Financial Anxieties and Long-run Neutrality of Money in Japanese Economy

Transcription

1 Financial Anxieies and Long-run Neuraliy of Money in Japanese Economy Md. Jahanur Rahman Ph.D. Suden, Faculy of Economic Sciences Hiroshima Shudo Universiy, Japan Toshihisha Toyoda Faculy of Economic Sciences Hiroshima Shudo Universiy, Japan Yoji Moria Faculy of Economics, Kyoo Gauen Universiy, Kyoo, Japan Conacing Auhor: Toshihisha Toyoda Professor and Dean Faculy of Economic Sciences Hiroshima Shudo Universiy 1-1-1, Ozuahigashi, Asaminami-u Hiroshima , Japan Phone:

2 Financial Anxieies and Long-run Neuraliy of Money in Japanese Economy Absrac Exising research shows he empirical evidence on he long-run neuraliy of money in Japanese economy. A surprising aspec is ha his hypohesis does no hold over he period when Japanese economy experiences some peculiar evens, which has never been observed before. In his paper, we invesigae why he sable long-run neuraliy of money disappear over his period. Financial anxieies are quanified o approximae he increased precauionary demand of money. Money is hen adjused by subracing his precauionary demand. Finally, he empirical evidence on long-run neuraliy of money has been recovered by using his adjused money over his period. Keywords: Long-run neuraliy, Fisher-Seaer mehodology, financial anxiey, precauionary demand. JEL Classificaion: E44, E51, E52. 1

3 1. Inroducion Long-run neuraliy (LRN) of money is a ey classical macroeconomic hypohesis whose modern heoreical foundaion was provided by Friedman (1969a, 1969b). Alhough here are many classical hypoheses o he efficacy of moneary policy, his is widely acceped among he economiss and policymaers. The main idea of LRN of money is ha changes in he money soc evenually change nominal variables, lie nominal prices and nominal wages, ulimaely leaving imporan real variables, lie real oupu, real consumpion expendiures, real wages, and real ineres raes, unaffeced. Since economic decision-maing is based on real facors, he long-run effec of injecing money ino he macroeconomy is ofen described as neural in he end, real variables do no change and so economic decision-maing is also unchanged. How long such a process aes, and wha migh happen in he meanime, are holy debaed quesions. A formal definiion of he LRN of money is ha a permanen, unexpeced (exogenous) change o he level of money supply has no effec on he level of real oupu in he long run. However, he permanen increase should cause permanen changes o price level. Under LRN, changes in he money supply may or may no have shor-run real effecs. There is a second hypoheical proposiion ha is relaed o he LRN of money is he long-run super neuraliy (LRSN) of money, which occurs when a permanen, exogenous changes in he growh rae of money supply leave he level of real economic variables unaffeced. In he world of moneary heory, nearly all models based on sandard economic assumpions embody some form of LRN of money. Mos liely his is because moneary heoriss generally hin LRN of money is sensible, and, herefore, hey build i ino heir models. Perhaps surprisingly, here are many plausible analyses ha sugges ha deparures from LRSN of money migh be consisen wih sandard economic heory. I is, in fac, relaively easy o produce such heories. Accordingly, when LRN of money is aen almos as an axiom of moneary economics, LRSN of money is far more circumspec. Empirical ess ha convincingly documened deparures from LRN of money herefore would be quie surprising (or quie suspeced) o moneary economiss. Whereas, an empirical es ha convincingly showed deparures from LRSN of money would no be oo surprising, since his resul is consisen wih a number of exising economic heories. These long-run neuraliy preposiions have been sudied exensively, boh heoreically and empirically, and are sill very conroversial opics among macroeconomic researchers. Alhough LRN of money is generally assumed o be rue in economic heory, he empirical evidence on i has been very mixed and far from convincing. There is a lenghy hisory of effors o es for LRN of money and various economeric procedures are available for esing 2

4 hese hypoheses. In he 1960s, he primary mehod in esing for LRN of money consised of simple regression, no aing ino accoun any properies of ime series daa. The resuls generaed were found o be inconclusive in mos cases. Lucas (1972) and Sargen (1971) argued ha his mehod was no hrough enough as i did no firs es o find evidence as o wheher or no he money soc had been affeced by a permanen change; in oher words, here was no esing for a uni roo. They showed ha if money soc did no conain uni roo, i.e., i did no conain permanen change; LRN of money could no be esed. This mared he beginning of firs esing for a permanen change or uni roo on variables being used in LRN of money analysis. In response o he Lucas and Sargen criics, Fisher and Seaer (1993), and King and Wason (1992, 1997) have advanced he dominan approaches for esing LRN of money. Fisher and Seaer (1993) employ a bivariae srucural vecor auoregression (SVAR) model o es long-run neuraliy and supper neuraliy. The orders of inegraion of he variables deermine he model s resricion, while he exogeneiy of nominal variable is a necessary condiion in heir framewor. Specifically, LRN of money ess are possible only if nominal money soc as well as real oupu variables are a leas inegraed of order one. The neuraliy of money hypohesis has been esed for numerous counries using heir mehodologies. For example, Boschen and Oro (1994), Olealns (1996), Serleis and Krause (1996), Haug and Lucas (1997), Coe and Nason (2003), Shelley and Wallace (2006) used Fisher and Seaer mehod, while Weber (1994), Jefferson (1997), Serleis and Kousas (1998, 2001) employed King and Wason mehods for esing he LRN and LRSN of money. In spie of he progress in research on LRN of money, only a very limied number of comprehensive sudies are available in Japan. Yamada (1997) shows ha moneary neuraliy holds in erms of real oupu, by applying Fisher and Seaer s (1993) procedure o he Japanese seasonally adjused quarerly daa from 1957q1 o 1995q1. Using wo ypes of daa ses for cenury-long annual daa covering 119 years from 1885 hrough 2003 as well as poswar seasonally adjused quarerly daa over he period 1955q2 2003q4, Oi e al. (2004) have found evidence supporing LRN of money. Alhough, previous researches suppor he LRN of money proposiion for Japanese economy, in his aricle, i has been shown ha his proposiion became disappeared during he period This sub sample covers he period when he Japanese economy experienced some peculiar evens which had never been observed before. Over his period, people s anxieies over he financial sysem rapidly increased due o he failure of several big bans and securiy companies. As a resul boh firms and household seem o ry o increase he money demand by heir precauionary moivaion. Therefore, he rise of his precauionary demand due 3

