The Commodity Price and Exchange Rate Dynamics

Transcription

1 Theoreical Economics Leers, 2017, 7, hp:// ISSN Online: ISSN Prin: The Commodiy Price and Exchange Rae Dynamics Liping Zou 1*, Boliang Zheng 2, Xiaoming Li 1 1 School of Economics and Finance, Massey Universiy, Auckland, New Zealand 2 School of Accounancy, Jiangxi Universiy of Finance and Economics, Nanchang, China How o cie his paper: Zou, L.P., Zheng, B.L. and Li, X.M. (2017) The Commodiy Price and Exchange Rae Dynamics. Theoreical Economics Leers, 7, hps://doi.org/ /el Received: Augus 23, 2017 Acceped: Ocober 17, 2017 Published: Ocober 20, 2017 Copyrigh 2017 by auhors and Scienific Research Publishing Inc. This work is licensed under he Creaive Commons Aribuion Inernaional License (CC BY 4.0). hp://creaivecommons.org/licenses/by/4.0/ Open Access Absrac This paper invesigaes he dynamic relaionship beween he commodiy price and he exchange rae in Ausralia and New Zealand. We focus on Ausralia and New Zealand. No only do heir primary commodiies accoun for significan shares of heir expors, bu also heir currencies share some disincive characerisics ha are unique from oher commodiy currencies. Using counry-specific commodiy price indices, we examine he relaionship beween he deparure of currency value from is fair value and fundamenal macroeconomic variables. Evidence of a srong and robus relaionship beween he exchange rae and he commodiy price has been found. Resuls indicae ha he commodiy price can be used o improve he forecas abiliy of he fuure exchange rae. Our commodiy-price-augmened exchange rae forecasing model consisenly ouperforms he random-walk model, for boh in-sample and ou-of-sample forecasing. These resuls shed some exra lighs on policymaking for counries ha rely on primary commodiy producion, and aemp o move owards floaing exchange rae regimes as par of heir global marke liberalizaion process. Keywords Exchange Rae, Commodiy Price, Forecasing 1. Inroducion In pas few decades, many aemps o invesigae he relaion beween fundamenal macroeconomic variables and exchange raes have been proven o be failure, no o menion numerous unsuccessful endeavours ha economiss have made on building various ypes of exchange rae forecasing models. In early 1970s, afer he pos-war Breon Woods sysem of fixed exchange raes col- DOI: /el Oc. 20, Theoreical Economics Leers

2 lapsed, a large number of indusrialized economies shifed o floaing exchange rae regimes. Thus, los of ineress have been pu on foreign exchange markes sudies. Majoriy of hese work focused on he developmen of macroeconomic variables based empirical models when forecasing fuure exchange raes. Meese and Rogoff [1] [2] concluded ha here are no exchange rae forecasing models ha could ouperform a simple naïve model, i.e. a random walk model, for shor o medium ime horizons. Subsequen aemps have been underaken o develop a suiable model o forecas exchange rae, bu many failed o do so. Therefore, a simple random walk model has become a benchmark when evaluaing exchange rae forecasing performance. Recen lieraures on exchange rae deerminaion and forecasing are in line wih he proposiions found decades ago. Many empirical resuls conclude ha he exchange rae follows a random walk process, and changes in exchange rae are unpredicable, and currencies for high-inflaion counries end o depreciae in long erm wih he magniude being approximaely he inflaion differenial. Movemens of he acual exchange rae appear o be someimes oversho and hen followed by a smooh adjusmen o he equilibrium [3] [4] [5] [6]. There has been a srand of research invesigaing counries wih differing exchange rae regimes and economic srucures. In paricular, counries wih Commodiy Currencies have been given more aenions, and evidences have been found ha here exiss a long-run relaionship beween he real exchange rae and he real commodiy price for commodiy-exporing counries 1 [7], including Ausralia, Canada, Chile, New Zealand 2, and Souh Africa. Similar resuls are confirmed by Chen and Rogff [8] for Ausralia, Canada and New Zealand. In addiion, Chen [10] found ha for hree major OECD (The Organizaion for Economic Cooperaion and Developmen) primary commodiy producers (Ausralia, Canada, and New Zealand), nominal exchange raes exhibi a robus response o movemens in prices of heir corresponding commodiy prices. Chen and Rogff [11] hen subsequenly idenified wo counries from he OECD regime Ausralia and New Zealand, as he excepions o he rule when considering exchange rae deerminaion and forecasing. They confirm and exend previous findings ha for major commodiy exporers wih marke-based exchange raes, he price of heir primary commodiy expors is an imporan and robus deerminan for he real exchange raes. However, he commodiy price iself ells lile abou subsequen exchange rae movemens. Bu when combining various macroeconomic fundamenals wih he commodiy price, resuls sugges ha i can help o predic quarerly exchange rae changes, however, no single specificaion emerges as he clear winner across boh counries and ime period. Oher lieraures for selecive developed and developing naions looking a he same issue include: Amano and van Norden [12], Gruen and Wilkinson 1 For counries ha are heavily relied on commodiy producs o gain expor earnings, he price movemens in he world commodiy markes would hen be he changes in he relaive demands for heir corresponding currencies. 2 For New Zealand in paricular, i refers o Luo and Planier [9], which focuses on he persisence of NZ dollar misalignmens relaive o PPP. DOI: /el Theoreical Economics Leers

