-0.25) (-1.07)) Panel (C): Subprime crisis whole period from July 2007 o April 2009 (528bs.) Series MKTF 1-VW-TBL 2-S&P-TBL SMB HML BONDS AAA BAA CAPM (0.09) (-0.47) CAPM+Bonds (0.09) (-0.48) FF (1.07) FF+Bonds 0.04 (1.00) 0.01 (0.09) 0.09 (0.98) -0.15* (-1.69) -0.14** (-1.98) *** (-2.01) (-1.00) (-1.08) 4.2 The dynamic relaion beween gold socks and bonds VA esuls: I is imporan o examine he lagged impac of sock reurns and corporae bond yields on he gold reurns. I will presen he resuls of using VA model ha akes ino consideraion he lagged impac of he gold lagged reurns as well as he sock reurn and bond lagged reurns 25

27 because in mos cases he negaive sock reurns are followed by posiive socks. This change in he direcion of socks reurns changes he relaion beween socks and gold compleely. In mos cases he negaive sock reurns a ime moivae invesors o invesors o purchase gold a ime +1 hen he gold prices may be affeced and he reurns on gold may drop a ime +1. The meri of using VA is ha i considers all he variables in he model as endogenous variables so ha we can know he effec of all variables in he lagged form on he dependen variable a ime. The following ables show ha here is no dynamic relaion beween gold socks and bonds. As I mensioned before If he gold is zero or negaively correlaed wih socks and bonds during exreme marke periods hen i will be considered as a safe haven agains hese wo markes. Table (16) presens he VA esimaes of he dynamic relaion beween gold socks and bonds for he enire sample period during all calm and ension saes of economy. VA accouns of he lagged impac of he socks reurns and corporae bonds yields on he reurns on gold. I include he lagged reurns up o 3 periods o accoun for he ime series effec and know how he socks and bonds reurns in affec he reurns of oday. The relaion beween hree markes is oally differen from he conemporaneous relaion beween he hree markes. The gold reurns on ime is negaively affeced by he gold reurns. The significan negaive relaion beween socks and gold s reurns is no longer exiss. However here is a significan negaive relaion beween he corporae bond yields in -1 and he gold reurns a ime. Tables (16) panel (b) repors he VA esimaes of he dynamic relaion beween gold reurns corporae bond yields and equiy reurn during wo marke crashes (Black Monday and Sepember 11) and wo global financial crises (Asian crisis and subprime crisis) respecively. Table (17) documens he VA esimaes of he hree markes during marke crash 1987 and Sepember 11 crash in During he Marke crash he gold prices is negaively affeced 26

28 by he sock reurns on -1 and -3 however here is a posiive relaion beween gold reurns a ime and socks reurns a ime -2. This resul implies ha gold was considered a safe haven agains risky change in corporae bond yields during Sepember 2001 crash. Table (16) VA Esimaes of he relaionship beween gold socks and bonds reurns (Enire Sample Period) The able (16) presens he VA parameer esimaes of esing he dynamic relaionship beween gold f excess reurn Sock f he sock reurns using SP500. bond f yields as esimaed by equaions (1) (2) and (3) during he overall and ension periods ( ). -saisics are in [ ]. All variables are readed as endogenous sae variables All raes are represened in annualized percenage poins.. The firs number in he hird column refers o he -saisic of he es and he second number refers o p value under he null hypohesis ha he gold can no be used as a hedge agains drop in he sock and bond reurns. Panel (a) Hedging: Whole sample f Sock f bond f Panel (b) Safe haven: ension periods gold _ Tension f Sock _ Tension f bond_ Tension k f f (-1) f (-2) f (-3) Sock f (-1) * [ ] [ ] [ ] [ ] [ ] [ ] [ ] *** [ ] [ ] *** [ ] [ ] *** [ ] gold _ Tension f (-1) gold _ Tension f (-2) gold _ Tension f (-3) Sock_ Tension j f (-1) * [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] Sock f (-2) [ ] [ ] *** [ ] Sock_ Tension j f (-2) [ ] *** [ ] *** [ ] Sock f (-3) [ ] [ ] [ ] Sock_ Tension j f (-3) [ ] [ ] [ ] bond (-1) bond f (-2) f bond f (-3) * [ ] [ ] [ ] ** [ ] [ ] [ ] *** [ ] [ ] [ ] bond_ Tension k f (-1) bond_ Tension k f (-2) bond_ Tension k f (-3) [ ] [ ] [ ] *** [ ] * [ ] [ ] *** [ ] [ ] [ ] 27