5 o financial anxieies seems o brea down he LRN of money over he period 1980q1-2006q1, which exised in he period 1955q2-1979q4. In his paper, we have invesigaed why he sable LRN of money disappear over he sub-sample period 1980q1-2006q1. Financial anxieies have been quanified o approximae he increased precauionary demand of money. Accordingly, as a remedial measure, money has been adjused in erm of precauionary demand and hen reexamined neuraliy proposiion wih respec o ha adjused money. So he objecives of his paper are (a) o quanify financial anxieies, (b) o model precauionary demand as funcion of financial anxieies, (c) o adjused money by subracing precauionary demand and finally, (d) o reexamine LRN of money in erms of adjused money. The empirical esabiliy of hese LRN hypoheses are imporan for policy formulaion and design, such as he effeciveness in moneary policy, while he deerminaion of he ime series properies of he daa are crucial for he exising mehods. So his paper presens he economeric reamens of ime series daa for esing he proposiion of LRN of money in Japanese economy. Fisher and Seaer (1993) approach is used o invesigae he LRN of money in Japan over he period 1955q2-2006q1. Consequenly, quarerly seasonally adjused ime series daa on real GDP, nominal moneary variables M1 and M2+CD are used in his paper o examine he robusness of previous empirical resuls in Japan. 2. Fisher and Seaer Mehodology Fisher and Seaer (1993) develop an economeric mehodology o measure he long-run effec of money on real oupu. They assume ha relaionship beween money supply and real oupu can be presened by a saionary, inverible bivariae vecor auoregressive (VAR) model. The focus variables are m and y, which are he naural logarihms of nominal money supply and real oupu respecively. Their model is given as follows: a( L)(1 L) d( L)(1 L) m y m y b( L)(1 L) c( L)(1 L) y m y u m w where a(l), b(l), c(l) and d(l) are polynomials in he lag operaor L wih a 0 = d 0 = 1 and b 0 and c 0 are unresriced. The symbol x sands for he order of inegraion of x; if x = 1, means ha x is inegraed of order one, I(1). The errors vecor (u w ) is assumed o be i.i.d., wih mean zero and covariance marix. For presenaional convenience, consans and rends are suppressed. The Wold s moving average (MA) or impulse-response represenaion of sysem in equaion (1) is given by: (1) 4

6 m y (1 L) (1 L) m y [ ( L) u [ ( L) u ( L) w ] ( L) w ] where (L) = d(l)/a, (L) = b(l)/a, (L) = c(l)/a and (L) = a(l)/a, wih A = a(l)d(l)- b(l)c(l). Then he long-run effec of money is measured hrough he long-run derivaive of oupu wih respec o permanen sochasic exogenous change in money soc: y u LRD y, m lim (2) m u The LRD is no defined for he case in which lim( m u ) = 0 implying here is no permanen changes in he moneary variable. Therefore, a necessary condiion for esing LRN and LRSN is ha here mus be permanen sochasic shocs o he money supply. The LRD for esing LRSN is derived from he same formula by replacing m by m. The limi of he raio in equaion (2) measures he ulimae effec of a (sochasic) money disurbance on real oupu relaive o ha disurbance s ulimae effec on he money variable. Money is said o be neural (superneural) when, following a permanen shoc o he level (growh) of money, LRD y,m is equal o zero (LRD y, m is equal o zero). Equaion (2) also shows ha if here have been no permanen shocs o real oupu, hen lim( y u ) = 0, and hence LRD is equal o zero. Therefore, if real oupu is I(0), LRN and LRSN canno be rejeced. The specific value of LRD depends on m and y, he orders of inegraion of m and y. Fisher and Seaer show ha equaion (2) can be wrien as: m y (1 L) ( L) L 1 LRD y, m (3) (1) Equaion (3) demonsrae ha he value of LRD depends on m - y. Of paricular ineres in examining LRN and LRSN of money are he cases m = y 1 and m = 2, y = 1 respecively. Specifically, when m - y = 0 or when m = y = 1, equaion (3) reduce o LRD y, m (1) c(1). (4) (1) d(1) Assuming he resricion ha money is exogenous in he long run (or, b(1) = 0), Fisher and Seaer (1993) demonsraed ha c(1)/d(1) can be consisenly esimaed as he coefficien from he OLS long-horizon regression lim, where is ( y y 1 ) ( m m 1 ) (5) 5

7 For he LRSN of money, which is esable if m is I(2) and y is I(1), he esimaor of LRD y m is he slope coefficien in ( y y 1 ) ( m m 1) (6) Sandard pracice is o esimae using OLS for each value of aing values of one hrough a predeermined upper limi. Wih T observaions, he 95-percen confidence inervals hen are consruced for he s from a -disribuion wih T/ degrees of freedom using sandard errors correced for serial correlaion by he Newey-Wes (1987) procedure. LRN of money is rejeced if zero lies ouside he confidence inerval as become large. In his paper, we used Fisher and Seaer (1993) mehod for esing he LRN of money. The imporance in esing for LRN or/and LRSN of money lies in esablishing wheher or no here is a permanen change (uni roos) in he variables being analyzed and wha causes hose changes. Thus, before using Fisher and Seaer mehods, firsly, we have o deermine he order of inegraion of he variables, which is he ey requiremen for esing neuraliy hypohesis. Secondly, we have o chec wheher nominal money and real oupu are coinegraed, because, he presence of coinegraion is by iself sufficien for rejecing he neuraliy proposiion. Also, for he validiy of he model in equaion (1), we need o assume ha boh are no coinegraed. Finally, we have o show ha real oupu has no long-run effec on nominal money, ha is, money is exogenous. This is he crucial assumpion in Fisher and Seaer (1993) mehod. In he nex secion, we inroduce he daa used in his paper and chec he above ime series properies horoughly. 3. Daa and Their Time Series Properies Our objecive is o invesigae he LRN of money hypoheses using poswar daa for Japan. Seasonally adjused quarerly ime series daa over he period 1955:q2 2006:q1 are used in his sudy. Quarerly real GDP and nominal money supplies M1 and M2 + CD are used as he real oupu and nominal money socs respecively. We use Ban of Japan (BOJ) monhly moneary saisics for M1 and M2+CD. Monhly series are used o form he quarerly daa. Tha is, he observaion for he following monh is used as a proxy for he end-of-quarer money. For example, our M1 figure for he firs quarer of 1990 is he value of M1 on March, Whereas, quarerly real GDP series is aen from he Sysem of Naional Accouns (SNA) 1. The following symbolic noaions are used hrough ou his paper. y = log(real GDP) m1 = log(m1) m2 = log(m2 + CD) 6

