The Recovery of Mexican Investment after the Tequila Crisis: Basic Economics or "Confidence" Effects?

Transcription

1 The Recovery of Mexican Invesmen afer he Tequila Crisis: Basic Economics or "Confidence" Effecs? Daniel Lederman, Ana María Menéndez, Guillermo Perry and Joseph Sigliz The World Bank May, 2000 Absrac The sylized facs show ha invesmen was a he hear of he Mexican economic recovery afer 995. This sudy provides an empirical invesigaion of he deerminans of he growh rae of fixed invesmen in Mexico, wih special aenion given o he afermah of he crisis. I uses quarerly daa for Mos of he analysis is based on a srucural model of Mexican invesmen, bu addiional model-free analyses are also used o furher explore some key issues. The paper uses he Generalized Mehod of Momens esimaor wih insrumenal variables o deermine if he behavior of Mexican invesmen can be explained wih a sandard invesmen funcion. The only addiion o sandard models of invesmen in developing counries (which consider cos, scale and uncerainy facors) is he assumpion ha radable and non-radable oupu growh have differen muliplier effecs on Mexican invesmen growh. The auhors search for "confidence" effecs, o assess wheher he basic model of invesmen was operaional during imes of crisis. The main findings are ha he basic model does quie well in predicing he recovery of invesmen during 996 and 997; "confidence" effecs only seem o be presen during The sory ha emerges is ha Mexican invesmen growh declined during 995 due o he currency depreciaion's impac on he relaive price of capial goods, bu evenually he recovery was driven by he high muliplier effec from he growh of radable oupu and declining real ineres raes. Sigliz was Chief Economis and Vice-Presiden for Developmen Economics of he World Bank. Perry is Chief Economis and Direcor for Lain America and he Caribbean Region of he World Bank. Lederman and Menéndez are Economiss in he Office of he Chief Economis for Lain America and he Caribbean Region of he World Bank. The findings and opinions expressed in his paper should no be aribued o he World Bank, is Board of Direcors, or he counries which i represens. We are graeful o Luis Servén, Eliana Cardoso, William Maloney and wo anonymous reviewers for insighful commens on an earlier version of his paper.

2 INTRODUCTION The Mexican economy experienced a sharp conracion in 995, afer he Peso devaluaion of December 22, 994. The GDP growh rae reached an annualized rae of -6.2% in 995, and became posiive again during he firs hree quarers of 996. In 996 and 997 he Mexican economy grew a healhy raes of 5.% and 6.8% respecively. I has been suggesed by some auhors ha he decisive response of he fiscal and moneary auhoriies suppored by a generous financial package of billions of dollars announced in March 9, 995, were crucial for a rapid recovery of los invesor confidence. These observers furher argue ha hese policies promped a rapid sabilizaion of he currency and a urnaround in invesmen and economic aciviy. The aim of he presen paper is o examine if he behavior of invesmen in Mexico during can be explained by simple economic models. Secion I begins by esablishing he sylized facs regarding he conribuion of fixed invesmen o GDP growh rae since 993. A simple decomposiion analysis shows ha economic growh in Mexico afer he equila crisis recovered primarily due o he behavior of fixed invesmen. The invesmen growh rae declined drasically during he crisis year, bu subsequenly rose o levels exceeding pre-crisis levels. Hence he remaining analyses focus on he deerminans of he growh rae of fixed invesmen. Secion II focuses on possible deerminans of he fixed invesmen growh rae in Mexico, and proposes a sandard model of invesmen behavior, where he only addiion o sandard models is ha aggregae invesmen depends differenly on he oupu of radable and For example, he IMF (996: 25) wroe ha "In Mexico, igh fiscal and moneary policies helped reduce acual and expeced inflaion in he firs half of [995], conribuing o furher gains in confidence, declines in ineres raes, and he sabilizaion of he peso." 2

3 non-radable goods. To complemen his analysis, a echnical appendix included a he end of he paper develops a basic model of invesmen behavior by a represenaive firm, which shows ha he magniude of he income acceleraor effec depends on he capial inensiy of he firm's oupu. Secion III describes he daa used in he empirical implemenaion of he basic model. Secion IV reviews he esimaion sraegy, which relies on he General Mehod of Momens (GMM) esimaor o conrol for he possible endogeneiy of he explanaory variables. In addiion, he esimaion sraegy includes hree specificaion ess: A Wald es is used o assess wheher he secoral mulipliers are saisically differen; Hansen's J-saisic is used o deermine he validiy of he se of insrumenal variables used in he GMM regressions; and he Q-saisic is used o deermine if he model specificaions suffer from serial correlaion among he errors. In addiion, his secion presens resuls of seasonal uni roo ess for all variables in levels, following he mehodology proposed by Hylleberg e al. (990). Mos of he variables in levels seem o have eiher zero-frequency uni roos or some form of seasonal roos. The excepions are he real ineres raes, he volailiy of he real exchange rae (RER), and he RER iself. These resuls influenced he esimaion sraegy by saisically jusifying he use of he year-on-year differences of he variables. The GMM regression resuls are presened in Secion V. The basic invesmen funcion has as he dependen variable he growh of Mexican of fixed invesmen. The explanaory variables include he growh of he secoral GDPs, he domesic real ineres rae, changes in RER volailiy, variaions in he relaive price of capial, and he lagged dependen variable. This funcion is esimaed for he whole sample ( :Q3) and for a resriced sample ( ) used o conduc an ou-of-sample forecas. The basic model is hen exended along four 3