3 [13], De Gregorio and Wolf [14], Chinn and Johnson [15], and Moniel [16]. However, none was able o offer conclusive evidence o answer why hese commodiy currencies are unique and may require furher invesigaion. Therefore, our sudy serves o answer hese quesions. Ausralia and New Zealand are wo open economies in he OECD regime. They boh share a number of disincive feaures in economic srucures and policy seings. These special feaures have made he wo naion s currencies demonsrae a remarkable srong commodiy currency phenomenon and o sand ou among oher commodiy currencies 3. Ausralia is an expor-oriened economy, whose expors in primary commodiy producs ake a significan porion of heir gross domesic producion. Characerized by mainly mineral (or hard ) commodiy exporing, he primary commodiy producs expored by Ausralia include iron ore, meallurgical coal, hermal coal, gold and various oher meal producs. These non-energy producs, along wih oher commodiies accoun for more han 60% of Ausralian oal expors. New Zealand, in conras, is expors are heavily dependen on agriculural producs such as dairy producs, wools, meas and oher sof commodiies. Commodiy producs exporing conribue abou 67% of is oal expors in he lae 80s. Primary commodiy producs sill ake abou 50% of New Zealand s oal expors oday. Moreover, New Zealand is a well-known key player of is commodiy producs in he global marke, in spie of is relaively small economy. I supplies nearly 50% of he oal world expors of lamb and muon where only a fracion of less han 20% of is mea producion is consumed domesically. Given boh counries are heavily dependen on commodiy producs o gain expor earnings, he price movemen in world commodiy marke would hen have an impac on he relaive demand for he corresponding currency. Neiher of he wo counries is big enough o influence he world marke, hey boh also have adoped a sufficienly long period of a floaing exchange rae regime under he inflaion argeing sysem 4, cenral banks normally have very limied conrols and inervenions over exchange rae movemens. Commodiy price flucuaions can hen essenially represen a source of exogenous shocks o heir erms-of-rade, which evenually channel up and rigger he exchange rae responses. Therefore, he inroducion of he counry-specific commodiy prices indices in Ausralia and New Zealand s commodiy exporing in early 1980s, has offered an opporuniy o idenify and measure exchange rae flucuaions using hese indices. Moivaed by Chen and Rogff [8] [11] sudies, we ry o invesigae if movemens in commodiy prices can explain he flucuaion in exchange raes. We develop an exchange rae forecasing model by puing he commodiy price ino he srucural forecasing model and o examine if his improves he accuracy of 3 Cashin, Cespedes, and Sahay [7] have classified 58 counries as he commodiy economies ha have a significan commodiy dependency. In addiion o Ausralia and New Zealand, hese counries include Finland, Iceland, Norway, and numerous oher developing counries. 4 Ausralia and New Zealand abandoned heir exchange rae pegs in 1983 and 1985, respecively, as par of he economic reform effors o revialize heir domesic economies. Moreover, around 1990, hey also adoped some varian of inflaion-argeing moneary policy [8]. DOI: /el Theoreical Economics Leers

4 exchange rae forecasing. A hree-sep process is followed in his paper: firsly, we es if he counry-specific commodiy price index has explanaory power when modelling exchange raes movemens in Ausralia and New Zealand. Our resuls indicaed ha he counry-specific commodiy price index does have some explanaory power and our resuls are robus. We herefore conclude ha he commodiy price index should be considered as one of he deerminans when forecasing exchange raes. Secondly, we use he cross-rae beween Ausralian Dollar and New Zealand Dollar o examine he relaionship beween he cross rae and he commodiy price. Resuls indicae ha he commodiy price offer an excepional explanaory power when forecasing he cross-exchange rae movemens. Thirdly, once including commodiy prices index ino he forecasing model, he forecasing performance of he modified model can be evaluaed boh in-sample and ou-of-sample agains a simple random walk model. Our resuls sugges a remarkable in-sample forecasing performance. For ou-of-sample forecass, he modified exchange forecasing model is also able o provide more consisen forecasing performance han he benchmark random walk model. Therefore, our sudy conribues o he exising lieraures ha he commodiy price is an imporan deerminan of he curren exchange rae and he fuure exchange rae movemens in Ausralia and New Zealand. 2. Commodiy Price Indices in Ausralia and New Zealand 5 Figure 1 and Figure 2 below presen a graphical view of he relaionship beween Figure 1. Relaionship beween he Ausralian Commodiy Price and he Value of Ausrian Dollar. Figure 1 illusraes he relaionship beween he Ausralian commodiy price and he value of Ausralian Dollar from 1986 o 2010, he base level is se a 100 as of January See Appendix A for more informaion abou he commodiy indices in Ausralia and New Zealand. DOI: /el Theoreical Economics Leers

5 Figure 2. Relaionship beween he NZ Commodiy Price and he Value of NZ Dollar. Figure 2 illusraes he relaionship beween he NZ commodiy price and he value of New Zealand Dollar from 1986 o 2010, he base level is se a 100 as of January Ausralian and New Zealand commodiy price index and heir corresponding quoed exchange rae agains US dollar from 1986 o 2010, respecively. The commodiy index is he monhly ime series and is base value is se a 100 as of 1 s January In Figure 1, we observe ha he Ausralian commodiy price and he AUD/USD exchange rae appear o have an inverse relaionship. This suggess ha when he commodiy price is moving up, he AUD/USD exchange rae is going down. Thus, Ausralian dollar is appreciaing when he Ausralian commodiy price goes up. This implies ha he value of he Ausralian dollar is moving in line wih prices of is commodiy exporing. Similar paern is observed from Figure 2 for New Zealand. The New Zealand commodiy price index also appears o have an inverse relaionship wih he value of is currency, especially afer lae 1990s 6. Figure 3 below illusraes he relaionship beween Ausralian and New Zealand commodiy price indices. The naure of he underlying commodiies which wo indices represen is fundamenally differen. However, he commodiy prices in Ausralia and New Zealand are having a high level of co-movemens from mid 1980s o early 2000s. From lae 1990s o 2004, he price of New Zealand sof commodiies enjoyed a relaively higher growh rae compared o Ausralian commodiy prices. However, wih he increasing demand for meal producs from developing counries (China in paricular) and global economic expansions, Ausralia has enjoyed he rapid growh in heir commodiy prices since Previous sudies have found ha some economics models, i.e. PPP, UIP, can only provide valid informaion when forecasing exchange raes in he long run. Resuls from hese models usually are no beer han a simple random walk 6 Please see Cashin, Cespedes and Sahay [7] for more deails. DOI: /el Theoreical Economics Leers

