CESSA WP 208-0 Deermnans of frm exchange rae predcons: Emprcal evdence from survey daa of Japanese frms Th-Ngoc Anh NGUYEN Yokohama Naonal Unversy Japan Socey for he Promoon of Scence May 208 Cener for Economc and Socal Sudes n Asa (CESSA) Workng Paper Downloadable from: hp://www.econ.ynu.ac.jp/cessa/publcaon/workngpaper.hml Cener for Economc and Socal Sudes n Asa, Deparmen of Economcs Yokohama Naonal Unversy
Deermnans of frm exchange rae predcons: Emprcal evdence from survey daa of Japanese frms Th-Ngoc Anh NGUYEN * May 208 Absrac Ths leer uses Bank of Japan s Tankan survey daa coverng 0 manufacurng ndusres o nvesgae he characerscs as well as he deermnans of frms exchange rae predcons. Evdence of asymmery s found regardng he way frms adjus her exchange rae predcons n response o dfferen currency regmes. Japanese exporng frms are more lkely o updae her predcon of fuure exchange raes n response o unexpeced yen apprecaon han o unexpeced yen deprecaon. Furhermore, frm exchange rae predcon updang decsons are affeced by oher frm characerscs such as expor dependency and profably. JEL classfcaon: D22, D84, F3, G3 Keywords: Exchange rae pass-hrough (ERPT), prcng-o-marke (PTM), exchange rae expecaon, Japanese expor, predcon adjusmen speed, Tankan daa * Ths sudy s fnancally suppored by he Japan Socey for he Promoon of Scence (JSPS) KAKENHI Gran Number 6J988. Graduae School of Inernaonal Socal Scences, Yokohama Naonal Unversy and Research Fellow of he Japan Socey for he Promoon of Scence (JSPS). Emal: nguyen.hngocanh208@gmal.com 0
. Inroducon Exchange rae expecaons have long been an mporan research opc n nernaonal macroeconomcs; ye, he queson of how marke parcpans predc fuure exchange rae levels remans unresolved. A few sudes only, such as hose by Io (989) and Takag (99), have used survey daa o analyze hs queson emprcally. Accordngly, he lmed daa avalably on expeced fuure exchange raes has prevened furher rgorous examnaon of hs unresolved research queson. Regularly, he Bank of Japan (BOJ) publshes frm predcons of he yen/ US dollar blaeral normal exchange rae when expor by frm sze and by ndusry, whch hey oban by conducng large-scale surveys of Japanese frms. Fgure presens he exchange rae predcon error for boh large-sze and small-sze frms as well as he acual (quarerly averaged) yen/ US dollar blaeral nomnal exchange rae, where he predcon error s defned as he dfferences beween he acual and predced exchange rae. I can be observed ha he magnude of he predcon error s larger wh a longer duraon n shaded yen deprecaon perods han n non-shaded yen apprecaon perods. 2 In addon, n mos perods, he graph of large-sze frms (black x marked lne) s above ha of small-sze frms (red doed lne) mplyng ha predcon errors by large-sze frms are larger (smaller) han hose of small-sze frms durng yen deprecaon (apprecaon) perods. Thus, quesons are rased abou why he degree of predcon errors dffers () beween yen apprecaon and deprecaon perods, and (2) dependng on frm sze. Ths leer ams o use he BOJ s Tankan survey daa o reveal he characersc as well as he deermnans of frm-level exchange rae predcons. The analyss revealed ha, frs, Japanese frms adjus her exchange rae predcon errors more quckly (slowly) n yen apprecaon (deprecaon) perod. Second, he adjusmen speed of exchange rae predcon s posvely correlaed wh a frm s degree of expor dependency n he yen deprecaon perod. Thrd, n conras, he adjusmen speed s found o have negave relaonshp wh a frm s profably performance n he yen deprecaon perod, wh small-sze frms endng o ncrease he adjusmen speed n response o an unancpaed deprecaon of he yen. The leer s organzed as follows. Secon 2 nvesgaes he characerscs of exchange rae predcon and adjusmen speed. The relaonshp beween he adjusmen speed and frm BOJ s Shor-Term Economc Survey of Enerprses n Japan (TANKAN) s publshed n every quarer. As of March 208, for example, he BOJ sen ou quesonnares o 0,020 frms and presens he averaged resuls by hree ypes of frm sze (large, medum-szed, and small frms) and by ndusry (7 manufacurng and 4 non-manufacurng ndusres). 2 In hs leer, he crera of dvdng currency regmes are based on predcon errors of exchange raes,.e., he perod s couned as yen deprecaon perod f he yen deprecaes more han frms expecaon (predcon error >0) and yen apprecaon oherwse. In fac, hs currency regme specfcaon s reasonable when comparng wh he movemen of acual yen/ US dollar exchange rae (blue sold lne) n Fgure.