8 Alhough m2 is he major indicaor of board moneary aggregae, consideraion of wo measures of money supply, namely m1 and m2, serves as a sensiiviy analysis of he poenial effecs of money on real oupu. I is shown in Bullard (1994) and Olealns (1996) ha he oucome of he es for LRN is sensiive o he measure of money involved. 3.1 Tess for Uni Roos Time series properies of he daa are crucial imporan for esing LRN of money. So required ime series properies would be invesigaed in his secion. Idenificaion of he orders of inegraion of nominal money and real gdp is an imporan issue before esing LRN of money using he mehodology discussed above. Unforunaely, i is well nown ha uni-roo ess have low power and ha resuls can vary wih he ype of es used and on he number of lags included in he es equaion. For his reason, i becomes a sraegy among he researchers o examine he resuls of several es procedures in order o draw conclusions regarding variable inegraion. Wih his in mind, hree uni roo ess are performed: (i) mos widely used Augmened Dicy-Fuller (ADF) es of Dicy and Fuller (1979, 1981) (ii) he asympoically mos powerful DF-GLS es of Ellio e al. (1996) and (iii) he Kwiaowsi e al. (1992) LM es (KPSS). The null hypohesis of ADF and DF-GLS ess is ha a ime series variable has a uni roo while ha of KPSS es is ha a variable is saionary. A common sraegy is o presen resuls of boh ADF/DF-GLS and KPSS ess, and show ha he resuls are consisen (e.g., ha he former rejec he null while he laer fail o do so and vice-versa). The lag lengh is seleced by using he Aaie Informaion Crieria (AIC), Hannan-Quinn Informaion Crieria (HIC) and Schwarz Informaion Crieria (SIC), seing he maximum lag a 12. Before beginning he formal ess, he variables should be ploed agains ime o visually deermine if a rend exiss in he ime series. The necessiy of his sep is simply due o he fac ha he criical values of he ess depend on he sample size and he inclusion of deerminisic componens, i.e., he inclusion of a consan and a ime rend. All variables in level have been graphed agains ime in Figure 1 over he period 1955q2-2006q1. By referring Figure 1, i is visually eviden ha m1, m2 and y presen upward rends bu i is difficul o guess wheher he rends are linear or quadraic. From he beginning of he ime lengh under consideraion, Japanese economy was evolving lie a developing counry and gradually he economy experienced very unusual phenomena including he exchange rae flucuaions, a highly growh rae counry, he bubble economy, a burs of he bubble and a deflaionary economy hrough he period. This hisory of Japan indicaes he possibiliies for exising quadraic rends in he economic variables 2. To chec a quadraic rend we have applied ADF es on he firs 7

9 differences of m1, m2 and y wih he presence of a consan and a rend erm. Table 1 shows he ADF es resuls on firs difference ime series under consideraion. Highly significan linear rends are deeced by checing he -saisic of ADF ess. Afer recognizing a linear rend in firs difference of each series, we have applied DF-GLS es wih he null hypohesis of uni roo wih a drif and alernaive hypohesis of a saionary process wih a linear rend. DF-GLS es rejecs he null hypohesis for all firs difference series m1, m2 and y respecively. We have also applied KPSS es on m1, m2 and y, where he null hypohesis of a saionary process wih a linear rend canno be rejeced. The resuls of ADF, DF-GLS and KPSS ess have been repored in Table 2. Thus, we can conclude ha m1, m2 and y are saionary wih a linear rend by all ADF, DF-GLS and KPSS ess. This implies ha he level of m1, m2 and y conain quadraic rends. Nex sep is o invesigae wheher he level of m1, m2 and y are saionary or nonsaionary wih a quadraic rend. Since he criical values of he above ess are no available for he presence of quadraic rend in he models, we have applied anoher mehod. Coinegraion analysis by Johansen (1995) gives us an easy deecion mehod abou hese problems wih a quadraic rend. Coinegraion ess can deec wheher a linear combinaion(s) of ime series variables is(are) saionary or nonsaionary wih a linear or a quadraic rend. Since Johansen s mehod is based on a ran condiion of a marix, i is applicable o deec wheher a single variable is saionary. Therefore, if a single ime series variable is shown o be non-coinegraed wih a quadraic rend, hen we can conclude ha he series is nonsaionary (or conain uni roo) wih a quadraic rend. We have used Johansen s coinegraion mehod on each level series for esing he null hypohesis of nonsaionary wih a quadraic rend. Tes resuls in Table 3 show ha m1, m2 and y are nonsaionary wih quadraic rend. Summarizing he above es resuls, we conclude ha all variable are inegraed of order one, I(1). This implies ha permanen componens exis in y, m1 and m2 respecively. Our es resuls are consisen wih Oi e. al. (2004), where hey used a consan and a linear rend boh in level and firs difference level of he variables. 3.2 Uni Roo Tes wih a Srucural Change Our uni roo ess used above do no allow for a srucural brea in he ime series daa, which migh no accuraely describe inegraion properies of daa in he case ha he sudied period migh conain a srucural brea poenially. Perron (1989) argues ha if here is a brea in he deerminisic rend, hen he uni roo ess will lead o a misleading conclusion ha here is a uni roo, when in fac here is no. 8