4 dimensions: We firs add he inflaion differenial beween radable and non-radable goods o he se of explanaory variables, and secondly we include he US real ineres rae. Third, we include wo measures of credi availabiliy (he growh raes of oal domesic credi and credi o he privae secor issued by commercial banks), in order o assess wheher Mexico has suffered from credi raioning. Finally, we es he sabiliy of he coefficien on he domesic real ineres rae during periods of crisis. The resuls indicae ha he radable secor has a higher muliplier effec on Mexican invesmen han he non-radable secor bu his resul is no always saisically significan. An imporan finding is ha he domesic real ineres is a significan deerminan of Mexican invesmen, bu he US prime lending rae deflaed by he inflaion rae of US impors is also significan. The relaive price of capial and he volailiy of he real exchange rae usually have negaive effecs on invesmen. The basic model provides a good explanaion of he growh of fixed invesmen in Mexico, especially in he afermah of he 995 crisis. Evidence of his is provided by he ou-of-sample forecas of he growh rae of fixed invesmen. An specificaion of his model ha includes variables ha capure credi availabiliy insead of real ineres raes presens similar resuls. We were unable o confirm or rejec he hypohesis of credi raioning bu when we include credi availabiliy insead of real ineres raes, his variable has a posiive effec on he rae of growh of fixed invesmen. Finally, resuls on he sabiliy of he coefficien on he domesic real ineres rae during periods of crisis indicae ha a confidence effec may have been presen only during he crisis bu no in 995. Secion VI presens model-free, or non-srucural, analyses ha address hree key quesions: () Was he radable secor a leading secor? (2) Wha was he dynamic relaionship beween fixed invesmen and he real exchange-rae? (3) Were here shor-erm subsiuion 4

5 effecs of RER variaions? The firs quesion is addressed by Granger causaliy ess ha aim o assess wheher he growh of radable oupu "Granger-causes" he growh of boh non-radable oupu and fixed invesmen. The resuls show a high level of endogeneiy on his group of variables and as a resul we canno conclude ha he radable secor was a leading secor. The second and hird quesions are addressed by examining he impulse-response funcions (IRFs) of invesmen and secor oupus o exogenous shocks of he RER. For example, in order o capure dynamic effecs of he real exchange rae on he growh rae of fixed invesmen, we esimaed a Vecor Auo-Regressive (VAR) sysem ha included several exogenous variables as well as he hree endogenous variables (he differenced log of he level of fixed invesmen, he relaive price of capial and he RER). The resuls show ha an appreciaion of he RER iniially has an expansionary effec, bu is effec becomes conracionary afer abou 5 quarers. An appreciaion also seems o have significan shor-erm subsiuion effecs in favor of he radable secor. More generally, he sory ha emerges is ha Mexican fixed invesmen growh declined during 995 due o he devaluaion's impac on he relaive price of capial and he negaive income effec, bu he evenual recovery was driven by he high muliplier effec from he radable secor and declining ineres raes. I. Growh Decomposiion: The Sylized Facs The decomposiion of he GDP growh rae ino he conribuions of is aggregae demand componens comes from a ransformaion of he basic macroeconomic ideniy: Y=CIX-M, where Y is he level of oupu, C is he sum of privae and public consumpion, I is he level of invesmen, X sands for expors, and M for impors. By firs differencing each elemen, dividing 5

6 by he ex-ane level of Y, and some simple manipulaions, he growh of oupu can be decomposed ino he conribuions of he growh raes of each of is componens: () Y/Y - = (C - /Y - ) C/C - (I - /Y - ) I/I - (X - /Y - ) X/X - - (M - /Y - ) M/M -. This expression can be easily expanded o include he conribuion of more disaggregaed demand componens. For example, he conribuion of gross invesmen can be furher decomposed ino he sum of he conribuion of fixed invesmen and invenory accumulaion. Figure shows he growh raes of GDP and fixed invesmen, as well as he laer's conribuion o GDP growh, on he basis of year-on-year (i.e., seasonally-adjused and annualized) growh raes. The graph shows ha fixed invesmen played a key role during and afer he equila crisis. Fixed capial accumulaion declined more han any oher demand componen during 995, bu also recovered briskly aferwards. 2 Moreover, he rae of growh of invesmen during was significanly higher han during Alhough hese sylized facs are illusraive, he growh decomposiion exercise suffers from he fac ha he growh raes of he demand componens and GDP are all endogenous. In he following secions we analyze he deerminans of he growh rae of fixed invesmen in a simulaneous equaions framework. II. A Mexican Invesmen Funcion The growh decomposiion exercise indicaes ha fixed capial formaion was a he hear of he economic recovery in Mexico afer 995. Also, here seems o be some evidence ha Mexican invesmen may be linked o he performance of he radable secor (Krueger and 2 A companion paper (Lederman e al. 2000) compares he conribuion of fixed invesmen wih hose of oher demand componens. 6

7 Tornell 999). We can model Mexican invesmen as a funcion of he oupu of radables and non-radables, where we expec o find differen secor muliplier effecs: (2) I = I m m I ( RIR, p, σ ) T Y T NT Y NT K RER where I is he (log of he) level of privae fixed invesmen a consan prices; I is a consan, minimum level; and he m s are he corresponding secor mulipliers. The las erm on he righhand side is he porion of he invesmen funcion ha is deermined by cos facors, including he real ineres rae (RIR), he relaive price of capial goods (p K ) and an uncerainy variable, which we idenify wih he volailiy of he real exchange rae ( σ RER ). This simple model is broadly consisen wih sandard empirical models of invesmen behavior in developing counries (see Rama 993), excep for he assumpion regarding he wo secors' mulipliers. The Appendix presens a sylized model of a represenaive firm's invesmen behavior, which highlighs he effec of firm's facor inensiy of producion on is income muliplier. The model also shows how invesmen cos facors (such he discoun rae and relaive price of capial) affec he firm's decisions. According o Servén (998), here are no predominan heoreical argumens o expec a paricular sign for he relaionship beween uncerainy measures and invesmen. Noneheless, in his empirical exercise his auhor finds a srong negaive effec of real exchange-rae uncerainy on invesmen-oupu raios in a cross-counry panel framework. This effec could be an imporan explanaion of he fall of invesmen during 995 in Mexico, and even of he downfall of invesmen during he deb crisis of Also, i is necessary o consider his effec in order o be able o isolae he direc effec of changes in he level of he real exchange rae on 7

