Country Fund Discounts and the Mexican Crisis of December 1994: Did Local Residents Turn Pessimistic before International Investors?

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 563 September 1996 COUNTRY FUND DISCOUNTS AND THE MEXICAN CRISIS OF DECEMBER 1994: DID LOCAL RESIDENTS TURN PESSIMISTIC BEFORE INTERNATIONAL INVESTORS? Jeffrey A. Frankel and Sergio L. Schmukler NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to InternationalFinance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors.

ABSTRACT It has been suggested that Mexican investors were the “front-runners” in the peso crisis of December 1994, turning pessimistic before international investors. Different expectations abouttheir own economy,perhaps dueto asymmetric information, prompted Mexican investors to be the first ones to leave the country. This paper investigates whether data from three Mexican country funds provide evidence that supports the “divergent expectations” hypothesis. We find that, right before the devaluation, Mexican country fund Net Asset Values (driven mainly by Mexican investors) dropped faster than their prices (driven mainly by foreign investors). Moreover, we find that Mexican NAVS tend toGranger-cause thecountry fundprices. This suggeststhat causality,in some sense, flows from the Mexico City investorcommunity to theWall Streetinvestorcommunity.

CountryFundDiscountsandtheMexicanCrisisofDecember1994: DidLocalResidentsTurnPessimisticBeforeInternationalInvestors? Jeffrey A. Frankel and SergioL. Schrnuklerl I.Introduction “The available data show that the pressure on Mexico’s foreign exchange reserves during 1994, and inparticular just prior to the devaluation, came not from theflight offoreign investors, butfrom Mexican residents.” — International CapitalMarkets, InternationalMonetary Fund, 1995. “YOUstate that ‘thefirst to jlee were not fickle foreign investors but wellinformed Mexicans. I have yet to see a serious methodology that in effect distinguishes between national andforeign por~olio investors.” — Letter to theEditor, TheEconomist, (11/11/95), Jaime Serra, Former Mexican FinanceMinister. 1Theauthorsarerespectively:professorandgraduatestudentattheDepartmentofEconomics, Universityof Californiaat Berkeley.The paperwasfinishedwhileSergioSchmuklerwas an Intern in the Divisionof InternationalFinance,Boardof Governorsof the FederalReserve System.Wewouldliketo thankBradfordDeLong,CarmenReinhart,RichardLyons,Maurice Obstfeld,ThomasRothenberg,and the participantsof internationalfinance seminarsat the Federal ReserveBoard,the IMF ResearchDepartmentand the NBERSummerInstitutefor helpfulcommentsand suggestions.We also thankCharlesKramerandR. ToddSmith,of the IMFResearchDepartment,andDonaldCassidy,ofLipperAnalytical,forthecountryfunddata. However,no onebesidesus is responsiblefor anyremainingerrors.Thispaperrepresentsthe viewsoftheauthorsandshouldnotbeinterpretedasreflectingthoseoftheBoardofGovernors of the FederalReserveSystemor othermembersof its staff.E-mailaddressesof the authors: frankel@econ.berkele.yeduandsergio@econ.berkele.yedu.

The Mexican crisis in December 1994 posed a question regarding how international financial markets work, among many others. It has been suggested that domestic residents were “closer to information” and thus had better, or at least different, expectations about local economic events in the pre-crisis period. The International Monetary Fund (IMF) in its annualCapitalMarkets Report (1995)expresses the view that “...resident investors in emerging market countries tend to be front-runners in a currency crisis...” (page 7). Under this hypothesis, local investors led the stampede out of Mexican assets in December 1994 much as theyhad done in the earlier crisis of 1982(engagingin massive capital flight at a time when U.S. banks were still pouring money into Latin America). Three Mexican closed-end country fundshave been established as vehicles to hold Mexican equities. They are the Mexico Fund (MXF), Mexico Equity and Income Fund (MXE), and Emerging Mexico Fund (MEF). The first one was established in 1981,and the other two in 1990. They offer a valuable opportunity to study the dynamics of the crisis.z The Net Asset Values (NAVS) of the funds are the aggregate value of the constituency equities, evaluated at local market prices, thoughtranslatedintoU.S. dollars. 2Countryfundsareideallysuitedtohelpinvestigateseveralquestions.Asa secondconcern,the crisis also generatednew interestin the contagioneffects of crises. In the presentpaper we concentrateon the “asymmetricinformation”hypothesis,whilewe studycontagionin Frankel andSchmukler(1996). 2

If markets were perfectlyefficient and internationallyintegrated,then the price of the fund would be equal to the NAV. However this is not the case. We argue that the price of the country fund, which is traded on Wall Street, reflects relativelybetter the information and expectationsheldbyinternationalinvestors,whiletheNAV, whichisdeterminedinMexico City, reflects relativelybetter the information and expectationsheld by local investors. In otherwords, the countryfund discount,which is thepercentagedifferencebetween the two prices, reflects the relative optimism of domestic versus international. A large discount indicates that domestic residents have relativelymore favorableexpectations. A premium indicates that foreigners have relatively more favorable expectations. The present paper focuses on what might variously be called the hypothesis of “divergent sentiments,” “heterogeneous expectations,” “asymmetric information,” or “closenessto information.”3 Anticipating the most interesting fact in this paper, Figures 1-3plot Mexican fund prices, NAVSand percentage discounts beforethe devaluation.They appearto supportthe claim of the IMF (1995), that Mexican investors were the front-runners in the crisis. The NAVSin Mexico City fell sharply relative to prices in New York in December 1994.In 3Frankel(1994b,p.254)containeda warning,basedonpremiainsuchcountryfundprices,thata repeatofthe1982crisismightbecominginLatinAmerica: “FluctuationsinthepremiumoftheU.S.priceofthefundoverthenetassetvaluecouldbe a measureoffluctuationsinthedifferenceinexpectationsof U.S.versuslocalinvestors. Formostof thesefundsthispremiumhasbeenhigher(orthediscounthasbeenlower) duringthe period 1990-1992than duringthe precedingthreeyears,suggestingbullish sentimentonthepartofforeigninvestors....MexicoandBrazilshowaclearlyhigherlevel ofrelativeU.S.investorconfidenceinthethreeyearsfrom1990...Ifourinterpretationof the datais correct,thattheyrepresentthe confidenceof U.S. investorsrelativeto local investors,thesefourgraphssuggesta possiblereplayof theperiodleadingup to 1982, whenLatinAmericanresidentsturnedpessimisticregardingtheirowncountrieswhileU.S. bankswerestillbullish.” ThesamepointwasmadeinFrankel(1994a,p.17). 3