6 AUS & NZ CP INDICES 1/01/1986 1/03/1987 1/05/1988 1/07/1989 1/09/1990 1/11/1991 1/01/1993 1/03/1994 1/05/1995 1/07/1996 1/09/1997 1/11/1998 1/01/2000 1/03/2001 1/05/2002 1/07/2003 1/09/2004 1/11/2005 1/01/2007 1/03/2008 1/05/2009 NZ CP AUS CP Figure 3. Relaionship beween NZ and Ausralian Commodiy Prices. This figure illusraes he relaionship beween he commodiy price in Ausralia and New Zealand over ime, he base level is se a 100 as of January 1986 for boh counries. model. Since he erm Commodiy Currency has been inroduced and brough o he aenion of inernaional finance, researchers have found some promising evidences ha commodiy price may play some roles in deermining and forecasing exchange raes 7. A recen sudy by Luo and Planier [9] has esimaed he half-life of NZD/AUD cross raes deviaion is abou 0.6 years (he ime i akes for acual exchange rae o reurn o is heoreical fundamenal value, in his case is PPP value). Whereas a sudy conduced by Chen and Rogoff [8] repors a more han 1.6 years half-life for he NZD/USD pair and an average of 3 o 5 years repored for various pairs of he commodiy currencies in simple PPP-based regression models. If a random walk is wha he exchange rae movemen follows, hen effecs of innovaions on he exchange rae are highly persisen and he ime series can flucuae wihou bounds [7]. Figure 4 below presens he exchange rae behaviour of NZD/AUD cross rae from 1991 o The graph indicaes ha he acual exchange rae moves wihin he 10% bounds of is Fair value 8 for majoriy of he sample period. Moreover, here is a clear endency of he acual crossexchange raes o reurn o heir long-erm Fair Value over ime. Since he Fair Value is usually deermined by fundamenal macroeconomic variables, i is herefore reasonable o believe ha NZD/AUD cross raes may be prediced if shocks are idenified and included ino an exchange rae forecasing model wih possible srucural breaks. In Figure 4, he NZD/AUD long-erm Fair Value (or he long-erm Equilibrium Value ) is wihin he 0.8 and 0.85 NZ dollar per Ausralia dollar range in he pas 20 years. The acual exchange raes, on he oher hand, flucuae from as 7 For more comprehensive background informaion and empirical evidences, see Chen [10], Chen and Rogoff [11], Chen, Rogoff and Rossi [17]. 8 The fair value is defined as he exchange rae ha would be prediced according o PPP and UIP, as illusraed by Equaion (11). DOI: /el Theoreical Economics Leers

7 Figure 4. The NZD/AUD Exchange Rae and he Corresponding Fair Value. This figure illusraes he NZD/AUD cross exchange rae over ime. I also presens he Fair Value of he cross rae deermined by fundamenal macroeconomic variables. The horizonal axis represens he value of New Zealand dollar per Ausralian dollar. high as 0.7 around 1992 o as low as 0.95 in 1995 and In addiion, he fair value appears o be relaively consan overime and failed o pick up any major break of acual exchange rae movemens. The lacking correlaion beween he nominal exchange rae and is underlying long-erm equilibrium value from exising lieraures may be due o he fac ha many aemped o model movemens of nominal exchange raes using long-erm equilibrium value as he explanaory variable. Therefore, in our model, we propose o use he deparure from currency s long-erm equilibrium value (he difference beween he nominal exchange rae and is long-erm equilibrium value) raher han he long-erm equilibrium value iself. This makes our mehodology superior o previous sudies. 3. Daa Descripions The exchange rae and he commodiy price index are monhly ime series for he period from 1986 o Daa are obained from hree sources: Federal Reserve of he U.S., DaaSream daabase and he Reserve Bank of Ausralia. The Ausralian and New Zealand direc exchange raes agains he US dollar are monhly mid-rae and obained from he Federal Reserve Bank of he Unied Saes. The cross exchange raes beween AUD and NZD are fixed a 4:00 p.m. (6:00 GMT) and loaded a approximaely 4:30 p.m. (6:30 GMT) for daily records. The monhly series is he mid-poin deermined by he Reserve Bank of Ausralia on he basis of quoaions in he Ausralian Foreign Exchange marke (Source: Reserve Bank of Ausralia). The commodiy price indices used in his sudy are he counry-specific indi- DOI: /el Theoreical Economics Leers

8 ces of commodiy expor prices. These indices are consruced in a way o reflec he specific characerisics of he respecive counry s commodiy rading wih he res of he world. The CBA NZ Commodiy Price Index and The RBA Commodiy Index SDR are he wo commodiy price indices for New Zealand and Ausralia. They were developed by he Commonwealh Bank of Ausralia (CBA) and The Reserve Bank of Ausralia (RBA) in 1980s. For he US, we use S & P GSCI Non-Energy Spo Price Index 9. There are oher indices poenially can be used as well. These include, he price of individual primary commodiies, erms of rade indices and aggregae (non-counry-specific) indices of commodiy-price. The counry-specific commodiy expors price index is used due o he following reasons. Firsly, neiher New Zealand nor Ausralia has he expor price of a single commodiy can well mirror he movemens of overall commodiy-expor prices. Secondly, erms-of-rade indices could be affeced no only by he composiion of he counry s expors bu also he composiion of he counry s GDP. This is because erms-of-rade indices are ypically calculaed using expors and uni values. Thirdly, prices of individual commodiies do no end o move ogeher on global-commodiy-markes; he movemens in aggregae commodiy-price indices are likely o be a poor proxy of movemens compared o he counry-specific commodiy expor prices [7]. 4. Theoreical Framework and Mehodology Purchasing Power Pariy (PPP) saes a relaionship beween he nominal exchange rae and he inflaion (price level) differences beween wo counries. Such a relaionship can be expressed in a basic form as he following: S = P P + ε (1) counry A counry B where S is he nominal exchange rae; P is he price level, boh in logarihm forms. Equaion (1) simply defines ha he change of he exchange rae can be approximaed by he difference of he price levels beween wo counies and he purchasing power pariy holds o a cerain degree of error. The Uncovered Ineres Pariy (UIP) condiion gives an approximaion ha: counry A counry B ( 1 ) ( ) E S + = S + i i (2) where i is he nominal ineres rae; E represens he expeced spo exchange rae a ime. Noes his seing is only valid under he risk-neural assumpion. If we relax his assumpion and we ge Equaion (3): ( ) counry A counry B E S+ 1 = S + i i + RP (3) where RP is he risk premium on counry A s ineres bearing asses over counry B s asses. As expecaions are no necessarily raional, we herefore rearrange 9 As in Chen and Rogoff [8], we focus only on non-energy commodiy due o is underlying economics complexiy. In addiion, for non-economic-relaed causes, such as inernaional securiy concerns, ofen conribue o boh global energy price and currency flucuaions, and are likely o complicae inerpreaions. This may poenially explain in par why higher energy prices, a imes, appear o lead o a depreciaion of he currency relaive o he US dollar. DOI: /el Theoreical Economics Leers