8 Mexican fixed capial formaion. Moreover, he consideraion of RER uncerainy is consisen wih wo plausible assumpions. Firs, domesic invesors can be risk-averse, hus uncerainy may adversely affec privae invesmen. Second, i is also possible ha a leas porions of privae invesmens are irreversible and conribue o sunk coss. Under hese circumsances, macroeconomic uncerainy can be associaed wih swings in he value of privae firms, hus hampering producive invesmen by firms (Pyndick 988). If he radables secor is a leading secor, hen he uncerainy of he RER could also have an indirec effec on invesmen hrough is effec on he oupu of radables. As argued by Maloney and Azevedo (995), uncerainy abou expeced reurns of producing for domesic versus expor markes will affec he composiion of oupu. If he oupu mulipliers are consan over ime, hen in erms of growh raes, or in differences of he logs, he invesmen funcion can be re-wrien as follows: (3) g = m g m g g( RIR, g, σ ) I T T NT NT pk RER where he g s denoe growh raes of he corresponding variables. In his specificaion, he mulipliers are analogous o he sandard income acceleraors ha are sandard in empirical invesmen funcions (Rama 993, Servén 998). The main complicaion of his funcion is ha he growh raes of Y T, Y NT, and PK are deermined simulaneously by cerain common facors. For example, he growh rae of he radable secor can be expressed as a funcion of changes in he real exchange rae, and foreign demand (mosly US income in he case of Mexico): 8

9 (4) g = g( g, g ) T RER USY In conras, he growh of non-radable income can be specified as a funcion of relaive prices, such as he real exchange rae and he real ineres rae, plus domesic aggregae demand facors: (5) g = g( g, RIR, g, g, g ) NT RER T I G where g G is he growh rae of real governmen consumpion. In addiion, as is common in developing counries ha impor capial goods, Mexican invesmen could be sensiive o changes in he relaive price of capial goods. In fac, Mexico's average raio of impors of capial goods o gross domesic fixed invesmen during was 37.4 percen. 3 The growh rae (or changes in he log) of he relaive price of capial will, in urn, depend on changes in impor ariffs and he real exchange rae: (6) g = g( g, g ) pk ariff RER In a previous sudy of Mexican invesmen during he crisis, Warner (994) found ha he fall of invesmen was indirecly associaed wih a decline in Mexico's erms of rade. In our model, his effec is capured hrough he variaions in he RER. Following Warner, in he empirical implemenaion of he invesmen funcion we use a measure of Mexico's erms of rade as an insrumenal variable. 9

10 III. The Daa The empirical analyses rely purely on publicly available daa, primarily from he Mexican saisical agency (INEGI) and he Inernaional Moneary Fund. A descripion of he variables (and heir sources) is presened in Table. All he daa has a quarerly frequency and covers he period beween 980 and he hird quarer of 999. The dependen variable. As menioned, he main variable o be explained is he growh rae (e.g., log differences) of he quarerly level of gross fixed invesmen a consan prices. The daa is available hrough he World Wide Web in INEGI's homepage. The explanaory variables. The growh of radable and non-radable oupu, as well as he relaive price of capial, appear as explanaory variables of he growh of fixed invesmen. As a proxy for he oupu of radables we use INEGI's GDP series for manufacures, mining and agriculure. For non-radables, we use consrucion, energy and waer services. All series were expressed a consan prices in local currency unis. Regarding he relaive price of capial, we use he raio of he price index of capial goods (which is one of he componens used by he Mexican auhoriies o calculae heir producer's price index) relaive o he consumer price index. The empirical invesmen funcion includes a "naïve" measure of volailiy of he real exchange rae on he righ-hand side, as proposed by Servén (998). This variable measures he sandard deviaion of he change in he naural logarihm of he real exchange rae during six monh periods. 4 3 Serven (999) shows ha he average for developing counries in 990 was 3. percen and 29.3 percen for indusrial counries. 4 See daa descripion in Table. 0

11 Real ineres raes are expeced o have a negaive effec on invesmen. Ye i is commonplace o find no significan relaionship beween real ineres raes and invesmenoupu raios in cross-counry sudies (Agosín 996, Servén 998), as well as in sudies of Mexican invesmen (Warner 994). 5 One possible explanaion for his common finding is ha i is difficul o ascerain which real ineres rae is he one being used by domesic borrowers, and herefore he usual average raes repored by Cenral Bank may no provide an accurae indicaor of he real cos of borrowing. In he Mexican case, for example, i is possible ha large Mexican firms acually finance heir invesmen aciviies by apping U.S. financial markes direcly (Krueger and Tornell 998). For his reason, he basic empirical model is exended o consider boh he average quarerly domesic rae and he corresponding U.S. rae. As a proxy for he domesic nominal ineres rae, we use he money marke rae as repored by he IMF's Inernaional Financial Saisics. For he U.S. rae, we use he prime lending rae. To derive he real ineres raes, he money marke rae was deflaed by he inflaion of consumer prices, and he U.S. rae was deflaed by he inflaion of U.S. impor prices. In an exension of he basic model, we also include he difference beween he inflaion of radable and non-radable oupu o assess wheher variaions in his relaive price have an effec on Mexican fixed invesmen. As a proxy of he uni price of radable and non-radable goods we use he corresponding GDP deflaors. In searching for "confidence" effecs, we exend he basic model o include crisis dummy variables ineraced wih he domesic real ineres rae. During he period under analysis, Mexico had wo crisis: he deb crisis (982:Q3 o 983:Q4) and he Tequila crisis (995:Q o995:q4). 5 Mussalem (989) did find a negaive relaionship beween real domesic ineres raes and privae invesmen in Mexico wih annual daa for This auhor used he "ex-pos afer ax average real ineres rae on banking insrumens."