Figure 2,the decline began two weeks before the devaluation. This seems to constitutethe sort of evidencethe existenceof which was questioned by Jaime Serra in the quote at the headofthispaper. Section II looks at the long-run and short-run relationships between the Mexican fund prices in New York, on the one hand, and the NAVS of their portfolios in Mexico City, on the other. Its purpose is to explain the behavior of discounts in the short run and long run, given the barriers to arbitrage that must exist. We also explain how the “divergent expectations” hypothesis is a useful complement to the “investor sentiments” and the “loss-aversion” models of country funds suggested by earlier researchers. Section III investigates whether the evidence is consistent with the “divergent expectations” hypothesis. II.Short-RunandLong-RunBehaviorofCountryFundDiscounts. a) CountryFund Discounts. Existing Hypotheses: A closed-end country fund (hereinafter country fund) consists of a fixed number of shares that are invested in a set of stocks from a particular country. Unlike open-end funds, once the fund is established new shares cannot be issued, while existing shares cannot be redeemed. Investors willing to buy or sell country-fund shares need to trade them on secondary security markets. Country funds are traded in New York at their U.S. 4

dollar price. As already noted, if markets were efficient, frictionless, and perfectly integrated internationally, the price of a fund should be equal to its NAV - which is the sum of the U.S. dollar market value of the individual equities at the home country. In practice, however, this is seldom the case. The gaps between prices and NAVSare both large andvariable. . ,.. ;, c-.. .. . . .. - , . .’ . -. It is well known that country funds, as well as domestic closed-end funds, trade at an average discount. Discounts are equal to log(NAVt/pricet). Various papers, such as ., Hardouvelis. La Porta and Wizman (1994), Diwan, Errunza and Senbet (1993, 1994)and Lee,.and Shleifer and Thaler (1991), document that domestic and country fund discounts are large, and also different from zero on a~erage. Several hypotheses have been suggested to explain this phenomenon. Any explanation must include the existence of market frictions that prevent perfect arbitrage. Frictions may be caused by various factors such as transaction costs, illiquidity of assets, capital gains tax liabilities, risks involved inthe arbitrage process, andbarriers to capitalmovements. Hardouvelis, La Porta and Wizman (1994) test the “investor sentiment” hypothesis for the case of closed-end country funds. They argue that country funds are a better indicator of investor sentiments than domestic closed-end funds. “Sentiments” here refer to generalized optimistic orpessimistic animal spirits, notbased on fundamentals. h the case of holdings of American equities, a change in U.S. investors’ sentiments is reflected in both U.S. NAVS and U.S. fund prices. On the other hand, in the case of

holdings of emerging market equities, a change in U.S. investors’ sentiments is reflected only in country fund prices, and not in their local NAVS (the prices of their underlying assets that are traded in each particular country). In other words, the co-movements of country funds are more likely to reflect U.S. investors’ sentiments, since the underlying assets of each of the funds are located in different countries with less common factors. The changes incountry fund NAVSmore likely reflect changes ineach particular market. Hardouvelis et al. find evidence that the noise trader model is consistent with the existence ofcountry fund discounts. Once cross-border restrictions are taken into account, they find that country funds trade at an average discount. U.S. investors, who mainly set funds’ prices, tend to underestimate the fundamental value of the funds. While our interpretation has something in common with Hardouvelis et al., we believe that the fund prices capture U.S. investor sentiment with respect to the country in question, rather than diffuse undifferentiated bullishness. Kramer and Smith (1995) challenge the investor sentiment hypothesis. Mexican funds and other Latin American funds turned from discounts to premia after the Mexican devaluation in December 1994. They claim that the investor sentiment hypothesis can onlyjusti~ these prernia by suggesting that international investors were optimistic about Latin America after the devaluation. Since optimism at that time seems implausible, they propose an alternative explanation. They hypothesize that the observed premia are evidence of loss-averse investors. When fund prices fell after the devaluation, investors ..

did not want to realize paper losses on their closed-end fund shares. They were willing to pay a premium for the country funds, even though they were pessimistic about these funds. They were not willing to sell when prices fell, because the marginal disutility of a loss is relatively high for loss-averse investors. Our response to the ~amer-smith argument isthatthe post-crisis premia areconsistent with pessimism by foreign investors, provided that Mexican investors turned pessimistic faster. b)Reconciliation ofHypotheses Regarding CountryFund Discounts: ~ . ,, 0 First, reasoning from the observed fact of wide disparities between prices and NAVS,we infer that arbitrage is not automatic. It is important to realize that in practice it is virtually impossible in this setting to engage in pure (riskless) arbitrage. The following summary sheds some light on why one perhaps should not expect perfect arbitrage. It describes a set of possible “arbitrage” strategies where each one has its own serious limitation. In addition, there exist general limitations to all of the strategies. The chart shows that arbitrage may be not only risky but also sometimes infeasible. Most of these general limitations havebeen pointed out in previous studies, such as Diwan, Errunza and Senbet (1993), Errunza (1991), andHardouvelis etal. (1994).