10 Perron (1989) inroduced a dummy variable o ADF ess o accoun for srucural change. In Perron's procedure, he null hypohesis is ha a series has a uni roo wih an exogenous srucural brea occurring a a paricular ime T B versus he alernaive hypohesis ha he series is saionary wih an exogenous change in he rend a ime T B. However, Perron's sudy has been criicized on he grounds ha his procedure is condiional on a nown brea poin. The pre-deermined period for he dummy variable requires full informaion abou he srucural brea poin, which raises he ris associaed wih he wrong period selecion. Zivo and Andrews (1992) improve Perron's approach by accommodaing a srucural brea wihou predeermining he brea poin ime. Differen from Perron's approach ha reas srucural brea as exogenous, hey endogenize he choice of a brea poin ino he procedures hrough he esimaion of a brea poin using sequenial mehods. The null hypohesis in he Zivo and Andrews es is a uni roo wihou any srucural change. The alernaive hypohesis is a saionary process ha allows for a one-ime unnown brea in he rend. Following Zivo and Andrews (1992), we can wrie he following augmened regression equaion: y 1 y 1 DU ( ) i 1 i y i (7) This equaion is similar o he ADF uni roo es equaion wih a dummy erm added in, where 1 for T DU ( ) and T B T. 0 oherwise Here, is he brea fracion, ranging from 2/T o (T - 1)/T. Through an ieraion procedure, he es obains T - 2 regressions and -saisics corresponding o he coefficien s. The number of lags, required for he ADF regressions, is allowed o vary for each choice of. We follow he procedure used in Zivo and Andrews (1992) o deermine he number of : saring bacwards from a maximum lag of 12, he appropriae number of lags is deermined when he value of i is chosen such ha -saisic of i is greaer han 1.6 in absolue value, and he saisic for I+n for n > 0 is less han 1.6. The rejecion of he uni roo hypohesis is deermined by criical values generaed for he value of in each case. Zivo and Andrews (1992) provide asympoic disribuions of he minimum -saisics and criical values for rejecing he null hypohesis for each fixed. We apply he Zivo and Andrews (1992) es o our daa. Table 4 provides he resuls along wih he brea dae. The esimaion resuls fail o rejec he null hypohesis of a uni roo for all variables in he level form. On he oher hand, he null hypohesis is rejeced for all he variables in firs difference form. These resuls are consisen wih he resuls of Oi e.al. (2004), 9

11 wherein hey used Perron (1989) ess. Thus we conclude ha m1, m2 and y are inegraed of order one. 3.3 Tess for Coinegraion Anoher relaed cavea is ha, even if i were nown ha money and real gdp are indeed nonsaionary, he esing procedure sill relies criically on he variables being noncoinegraed. This is imporan no only because a violaion maes he VAR model misspecified, hus maing heir esimaes suspec, bu also because he presence of coinegraion is by iself sufficien for rejecing he neuraliy proposiion. So, we also es for coinegraion beween y and m1, and y and m2 respecively. We es he null hypohesis of no coinegraion using boh he Engle and Granger (1987) wo-sep procedure as well as Johansen s (1988, 1991) maximum lielihood echnique. Table 5 summarize he oucome of he coinegraion analysis. The Johansen coinegraion race es saisic and maximum eigen-value es saisic can no rejec he null of no coinegraion (r = 0), as well as he null of a mos one coinegraion vecor (r 1) a he 5% level of significance for he money (boh m1 and m2) and real gdp. Accordingly, Engle-Granger ADF es saisic indicaes ha he residual from he OLS saic regression for such a relaion conain a uni roo, which implies no evidence of coinegraion beween such variables. So money has no impac on real gdp in he long run, as prediced by basic classical economics. 3.4 Coinegraion Tes wih a Srucural Change We also apply Gregory and Hansen (1996) es for considering srucural change in coinegraion relaions. While Zivo and Andrews (1992) ess allow a single unnown srucural brea in uni roo es, Gregory and Hansen (1996) ess accommodae a single unnown srucural brea in coinegraion analysis. The Gregory and Hansen ess are under he framewor of residual-based ess for coinegraion, wherein he srucural brea is capured. Similar o he Zivo and Andrews es, he selecion of brea poin is no predeermined, bu decided empirically. Following he regime shif model in Gregory and Hansen (1996), we have hree ypes of srucural brea forms. The simples case of srucural change is a level shif in he coinegraion relaionship. Namely, here is a change in he inercep while he slope coefficien eeps consan. This can be modeled as Level Shif (C) y 1 2 x (8) where is a fracion brea o represen srucural brea, ranging from 0 o 1, 10

12 0 1 if [ T ]. if [ T ] In his equaion, 1 is he inercep before he shif, and 2 denoes he change in inercep a he ime of shif. The second case is o include a ime rend ino he shif model: Level shif wih rend (C/T) y 1 2 x (9) The hird case is he regime shif model, wherein i allows boh he inercep erm and slope erm o change: Regime shif (C/S) y 1 2 1x 2x (10) where boh 1 and 2 have he same meaning as in he level shif model. 1 denoes he coinegraion slope coefficiens before he regime shif, and 2 represens he change in he slope of coefficien. Afer obaining he residuals in he equaion for each, we conduc he ADF es for he residuals o deermine if hey are saionary process. For each, is chosen by he procedure in Perron and Vogelsang (1992). They sar bacwards from he maximum max equal o 6 unil he las lag of he firs difference included is significan a he 5 percen level using normal criical values. The null hypohesis of Gregory and Hansen ess is ha he residuals conain a uni roo and hence here is no coinegraion. The alernaive hypohesis is ha residuals do no conain a uni roo and hence here is coinegraion wih a single unnown brea. Thus is chosen o minimize he ADF -saisics, and he brea year is seleced o be he year corresponding o he minimum -saisic. Gregory and Hansen (1996) provide asympoic disribuions and abulae criical values for minimum ADF saisic. We apply he Gregory and Hansen es (1996) o our daa o invesigae coinegraion relaions beween (y, m1) and (y, m2). Table 6 represens he es resuls of Gregory and Hansen (1996) ess as well as he esimaed breapoin. We can see ha he convenional ADF - saisic fails o rejec he null hypohesis of no coinegraion beween (y, m1) and (y, m2). Oi e. al. (2004) maes he same conclusion using he daa over he period 1955q2-2003q4. Thus, we conclude ha boh (y, m1) and (y, m2) are no coinegraed. 3.5 Tesing for Exogeniy Fisher and Seaer (1993) mehodology also requires money should be exogenous. The validiy of his assumpion is assessed by Granger-causaliy es using he mehodology developed in Toda and Yamamoo (1995). According o heir mehod, VAR model in levels is 11