9 Equaion (3) as following: counry B counry A ( + 1 ) ( ) S = E S + i i RP (4) Equaion (4) differs from Equaion (2) as i akes ino consideraion of he risk premium ha he invesor seeks in order o be willing o hold foreign asses. Equaion (4) can be carried forward for he infinie fuure and can hen be presened as: he curren exchange rae equals he curren ineres rae differenials subraced from he curren risk premium and plus he fuure expeced exchange rae. For example, he erm S can be furher expressed as following: 2 counry B counry A ( ) ( ) S = i i RP + E S (5) S + hen can be furher expressed as: counry B counry A ( ) ( ) S = i i RP + E S (6) We herefore derive he following equaion: ( ) ( ) counry B counry A counry B counry A = + + k + k + k + k + K = 1 S i i RP E i i RP E S (7) The erm S above can be represened by he PPP condiion as Equaion (1) and prevail in he infinie fuure under he assumpion ha here was no ineres rae differenial beween wo counies. In his case S can be expressed as: counry A counry B S P P ε = + (8) Furhermore, we can hen express he expeced fuure PPP exchange rae as he sum of he curren (oday s a ime ) PPP exchange rae plus he sum of fuure expeced inflaion rae differenials and any expeced changes of ε : ( ε ) counry A counry B counry B counry A = + ε+ + k + k + k + + k K = 1 S P P E P P (9) If we subsiue Equaion (9) ino Equaion (7) and rearrange he equaion, we can hen derive he curren exchange rae S : counry B counry A counry A counry B S = i i + P P RP + ε ( ) ( ) ( ) counry B counry A + E+ k( i+ k i+ k RP+ k) K = 1 K = 1 counry A counry B ( ε ) + E P P + + k + k + k + k Equaion (10) above represens a dynamic relaionship beween wo counries exchange raes, ineres raes and price levels (inflaion). The spo exchange rae is dependen on a number of observable facors as well as various unobservable facors. We define observable facors (ineres raes and price levels in Equaion (10)) as he Fair Value described in Equaion (11) below: ( i counry B i counry A ) ( P counry A P counry B ) ( RP ε ) (10) Fair Value = + + (11) All daa are monhly and in logarihm forms excep he average risk premium. DOI: /el Theoreical Economics Leers

10 i is he monhly average one-year inerbank swap raes 10 from Jan 1986 o Mar P is he logarihm form of monhly CPI over he same period, RP is he average risk premium beween he wo counries, in his case, he difference in he ineres raes. The erm ε is serving as a normalizaion facor. We calculae he erm Deparure from Fair Value by subracing Equaion (11) from Equaion (10) and arrives a he following reduced form: = ( RP + ε) ( RP + ε) counry B counry A counrya counryb ( ) + ( + ε ) Deparure from Fair Value E i i RP E P P + k + k + k + k + k + k + k + k K= 1 K= 1 Define Error as he deparure of he acual exchange rae from is long erm fair value a ime, derive from Equaion (12), we hen use he following regression specificaion o examine he relaionship beween Error and he following independen variables: ( ) ( ) (. ) Error = α + β Error + β Error + β Ral GDP counry A counry B ( Comm. Pr ) ( Comm. Pr ) + β + β + ε Daa in Equaion (13) are monhly daa and in logarihm forms excep he relaive GDP which is he logarihm difference beween wo underlying counries GDP growh raes and he daa is quarerly. α is he consan erm. β ( Error 1 ) is he momenum erm, where if he exchange rae appreciaes in one monh, i is more likely o coninues he appreciaion in he subsequen monh, β ( Error 1 ) represens he reversion-o-fair-value erm, where i capures he endency of currency value reverse back owards is long erm fair value over ime 11. β ( Ral. GDP 3 ) capures he effecs of he difference in relaive GDP growh raes beween wo underlying counries. Noes i is lagged 3 monhs as he GDP daa is quarerly oher han monhly. β ( Comm. Pr 1 ) capures he change of commodiy prices. In order o esimae significan shocks ha cause he Error (he deparure from he fair value), we ake a reduced-form approach by adoping a general-o-specific search among poenial variables ha may have an impac on he currency value. We invesigae a few macroeconomic variables ha are relevan and significan boh economically and saisically in deermining he nominal value of NZD/AUD cross exchange rae. The specificaion for approximaing NZD/AUD cross exchange rae is marginally differen from he specificaion for NZD/USD and AUD/USD exchange raes esimaions. This is due o differences in economic srucures, policy marking and governmen regulaions beween Ausralia, New Zealand and he U.S. Therefore, impacs from shocks on under- 10 The selecion of represenaive ineres rae has aking ino consideraions of he fac ha normally he shor-erm ineres rae variaion is ofen anicipaed by he marke, and any fully anicipaed moneary policy on he ineres rae changes should no affec he value of exchange rae. Moreover, we have realized ha a longer erm ineres rae maybe even beer, afer considering oher facors, i.e. daa availabiliy and liquidiy (paricularly in NZ marke), one-year inerbank swap rae was chosen. 11 The momenum erm and he reversion-o-fair-value erms in equaion 13 represen overshooing effecs in exchange rae dynamics. (12) (13) DOI: /el Theoreical Economics Leers