“Arbitra~e”Strategies: Particular Limitations: 1)Large outside investor could 1‘)Requires that investor has alot buy entire fund andliquidate. of capital, andthat thelocal market is so liquidthat large salesdo not drive theprices down. 2)Fund manager could convert 2’) Itmaybe difficult to get allof to anopen-end fund, generating thenecessary parties to agree to an immediate capital gainto open-end it.If the manager wanted share-holders. to deal with fluctuating inflows and redemptions, requiring new investments orliquidations, she would have started an open-end fund inthefrostplace. 3)Individual investors could 3’) Short-selling is difficult (or buy the fund and sellindividual even prohibited)in many ofthese shares short. markets, especially if itmeans trying simultaneouslyto sellshort a largenumber of holdings. 4) Individualinternational 4’) This factor (likenumber 3) investorswill have a lower will indeedput downward pressure demand for localshares onlocal shareprices and upward thanthey would otherwise, pressure on country fundprices; andinvestors willhave a butthere isno reason to think the higher demand forthe influenceshouldbe greatenough country fundthan they to eliminatetheprice disparity. would otherwise. 8

GeneralLimitations toAll Strategies: a)Markets may be illiquid.For example, Vanguard (1995)notesthat acountry’s entire market value,or capitalization,maybe less thanthat of asingle largeU.S. company. In many countries, the sharesheldby thecountry funds constitute a large fraction of the shares outstanding.Some shares turn over infrequently. b) Exchange rate risks areinvolved, sincecountry funds aretraded in U.S. dollars while theindividual sharesaretraded ineach country’scurrency. c) Markets donot trade atthe same time,making simultaneoustransactions sometimes infeasible. d)Transaction costs are larger than in standard U.S. securitiesmarkets. For instance,Vanguard estimates thatoveralltransaction costsforbuying abasket of emerging market stocks areexpected to be over 270. some countries there stillexistbarriers to capitalmovement. e) In The series of obstacles to arbitrage imply that expected discounts are different from zero. More properly, the observed fact of these disparities implies that the obstacles must exist. Even though perfect arbitrage is not to be expected, a large enough NAVprice difference should generate some kind of arbitrage. We suggest that discounts fluctuate inside bands before prompting much arbitrage activity.If discounts move below or above the band, rational investors will seek to profit from the NAV-price difference because theexpected gainsare substantial. Given thatbarriers to arbitrageexist,partially segmentingthemarkets, theprice in Mexico must reflect relatively more closely the asset demands of Mexican residents, and

that in New York the demands of foreign residents. It follows that, to whatever extent Mexican investors have different expectations from foreign investors, the country fund discount willto adegree reflect thedifference in expectations. Discounts appear tobe mean-reverting. Therefore shockshave larger effects in the short run than in the long run.4 Some of the limitations to arbitrage, such as market illiquidity and exchange rate risk, explain the limited speed of mean reversion. Since it takes time to find buyers in local markets for large blocks of stocks, without pushing down the price. the short run may display large gaps. Over a longer horizon, buyers can be found, and discounts shrink. Moreover returns are more uncertain in turbulent periods than in periods of tranquility, allowing discounts to deviate from their long-run equilibrium level. The dynamics of discountscan be summed up in the following way. There exists a stationary long-run relationship between each price and its NAV. Given a constant average discount, an innovation inthefund’NsAV is expected to be fully transmitted to the fund’s price in the long run. On the other hand, a change in a NAV is expected to be only partially transmitted to its price, changing the average short-run discount. In other words, the short-run elasticity of price with respect to NAV is expected to be less than one, while thelong-run elasticity isexpected to be close to one. 4Tests of stationarityin discountsare reportedbelow, concludingthat discountsare meanrevertingprocesses.Hardouvelisetal.(1993)alsofindstationarydiscounts. 10

The existing hypotheses (“investor sentiment” and “loss-aversion”) do not explain the complete story. They partly explain the magnitude and persistence of the premia. However, it remains to be understood why NAVSand prices reacted in different ways to the Mexican crisis. These hypotheses do not predict why discounts turned into premia around the time of theMexican devaluation.This paper arguesthatdifferent expectations, on the part of Mexican vs. American investors, may be present. The different expectations hypothesis complements the existing explanations. If Mexican investors foresaw the crisis, NAVS fell first and/or fell more rapidly than country fund prices. Therefore, discounts turned into premia even though investors were pessimistic about Mexico. c)Empirical Testing: In this subsection we estimate the short-run and long-run relationships between the three Mexican prices and their respective NAVS.5We first determine the stationarity of the series and the long-run relationships from the cointegrating vectors. Then we study the short-run adjustment towards the long-run equilibrium by error-correction models. The relationship between prices and NAVS, as well as that between discounts and exchangerates are analyzed. 5Thefunds’descriptionsaredetailedinAppendix1,aswellasthedifferencebetweenpricesand NAVS. 11

Unit root tests, displayed in Part a of Table 1, fail to reject that all prices and NAVS are non-stationary in levels. The null hypothesis tested is that the level of the variables contain a unit root. We perform three unit root tests, Weighted Symmetric, Dickey-Fuller, and Phillips-Perron, to check robustness of the tests. The Weighted Symmetric test tends to have higher power than the Dickey-Fuller one. The Phillips- Perron test calculates aresidual variance “robust” to autocomelation.The numbers of lags used in each case have been determined by the Akaike Information Criterion (AIC). Only the t-statistics for the optimal number of lags are tabulated. The critical values used take into account the finite sample properties. First differences of all the variables yield stationary results although they are notreported. Even though the levels of the series are non-stationary, there may exist stationary linear combinations of them, called cointegrated vectors. Part b of Table 1 displays unit root tests on discounts, testing whether discounts are stationary. In other words, we restrict the cointegrating vectors to be (1, -l), and perform tests on their residuals. The restrictions are not arbitrary; they are based in hypotheses of how prices are linked to NAVS.Table 1shows that two out of the three funds reject non-stationarity, according to the Phillips-Perron test. When we include the exogenous dummy variables damexdev and dpolstab, most of the tests yield stationarity. These variables control for events that particularly drove the discounts away from their long-run relationships. In some cases non-stationarity cannot be rejected,but these results maybe dueto not very powerful unit root tests. 12