characerscs s analyzed emprcally n Secon 3. Secon 4 concludes. 2. Adjusmen speed of exchange rae predcon The adjusmen speed ( α ) of exchange rae predcon s calculaed by dvdng he change n frm s predced exchange rae by he exchange rae predcon error n he prevous perod: α = (E s + -E - s ) (s -E - s ), where s s he naural log of blaeral nomnal exchange rae a perod, and E s an expecaon operaor condonal on all avalable nformaon a perod. Alhough frms use all he avalable nformaon o predc he fuure exchange rae, s mpossble o esablsh perfec expecaons. Thus, he hypohess proposed here s ha frms end o deermne he predced exchange rae conservavely when makng an expor plan. The addonal hypohess s ha he magnude of predcon errors depends on frm characerscs such as expor dependency and profably. To es he above hypohess, he adjusmen speed, α, whch measures how quckly frms updae her predced exchange rae by observng curren (realzed) exchange rae, s esmaed. To allow for possble asymmerc predcon of fuure exchange raes, he followng hreshold regresson model s esablshed: E ( s E s ) + ( I ) α2 ( s E s ) v s+ E s = I + + α () where he dummy I akes he value of (yen deprecaon regme) f he predcon error s posve, and akes 0 (yen apprecaon regme) oherwse. Quarerly daa from 997Q2 o 206Q for hree frm szes n 0 manufacurng ndusres are used for he above esmaon. 3 Table presens he resuls of α and α 2 as well as he resuls for he Wald-es for he null hypohess α = α2. An asymmerc paern of adjusmen speeds can be observed among almos all large and medum enerprses and n four ou of 0 ndusres for small enerprses, where he adjusmen speed s hgher n he yen apprecaon regme han n he yen deprecaon regme. One explanaon for regme-dependen adjusmen speed could be ha exporng frms mgh need o revse her expor plan quckly n response o an unexpeced apprecaon of he 3 Followng Nguyen and Sao (207), hs leer consrucs he daase from BOJ s Tankan daa. Alhough daa on 7 manufacurng secors are avalable, as foonoe shows, we fnally chose all manufacurng and nne manufacurng ndusres. See he foonoe of Table for ndusry nformaon. Acual yen/ US dollar blaeral nomnal exchange rae (quarerly average) s s aken from he IFS. 2
yen, especally when hey use foregn currency nvocng n her expors. 4 In conras, durng he yen deprecaon regme, Japanese frms could enjoy unexpeced exchange gans, whch provdes less ncenve o change her expor plan and, hence, o revse her predced exchange rae. 3. Deermnans of adjusmen speed Nex, we nvesgae wha facors deermne he adjusmen speed. The BOJ provdes noable daa of forecased (planned) expor amouns, oal sales, and ordnary profs as obaned from her Tankan surveys. These daa are used o calculae he forecased expor rao as well as he forecased prof rao. 5 To nvesgae he effec of frm sze on adjusmen speed, frm-sze dummes and her neracons wh explanaory varables are ncluded n he followng fxed-effec model esmaon: spd j = c + S + α exp_ rao + β prof _ rao j + j + α S exp_ rao β S prof _ rao j j + γ + η + ε j + j (2) where spdj, exp_ raoj, and prof _ rao j denoe he adjusmen speed, expor rao, and prof rao of frm-sze-level n ndusry j a me, respecvely. S denoes dummes for large or small enerprses. Fxed effecs of ndusry-frm-sze and me are capured by γ j and η, respecvely. Predcon error, er, s used o dsngush currency regmes. The sample perod ranges from 999Q2 o 206Q. A frm s expor rao has wo conflcng effecs on s adjusmen speed. One effec s an exposure effec ha occurs when he expor rao s reaed as a proxy for frms exchange rae exposure. Gven he assumpon of a posve relaonshp beween he adjusmen speed and ERPT, he sgn on α s expeced o be posve. 