11 lying exchange raes may differ subsanially. Specifically, we examine how do flucuaions in counry-specific commodiy prices ranslae ino movemens in he exchange rae? We give special emphasis on he deerminaion of NZD/AUD cross-exchange-rae and he imporance of is corresponding commodiy prices. To illusrae his, we modify equaion 13 as follows: NZ & AU 1( 1) 2 ( 1) 3 (. 3 ) NZ ( NZ. ComPr ) ( AU. ComPr ) ( Migraion ) Error = α + β Error + β Error + β Ral GDP (14) + β + β + β + ε The above specificaion follows an error-correcion framework; i is conduced in a way ha ensures he consisency of he robusness relaionship of dependen variables wih he exchange rae. The equaion, herefore, esablished o esimae coefficiens ha are no sensiive o minor variaions for he chosen sample period. 5. Resuls and Discussion 5.1. Regression Resuls Table 1 below repors parameer coefficiens and corresponding -saisics from Equaion (14), hree models have been esimaed using NZD/AUD, NZD/USD and AUD/USD exchange raes as dependen variables. Resuls are remarkably consisen across all currency pairs. Firs, he effecs of momenum and commodiy prices movemens are uniformly srong and consisen in all hree specific models. Second, majoriy of coefficiens from respecive regression models are in he heoreically correc sign and are saisically significan, wih only wo excepions which is he US commodiy price and he GDP change beween US and New Zealand. Third, here is no evidence o suppor he significance of US commodiy prices ha could be relaed o he movemens of eiher he NZD/USD or he AUD/USD exchange rae dynamics. Therefore, hese resuls confirm he hypohesis ha Ausralian and New Zealand commodiy prices should be applied o measure he exchange rae movemen. In addiion, resuls indicae ha when puing macroeconomic variables ino he regression model, he NZD/AUD cross rae model does provide superior resuls han ohers. In Table 1 he momenum coefficiens are posiive and ranging from 0.94 o 0.98 across all models, indicaing a srong momenum effec in exchange rae movemens. The New Zealand commodiy price coefficien in NZD/AUD cross rae model is negaive and saisically significan a he 1% level, indicaing a negaive relaionship beween corresponding commodiy price dynamics and he currency deviaion from is long-erm fair value. This negaive relaionship suggess ha an increase in NZ commodiy price may decrease he error, herefore resul an appreciaion of New Zealand dollar. The coefficien for Ausralian commodiy price is posiive and saisically significan a he 1% level, indicaing ha here exiss a posiive relaionship beween he Ausralian commodiy price and he error, wih increasing of Ausralian commodiy price, he error increases, hus, Ausralian dollar appreciae in value. DOI: /el Theoreical Economics Leers

12 Table 1. Exchange Raes and Commodiy Prices Dynamics: General o Specific Models. This able repors coefficiens and -saisics from Equaion (14). We conduc boh he general model (Columns 2, 4 and 6) and he specific model (columns 3, 5 and 7). We eliminae variables ha are insignifican from he general model unil all remaining variables are significan in he specific model. The dependen variable is he acual exchange raes beween NZD/AUD, AUD/USD and NZD/USD. The corresponding -saisics are repored in parenheses. All variables are in logarihm forms. We have included NZ Migraion as an explanaory variable because of he general-o-specific search indicaed ha i is saisically significan. The relaive GDP is lagged for hree periods as GDP daa is quarerly. *, ** and *** indicae saisical significan a he 10%, 5%, and 1% level, respecively. Variables NZD/AUD AUD/USD NZD/USD Momenum *** *** *** ** *** *** ( ) ( ) ( ) ( ) ( ) Mean Reversion ** ** Ral.GDP.NZ&AU ** ** ( 2.437) ( 2.077) ( 0.957) ( 0.794) ( 2.187) ( 2.178) Ral.GDP.AU&US * ** ( 1.955) ( 1.976) Ral.GDP.NZ&US ( 1.304) NZ Commodiy Prices *** *** *** *** ( 3.560) ( 3.748) ( 3.923) (-4.583) AU Commodiy Prices *** *** *** *** ( 3.771) ( 3.526) ( 5.323) ( 5.995) US Commodiy Prices ( ) ( 0.078) NZ Migraion *** ( 2.727) Adj. R-squared Sample period 1986M1-2010M1 1986M1-2010M1 1986M1-2010M1 1986M1-2010M1 1986M1-2010M1 1986M1-2010M1 No. obs Durbin-Wason The coefficiens for mean reversion and he relaive GDP growh in Table 1 are saisically significan a he 5% level for he NZD/AUD cross rae model (for boh general and specific specificaions). However, in NZD/USD models, here are no saisical evidences which suppor he exisence of eiher he mean reversion effec or he effec of relaive GDP growh. For AUD/USD models, he effec of relaive GDP growh beween Ausralia and he US is saisically significan a he 10% level in he general model, and a he 5% significan level in he specific model. We also pu he Ne Migraion variable o measure he effec of New Zealand Ne Migraion o Ausralia. The Ne Migraion coefficien is saisically significan a he 1% level in he NZD/AUD cross rae model. One DOI: /el Theoreical Economics Leers