12 To explore he possibiliy ha confidence effecs were presen during hese imes, we inroduce hree dummy variables for crisis periods one a a ime, in each case ineraced wih he domesic ineres rae. The firs crisis dummy variable is riggered during boh periods, while he oher wo are specific o each crisis. The analysis assesses wheher he coefficien of he real ineres rae changes sign (i.e., becomes posiive) during he crises. Tha is, he presence of confidence effecs is acceped when he esimaed coefficien of he ineraced erm is posiive, saisically significan, and larger in absolue value han he esimaed coefficien of he domesic real ineres rae. IV. Esimaion Sraegy and Specificaion Tess One of he firs decisions aken was o choose beween he esimaion of he invesmen funcion in levels (equaion ) and he funcion in difference (equaion 3). An imporan consideraion is wheher he variables in levels exhibi some form of uni roos. Wih his in mind, we conduced he zero-frequency and seasonal uni roo ess suggesed by Hylleberg e al. (990). The so-called HEGY OLS regression esimaed for each variable was: n i= (7) y = c α π z π z π z π z β y i µ ε The c is a consan and sands for he ime rend. The µ is a seasonal effec. I was an empirical quesion wheher he daa generaion process for each variable includes hese wo elemens. The inclusion of hese elemens affecs he disribuion of he -saisics for he π 's, which are he parameers o be esed. The z's in (7) are lagged polynomial represenaions of each variable (y) in levels: 2

13 (8) z = y y 2 y 3 y 4 ; (9) z 2 = y y 2 y 3 y 4 ; (0) z 3 = y 2 y 4 ; and () z 4 y y 3. = There are no seasonal uni roos if π 2 and eiher π 3 or π 4 are differen from zero. For a series o be saionary (i.e., wihou any uni roo), eachπ mus be differen from zero. A series has no annual uni roo if he F-saisic for he null haπ = π 0 is saisically significan. Hylleberg 3 4 = e al. (990) provide he criical values for he -saisics corresponding o he esimaed π 's and for he aforemenioned F-saisic. The criical values depend on wheher he esimaed models include he inercep and/or he deerminisic ime rend and/or seasonal dummies. The HEGY uni roo es resuls are presened in Table 2. Only for a few variables we can rejec he null hypohesis ha hey have some form of uni roo. These are: he real ineres raes, he US real ineres raes, he real exchange rae (and is difference), and RER volailiy (and is difference). Given his evidence, he invesmen funcion o be esimaed is he one specified in year-on-year differences of he variables. The building block of he model is he invesmen funcion expressed in equaion (3). The model is "naïve," because i neglecs money demand and/or supply, spreads, deb and oher variables relaed o he financial secor. Our objecive is simply o assess wheher he recovery of Mexican invesmen afer he Tequila crisis can be explained wih real variables. In urn, we also es wheher he basic model of invesmen behavior is able o predic he recovery of invesmen growh when i is esimaed wih daa up o 994. In oher words, we also presen he ou-ofsample forecas of invesmen behavior for , and compare i o he acual observaions. 3

14 The economeric models o be presened below were esimaed using he General Mehod of Momens (GMM) esimaor. This approach allows for he esimaion of he srucural coefficiens using as insrumens any exogenous and pre-deermined variables, including he lagged dependen variables. One advanage of hese esimaes is ha hey are robus o heeroskedasiciy and auocorrelaion of unknown form. However, esing he validiy of he momen condiions is crucial o ascerain he consisency of GMM esimaes. If he regression specificaion "passes" he es, hen we can safely draw conclusions aking he momen condiions as given. For insance, we can accep he saisical and economic significance of he esimaed coefficien of he growh of he GDP of radable goods and services as effecs going from his variable o invesmen; alernaively, we can safely discard he possibiliy ha his effec is due o some omied variable correlaed wih he insrumens. The specificaion es we use is he es of overidenifying resricions inroduced in he conex of GMM by Hansen (982) and furher explained in Newey and Wes (987). Inuiively, he fac ha we have more momen condiions (insrumens) han parameers o be esimaed means ha he esimaion could be done wih fewer condiions. The GMM approach uses his fac o esimae he error erm under a se of momen condiions ha excludes one insrumenal variable a a ime, and hen he procedure ess he validiy of he null hypohesis ha each esimaed error erm is uncorrelaed wih he insrumenal variable excluded from he corresponding insrumen se. In oher words, he null hypohesis of Hansen's es is ha he overidenifying resricions are valid, ha is, he insrumenal variables are no correlaed wih he error erm. The es saisic is simply he sample size imes he value aained for he objecive funcion a he GMM esimae (called he J- saisic). Hansen's es saisic has a chi-square disribuion wih degrees of freedom equal o he number of momen condiions (insrumens) minus he number of parameers o be esimaed. 4

15 The level of oupu is expeced o have a muliplier effec on he level of invesmen. However, in his case, he radable secor is expeced o be a leading secor, perhaps due o differences in he capial inensiy of producion -- see he model presened in he Appendix. In oher words, we expec ha he magniude of he muliplier from he radable secor's oupu will be larger han he effec from he non-radable secor. To es his hypohesis direcly, we rely on Wald ess of he specificaion resricion ha he secoral mulipliers are equal (i.e., m = m ). T NT More specifically, he ables wih regression resuls repor he p-value of he F-saisic for he null hypohesis ha he secoral mulipliers are equal. Therefore a low p-value (below 0.0) indicaes ha he mulipliers are no equal. Due o he possibiliy ha we have omied variables in he specificaion of he Mexican invesmen funcion, he presenaion of he regression resuls includes he p-value of he Q- saisic for serial correlaion. The Q-saisic is a es of he null hypohesis ha he correlaion among error erms is zero. Therefore a high p-value (greaer han 0.0) suppors he specificaion of he model under examinaion. The ables wih our regression resuls presen he p-value of he Q-saisic corresponding o he correlaion beween he curren error and is counerpar lagged 4 quarers. This choice was moivaed by he quarerly frequency of he daa, which may impose seasonal correlaion among he errors. For his reason, mos of our specificaions of he empirical model also include seasonal dummies. In mos of he models esimaed in his paper, he insrumens for endogenous variables are he lagged values of he variables hemselves, plus a se of exogenous variables ha includes he U.S. real ineres rae, he growh of U.S. GDP, and variaions in Mexico's erms of rade. A difficuly arises when choosing he number of lags of he endogenous variables ha are used as insrumenal variables. We considered ses of, 4 and 8 lags for all insrumenal variables in 5