29 Table (17): VA Esimaes - The Black Monday crash and Sepember 11 crash The able (5) presens he VA parameer esimaes of esing he dynamic relaionship beween gold excess reurn f and he sock reurns ( Sock f and bond f yields as esimaed by equaions (1) (2) and (3) during he crashes periods. -saisics are in [ ]. All variables are readed as endogenous sae variables All raes are represened in annualized percenage poins.. The firs number in he hird column refers o he -saisic of he es and he second number refers o p value under he null hypohesis ha he gold canno be used as a hedge agains drop in he sock and bond reurns. Marke crash f Sock f bond f Sepember 11 crash f Sock f bond f f (-1) * [ ] [ ] [ ] f (-1) *** [ ] [ ] [ ] f (-2) [ ] [ ] [ ] f (-2) [ ] [ ] [ ] f (-3) [ ] [ ] [ ] f (-3) * [ ] [ ] [ ] Sock f (-1) [ ] [ ] [ ] Sock f (-1) [ ] [ ] [ ] Sock f (-2) [ ] [ ] [ ] Sock f (-2) [ ] [ ] ** [ ] Sock f (-3) [ ] [ ] [ ] Sock f (-3) [ ] [ ] [ ] bond (-1) f [ ] [ ] [ ] bond (-1) * f [ ] [ ] ** [ ] bond (-2) f [ ] *** [ ] [ ] bond (-2) *** f [ ] [ ] [ ] bond (-3) *** f [ ] *** [ ] [ ] bond (-3) *** f [ ] [ ] [ ] Table (18) documens he VA esimaes of he hree markes during Asian crisis and recen global financial crisis. During he Asian crisis he gold prices are negaively affeced by he sock reurns and bonds reurns on -1-2 and -3. Overall here is no significan relaion beween gold and sock and bonds reurns during crisis and consequenly he gold can be considered as a safe haven even afer aking ino consideraion he dynamic effec beween he hree markes and aking he lagged reurns effec ino consideraion. 28

30 Table (18) VA Esimaes he Asian Crisis and he Subprime Crisis The able (5) presens he VA parameer esimaes of esing he dynamic relaionship beween gold excess reurn f and he sock reurns Sock f and bond yields bond f as esimaed by equaions (1) (2) and (3) during he crisis periods. -saisics are in [ ]. All variables are readed as endogenous sae variables All raes are represened in annualized percenage poins.. The firs number in he hird column refers o he -saisic of he es and he second number refers o p value under he null hypohesis ha he gold canno be used as a hedge agains drop in he sock and bond reurns. Asian Crisis f Sock f Subprime crisis f bond f Sock f bond f f (-1) *** [ ] [ ] [ ] f (-1) [ ] [ ] [ ] f (-2) [ ] ** [ ] [ ] f (-2) [ ] [ ] [ ] f (-3) [ ] [ ] [ ]*** f (-3) * [ ] [ ] [ ] Sock f (-1) ** [ ] [ ] [ ]** Sock f (-1) [ ] ** [ ] [ ] Sock f (-2) [ ] ** [ ] [ ] Sock f (-2) [ ] [ ] [ ] Sock f (-3) [ ] [ ] [ ] Sock f (-3) [ ] [ ] [ ] bond (-1) f [ ] [ ] ** [ ] bond (-1) f [ ] [ ] ** [ ] bond (-2) f [ ] [ ] [ ] bond (-2) f [ ] [ ] [ ] bond (-3) f [ ] [ ] [ ] bond (-3) f [ ] [ ] [ ] IF esuls: Using Impulse response Funcion echnique (IF) figure (3) shows he response of gold reurns o shocks in he sock reurns and corporae bonds yields during Ocober 1987 crash as well as he Sepember aack. The figure shows ha gold prices a ime commove agains he bonds and socks reurns shocks on he day -1. However gold prices a ime commoves wih he bonds and sock prices a ime -2 and -3 and hen for he res of he rading 29

31 monh (20 days) afer he crash he chocks in he socks and bonds do no affec he gold reurns. During Sepember 11 crash he response of gold is relaively higher o chocks in bonds holding socks reurns consan. The resuls show ha gold move agains chocks in bonds reurns. This figure along wih he able (17) proves ha here is a significan negaive relaion beween gold reurns and corporae bonds yields chocks during he Sepember 11 crash. Figure (4) shows he resuls of he impulse response funcion during he 20 days period afer he sar of he Asian crisis and he enire crisis period. I is obvious ha here is negaive or zero relaion beween he gold and he sock marke bu here is a posiive relaion beween gold prices and he sock prices over he Asian crisis period. Figure (5) presens he IF resuls during he subprime crisis and he figure shows ha here is a negaive bu insignifican relaion beween gold prices a ime and all he lagged reurns of he sock and bonds reurns excep he sock reurns a -1 and he bonds reurns a