13 possible explanaion is ha he flow of migran funds may affec supply and demand of boh currencies, hence he ne effec of migraion is expeced o play an acive role in deermining he cross-exchange rae dynamic 12. Munroe [18] found ha migraion flows (beween New Zealand and Ausralia) have an ousized effec on he housing marke, which causes large flows of funds, and affec he exchange rae indirecly via ineres raes. Despie he fac ha New Zealand ne migraion owards Ausralia is a saisically significan variable in he model, we excluded i from our furher analysis as i is deemed o be economically insignifican because he coefficien is Regression Models wih Srucural Breaks We now urn o examine possible srucural breaks during our sample period. Two possible srucural breaks are chosen o reflec wo mos recen major financial crises; early 2001 and lae We simply give he ime-dependen dummy variable a value of one for he period before he break dae and a value of zero for he period afer he break dae. Resuls in Table 2 and Table 3 sugges ha coefficiens of ime dependen variable provide relaively lile evidence of any srucural break. We herefore conclude ha we may have some possible small parameer shifs for he sample period, and he expeced sign and magniude of coefficiens are sable over he enire sample period wihou sysemaic srucural breaks. We now urn o explore in-sample and ou-of-sample forecasing performances for he NZD/AUD cross rae model In-Sample and Ou-of-Sample Forecasing 13 In his secion, he following regression equaion is going o be used o perform he in-sample forecasing for NZD/AUD cross rae: NZ & AU ( ) ( ) (. ) Error = α + β Error + β Error + β Ral GDP + k 1 + k k k 3 ( NZ. ComPr ) ( AU. ComPr ) + β + β + ε 4 + k k 1 + k Figure 5 below presens he performance of he commodiy-price-augmened regression model we developed for forecasing NZD/AUD cross exchange rae movemens (Equaion 15). I provides us a visual observaion for he in-sample NZD/AUD cross rae forecasing es 14. We include resuls from a random walk model and he acual NZD/AUD cross marke exchange rae. All parameers are esimaed using daa from January 1986 o January The movemen of he blue-line (represening he in-sample one-monh-ahead forecased NZD/AUD cross exchange raes) is remarkably close o he red line (he acual NZD/AUD marke cross exchange rae). This indicaes an excellen in-sample forecasing of our NZD/AUD cross rae model. This model no only picks up every major 12 According o Saisic NZ, he bigges desinaion for New Zealand emigrans is by far is bes neighbor Ausralia. Moreover, Ausralia is he second in line among he counries ha lised as he bigges source of New Zealand s immigrans wih accoun for 20% of all migran arrivals. 13 See Appendix B for an illusraion of he ou-of-sample forecasing in deails. 14 In-sample forecasing period is from he 1 s monh of year 1986 o 1 s monh of 2010, of which 219 observaions are included. DOI: /el Theoreical Economics Leers (15)

14 rends of he exchange rae movemen in pas wo decades, bu also accuraely measures exchange rae urning poin, represened by peaks and roughs in Figure 5. Resuls from he Random Walk (RW) model (represened by he green line) show no signs of any predicabiliy of acual NZD/AUD cross raes over he enire sample period. From early 1987 o he beginning of 1991, he Random Walk model compleely mis-specified he underlying exchange rae movemen. The Random Walk model also underesimaes he volailiy of NZD/AUD cross rae over ime. I offers no predicion of he changes of direcion on he currency movemen afer However, despie he superioriy of our NZD/AUD cross rae in-sample forecasing model, i is necessary for us o use conemporaneous daa when conducing such esimaes. Therefore, an ou-of-sample forecas performance es is o be conduced. For ou-of-sample forecasing, we firs adop sandard quaniaive procedures involved in forecasing he deparure from he fair value from Equaion (12), where he forecasing errors are defined as: FE + k= S+ k S+ k, where k 1 and S + k represens he k-sep-ahead forecas. We measure forecasing errors for four ime inervals, 1-monh-ahead, Table 2. OLS wih Possible Srucural Break. This able provides he regression resuls for possible srucural breaks wih he break poin se a January This break poin is seleced as a he firs monh of year 2001 o reflec he mos recen major global financial crisis. *, **, and *** indicae significance a he 10%, 5%, and 1% levels, respecively. NZD/AUD NZD/USD AUD/USD Coefficien Sd. Error -Saisic Coefficien Sd. Error -Saisic Coefficien Sd. Error -Saisic Momenum *** *** *** D*Momenum Mean Reversion ** D*Mean Reversion Ral.GDP.NZ&AU ** D*Ral.GDP.NZ&AU ** Ral.GDP.AU&US ** D*Ral.GDP.AU&US Ral.GDP.NZ&US D*Ral.GDP.NZ&US NZ Commodiy Prices *** *** D*NZ Commodiy Prices AU Commodiy Prices *** *** D*AU Commodiy Prices US Commodiy Prices D*US Commodiy Prices Adj. R Square No. obs DOI: /el Theoreical Economics Leers

15 Table 3. OLS wih Possible Srucural Break. This able provides he regression resuls for possible srucural breaks wih he break poin se a December This break poin is seleced as a he las monh of year 2007 o reflec he mos recen major global financial crisis. *, **, and *** indicae significance a he 10%, 5%, and 1% levels, respecively. NZD/AUD NZD/USD AUD/USD Coefficien Sd. Error -Saisic Coefficien Sd. Error -Saisic Coefficien Sd. Error -Saisic Momenum *** *** *** D*Momenum *** Mean Reversion ** D*Mean Reversion ** *** Ral.GDP.NZ&AU ** D*Ral.GDP.NZ&AU Ral.GDP.AU&US ** D*Ral.GDP.AU&US Ral.GDP.NZ&US D*Ral.GDP.NZ&US NZ Commodiy Prices *** *** D*NZ Commodiy Prices ** AU Commodiy Prices *** *** D*AU Commodiy Prices US Commodiy Prices D*US Commodiy Prices *** Adj. R Square No. obs Figure 5. One-monh-ahead in-sample regression forecas for NZD/AUD. DOI: /el Theoreical Economics Leers