16 view of he seasonaliy of he quarerly ime series. Regressions wih insrumens wih 8 lags could no be run for all specificaions of he model due o he limied number of observaions relaive o he number of insrumens. Wih one lag, several specificaions did no pass neiher Hansen s es for overidenificaion nor he Q-es for serial correlaion. Consequenly we were lef wih 4 lags of he endogenous variables o be used as heir insrumens. However, a specificaion wih wo lags for he insrumenal variables is used for he ou-of-sample forecas exercise, because he reduced sample covering he period beween 980 and he end of 994 included only 50 observaions (compared o 67) and hence we reduced he number of insrumens o 5 from 25. Our basic invesmen funcion has as he dependen variable he growh of Mexican fixed invesmen. The explanaory variables include he growh of he secoral GDPs, he domesic real ineres rae, RER volailiy, variaions in he relaive price of capial, and he lagged dependen variable. This funcion is esimaed for he whole sample ( :Q3) and for a resriced sample ( ) used o conduc an ou-of-sample forecas. The basic model is hen exended along four dimensions: We firs add he inflaion differenial beween radable and non-radable goods o he se of explanaory variables, and secondly we also include he US real ineres rae. Third, we include wo measures of credi availabiliy (he growh raes of oal domesic credi and credi o he privae secor issued by commercial banks). Finally, we es he sabiliy of he coefficien on he domesic real ineres rae during periods of crisis. We repor he hree specificaion ess (Hansen's J es for overidenifying condiions, he Wald es for equaliy of he secoral mulipliers, and he Q es for serial correlaion) for all he exensions of he basic model. The following secion presens he esimaion resuls. 6

17 V. Resuls A. Basic Resuls The J-es of he invesmen funcion (firs column, las row in Table 3) shows ha he insrumenal variables are no correlaed wih he error erm: he p-value of he null hypohesis ha he insrumens are no correlaed wih errors is Therefore we can inerpre he coefficiens as being he impac from he explanaory variables o he dependen variable. An imporan resul is ha he muliplier effec of he radable secor on invesmen clearly surpasses ha of he non-radable secor. Moreover, he p-value of he F-saisic for he null hypohesis ha he wo mulipliers are equal is 0.0. The volailiy of he real exchange rae has a negaive coefficien, bu he magniude of his effec is no significanly differen from zero. The relaive price of capial (e.g., he raio of he price index of capial goods over he CPI) has a negaive sign and is significan a he 5 percen level. The coefficien implies ha an increase of one percen in he growh rae of his relaive price "causes" a reducion of 0.33 percen in he growh of fixed invesmen. The relaionship beween he variaions in he RER and he invesmen growh is examined furher in Secion IV below. The coefficien of he domesic real ineres rae has he expeced negaive sign in he esimaion of he invesmen funcion using he whole sample (firs column in Table 3), and i is saisically differen from zero. The esimaed coefficien implies ha a one percenage poin increase in he real ineres rae leads o a 0.08 percenage poins decline in he growh rae of fixed invesmen. The resuls for he same regression esimaed only for he period is presened in he second column of Table 3. This model uses a reduced number of insrumenal variables due o he also reduced sample size. The resuls are quie similar o hose repored for he full sample. An imporan difference, however, is ha he esimaed negaive coefficien of he real 7

18 ineres rae is larger han he one repored in he firs column. Wheher he coefficien on he domesic real ineres rae is significanly differen during crisis episodes will be explored in Secion VD below. Anoher noable difference is ha he esimaion wih he resriced sample shows a Wald es ha does no rejec he null hypohesis ha he secoral mulipliers are equal -- is p-value is now To provide a visual illusraion of he capaciy of his basic model o explain he behavior of Mexican invesmen during and afer he 995 crisis, Figure 2 presens he forecased and acual values for he year-on-year growh rae of Mexican fixed invesmen. The forecas presened here uses he dynamic approach o forecasing which calculaes muli-sep forecass saring from he firs period in he forecas sample (firs quarer of 995). 6 Figure 2 shows ha his naïve model quie successfully explains he behavior of Mexican fixed invesmen, especially afer 995. The Theil inequaliy coefficien comparing he forecas wih he observaions is relaively low a 0.7 (a zero would indicae a perfec fi). More imporanly, almos 80 percen of his inequaliy is due o he covariance beween forecas and acual errors (or deviaions from he corresponding means). This means ha he lion's share of he inequaliy beween he forecas and he acual observaions is due o unsysemaic error covariance. A his poin, i remains o be explored wheher he recovery of Mexican fixed invesmen growh was due o he financial linkages wih he U.S., as argued by Krueger and Tornell (999). This quesion is furher explored in he following secion. 6 This mehod is less accurae han he saic approach, which calculaes a sequence of one-sep-ahead forecass using acual, raher han forecased values for he lagged values of fixed invesmen growh. 8