6 The oher effec s a marke share 4 As shown by Io e al. (208), Japanese frms end o nvoce expors n he desnaon currency (manly n US dollars) when exporng producs o advanced counres. Even for expors o Asan counres, he share of US-dollar-nvoced expors s hgher han he share of yen-nvoced expors. Gven such a large share of foregn currency nvocng n expors, Japanese frms shoulder foregn exchange loss when he yen apprecaes unexpecedly. 5 Expor (prof) rao s calculaed as he rao of frm s expor (ordnary prof) amoun compared wh oal sales amoun. The wo raos correlaon coeffcen s 0.33. 6 Floden e al. (2008) shows ha hgher exchange rae exposure leads o a hgher ERPT level. 3
effec ha arses when he expor rao s consdered as a proxy for frms sales dependency on he nernaonal marke. The fac ha frms depend heavly on overseas sales mgh gve hem an ncenve o manan (boos) her marke share by decreasng (ncreasng) he level of ERPT and hence o slowly (quckly) revse exchange rae predcons when he yen apprecaes (deprecaes). 7 Therefore, aken ogeher, α s expeced o be posve when he yen deprecaes, bu remans ambguous when he yen apprecaes. The prof rao s used o proxy for frms busness performance and s hghly relaed o a frm s mark-up level. Dependng on ha frm s expor compeveness, he relaonshp beween frm profably and adjusmen speed can be eher posve or negave. 8 Columns (4) (6) n Table 2 repors he resuls of regresson (2). 9 The coeffcens of boh he expor and prof raos are found o be sgnfcan n he regme-free as well as yen deprecaon regme, bu no n he yen apprecaon regme. In he yen deprecaon regme, he coeffcen of he expor rao s sgnfcanly posve as expeced, wh % ncrease n he expor rao leadng o a % hgher adjusmen speed. However, frm sze self does no have a sgnfcan effec on adjusmen speed, ndcang a lnear effec of he expor rao on adjusmen speed among all ypes of enerprses. Meanwhle, he prof rao shows a srongly sgnfcan nonlnear relaonshp wh he adjusmen speed n he yen deprecaon regme. For benchmark medum-sze enerprses, profably negavely affecs adjusmen speed, whereas he neracon beween frm sze and profably s sgnfcanly posve for small-sze enerprses, makng he pon esmae 34.004. Thus, he smaller he frm sze, he more lkely s ha frms wh hgh profably wll ncrease her adjusmen speed n he yen deprecaon regme. As dscussed earler, frms wh hgh profs end no o revse her predcons of fuure exchange raes quckly n response o an unexpeced deprecaon of he yen. However, he small-sze frms, ofen regarded as less compeve, are lkely o have a sronger ncenve o updae her predced exchange rae o decrease her expor prce, hereby mprovng her expor prce compeveness durng he yen deprecaon regme. Ths fndng s conssen wh he emprcal resuls of Berman e al. (202) and Nguyen and Sao (207), whch demonsraed an asymmerc relaonshp beween frm busness performance and ERPT. 4. Concludng remarks 7 Krugman (987) and Kneer (993) dscuss possble ERPT asymmery due o marke power. 