16 3-monh-ahead, 6-monh-ahead and 12-monh-ahead. The esimaion involves he predicion of he long erm Fair Value (FV) of NZD/AUD cross rae, his FV is no consan over ime. We firs need o define he FV before he forecasing error can be derived. To ensure a heoreically sound process of approximaing FVs, we apply four mehods o predic FVs over he esing ime horizon. Firsly, we assume he Fair Value (FV) is known for he ou-of-sample period. Hence, FVs are direcly aken from he in-sample forecasing models which have been done in he previous secion 15. Secondly, FVs are assumed o be unknown for he ou-of-sample period, and are herefore prediced using an AR(1) model, an ARMA(1, 1) model and a random walk model, respecively. To evaluae he forecasing performance, four forecasing measures are applied: Roo Mean Square Error (RMSE), Mean Absolue Error (MAE), Mean Absolue Percenage Error (MAPE) and Theil Inequaliy Coefficien (THEIL). These measuremens offer a se of quaniaive values, which specifically measure how far he forecased fair values are away from he acual observed exchange raes. A lower value indicaes a beer fi, hence, higher accuracy of he forecasing model. The ou-of-sample forecasing involves re-esimaing he NZD/AUD cross rae for each hisorical period, using only daa ha would have been available o us a ha ime. Appendix B provides a deailed illusraion for a 3-monh-ahead ou-of-sample forecasing. Table 4 and Table 5 presen forecasing errors for he ou-of-sample performance. Resuls sugges ha our commodiy price augmened forecasing model remarkably bea he random walk model in every seing under 3- and 6-monh esing periods. For 9- and 12-monh period, resuls also sugges ha forecasing errors are consanly smaller in mos cases. This indicaes ha our NZD/AUD cross exchange rae forecasing model ouperforms he random walk model in majoriy of he esing ime periods. I is necessary o menion ha he Random Walk model has been widely considered as he benchmark for exchange rae forecasing models. Based on exising lieraures, he Random Walk model works superior compared o oher forecasing models. We pick up wo sample saring poins when conducing ou-of-sample forecasing performances: (1) February 1990; and (2) March These saring poins are chosen for he following reasons: New Zealand and Ausralia boh adoped inflaion-argeing moneary policy for period beween 1990 and Therefore, i is heoreically sound for fuure exchange rae forecasing o use he daa se available before and afer he exchange rae regime shif. In addiion o he exchange rae regime shif, he ou-of-sample period also spans a period of sharp exchange rae depreciaion as well as he susained appreciaion, which began in Thus, our exchange rae model is believed o beer faciliae a more comprehensive model evaluaion across he ou-of-sample forecasing period. Conrary o previous findings, our fundamenal macroeconomic variable exchange rae model has a beer forecasing accuracy in shor-ime horizons (3-monh and 6-monh) relaive o 15 The in-sample forecasing period is from he 1 s monh of year 1986 o 1 s monh of 2010, which conains enire sample of 219 observaions. DOI: /el Theoreical Economics Leers

17 longer ime horizons (9-monh and 12-monh). 6. Conclusion The relaionship beween he exchange rae and fundamenal macroeconomic variables has drawn los of aenions o many researchers for many years, ye, no clear consensus has been emerged. In his paper, we focus on a se of selecive currencies which are considered as he Commodiy Currencies, namely he New Zealand and he Ausralian dollars. We incorporae commodiy prices from hese wo counries ino he exchange rae forecasing model. Resuls shed exra ligh on wo main issues. Firsly, commodiy currencies are no he same across counries. Some are more of commodiy currencies han ohers, i.e. more pronounce in New Zealand and Ausralia han in he U.S. This is due o he fac ha Table 4. Ou-of-Sample Forecasing (from Mar. 1993). This able repors Ou-of-Sample forecasing performance compared o he Random Walk model. The sample period sars from March 1993 o May 2002, and forecasing period is from May 2002 o January The evaluaion resuls of RW model are repored in columns 3, 5, 7 and 9 and he buil in models (Equaion (15)) are repored in columns 4, 6, 8 and 10. Roo Mean Square Error (RMSE), Mean Absolue Error (MAE), Mean Absolue Percenage Error (MAPE) and Theil Inequaliy Coefficien (THEIL) offer a se of quaniaive values, which specifically measure how far away he model forecased exchange raes are o he acual exchange raes observed laer. A lower value indicaes a beer fi (higher accuracy) of he esed model. * indicaes forecasing superioriy over RW model. Mehod of FV Horizon RMSE_RW RMSE MAE_RW MAE MAPE_RW MAPE THEIL_RW THEIL In-Sample 3 m * * * * 6 m * * * * 9 m * * * * 12 m * * * * AR(1) 3 m * * * * 6 m * * * * 9 m * * * * 12 m * ARMA(1, 1) 3 m * * * * 6 m * * * * 9 m * * * * 12 m * * * * RW 3 m * * * * 6 m * * * * 9 m * * * * 12 m * * * DOI: /el Theoreical Economics Leers