19 B. Domesic and U.S. Ineres Raes, and Inflaion Differenials Table 4 shows he regression resuls for exended versions of he basic model ha include he difference beween radable and non-radable inflaion raes and he U.S. real ineres rae. These exercises aim o examine wheher he domesic ineres rae (he money marke rae) deflaed by domesic consumer price inflaion is he discoun rae ha is more ighly associaed wih Mexican invesmen behavior. Boh specificaions pass he es for serial correlaion. Also, in boh cases, he muliplier from he radable secor remains significanly higher han ha of he non-radable secor. All he oher explanaory variables reain heir expeced signs, and he inflaion differenial comes ou wih a negaive bu insignifican coefficien. When he U.S. real ineres rae (ha is, he prime lending rae deflaed by he inflaion of US impor prices) is inroduced ino he model, as shown in he second column of Table 4, he domesic ineres rae mainains is significance. The negaive and significan coefficien is also presen for he U.S. rae. This resul provides srong evidence in favor of he Krueger-Tornell hypohesis ha access o he U.S. financial marke by Mexico's firms operaing in he radable secor was a key feaure of he economic recovery afer 995. C. Conrolling for Credi Availabiliy Under credi raioning, invesmen could be driven by credi availabiliy and no by real ineres raes (Sigliz and Weiss 98). Table 5 presens he resuls of economeric models ha exend he basic model o examine he effec of credi availabiliy. The firs column shows he resuls for he specificaion ha includes he growh rae of oal domesic credi a consan prices o he model; he second column shows he resuls for he model ha includes he growh rae of credi o he privae secor issued by privae commercial banks. These variables conrol for 9

20 he quaniy of credi bu hey could reflec variaions in eiher supply and/or demand. Neiher specificaion passes he Q es for serial correlaion, hus indicaing ha he inroducion of credi quaniies produces a mis-specified model. Due o his specificaion problem, he esimaed coefficiens canno be easily inerpreed, because serial correlaion creaes biases in he coefficiens. The hird and fourh columns show he resuls for he specificaions ha include he growh rae of oal domesic credi a consan prices and he growh rae of credi o he privae secor issued by privae commercial banks (also a consan local prices) as explanaory variables. Resuls for he Q es show ha boh specificaions pass he es for serial correlaion and herefore hese models do no have specificaion problems. The availabiliy of boh ypes of credi has posiive coefficiens ha are significanly differen from zero. These coefficiens imply ha: () an increase of one percen in he growh rae of oal credi leads o an increase of 0.03 percen in he growh of fixed invesmen, and (2) an increase of one percen in he growh rae of credi o he privae secor leads o an increase of 0.04 percen in he growh of fixed invesmen. In boh cases, he muliplier from he radable secor remains significanly higher han ha of he non-radable secor, and he volailiy of he RER presens negaive coefficiens ha are significanly differen from zero. The relaive price of capial, on he oher hand, has posiive coefficiens bu wih magniudes no significanly differen from zero. Credi availabiliy could be considered an imporan deerminan of fixed invesmen in Mexico and he decrease in he availabiliy of credi during he crises negaively affeced he growh of fixed invesmen. However, we canno empirically disinguish beween he effecs of high real ineres raes and low credi availabiliy, because he model ha includes boh variables suffers from serial 20

21 correlaion as menioned earlier. Therefore we were unable o confirm or rejec he hypohesis ha Mexico has suffered from credi raioning. Afer having esed he basic invesmen growh funcion for Mexico by adding, sequenially, alernaive ineres raes, inflaion raes, and credi availabiliy, he following secion examines he sabiliy of he esimaed ineres-rae coefficien during periods of crisis. In oher words, he following models search for evidence in favor of so-called "confidence effecs." D. Searching for "Confidence" Effecs This secion aims o deermine wheher here is any evidence of he exisence of "confidence" effecs during imes of crisis. A confidence effec would be presen in imes when increases in ineres raes are associaed wih increases in he growh rae of invesmen. This phenomenon can heoreically arise when high ineres raes signal ha he moneary auhoriies are willing o defend he value of he currency, hus proecing he ne worh of firms wih liabiliies denominaed in foreign currency. As explained by Bernanke and Gerler (989), if invesmen is limied by he wealh of firms (i.e., hey canno be infiniely leveraged), hen a shock ha deerioraes heir ne worh will lead o a fall of invesmen. Krugman (999) suggess ha his effec may have aided he recovery of Mexico afer 995 and Korea afer 998. The resuls are presened in Table 6. The esimaion sraegy is o include he crisis dummy variables boh as mean shifers as well as ineraced wih he domesic ineres rae. This approach is superior o he inclusion of he ineraced erm alone, because i prevens capuring a shif in he 2

22 mean growh rae of fixed invesmen during crises ha may no be due o a srucural change in he ineres-rae coefficien. 7 In he firs column, he coefficien for he ineraced erm wih he dummy variable for boh crises (982:Q3-983:Q4 and 995) shows a posiive and significan sign. The large size of he coefficien implies ha he overall effec of increases in he real ineres rae is acually posiive during he crisis periods. For he res of he sample, he model predics a negaive and significan effec of real ineres raes on invesmen. All he oher coefficiens for he invesmen funcion have he expeced signs, excep RER volailiy. However, in his specificaion he esimaed coefficien of he growh of radable GDP is significanly smaller han he growh of non-radables. However, he Q es indicaes ha his is no a valid specificaion. The second and hird columns of Table 6 show he resuls for specificaions wih each ineraced crisis dummy inroduced separaely. Resuls for he ineraced dummy of he 982 crisis are consisen wih hose discussed above; ha is, here is evidence of a confidence effec. On he oher hand, resuls for he 995 crisis show a differen sory: he sign of he ineraced dummy coefficien is negaive. This suggess ha here was no confidence effec during his crisis. Acually, he resuls would indicae ha during his crisis invesmen was even more sensiive o increases in ineres raes han during normal imes. However, he Q es for serial correlaion indicaes ha his model is also wrongly specified. Indeed, only he second column in Table 6 presens a model suppored by all of our specificaion ess. Therefore he main conclusion ha can be derived from his evidence is ha we canno rejec hypohesis of he 7 In any case, regressions wihou he crisis mean shifer produced very similar resuls. 22