The variable dpolstab takes the value 1the week that N~A was approved and the week President Zedillo was elected. Ittakes the value -1 in the weeks that the markets received bad political and economic news from Mexico, namely when the two political (Colosio and Ruiz-Massieu) assassinations took place, the week of the Chiapas uprising, and the week of the peso devaluation. Otherwise, it contains the value O.Therefore, this variable controls for the good news and bad news shocks on the country funds. The variable &mexdev takes the value 1 for the six months following the devaluation, otherwise ittakes the valueO. Table 2 reports the results from cointegration tests. In this case, we do not impose any particular value for the cointegrating vector. (Even though we believe that the constraints arejustified on apriori grounds, we gothrough thecointegration tests because some readers will expect to see them.) The table tabulates the Engle-Grangerb and the Johansen-trace7 (maximum likelihood) cointegration tests along with the estimated normalized cointegrating vectors. The cointegrating vectors are interpreted as the longrun relationships between the variables. When no other variables are taken into account, theJohansen testfinds onecointegrating relationship for the fund MXF.8 GThe Engle-Grangercointegrationtests is a Dickey-Fullertest on the residuals from the cointegrationregression. 7TheJohansen-tracealgorithmtests,in severalsteps,nullhypothesesofncointegratingvectors againstalternativehypothesesofn+l cointegratingvectors. 8Whenothervariablesareincluded,theJohansentestsfindcointegrationforthefundsMEFand MXE.Inallofthecasesitcannotberejectedthatonlyonecointegratingvectorexists.

We also test the hypothesis that the cointegrating vector between prices and NAVSis (1, -l). We cannot tell that directly from the cointegrating vector, because of the presence of nuisance parameters. Since the residuals are autoconelated, the fact that there is cointegration is not sufficient to imply that the errors are i.i.d. Normal. As a consequence, we need to correct the statistics so that they are asymptotically Normal. We perform the correction, running two OLS regressions, according to the method of Stock and Watson (1993). The usual t-statistic is multiplied by the first regression’s standard error, and divided by the second regression standard error over 1 minus the autocorrelation coefficients. Table 3 shows that two of the three funds cannot reject the hypothesis that the cointegrating vector is (1, -l). A normalized coefficient of 1 implies, as expected, that a change in the NAV is fully transmitted to its price in the long run. A change in fundamentals, which shifts the NAV, ultimately shifts its price by the same magnitude. Once we have studied the cointegrating vectors, we calculate the speed of adjustmenttowards the long-run relations. The speed of adjustment determines how much time is necessary for the price to adjust to the long-run relationship with itsNAV. In other words, it expresses by how much prices adjust, in the short run, given a departure from the long-run equilibrium. Since all the estimated speeds of adjustment are positive, a discrepancy from the long-run equilibrium means an adjustment of theshort-run values of prices towards the long-run values.9 9Thecloserthespeedof adjustmentis to zero.theslowertheconvergence.Whenthespeedof adjustmentisequalto 1,theconvergenceisinstantaneous.Notethatthespeedsofadjustmentare definedasthenegativeofthecoefficientsthatappearinthetables. 14

Tables 4 and 5 display different error-correction models, estimating the speed of adjustment. The adjustment factors have been calculated by a seemingly unrelated regressions (SUR), using the Engle-Granger two-step estimator.~”The first step yields super-consistent estimators of the cointegrating vector. Therefore, efficient estimates are obtained in the second step. The lagged residuds from the first step stand for the deviations from the long-run relationships. Table 4 assumes stationary long-run relationships between prices and NAVS. However we allow the long-run relationships to differ across funds. A SUR is run in the second step, constraining the adjustment factor to be the same for each fund. Assuming that the constraint is valid, the second step yields efficient and unbiased estimators of the error-correction model. The estimated adjustment coefficient is 0.15 per week, and is statistically significant. However, notice that if indeed there is no cointegration, the residuals arenon-stationary, making the usualt-tests inappropriate. The above results show that adjusting to the long-run relationship may take some time, especially in a period when successive shocks occur. These results can also be looked atin a different light. Sudden gaps such asthe onethat in December 1994opened up in the Mexican funds may routinely and mechanically reflect the short-run impact of changes in the exchange rate. After all, equities in Mexico City are priced in pesos and the country funds in New York are priced in dollars. Hardouvelis et al. documented that 10SeeBanerjeeetal.(1994).

exchange rate changes have such effects on country funds in general. We can isolate the effect associated with the exchange rate per se by estimating the normal relationship between changes in theexchangerate andcountry fund discounts. Results from table 5 also show that the Mexican devaluation of 1994 may have been different from other exchange rate changes. It shows that changes in discounts can be only partially explained by changes in the exchange rate. The dummy variable &mexdev is negative and statistically significant, explaining the unusual premia observed after the devaluation. In other words, the fall in the discount in December 1994 was greater than would be expected from the magnitude of the devaluation and the usual pattern associated with exchange rate changes. We interpret this as a loss in confidence by Mexican investors (relative to U.S. investors). But, perhaps the most convincing piece of evidence supporting this hypothesis was already evident in Figures 1-3: the change came a few weeks before the devaluation. This supports the hypothesis that the change in discounts was partly due to less optimistic Mexican investors, and not simply to the devaluation itself. In the second step, we do notconstrain the adjustment coefficients to be theequal to each other, in order to check how different they vary. The results displayed in Table 5 show that the short-run elasticities are not very different from the previous one. It takes some time to goback to the long-run relationships. In this case, the coefficients vary from 1370to22%, implying ahalf life of around 3to 5 weeks.