18 Table 5. Ou-of-Sample Forecasing (from Feb. 1990). This able repors Ou-of-Sample forecasing performance compared o he Random Walk model. The sample period sars from Feb o May 2002, and forecasing period is from May 2002 o January The evaluaion resuls of RW model are repored in columns 3, 5, 7 and 9 and he buil in models (Equaion (15)) are repored in columns 4, 6, 8 and 10. Roo Mean Square Error (RMSE), Mean Absolue Error (MAE), Mean Absolue Percenage Error (MAPE) and Theil Inequaliy Coefficien (THEIL) offer a se of quaniaive values, which specifically measure how far away he model forecased exchange raes are o he acual exchange raes observed laer. A lower value indicaes a beer fi (higher accuracy) of he esed model. * indicaes forecasing superioriy over RW model. Mehod of FV Horizon RMSE_RW RMSE MAE_RW MAE MAPE_RW MAPE THEIL_RW THEIL In-Sample AR(1) ARMA(1, 1) RW 3 m * * * * 6 m * * * * 9 m * * * * 12 m * * * * 3 m * * * * 6 m * * * * 9 m * * * 12 m * * 3 m * * * * 6 m * * * * 9 m * * * 12 m * * 3 m * * * 6 m * * * * 9 m * * * * 12 m * * * Ausralia and New Zealand are commodiy-expor dependen counries. They also share some disincive economic srucures and policies comparing o oher counries. In his regard, consideraions should no be solely given o commodiy expor componens, bu various oher facors as well. Resuls in our sudy confirmed ha here is a srong and robus relaionship beween he exchange rae and commodiy prices in Ausralia and New Zealand. Secondly, evidences have indicaed ha he cross exchange rae beween NZD and AUD can be beer measured and forecased han he direc quoed (agains USD) exchange rae pairs. As illusraed in his paper, when including commodiy prices as independen variables in our model, he NZD/AUD cross rae model no only ouperforms oher wo models in almos every aspec, bu also offers a remarkable forecasing abiliy. The commodiy-price-augmened exchange model (NZD/AUD cross rae model) developed in his paper consisen- DOI: /el Theoreical Economics Leers

19 ly ouperforms a random walk a shor horizons according o four evaluaing mehods. However, here are some evidences sugges ha, in ou-of-sample forecass, he superioriy of such a model does slighly deeriorae as he forecasing horizon increases. The aribues behind our empirical resuls are likely o be he Enhanced Commodiy Currencies Phenomenon, represened by he fac ha boh underlying currencies (NZD and AUD) are commodiy currencies, and hey are boh subjec o he Commodiy Currencies Phenomenon discovered in recen lieraures. In addiion, here is a unique underlying economic relaionship beween hese wo OECD naions: close geographically, free from rade resricions, high co-movemen in financial markes, and being conribuing o our findings. Resuls in his paper have provided furher undersanding of exchange rae dynamics o commodiy prices. In addiion, he superior and poenial exploiable forecas abiliy of commodiy-price-augmened exchange rae forecasing model may provide imporan informaion for naions ha heavily rely on primary commodiy producion, and wish o develop he capial marke liberalizaion by moving owards floaing exchange rae regimes. The model developed in his paper can also benefi porfolio managers in beer modeling NZD/AUD cross raes, hence, making beer sraegic decisions on currency rading and porfolio rebalancing. References [1] Meese, R.A. and Rogoff, K. (1983) Empirical Exchange Rae Models of he Sevenies: Do They Fi ou of Sample? Journal of Inernaional Economics, 14, [2] Meese, R.A. and Rogoff, K. (1983) The Ou-of-Sample Failure of Empirical Exchange Rae Models. In: Frenkel, J.A., Ed., Exchange Raes and Inernaional Macroeconomics, Universiy of Chicago Press, Chicago, [3] Mussa, M. (1979) Empirical Regulariies in he Behavior of Exchange Raes and Theories of he Foreign Exchange Marke. In: Brunner, K. and Melzer, A.H., Eds., Policies for Employmen, Prices, and Exchange Raes, Norh-Holland Pub. Co., New York, [4] Calvo, G. and Rodriguez, C. (1977) A Model of Exchange Rae Deerminaion under Currency Subsiuion and Raional Expecaions. Journal of Poliical Economy, 85, hps://doi.org/ / [5] Sockman, A.C. (1980) A Theory of Exchange Rae Deerminaion. Journal of Poliical Economy, 88, hps://doi.org/ / [6] Lucas, R. (1982) Ineres Raes and Currency Prices in a Two-Counry World. Journal of Moneary Economics, 10, [7] Cashin, P., Cespedes, L. and Sahay, R. (2004) Commodiy Currencies and he Real Exchange Rae. Journal of Developmen Economics, 75, [8] Chen, Y. and Rogoff, K. (2003) Commodiy Currencies. Journal of Inernaional Economics, 60, [9] Luo, C. and Planier, C. (2003) The Persisence of NZ Dollar Misalignmens Relaive o Purchasing Power Pariy (Research Paper Series, Paper ). Auckland Universiy of Technology. DOI: /el Theoreical Economics Leers

20 [10] Chen, Y. (2004) Exchange Raes and Fundamenals: Evidence from Commodiy Economies. Universiy of Washingon, Working Paper. [11] Chen, Y. and Rogoff, K. (2006) Are he Commodiy Currencies an Excepion o he Rule? Universiy of Washingon, Working Paper. [12] Amano, R. and Norden, S. (1995) A Forecasing Equaion for he Canada U.S. Dollar Exchange Rae. The Exchange Rae and he Economy (207-65), Bank of Canada, Oawa. [13] Gruen, D.R. and Wilkinson, J. (1994) Ausralia s Real Exchange Rae Is I Explained by he Terms of Trade or by Real Ineres Differenials? Economic Record, 70, hps://doi.org/ /j b01839.x [14] De Gregorio, J. and Wolf, H.C. (1994) Terms of Trade, Produciviy, and he Real Exchange Rae (Working Papers 94-19). New York Universiy. [15] Chinn, M. and Johnson, L. (1996) Real Exchange Rae Levels, Produciviy and Demand Shocks: Evidence from a Panel of 14 Counries. Working Paper 5709, NBER. [16] Moniel, P.J. (1997) Exchange Rae Policy and Macroeconomic Managemen in ASEAN Counries. In: Hinklin, J., Robinson, D. and Singh, A., Eds., Macroeconomic Issues Facing ASEAN Counries, Inernaional Moneary Fund, Washingon DC, [17] Chen, Y., Rogoff, K. and Rossi, B. (2008) Can Exchange Raes Forecas Commodiy Prices? Working Paper 13901, NBER. [18] Munroe, A. (2004) Wha Drives he New Zealand Dollar. Reserve Bank of New Zealand Bullein, 67 (June). DOI: /el Theoreical Economics Leers