8 Berman e al. (202), Nguyen and Sao (207) sugges a nonlnear relaonshp beween frm performance (or ndusry compeveness) and ERPT. 9 For reference, he resuls of he model whou frm-sze dummes (model (3)) are repored n column () (3) of Table 2. Only he expor rao coeffcens are found o be sgnfcan n regme-free and yen deprecaon regme. spd c + α exp_ rao + β prof rao + γ + η + ε (3) j = j _ j 4 j j
Usng he BOJ s Tankan daa, we have nvesgaed emprcally he characersc and deermnans of frm exchange rae predcons. The novel fndngs are hree-fold. Frs, Japanese exporng frms are more lkely o updae her exchange rae predcons n response o unexpeced yen apprecaon han n response o unexpeced yen deprecaon. Second, he performed panel esmaon demonsraed ha a frm s expor dependency, as well as s profably, has sgnfcan effec on he speed of updang predced exchange raes n response o an unexpeced yen deprecaon. Fnally, unlke he negave effec of profably on predcon updang found among large- and medum-sze frms, small-sze frms end o updae her predcon of fuure exchange raes n response o unexpeced yen deprecaon, whch suggess ha frm sze, ofen assocaed wh frm expor compeveness, does maer n he predcon of fuure exchange raes. These emprcal fndngs wll movae furher research on frm-level predcons of fuure exchange raes. 5
References Berman. N., Marn, P., Mayer, T., 202. How do dfferen exporers reac o exchange rae change? Q. J. Econ., 27, 437 492. Flodén, M., Smbanegav, W., Wlander, F. 2008. When s a lower exchange pass-hrough assocaed wh greaer exchange rae exposure. J. In. Money Fnance, 27, 24 39. Io, T., 990. Foregn exchange rae expecaons: mcro survey daa. Am. Econ. Rev., 80, 434 449 Io, T., Kobuch, S., Sao, K., Shmzu, J. 208. Managng Currency Rsk: How Japanese Frms Choose Invocng Currency, UK: Edward Elgar, n press. Kneer, M. M., 993. Inernaonal comparsons of prce-o-marke behavor. The Am. Econ. Rev., 83, 473 486 Krugman, P., 987. Prcng o marke when he exchange rae changes, n: Arnd S.W. Rchardson J.D. (Eds.), Real-Fnancal Lnkages among Open Economes, Cambrdge: MIT Press. Nguyen, A. T. N., Sao, K., 207. Frms predced exchange raes and nonlneares n prcngo-marke presened a 207 Asan Meeng of Economerc Socey, The Chnese Unversy of Hong Kong, Hong Kong, June 3 5. Takag, S., 99. Exchange rae expecaons: A survey of survey sudes. IMF Economc Revew Saff Papers, 38, 56 83. 6
Fgure : Predcon error of large and small Japanese manufacurng exporers Noe: Predcon error s calculaed as he dfference beween naural log of he acual exchange rae and naural log of he predced exchange rae, er ln s ln E = s, ndcang he devaon rae of he realzed exchange rae from he expeced one. Error (Large Frms), Error (Small Frms) are calculaed usng he expeced exchange rae for large- and small-sze frms, respecvely. The shaded (non-shaded) yen deprecaon (apprecaon) perods are perods n whch he predcon errors are larger (smaller) han 0. Source: Bank of Japan, Shor-Term Economc Survey of Enerprses n Japan (TANKAN); IMF, Inernaonal Fnancal Sascs. 7