In summary, the results show that it takes a few weeks for the short-run variables to filly adjust to the long-run relationships, assuming that no new discrepancies arise. Namely, shocks that drive prices and discounts away from their long-run relationships have only a partial effect in the short run. If no new shocks occur, the prices and discounts adjust at rates ranging from 13 to 22 percent of the gap each week. Since the cointegrating coefficients for NAVSare close to one, a change in a NAV means that its price will change by the same amount in the long run. Even after the initial devaluation, on December 20, the discrepancy remained for several months, suggesting that Mexican residents were more aware than foreigners of the negative implications of the crisis for theMexican economy. III.WereNAVSandPricesDrivenbyDivergentExpectations? In the previous section we demonstrated mean-reversion in country fund discounts. We also argued that the divergent expectations hypothesis helps explain the prernia observed after the devaluation of December 1994. As the IMF capital markets report argued, Mexican investors reacted first to economic and political local events, i.e., the Mexican investors were thefront-runners in the devaluation. In the present section we test that divergent expectations drive country fund NAVS and prices throughout the sampleperiod. 17

First, the variables are plotted against time to observe the reactions of country funds before the devaluation. Second, two econometric approaches are followed. Granger-causality tests are computed to search for causality in the variables. Then, a regression is calculated by SUR to obtain point estimates of how different prices and NAVSare statisticallyrelated. The plots of the three stocks contain some information about expectations. Figures 1-3 show that both NAVS and prices went up, reflecting more positive expectations from local and foreign investors when markets received good news about Mexico. The two clearest cases are the N~A approval in November 1993 and President Zedillo’s victory in the presidential election of August 1994.11Country funds are sensitive to changes in sentiments andexpectations. As noted, the figures also show that both prices and NAVS started falling before the devaluation in December 1994.Finally and crucially, the figures provide evidence of divergent expectations before the devaluation. The MXE discount turned into premia a week before the devaluation. In the case of the two other funds, MXF and MEF, both NAVS and prices fell before the devaluation. However, NAVS fell much more rapidly, showing thatdiscounts started falling sharplybefore thedevaluation. 11Suchpoliticaleventshada statisticallysignificanteffectonMexicaninterestratesduringthe yearandahalfprecedingthecrisis,asnotedinFrankelandOkongwu(1996). 18

As a first econometric approach, Granger-causality tests are estimated to determine whether changes in NAVS preceded changes in prices, not just in December” 1994,but in general. We run the VAR process in first difference form, since the typical Granger-causality test does not have its standard distribution when the variables are 1(1).12Four alternative hypotheses may be tested from these tests: prices Granger-cause NAVS, NAVS Granger-cause prices, neither of them cause the other, or they are simultaneously determined. Table 6 shows the results.13The table and corresponding . Ilgure only report the cases where one-direction Granger-causality was found. Figure 4 displays the results in a different way. It indicates the causality relationships with arrows. All three Mexican NAVSGranger-cause one of thethree prices in New York. Moreover, both within Mexico and within New York, the biggest Mexican fund, MXF, affects the other funds. The arrow sizes ofFigure 4 havebeen chosen in an arbitrarybut readily-perceived way. The thick arrow indicates that both of two results hold. First, the probability of accepting the null hypothesis of no Granger-causality is less than 5 percent. (More properly, the probability of rejecting the null is higherthan 95 percent.) At the same time, the thick arrow means that the difference between the probabilities of accepting the null hypothesisis atleast of 50 percentagepoints. In other words, the difference in probability values (P-values) is at least 0.50, so we are very confident that Granger-causality only 12Schmukler(1996)performsothererogeneitytests,whichincorporatecointegratingvectorsin theestimation.Thoseresultsareverysimilartotheonesreportedhere. 13Sincethe Granger-causalitytest can be very sensitiveto the choiceof lag length,different specificationshavebeentried,withoutsubstantiallychangingtheresults.Onlyonespecification isdisplayedhere. 19

goes in one direction, because we accept and reject tie null hypotheses strongly.The thin arrow means that the probability of accepting the null hypothesis is less than 5 percent, and that the difference between probabilities is greater than 10 and less than 50 percentage points. Having tested that causality goes from Mexico to New York, we estimate, as a second econometric approach, a SUR/VAR. In this case, we are interested in how prices are affected by other variables. We report only one representative SUR estimation in Table 7. It shows the contemporaneous relationship between NAVS and prices. The variables are in first differences, to avoid the spurious regression problem and to use Normal limiting distributions. The estimates are calculated by nonlinear least squares, imposing constraints for equal coefficients, but allowing for different constants. The dependent variables are the country fund prices. The independent variables are the fund NAVS, the Mexican exchange rate, the international interest rate, and the dummy for political stability. Under the assumption that the constraints are not too restrictive, the SURestimation enhances efficiency. The regression output shows that NAVS are significant in explaining changes in prices, confirming the results obtained with the Granger-causality tests. We also include laggedprices, since we found Granger-causality within prices. In this sense the regression displayed in Table 7 is a VAR, with other exogenous variables. The exchange rate is statisticallysignificant and of the right sign.A drop in the value of the peso is reflected as 20

afall in the value of the underlying assets in terms of dollars. The dummy variable that reflects political stabilityis alsoof theright sign, and significant. The international interestrate is expected to have a negativeeffect. since adrop in the international interest rate results in anincrease in demands and prices for many assets, including Mexican country funds. The regression yields the right sign. although the variable does not turn out to be significant. A negative coefficient for the interest rate agreeswith other studiesof foreign investor demand in emerging markets more generally, such as Calve, Leiderman and Reinhart (1993). Chuhan, Claessens and Mamingi (1994), Dooley et al. (1994), Fernandez-Arias (1994). Frankel and Okongwu (1996). and Schadler etal. (1993). IV.SummaryofConclusions In thepresent paper weuse thethree Mexicancountry fundsto show evidencethat local and foreign investors had divergent expectations during the Mexican crisis of December 1994.The asymmetric information hypothesis was suggested in the aftermath of the crisis, implying thatMexicaninvestorresactebdeforeinternatioinnavlestotros newsabouttheMexicaneconomy.Thisstatemenctanbeinterpreitnetdwoways:either domestiacndinternatioinnavle”storresceivetdwodifferesnettsofinformatioonr,the locailnvestowresremorealeratndsensititvoepotentiwaalrningsignals.