13 used for esing Granger-causaliy even if he processes may me inegraed or coinegraed of an arbirary order (as far as, he order of inegraion of he process does no exceed he rue lag lengh of he model). Wha we need o now o apply his mehod is he lag lengh () and maximum order of inegraion (d max ) of he variables used in level VAR model. Having chosen a lag lengh, we hen esimae a wo variable VAR model of order ( + d max ) for esing he null hypohesis of non-causaliy of real gdp o nominal money. The coefficien marices of las d max lagged vecors in he model are ignored and we can es for causaliy based on he firs coefficien marices using he sandard asympoic heory. The lag lengh is deermined using he AIC, SIC and HQ crieria, wih a maximum of 12 lags considered. We es causaliy for boh d max = 1 and d max = 2 and he resuls are shown in Table 7. As able shows, he null hypohesis of real gdp does no Granger-causes money canno be rejeced for each of money and real gdp series. Therefore, we conclude ha money is exogenous. So, all ime series daa saisfy he required ime series properies and we can proceed o apply he Fisher and Seaer (1993) echnique o assess more deeply he ey classical hypohesis for he income-money relaionship in Japanese economy. 4. Empirical Resuls for LRN of Money In he previous secion, we showed ha nominal money socs m1 and m2 and real gdp y are inegraed of order one and are no coinegraed. Also, Granger-Causaliy ess indicae m1 and m2 are exogenous wih respec o y. Now we are ready o precede wih Fisher and Seaer (1993) mehodology for esing LRN of money in Japanese economy over he period 1955q2-2006q1. As he growh rae of real gdp shows upward end shown in Figure 1, a rend erm is added o he Fisher-Seaer regression in equaion (5) as follows 3 : ( y y 1 ) ( m m 1) (11) Equaion (11) for wo nominal moneary variables m1 and m2 are separaely esimaed when y is dependen variable for values of = 1, 2,, 30. Each coefficien,, presens he esimaed response of he change in log real gdp o he change in logged money over +1 periods. The residuals from he regression for he various lags may be non-spherical, possibly leading o biased -raios and oucomes of he LRN ess. Following Fisher and Seaer, sandard errors of he esimaed coefficien are adjused using he Newey-Wes (1987) procedure. Esimaes of and he corresponding sandard errors for differen lags are presened in Table 8. The - disribuion wih n/ degrees of freedom is used o consruc he 95% confidence inervals of. LRN of money is rejeced if zero lies ouside he confidence inerval as become large. Tes 12

14 oucomes for LRN of money can be observed by examining a plo of he esimaes of and is 95% confidence inerval agains he lag lengh. In Figure 2(a), we firs presen he graph of he coefficien and is 95% confidence inervals when he (+1) difference of nominal money m2 is he explanaory variable in equaion (11). The value of is ranging 1 o 30 over he period 1952q2 2006q1. As can be seen from he graph, he esimaed coefficiens are significanly posiive for he values of less han 14 quarers, suggesing a shor-run posiive effec of moneary policy using m2. Bu hese shor-run effecs disappear afer abou hree years. Tha is, as he lag lengh increases, here is an obvious downward rend in he plo of he esimaes. The confidence inerval bands include zero line for high values of 13. This suggess ha m2 does no affec real GDP in he long run. So i is observed ha he LRN hypohesis is suppored by he nominal money m2. Since money supply m2 is no inegraed of order 2, he LRSN of m2 canno be analyzed. Figure 2(b) shows he plo of coefficien and confidence inervals agains he lag lengh when nominal money m1 is he independen variable in equaion (11) over he period 1955q2 2006q1. For all values of, lower limi of 95% confidence inerval shows waving paern on he zero line, suggesing somewha a boarder line rejecion of neuraliy hypohesis in erms of m1. The confidence inerval conains zero line in he shor-run ( 5) and in he medium-run (20 26). I shows shor-run neuraliy as well as mixed effecs in he long run. This conclusion is no robus lie he case of m2. This resul indicaes he sensiiviy of he LRN of money resuls o he ype of money supplies considered. Since m1 is I(1) and he LRN hypohesis is rejeced using m1 as he measure of money supply, he LRSN hypohesis using m1 is also rejeced, ha is, he hypohesis ha permanen sochasic changes o he growh rae of m1 ulimaely leave he level of real GDP unchanged is rejeced. Thus we can conclude ha he LRN hypohesis is suppored using m2 as he money supply, while he use of m1 leads o he LRN hypohesis inconclusive for Japan. Our resuls are no compleely consisen wih he former findings. Yamada (1996) and Oil e. al. (2004) go evidence for LRN of money for boh m1 and m2. Alhough more careful comparisons will be necessary in fuure, we can enaively say ha our resuls have been obained by using one of he mos up-o-dae mehodologies and by including more recen daa compared wih he former resuls. Finally, o chec he robusness of our resuls in erms of m2, we examine he effecs of including he daa under he bubble economy (before 1989), afer he burs of he bubble (afer 1989) and he recen zero ineres raes (afer 1995) during he sub sample period 1981q1-13

15 2006q1. This sub-sample period was no considered in he previous researches for esing neuraliy hypohesis in Japanese economy. 4.1 Long-run Neuraliy on Sub Sample Periods In his secion, we spli he full sample ino wo sub sample periods; 1955q2 1979q4 and 1980q1 2006q. The former sub sample conains he period when Japanese economy possessed almos sable relaionships among he economic variables excep he oil shoc in The laer sub sample covers he period when Japanese economy experienced some peculiar evens which had never been observed before, such as, bubble economy (before 1989), burs of he bubble (afer 1989), failures of several big financial insiuions (1997), abou zero ineres raes (afer 1995) ec., The financial disress and deflaion is rooed in he so-called bubble economy of he laer half of he 1980s when he economy has experienced he expansion of bubbles in asses prices. The real GDP recorded negaive growh for five consecuive quarers, from he fourh quarer of 1997 on a quarer-o-quarer basis. The recession is characerized by rapid decline in asses prices, which subsanially accumulaed he non-performing loans. The mouning non-performing loans, especially in he financial secors, hampered he normal funcions of financial inermediaries and Japan s economy came o he verge of financial panic. So i is imporan o invesigae he neuraliy preposiion over hese sub samples. Thus, Fisher and Seaer (1993) mehod is used o chec he neuraliy proposiion over hese sub samples and he resuls are shown in he Figures 3(a) and 3(b). Figure 3(a) shows ha he zero line lies inside he 95% confidence inerval for he sample period 1955q2-1979q4. This implies ha LRN of money holds in erm of m2, which is consisen wih he full sample resul. On he oher hand Figure 3(b) shows he opposie resuls. Lower limi of 95% confidence inerval lies above he zero line for he sample period 1980q1-2006q1. Tha is, a posiive relaionship beween m2 and real GDP exiss in he long run. So he proposiion of long-run neuraliy of money is rejeced over he sub period 1980q1 2006q1. Why does he sable LRN of money disappear over he period 1980q1-2006q1? The reason seems o be relaed o he rise of he financial anxieies during his period. Then we consider he financial anxieies, which occurred afer he collapse of he bubble economy. Nex secion invesigaes how financial anxiey is responsible for disappearance of he LRN of money. 5. Quanifying Financial Anxieies The busing of he bubble caused he rapid decline of asse price, especially land price. The coninuous drop of land price decreases he collaeral value of land and accumulaes he 14