Table : Adjusmen speed of exchange rae predcons o an error n he prevous perod Frm-sze level Large enerprse Medum enerprse Small enerprse Indusry Depre. Appre. Symmery Depre. Appre. Symmery Depre. Appre. Symmery Manufacurng 0.35*** 0.632*** 0.000*** 0.357*** 0.542*** 0.000*** 0.397*** 0.480*** 0.04 (0.026) (0.052) (0.023) (0.037) (0.030) (0.04) Texle 0.43*** 0.573*** 0.03** 0.442*** 0.578*** 0.0** 0.509*** 0.6*** 0.87 (0.038) (0.527) (0.03) (0.042) (0.046) (0.062) Chemcal 0.38*** 0.573*** 0.00*** 0.386*** 0.565*** 0.006*** 0.345*** 0.50*** 0.05** (0.030) (0.049) (0.034) (0.053) (0.032) (0.058) Iron and seel 0.478*** 0.62*** 0.04** 0.354*** 0.543*** 0.07** 0.480*** 0.532*** 0.539 (0.045) (0.063) (0.040) (0.066) (0.053) (0.065) Non-ferrous meal 0.377*** 0.590*** 0.0** 0.49*** 0.63*** 0.00*** 0.500*** 0.586*** 0.24 (0.04) (0.07) (0.032) (0.049) (0.047) (0.056) Processed meal 0.370*** 0.523*** 0.036** 0.278*** 0.673*** 0.000*** 0.256*** 0.77*** 0.000*** (0.039) (0.060) (0.03) (0.077) (0.047) (0.09) General 0.335*** 0.656*** 0.000*** 0.345*** 0.573*** 0.000*** 0.380*** 0.628*** 0.00*** (0.032) (0.069) (0.025) (0.046) (0.034) (0.063) Elecrc 0.25*** 0.65*** 0.007*** 0.32*** 0.623*** 0.000*** 0.396*** 0.53*** 0.04** (0.049) (0.37) (0.027) (0.054) (0.03) (0.047) Transpor 0.326*** 0.699*** 0.000*** 0.256*** 0.536*** 0.000*** 0.373*** 0.352*** 0.784 (0.034) (0.076) (0.025) (0.05) (0.054) (0.052) Oher 0.367*** 0.534*** 0.04** 0.40*** 0.370*** 0.58 0.328*** 0.275*** 0.527 (0.034) (0.057) (0.04) (0.045) (0.058) (0.060) Noe: (a) Manufacurng, General, Elecrc, Transpor, and Oher denoes All manufacurng ndusres, General machnery (ncludng he hree secors of Generalpurpose machnery, Producon machnery, and Busness orened machnery ), Elecrcal machnery, Transporaon machnery, and Oher manufacurng, respecvely. (b) Depre. and Appre. denoe he yen deprecaon and apprecaon regmes, respecvely. (c) Symmery ndcaes p-values of he Wald es for he hypohess of equaly n adjusmen speed ( α = α ) n he wo regmes. 2 8
Table 2: Relaonshp of adjusmen speed wh oher frm characerscs. dependen var: adjusmen speed no regme deprecaon apprecaon no regme deprecaon apprecaon () (2) (3) (4) (5) (6) exp_rao 5.462** 9.532*** -2.788 6.955**.52*** -.590 (2.543) (2.982) (4.87) (3.472) (4.45) (5.657) prof_rao -8.624-7.355 -.378-7.628** -25.75** -6.280 (5.589) (6.503) (9.297) (8.252) (0.83) (2.942) large*exp_rao -.592 -.325-3.406 (5.098) (5.937) (8.463) small*exp_rao -2.079 0.390.427 (6.858) (9.024) (0.0) large*prof_rao 0.383 9.538 7.562 (9.694) (2.063) (5.004) small*prof_rao 25.820** 59.79*** 8.647 (2.22) (6.058) (7.962) Consan -0.590 -.482* 2.378 0.073-0.504 4.025 (0.84) (0.899) (3.378) (0.840) (0.878) (3.47) large enerprse -0.68-0.889 -.430 (.066) (.2) (.823) small enerprse -.099 -.440-3.06* (.03) (.294) (.585) me-fxed effec YES YES YES YES YES YES cross-secon fxed effec YES YES YES YES YES YES degree of freedoms 908 074 77 904 070 767 R-squared 0.067 0.232 0.28 0.069 0.242 0.29 No. of obs 2006 59 847 2006 59 847 Noe: */**/*** denoes sgnfcance a 0%, 5% and %, respecvely. 9