More generally, we incorporate severai elements from the literature on country funds into a new picture of how these funds behave. Even though indirect arbitrage may exist, it faces several obstacles. We suggest that the nature of these barriers may explain mean reversion, or different reactions of country fund discounts in theshort run and in the long run. k the short run, discounts may be large due to market iiliquidity or because of increased obstacles to arbitrage. In the long run, they tend to narrow. This hypothesis complements existing models such as the investor-sentiments and the loss-aversion interpretations. On the empirical side. Section Hshowed that although a change in a NAV is fully transmitted to the country fund’s price in the long run, it is only partially transmitted in the short run. It also showed that the rates of adjustment towards the long-run relationship, estimated by error-correction models, are around 0.15 per week, depending on the case. They imply that 5070of the adjustment takes place in around 3 to 5 weeks. A similar estimate was found for the adjustment of discounts, towards their long-run relationship with the exchange rate. A slow rate of convergence plus the divergent expectations hypothesis suggests a reconciliation between the investor sentiment hypothesis and theloss-aversion one. SectionIII provided supportfor the asymmetric expectations hypothesis.The most simple andimmediate proof of heterogeneous expectations is in Figures 1-3,which show 22

that NAVS fell first or faster relative to prices right before the devaluation. Grangercausality tests, a SUWAR confirm that observation more generally. Several extensions of this work are desirable. First, the results could be enriched by a larger data set covering more countries, as well as by higher frequency data, if the datacan be obtained. Second, there is aneed for valid instrumental variables to cope with potential endogeneity. Third, a theoretical model needs to be constructed to explore further some of theideas expressed in this paper. 23

References Banerjee, Anindya,Juan Dolado, John W. G~brtith, ad David F=HendrY,1994,Q Inte~ration,Emor-Comection,andthe Econometric Analysisof Non-Stationm Data, O#ord Universi~ Press. Calve,Guillermo,Leonardo Leiderman, and Carmen Reinhart, 1993,“CapitalInflows and RealExchange Rate Appreciation inLatin America: theRole of External Factors,” ZMFStaff Papers, 40(l), 108-150. ChuhanP,unamS, tijnClaessens,andNlanduMarningi, 1994,“Equity and Bond Flows to Latin America and Asia: theRole of Global and Country;’ unpublished manuscript, me WorldBank. Diwan, Ishac, Vihang Errunza, and Lemma Senbet, 1993,“Coun~ Funds For Emerging Economies,” in Stijn Claenssens and Sudarshan Gooptu (eds.) Portfolio Investment in Develo~in~Countries, Washington: me WorldBank. Diwan, Ishac, Vihang Errunza, andLemma Senbet, 1994,“Diversification benefits of country funds,” in Investing inEmer~in~Markets, Euromoney Books and me WorldBank. Dooley, Michael, Eduardo Femandez-Arias, and Kenneth Kletier, 1994,“RecentPrivate CapitalInflows to Developing Countries: Is the Debt Crisis History?;’ National Bureau ofEconomic Research WorkingPaperNo. 4792, July. Errunza, Vihang R., 1991,“Pricing of National Index Funds”, Review of Quantitative FinanceandAccounting, pp. 91-100. Femandez-Arias, Eduardo, 1994,“The New Wave of Private CapitalInflows: Push or Pull?,” Policy Research WorkingPaper 1312,Debt and InternationalFinance Division, International Economics Department, me WorldBank, June. Frankel, Jeffrey A., 1994a,“Introduction” to The Internationalization ofEauitv Markets, the Universityof ChicagoPress, Chicago, 1994. Frankel, Jeffrey A., 1994b,“Sterilization of Money Inflows: Difficult (Calve) or Easy (Reisen)?” in Afluencia deCa~italesYEstabilizacion en America Latina, edited by Roberto Steiner, Fedesarrollo, Bogota, pp. 241-267. Jeffrey FrankelA., and Sergio L. Schmukler, 1996,“Crisis, Contagion, and Country Funds:Effects on East Asia and Latin America:’ forthcoming in Mana~inR CapitalFlows and Exchan~eRates: Lessons from thePacific Basin, editedby Reuven Glick, Federal Reserve Bank of San Francisco. 24

Hardouvelis, Gikas A., Rafael LaPorta, and Thierry A. Wizman, 1994,“What Movesthe Discount on Country EquityFunds?,” National Bureau ofEconomic Research WorkingPaper4571, in J. Frankel, cd., The Internationalization of Equity Markets, the Universityof ChicagoPress, Chicago. InternationalMonetary Fund, 1995,International CapitalMarkets. Developments, ProsWcts, and PolicyIssues, August. Kramer, Charles, and R.Todd Smith, 1995,“RecentTurmoil in Emerging Markets and theBehavior of Country-Fund Discounts: Removing the Puzzle of thePricing of Closed-End MutualFunds,” InternationalMonetary Fund WorkingPaper 95/68, July. Lee, Charles, M.C., Andrei Shleiffer, andRichard Thaler, 1991,“Investor Sentiment and theClosed-end Fund Puzzle,” Journal ofFinance, Vol. 46, No.1,pp. 75-109. Schadler, Susan,Maria Carkovic, Adam Bennett, andRobert Kahn, 1993,“Recent Experiences with Surges in Capital Inflows”, OccasionalPaper 108, International MonetaryFund. Schmukler, Sergio, 1996,“Country Funds and Asymmetric Information,” unpublished manuscript, Universityof CaliforniaatBerkeley. Stock,James H., andMark W. Watson, 1993,“A SimpleEstimator of Cointegrating Vectors in Higher Order Integrated Systems,”Econometric 61:783-820. Vanguard Group, 1995,Plain Talk,“AboutInvesting in Emerging Markets,” bulletin. 25

Appendix1:Closed-endCountryFunds The three closed-end funds used are: Emerging Mexico Fund (MEF) Mexico Equity & Income (MXE) Mexican Fund (MXF) Net Asset Values (NAVS)are calculated at the local market close in U.S. dollars. Prices arerecorded on theday theNAVSwere calculated, usually Fridays. 26