16 non-performing loan he baning secor has. The firms, which rapidly increased heir liabiliy under heir bullish expendiure in he laer half of 1980s, had go ino he financially difficul siuaions. Banrupcies began o increase rapidly afer he burs of he bubble. The amoun of banrupcies in 1991 had risen more han 300 percen from he previous year. The baning secor has become increasingly vulnerable due o non-performing loan problems. I was he year of 1997 when serious financial problems had come ou in he Japanese economy. Several big bans and securiy companies had failed, including Hoaido Taushou Ban and Yamaichi Securiies. Though several financial insiuions had been failing afer he burs of he bubble economy in 1990, hey were he small sized insiues and acically deal by insurance deposi. However, he failure of wo big financial insiuions was quie differen from he former ban failures when he significance of heir role in he Japanese economy was pu ino consideraion. Furher heir failure riggered he rapid decline in he share prices of many financial insiuions. Japan premium was also imposed in he inernaional mare a he same ime. People s anxieies over he financial sysem rapidly increased. As a resul boh firms and household seem o ry o increase he money demand by heir precauionary moivaion. Therefore, he rise of his precauionary moivaion due o financial anxieies seems o brea down he long-run neuraliy of money over he period 1980q1-2006q1, which exised in he period 1955q2-1979q4. So i is crucial imporan o adjus money in erm of precauionary demand and hen reexamine neuraliy proposiion wih respec o ha adjused money. So, firs of all, we need o quanify financial anxieies over he period 1980q1-2006q1. Kimura and Fujia (1999) proposed a new variable o capure hese financial shocs as psychological change of people due o financial anxieies. They used he Corporae Financial Posiion Diffusion Index issued quarerly by Ban of Japan nown as TANKAN in order o quanify he unobservable variable over he period 1976q2-1999q3. They employed wo TANKAN diffusion index shown as: DI = <easy> - <igh> rae = <rise> - <fall> where DI is a rae of financial posiion such ha <easy> (<igh>) means he percenage wih which company feels financial posiion as easy (igh), and where rae is change of ineres rae wih which a company borrows money from ban such ha <rise> (<fall>) means he percenage wih which company feels ineres rae as rise (fall). Kimura and Fujia (1999) define a new variable rae as an ineres rae by accumulaing rae as: rae = rae 1 + rae rae. 15

17 They employed nonsaionary DI and rae in TARCH (Threshold Auoregressive Condiional Heeroscedasiciy) model and reaed he condiional variances as financial anxieies. However, due o rough reamen of nonsaionary daa, heir model had been affeced by unexpeced parameer values and sign problems and hence couldn explain he asymmeric properies properly. As a resul, heir model showed financial anxieies in he period of bubble as well as afer he bus of he bubble economy, which could no be explained in economic views. To ge rid of his problem, in our previous aricle (Rahman, e al. 2005), we improved he anxiey variable using he growh rae model for he same daa bu over he period 1976q2 o 2005q1 using TARCH model and our resuls shows financial anxieies only afer he bus of he bubble. The magniude and non-negaiviy condiions in esimaing our TARCH model are valid in saisical sense and our esimaion can exhibi he financial anxieies explicily over he economy, which is consisen wih economic views. We have reesimaed financial anxieies using our procedure over he period 1976q3-2006q1. Since DI and rae are nonsaionary, we have used he following growh rae model for condiional mean equaion 4 : DI rae rae (12) where a random error wih mean 0 and Var -1 ( ) = h wih Var -1 ( ) denoing he variance condiional on he informaion a ime -1. The financial anxieies have been capured by his condiional variance. Then he condiional variance is described by TARCH model wih asymmeric variance propery as: h h I (13) where I 1 = 1 if 1 < 0 = 0 oherwise. In his model, for TARCH effec, he asymmeric erm 0 and he condiion for nonnegaiviy will be 0 0, 1 0, 0 and The condiional variance h is subjec o an impac 1 from good news ( -1 0), while an impac ( 1 + ) from bad news ( -1 0). This ind of asymmeric propery corresponds o he siuaion such ha he psychological change of people due o he financial anxieies increases he precauionary demand and ha an easy financial posiion does no raise he precauionary demand. Esimaion resuls are shown in Table 9. The sign of all parameers seem o be reasonable in economic sense. Since a rise of DI implies easy financial posiion and a rise of rae means ha of ineres rae, should ae a negaive value. The 2 parameer of 1I 1 aes a posiive value and hence he condiional variance is 16

18 shown o exhibi asymmeric propery. Figure 5 shows financial anxieies by Kimura e al (1999) while our case is depiced in Figure 6 using TARCH model. Our resul in Figure 6 shows anxiey variable h raise a firs from (firs financial anxiey) when small credi unions and cooperaives failed because of an increase in he nonprofi loan caused by he rapid decline of soc and land price afer he bus of he bubble. The Japanese economy began o show he modes recovery in lae 1995, when real GDP began o increase and he official esimaion of NPLs decreased. However, he economy sharply decline in 1997 when Prime Miniser Ryuaro Hashimoo had declared he rise of he consumpion ax from 3 o 5 percen and he end of emporary income ax cu. Hoaido Taushou, one of he bigges bans and Yamaichi, one of he Big Four securiies had failed in November 1997 (second financial anxiey). People feel ha no financial insiue is immune from failure when governmen oo a very negaive view o use public funds o help affeced bans. People anxieies remendously increased, as indicaed in he rise of h in Then h rapidly decreases afer The Ban of Japan had adoped an aggressive moneary easing policy o reduce he iner-ban money rae o a low level in February Measuremen of Precauionary Demand and Adjusmen of Money In our previous aricle (Moria e al., 2006) we esimaed he precauionary demand as funcion of financial anxieies. Then we adjused real money by he precauionary demand o show he sable relaionship beween adjused real money, real GDP and spread of ineres rae afer 1998 and go he required relaionship. In his paper we follow he same procedure as Moria e al. (2006) used, jus using nominal money insead of real money. We define he precauionary demand (pd) and adjused money (m2 adj ) as follows: Precauionary demand pd = c h (14) Adjused money m2 adj = m2 c h (15) where c is unnown consan and h denoes financial anxieies. To idenify he precauionary demand, a sysem model is used o esimae he unnown consan c. According o Moria e al. (2006), we used he VEC model described by he se of variables m2 adj, y and r, where r is he spread of ineres rae defined as r = 10 years bond rae 3 monh cd rae 5. The unnown parameer c in equaion (14) is esimaed (esimaion procedure is shown in he Appendix) over he period 1980q1-2006q1. The esimaed value of c is and we approximaed he precauionary demand and adjused money as pd = h m2 adj = m h. 17