23 presence of a srucural break in he ineres-rae coefficien during he crisis. 8 In conras, we can rejec he hypohesis ha here was a similar break during VI. Non-srucural Analyses This secion examines he relaionship among key variables in model-free frameworks. The purpose is o aemp o answer hree key quesions ha are essenial for he inerpreaion of he srucural model resuls. A. Was he radable secor a leader? The resuls of he srucural model presened in he firs column of Table 3 indicaed ha he growh rae of radable oupu was a leader in he sense ha i is a significan deerminan of fixed invesmen wih a higher acceleraor effec han non-radable GDP. Here we ask ourselves if hese resuls are suppored in a framework ha does no impose srucural assumpions on he variables. For his purpose we conduced pairwise Granger causaliy ess, which are presened in Table 7. The resuls indicae ha we canno rejec he possibiliy ha he growh of radable oupu was caused by eiher he growh of fixed invesmen or he growh of non-radables oupu. Moreover, we canno confidenly rejec he possibiliy ha all hree variables are caused by he ohers. The use of he GMM esimaor for he empirical fixed invesmen funcion is clearly jusified by hese resuls. Neverheless, we canno conclude ha he radable oupu 8 However, we acknowledge ha his break may be due o oher effecs ha are no really "confidence effecs." For example, he banking sysem was naionalized during ha ime, and he change in he coefficien's sign during ha crisis may be due o ineres-rae conrols. 9 We also acknowledge ha we canno es for confidence effecs wih a duraion of less han one quarer, given he frequency of he available daa. 23

24 growh was a leader in he sense ha i precedes variaions in fixed invesmen and non-radables oupu. B. Wha was he dynamic shor-erm relaionship beween he real exchange-rae and invesmen growh? None of our previous models examine he direc relaionship beween he RER and fixed invesmen. Here we use he impulse-response funcion (IRF) derived from Vecor Auo- Regressions (VAR) o sudy he relaionship beween fixed invesmen and he RER. We ran a VAR wih he growh raes of fixed invesmen, he relaive price of capial goods, and he real exchange rae as endogenous variables. The choice of hese variables is inspired on exising evidence for Mexico ha indicaes ha he relaive price of capial may be an imporan channel hrough which variaions in he real exchange rae (and is deerminans, such he erms of rade) affec domesic invesmen (Warner, 994). We also included he following exogenous variables ha are likely o affec hese hree variables: he growh of U.S. GDP, variaions in public consumpion, variaions in he erms of rade, he U.S. real prime lending rae, he OPEN and NAFTA dummy variables, and a consan. Figure 3 shows he corresponding IRF assuming ha he real exchange-rae is more exogenous han he relaive price of capial, which, in urn, is assumed o be more exogenous han fixed invesmen. Twelve lags of he endogenous variables were included in he VAR, because he Akaike, Scwarz and Log Likelihood ess all indicaed ha his was he bes disribued lag specificaion, when compared o, 4 and 8 lags. The IRF illusraes he effec on invesmen of a one sandard deviaion innovaion in he real exchange rae. The resuls show ha he response of invesmen o an appreciaion of he real exchange 24

25 rae is posiive a firs bu becomes negaive and significan during he second year, afer approximaely five quarers. The finding ha an appreciaion of he RER in Mexico is associaed wih an expansion of he economy is consisen wih he conclusions of oher recen empirical sudies. For example, Kamin and Rogers (998) also find ha appreciaions (depreciaions) have posiive (negaive) effecs on oupu, bu hese auhors argue ha here are muliple channels hrough which his effec akes place, including is effec on governmen spending, moneary aggregaes, and capial flows. In fac, Kamin and Rogers only enaively rejec he hypohesis ha he channel is he inflaion rae. Four quarers afer he simulaed appreciaion he growh of fixed invesmen ends o decline. A plausible inerpreaion of hese dynamic effecs is ha an appreciaion has an iniially posiive effec ha could be due o income effecs (or posiive ne worh effecs) and he reducion of he relaive price of capial. 0 Laer, as he subsiuion effec akes hold, and he radable secor's influence on invesmen predominaes, invesmen growh declines as a consequence of he real appreciaion of he currency. C. Were here shor-erm subsiuion effecs of RER variaions? Based on pairwise VARs, he impulse-response funcions presened in Figures 4A-C aemp o evaluae he inerpreaion presened above by looking a he effec of he RER on he relaive growh of he radable secor. Figures 4A hrough C are differen only in he number of lags used in he VAR, because he Akaike, Schwarz and Log Likelihood ess provided suppor o differen lag specificaions. Figure 4A shows he IRF based on he one-lag specificaion ha 0 The companion paper (Lederman e.al. 2000) shows IRF for he effec of an appreciaion on secoral GDP growh raes. I finds ha an appreciaion ends o emporally raise he growh rae of he radable and non-radable secors, his effec is larger for non-radables. We inerpre his differenial as evidence of ransiory subsiuion effecs. 25

26 was suppored by he Schwarz crieria; 4B is based on eigh lags ha was suppored by he Akaike crieria; and 4C was derived from a VAR ha used 2 lags of he endogenous variables, which was suppored by he log likelihood es. All hree IRFs provide evidence ha variaions in he RER did have significan subsiuion effecs in he sense ha i affeced he composiion of oupu: an appreciaion decreases he rae of growh of radables over non-radables. CONCLUSIONS Since he erupion of he so-called Tequila crisis of 995, and afer he Asian crises of 997, much has been wrien abou he causes of financial crises in developing counries. Ye much less aenion has been given o he evoluion of he producive economy afer hese crises. The sylized facs abou he Mexican recovery show ha fixed invesmen was a he hear of he recovery. This sudy provides an empirical invesigaion of he deerminans of he growh rae of fixed invesmen in Mexico, wih special aenion given o he afermah of he so-called Tequila crisis of 995. Perhaps he only addiion o basic models of invesmen in developing counries (which consider cos, scale and uncerainy facors) is he assumpion ha radable and nonradable oupu growh have differen muliplier effecs on Mexican invesmen growh. The main finding is ha he basic model does quie well in predicing he recovery of invesmen during 996 and 997; evidence of "confidence" effecs only seem o be presen during , and high ineres raes ended o hamper invesmen during mos of he period under invesigaion. Naïve measures of real exchange-rae uncerainy were no robus deerminans of Mexican fixed capial formaion. Finally, he impac of variaions of he real exchange rae seem 26