Appendix2:DescriptionofVariablesandData~4 Country-Funds data have been provided kindly by R. Todd Smith of the International Monetary Fund, Research Department and by Don Cassidy of Lipper Analytical Services. Exchange rate data and Treasury bill rates data have been obtained from Data Stream. The datahave weekly frequency and go from 1/5/90to 3/8/96. Variables: - mefnavl, mefpricel, mefdisc, mxenavl, mxepricel, mxedisc, @avl, m~pricel, wfdisc; Correspond to the Mexican country funds described in Appendix 1. For each country fund, its NAV, price and discount are available. NAVS and prices are all expressed in logarithms, while discounts are differences of logarithms. - dpolstab: Qualitative variable that reflects political stability in Mexico. Contains 1s when President Zedillo was elected and when the NAFTA agreement was approved. Contains -1s when disturbing political events arose in Mexico, i.e. in Colosio and Ruiz- Massieu assassinations, under the Chiapas uprising andwhen thepeso devalued. Contains 0s otherwise. -dumexdev:Is adummy variable, with 1for the six months after the Mexican devaluation and Ootherwise. - mexerl: Mexican exchange rate in logarithms. Equals the log of the amount of dollars per peso. - tbilllml: One-month Treasury billrate in logarithms. 14Allthe model havebeen estimated using the statisticalpackages Econometric Views and TSP. 27

: u c z a 0 0

# . \ . o o 0 0 m 8

A ,- - . # . b l o . # # . * , # a l\ , ,# # . . # l . .’ , , . . 8 , # , , . 1 * ‘. . n . , “. # 1 1 ! . I ,- .“’ , l b 8 l i- . . . . <

Figure4:MexicanpricesandNAVS . NewYorkCip: dmef~ric~ + dmxfn/a - ~ Rob.ofac.eptin~<d5i%f,fe~e~~elnP~~bablli~l~~~fa~l~a~~5OP~rCenta~e~0intS. ~ ~~b.ofaccepticng5%,differeinnpcreobabilitiebsetween 10and50percentagepoints. 31

Table 1: UnitRootTeetsonMexicenCountry Funds SamplePertti: 1/S90-3/W I a)UnitRootTestaforMexicen NAVaandPrices. I b)UnitRootTestsforMexicen Discounts. I MEFNAVL: T-stat P-value Num.lag: I MEFDISCL: T-Stat P-valueNum.lags (23sobs.) Wtd.Sym. -1.97 0.67 3 I (23eobs.) Wtd.Sym. -2.47 0.31 7 Dickey-F -2.07 0.57 3 I Dickey-F -2.23 0.47 7 Phillips -6.22 0.73 3 I Phillips -32.96 “0.00 7 I MXENAVL: T-Stat P-value Num,lag: I MXEDISCL: T-Stat P-valueNum.lags (23eC4n) Wtd.Sym. -1.97 0.67 3 I (23sObs.) Wtd.Sym. -2.50 0.29 3 Dickey-F -2.07 0.57 3 I Dickey-F -2.27 0.45 3 Phllllps -6.22 0.73 3 1 Phillips -18.48 0.10 3 I MXFNAVL: T-Stat P-valueNum.lag: I MXFDISCL. T-Stat P-valueNum.lags (313Obs.) Wtd.Sym. -1.65 0.64 4 I (313Obs.) Wtd.Sym. -2.62 0.23 6 Dickey-F -1.64 0.69 4 I Dickey-F -3.11 0.10 6 Phillips -5.21 0.61 4 I Phillips 49.66 “0.00 6 I MEFPRICEL: T-Stat P-value Num.lag! I Thefollowingthreegroupscontrolforthe (23sas.) Wtd.Sym. -0.96 0.96 2 I exogenous variabka DAMEXDEV andDPOLSTAB. Dickey-F -1,60 0.79 2 I Ph!llips 4.24 0.67 2 I MEFDISCL: T-Stat P-value Num.lags I (23aObs.) Wtd.Sym. -3.23 ‘0.04 7 MXEPR/CEL: T-Stat P-value Num.lag: I Dickey-F -3.15 0.09 7 (230Obs) Wtd.Sym. -1.33 0.93 2 I Phillips -67.16*”1.45D-06 7 Dickey-F -1.05 0.94 2 I Phtllips -3.33 0.92 2 I MXEDISCL: T-Stat P-valueNum.lags I (23eobs) Wtd.Sym. -3.43 “0.02 3 MXFPRICEL: T-Stat P-value Num.lag: I Dickey-F -3.26 0.07 3 (3130bs) Wtd.Sym. -1.37 0.92 2 I Phillips -43.57 “0.00 3 Dickey-F -1.35 0.67 2 I Phillips 4.69 0.63 2 I MXFDISCL: T-Stat P-valueNum.lags I (3130bs.) Wtd.Sym. -3.09 0.07 6 I Dickey-F -3.60 ‘0.02 6 I Phillips -77.57**1.45D-07 6 UnitRootTestsConsistofWeightedSymmetricA,ugmentedDickey-FullearndPhillips-PerronTests NAVS:U.S.dollarpriceofunderlyingcountryfundassets. Price: U.S.dollarpriceofcountryfundinNewYorkCity. Discount: ln(NAVPrice). l(**)Denotesrejectionsofthehypothesisat5%(l%)significancelevel 32