19 Money, m2, and adjused money, m2 adj, are shown in Figure Recovering LRN of Money over 1981q1-2006q1 In his secion, we have checed he long-run neuraliy of money in erm of adjused money (m2 adj ) using he same mehod over he sub sample 1980q1-2006q1. The resuls are shown in Figure 8. The figure explicily shows ha zero line lies inside he 95% confidence inerval afer lag 23. So, long-run neuraliy canno be rejeced in erm of adjused money (m2 adj ) over he sub sample 1980q1-2006q1 6. While, long-run neuraliy does no hold over he same period in erm of unadjused money (m2). 8. Conclusion Fisher and Seaer s (1993) seminal research on he long-run neuraliy of money is adoped o es quarerly seasonally adjused Japanese daa over he period 1955q2 2006q1. The long-run moneary neuraliy hypohesis is suppored using M2+CD as he measures of money supply, ha is, changes in M2 have no effecs on changes in real oupu in he long-run. Since M2 is no inegraed o a sufficienly high order, here are no permanen sochasic changes o he growh rae, and hence he superneuraliy hypohesis using M2+CD canno be analyzed for he daase. However, he long-run neuraliy and superneuraliy hypoheses are rejeced using M1, in ha changes in M1 significanly affec changes in real oupu. These resuls are consisen wih he previous researches excep neuraliy of money in erms of M1. Bu, surprisingly, long-run neuraliy does no hold using M2+CD in he sub sample period 1981q1-2006q1, when he Japanese economy experienced some peculiar evens, which had never been observed before. Due o he verge of financial panic, people s anxieies over he financial sysem rapidly increased over his period. As a resul boh firms and households seem o ry o increase he money demand by heir precauionary moivaion. Financial anxiey over his sub sample was quanified and precauionary demand was esimaed as a funcion of financial anxieies. Then money was adjused by subracing his precauionary demand. Using his adjused money, i is shown ha long-run neuraliy hypohesis canno be rejeced for M2+CD over his sub sample. Thus we can conclude ha long-run neuraliy of money in erms of M2+CD canno be rejeced in Japanese economy, bu his hypohesis is inerruped by he financial anxieies over he period 1980q1-2006q1. 18

20 Foonoes: We are indeb o he anonymous referee for a number of helpful commens and insighs. Remaining errors are, of course, our responsibiliy. 1. For money soc, we use he ousanding of end of period daa. From April 1998 he coverage of he money supply saisics is widen o include foreign bans operaing in Japan. So, we adjus for differenial beween hem o mae he daa coninuous. Addiionally from 1979 he M2 daa includes CDs. Regarding he real gdp daa, we use SNA68, 1990 basis, from 1955q2-2001q1 and hen exend he period o 2006q1 using SNA93 afer adjusing for differenial. Each series are seasonally adjused by using X-12-ARIMA approach. The ARIMA models used for seasonal adjusmens were (0, 1, 1)(0, 1, 1) 12 for M1 and (2, 1, 2)(1, 1, 1) 12 for M2, and differen ARIMA models are used for he componens of real gdp. 2. A rend (eiher linear or quadraic) in ime series analysis is resriced o he ime inerval under consideraion. In fuure ou of he ime inerval, we are no sure wheher he rend is significan or no. 3. Since he firs differences of real gdp and money supply (boh m1 and m2) possess linear deerminisic rend, a rend erm is added o Fisher-Seaer regression. Shelley and Wallace (2006) showed ha inclusion of rend in Fisher-Seaer equaion does no bias esimae of he coefficiens. However, he esimae of and are biased; bu hese are only nuisance parameers ha are no used for inference in he Fisher-Seaer es. 4. Uni roo ess conclude ha boh TANKAN indexes DI an rae are inegraed of order one. 5. m2, y and r are inegraed of order one and hey possess one coinegraion relaionship over he period 1980q1-2006q1. Tes resuls are no shown o save he space. Resuls are available from he auhors on reques. 6. TANKAN daa is available from 1974 up o presen. So we canno adjus money for full sample period using financial anxieies. 19

21 APPENDIX: Ouline of Esimaion Procedure of Precauionary demand To idenify he precauionary demand, a sysem model is used o esimae he unnown consan c in Equaion (14). According o Moria e al. (2006), we used he following VEC model described by he se of variables m2 adj, y and r, where r is he spread of ineres rae: i i i adj, c m0 mec 1 cm m2adj, i d m y i em r i m, (16) i 1 i 1 i 1 m2 y c i i i y0 yec 1 c y m2 adj, i d y y i ey r i y, (17) i 1 i 1 i 1 r c ec i i i cr m2, i d r y i er r i r,, adj (18) r0 r 1 i 1 i 1 i 1 where ec -1 on righ hand side of above equaions is an error correcion erm defined by ec m adj, y y rr cons. All parameers should be esimaed under he crierion of minimizing he sum of squared residuals. Since he sysem equaions are nonlinear in parameer, iniial condiions for parameer esimaion are essenially imporan for he convergence of esimaion. The following ieraion procedures are followed o esimae he unnown consan c: Sep-1: Esimae an iniial value of c. Sep-2: Calculae adjused money as m2 adj = m c h Sep-3: Esimae VEC model of (m2 adj, y, r) in equaions (16) o (18). Esimaed parameers ogeher wih he iniial value of c in Sep-1 are regarded as iniial condiions for nonlinear minimizing procedure. Sep-4: Carry ou he nonlinear minimizaion procedure. If he esimaed value of c is sufficienly coincide wih he iniial or jus previously esimaed value of c, hen sop he procedure and consider i as he required esimaed value. Oherwise go o Sep-5. Sep-5: Readjus m2 using he esimaed value of c in sep-4 as m2 adj = m c h. Go o Sep- 4 considering all esimaed parameers in Sep-4 as iniial condiions. I should be noed ha he iniial value of c in Sep-1 is more imporan in nonlinear minimizaion procedure. Our esimaion of c in Sep-1 is as follows: Firs, VEC model wih resricion is esimaed for he variables m2, y, r and h. Since h is saionary, we resric wo error correcion erms ec 1, = m2 + c 1 y + c 2 r + c 3 h + c 4 and ec 2, = c 5 h + c 6. The iniial value of c in Sep-1 is given by c = - c 3. 20