27 o be dominaed by income effecs iniially, bu subsiuion effecs predominae in he medium erm. The sory ha emerges is ha Mexican invesmen growh declined during 995 due o he currency depreciaion's impac on he relaive price of capial goods and he income effec, bu evenually he recovery was driven by he high muliplier effec from he growh of radable oupu; declining domesic ineres raes and low lending raes in he U.S. See Perry and Lederman (999) for a comparison of he afermah of Lain American and Asian Crises in he 990s. 27

28 REFERENCES Agosín, Manuel "Relación de dos regiones: La inversión en la América Laina y en el Asia Orienal." El Trimesre Económico 63: Alfaro, Samuel, and Javier Salas "Evolución de la balanza comercial del secor privado en México: Evaluación con un modelo economérico." El Trimesre Económico 59: Bernanke, Ben, and Mark Gerler "Agency Coss, Ne Worh, and Economic Flucuaions." American Economic Review 79: 4-3. Furman, Jason, and Joseph E. Sigliz "Economic Crises: Evidence and Insighs from Eas Asia." Brookings Papers on Economic Aciviy 2: -36. Hansen, Lars P. 982, Large Sample Properies of Generalized Mehod of Momens Esimaors. Economerica 50: Hylleberg, S., R.F. Engle, C.W.J. Granger, and B.S. Yoo "Seasonal Inegraion and Coinegraion." Journal of Economerics 44: Inernaional Moneary Fund World Economic Oulook. Washingon, DC: Inernaional Moneary Fund. Kamin, Seven and John Rogers Oupu and he Real Exchange Rae in Developing Counries: An Applicaion o Mexico. Board of Governors of he Federal Reserve Sysem Inernaional Finance Discussion Papers #580. Washingon, DC. Krueger, Ann and Aaron Tornell The Role of Bank Resrucuring in Recovering from Crisis: Mexico Naional Bureau of Economic Research Working Paper Series # Cambridge, Massachuses. Krugman, Paul "Analyical Aferhoughs on he Asian Crisis." Mimeographed. hp://web.mi.edu/krugman/www/minicris.hm Lederman, D., A.M. Menendez, G. Perry, and J. Sigliz "Mexico: Five Years afer he Crisis." Paper prepared for presenaion a he Annual Bank Conference on Developmen Economics, April 6, The World Bank, Washingon, DC. Maloney, William and Rodrigo Acevedo Trade Reform, Uncerainy and Expor Promoion: Mexico Journal of Developmen Economics 48:67-89 Musalem, Albero Privae Invesmen in Mexico: An Empirical Analysis. Policy, Planning and Research Working Papers # 83. Washingon, DC, The World Bank. Newey, Whiney and Kenneh Wes Hypohesis Tesing wih Efficien Mehod of Momen Esimaion. Inernaional Economic Review 28: Perry, Guillermo, and Daniel Lederman Adjusmens afer Speculaive Aacks in Lain America and Asia: A Tale of Two Regions? Washingon, DC: The World Bank. Pindyck, Rober "Irreversible Invesmen, Capaciy Choice and he Value of he Firm." American Economic Review 78: Rama, Marín "Empirical Invesmen Equaions for Developing Counries." In Sriving for Growh afer Adjusmen: The Role of Capial Formaion, edied by L. Servén and A. Solimano. Washingon, DC: The World Bank. Servén, Luis Macroeconomic Uncerainy and Privae Invesmen in LDCs: An Empirical Invesigaion. World Bank Policy Research Working Paper No Washingon, DC, The World Bank. Servén, Luis "Terms-of-Trade Shocks and Opimal Invesmen: Anoher Look a he Laursen-Mezler Effec." Journal of Inernaional Money and Finance 8:

29 Radele, Seven, and Jeffrey D. Sachs "The Eas Asian Financial Crises: Diagnosis, Remedies, Prospecs." Brookings Papers on Economic Aciviy :-74. Sigliz, Joseph E., and Andrew Weiss. 98. Credi Raioning in Markes wih Imperfec Informaion. American Economic Review 7: Warner, Andrew M Mexico's Invesmen Collapse: Deb or Oil? Journal of Inernaional Money and Finance 3:

30 Appendix: Opimal Invesmen Rules and he Capial Inensiy of Oupu This secion presens a simple model of he decision o inves by a represenaive firm. The purpose of he model is o highligh he role played by he capial inensiy of oupu in he deerminaion of he opimal level of invesmen a a given poin in ime. More specifically, i aemps o show how he magniude of he income acceleraor depends on he capial inensiy of producion. The discussion begins wih a general framework and simplifying assumpions are inroduced sequenially. In general, managers make invesmen decisions in order o increase he marke value of he firm. The change in he marke value of he firm can be expressed as a funcion of curren profis and capial gains, minus he value of curren invesmen, plus he discouned value of fuure profis and capial gains. This can be expressed in erms of a simplified version of Rama's (993) general wo-period model as follows: () V = Z P K P K i K P I K K K I α K 2 ( ) p Q w L ( i ) The firs erm, Z, in equaion () represens he sum of curren profis and capial gains (losses). I is given a ime, and herefore i is considered o be a consan. The second erm in he righhand side corresponds o he value of curren invesmens, measured as a funcion of he curren nominal price of capial, K P, and is a convex funcion of he invesmen rae (i.e., he raio beween gross invesmen and he capial sock). The α parameer is assumed o depend on 30