22 References: Boschen, J. F. and C. M. Oro (1994), Long-Run Neuraliy and Superneuraliy in an ARIMA Framewor: Commen, American Economic Review, 84, pp Bullard, J. B. (1994), Measures of Money and he Quaniy Theory, Federal Reserve Ban of S. Louis Review, 76, pp Coe, P. J. and Nason, J. M. (2003), The Long-horizon Regression Approach o Moneary Neuraliy: How Should he Evidence be Inerpreed?, Economics Leers, vol. 78, pp Dicy, D. A. and W. A. Fuller (1979), Disribuions of he Esimaors for Auoregressive Time Series wih a Uni Roo, Journal of he American Saisical Associaion, 74, pp Dicy, D. A. and W. A. Fuller (1981), Lielihood Raio Saisics for Auoregressive Time Series wih a Uni Roo, Economerica, 49, pp Ellio, G., T. J. Rohenberg and J. H. Soc, (1996), Efficien Tess for an Auoregressive Uni Roo, Economerica, 64, pp Engle, R. F. and C. W. J. Granger (1987), Co-inegraion and Error Correcion: Represenaion, Esimaion and Tesing, Economerica, 55, pp Fisher, M. E. and J. J. Seaer (1993), Long-run neuraliy and superneuraliy in an ARIMA framewor, American Economic Review, 83, pp Friedman, M. (1969a), The Opimum Quaniy of Money and Oher Essays, Chicago: Aldine. Friedman, M. (1969b), The Role of Moneary Policy, American Economic Review, 59(1), pp Gregory, A. and B. E. Hansen (1996), Residual-Based Tess for Coinegraion in Models wih Regime Shifs, Journal of Economerics, 70, pp Haug, A. A. and R. F. Lucas (1997), Long-Run Neuraliy and Superneuraliy in an ARIMA Framewor: Commen, American Economic Review, 87, pp Jefferson, P. N. (1997), On he Neuraliy of Inside and Ouside of Money, Economica, 64, pp Johansen, S. (1988), Saisical Analysis of Coinegraion Vecors, Journal of Economic Dynamics and Conrol, 12, pp Johansen, S. (1991), Esimaion and Hypohesis Tesing of Coinegraed Vecors in Gaussian Vecor Auoregressive Models, Economerica, 59, pp Kimura, T. and S. Fujia (1999), Kinyufuan o mane, jiaieizai, bua no aneiniuie (The relaionship beween financial anxieies, money, real economy, and prices) (in Japanese only), Woring Paper 99-6, Ban of Japan. King, R. G. and M. W. Wason (1992), Tesing Long run Neuraliy, Woring Paper No. 4165, Boson: Naional Bureau of Economics Research. King, R. G. and M. W. Wason (1997), Tesing long run neuraliy. Economic Quarerly, 83, Federal Reserve Ban of Richmond, pp Kwiaowsi, D., P.C.B. Phillips, P. Schmid and Y. Shin (1992), Tesing he Null Hypohesis of Saionariy agains he Alernaives of a Uni Roo: How Sure Are We Tha Economic Time Series Have a Uni Roo?, Journal of Economerics, 54, pp

23 Lucas, Rober E., Jr, (1972), Expecaions and he Neuraliy of Money, Journal of Economic Theory, 4, pp Moria, Y., M. J. Rahman and S. Miyagawa (2006), Esimaion of Precauionary Demand Caused by Financial Anxieies, Journal of he Faculy of economics, KGU,15-3, pp Newey, W. K. and K. D. Wes (1987), A Simple, Posiive Semidefinie, Heerosedasiciy and Auocorrelaion Consisen Covariance Marix, Economerica, 55, pp Oi, Hiroyui, Shigenori Shirasua, and Toyoichiro Shiroa (2004), On Long-Run Moneary Neuraliy in Japan, Moneary and Economic Sudies, Ban of Japan, 22(3), pp Olealns, N. (1996), Some Furher Evidence on he Long-run Neuraliy of Money, Economics Leers, 50, pp Perron, P. (1989), The Grea Crash, he Oil Price Shoc and he Uni Roo Hypohesis. Economerica, 57, pp Perron, P. and T.J. Vogelsang (1992), Nonsaionary and Level Shifs wih an Applicaion o Purchasing Power Pariy. Journal of Business and Economic Saisic, 10, pp Rahman, M. J., S. Miyagawa and Y. Moria (2005), Financial Anxieies in Japanese Economy, Journal of he Faculy of economics, KGU, 15-1, pp Sargen, T. J. (1971), A noe on Acceleraionis Conroversy, Journal of Money, Credi and Baning, 3, pp Serleis, A. and D. Krause, (1996), Empirical Evidence on he Long-run Neuraliy Hypohesis Using Low-frequency Inernaional Daa, Economics Leers, 50, pp Serleis, A. and Z. Kousas (1998), Inernaional Evidence on he Neuraliy of Money, Journal of Money, Credi and Baning, 30(1), pp Serleis, A. and Z. Kousas (2001), Moneary Aggregaion and he Neuraliy of Money, Economic Inquiry, 39(1), pp Shelly, G. L. and F. H. Wallace (2006), Long Run Effecs of Money on Real Consumpion and Invesmen in he U.S., Inernaional Journal of Applied Economics, 3(1), pp Toda, H. Y. and T. Yamamoo (1995), Saisical Inference in Vecor Auoregressions wih Possibly Inegraed Process, Journal of Economerics, 66, pp Weber, Alex A. (1994), Tesing Long-Run Neuraliy: Empirical Evidence for G7 Counries wih Special Emphasis on Germany, Carnegie- Rocheser Conference Series on Public Policy, 41, pp Yamada, Kazuo (1997), Nihon ni oeru Kahei no Choi Churisusei (Long-Run Moneary Neuraliy in Japan), Osaa Universiy Economics Journal, 46(3), pp (in Japanese). Zivo, E. and D. W. K. Andrews (1992), Furher Evidence on he Grea Crash, he Oil-Price Shoc, and Uni-Roo Hypohesis, Journal of Business & Economic Saisics, 10(3), pp