Table2: CointegratioTnestsBetweenMexicanFundPricesandNAVS SamplePeriod:1/5/90-3/8/96 Dependentvariable:MEFPRICEL Engle-Granger(tau)tests Johansen(trace)tests NumlagfOpt:8 NumlagsOpt:2 alpha 0.91 ~igvall 0.03 TestStat -2.21 Eigva12 0.00 P-value 0.42 HO:r=O 8.12 Const 0.34 P-val- 0.61 t(Const) 8.83 HO:r<=l 0.59 Numobs 227.00 P-val- 0.74 LogLike 360.53 Numobs 233.00 AIC -3.10 LogLike 733.90 Varres 0.00 AIC -6.18 CointegratingvectMEFPRIM( EFNAVL 1 -0.9477 Dependentvariable:MXEPRICEL Engle-Granger(tau)tests Johansen(trace)tests NumIagfOpt:5 Num Opt:l alpha 0.97 Eigvall 0.02 TestStat -1.45 Eigva12 0.01 P-value 0.78 HO:r=O 7.37 Const 1.46 P-val- 0.68 t(Const) 21.36 HO:re=l 1.89 Numobs 230.00 P-val- 0.58 LogLike 390.41 Num 234.00 AIC -3.34 LogLike 762.20 Varres 0.00 AIC -6.43 CointegratingvectMXEPRlM~XENAVL 1 -0.5868 Dependentvariable:MXFPRICEL Eng/e-Granger(tau)tests Johansen(trace)tests NumIagfOpt:6 Num opt:2 alpha 0.91 Eigvall 0.06 TestStat -2.58 Eigva12 0.01 P-value 0.25 HO:r=O 22.02 Const 0.16 P-val- “0.01 t(Const) 3.94 HO:r<=l 3.90 Numobs 306.00 P-val- 0.30 LogLike 527.15 Num 310.00 AIC -3.40 LogLike 1054.7 Varres 0.00 AIC -6.71 CointegratingvectMXFPRIM( XFNAVL 1 -0.9906 *(**)Denotesrejectionsofthehypothesisat5%(l%)significancelevel 33

Table3: Testsof HO:Cointegrating Vector between PricesandNAVS=[1,-1]* Sample Period: 1/5/90-3/6/96 Equation1:MEFPRICEL CoefficientT-statistic CorrectedT-statistic c 0.28 6.85 Ho:coeff.ofMEFNAVL=1 MEFNAVL 0.90 61.35 T(Stock-Watson=) -1.58 DMEFNAVL(+2) -0.07 -0.55 DMEFNAVL(+l) 0.10 0.74 DMEFNAVL -0.05 -0.36 DMEFNAVL(-1) -0.14 -1.07 DMEFNAVL(-2) -0.07 -0.51 Numberofobsewations: 216 AdjustedR-squared 0.95 Std.errorofregression 0.09 Equation2:MXEPRICEL — CoefficientT-statistic c 1.43 19.03 CorrectedT-statistic MXENAVL 0.48 17.73 Ho:coeff.ofMXENAVL=1 DMXENAVL(+2) -0.41 -1.68 T(Stock-Watson)= -2.27 DMXENAVL(+l) -0.39 -1.52 DMXENAVL -0.13 -0.52 DMXENAVL(-1) -0.10 -0.42 DMXENAVL(-2) 0.04 0.15 Numberofobservations: 216 AdjustedR-squared 0.95 Std.errorofregression 0.09 Equation3:MXFPRICEL —— CoefficientT-statistic c 0.11 2.55 CorrectedT-statistic MXFNAVL 0.95 65.50 Ho:coeff.ofMXFNAVL=1 DMXFNAVL(+2) -0.10 -0.97 T(Stock-Watson=) -0.91 DMXFNAVL(+l) 0.17 1.65 DMXFNAVL -0.09 -0.87 DMXFNAVL(-)1 -0.22 -2.23 DMXFNAVL(-2) -0.23 -2.28 Numberofobservations: 270 AdjustedR-squared 0.94 Std.errorofregression 0.08 lThecorrected T-statistics arecalculated inasecondstage,usingtheadjustment suggesteda,mongothers,byStockandWatson(1993). ThecorrectedT-statisticshouldbecomparedwiththecriticavlaluesfromaN(O,l). 34

Table 4: ErrorCorrection ModelforMexicanFundPrices EstimatedbyIterativeSeeminglyUnrelatedRegression Engl*Granger TwoStepEstimator. SamplePeriod:1/5/90-3/8/96 ~“ CoefficientT-Statistic CoefficientT-Statistic c(1) 0.23 7.17 c(1) 0.00 -0.32 MEFNAVL 0.91 77.73 RESID1(-1) -o./5 -4.68 C(I1) 0.21 3.68 D(MEFPRICEL(-)l()EQ.1) -0.25 -4.04 MXENAVL 0.92 44.56 D(MEFPRICEL(-)2(E)Q.1) -0.17 -2.62 C(21) 0.16 4.40 D(MEFNAVL()-1)(EQ.1) 0.33 4.21 MXFNAVL 0.93 78.96 D(MEFNAVL(-2E))Q(.1) 0.22 2.77 C(l1) 0.00 0.32 D(MXEPRICEL(-I()E)Q.2) -0.18 -2.62 Equat}on1:MEFPRICEL D(MXEPRICEL(-2(E))Q.2) -0.02 -0.30 Observations: 237 S.E.ofregression 0.10 D(MXENAVL(-l)E)(Q.2) 0.36 3.85 AdjustedR-squared 0.941 Durbin-Watsosntat 0.29 D(MXENAVL(-2E))Q(.2) 0.05 0.59 C(21) 0.00 0.32 D(MXFPRICEL(-l()E)Q.3) -0.08 -1.22 Equation2:MXEPRICEL D(MXFPRICEL(-2(E))Q.3) -0.02 -0.32 Observations: 234 S.E.ofregression 0.11 D(MXFNAVL(-l))(EQ.3) 0.07 0.97 AdjustedR-squared 0.84 Durbin-Watsosntat 0.18 D(MXFNAVL(-2))(EQ.3) 0.15 2.25 Equation1:D(MEFPRICEL) Equation3:MXFPRICEL Observations: 228 S.E.ofregression 0.06 Observations: 314 S.E.ofregression 0.06 AdjustedR-squared 0.05 Durbin-Watsosntat 1.75 AdjustedR-squared 0.94 Durbin-Watsosntat 0.35 Equation2:D(MXEPRICEL) Observations: 221 S.E.ofregression 0.06 AdjustedR-squared 0.10 Durbin-Watsonstat 2.02 Equation3:D(MXFPRICEL) Observations: 294 S.E.ofregression 0.06 AdjustedR-squared 0.04 Durbin-Watsosntat 1.85 35

Table 5: ErrorCorrectionModel for Mexican Fund Discounts Estimated byIterative Seemingly Unrelated Regression EngbGrenger TwoStep Estimator SamplePeriod:115190-3/6/96 ~. EstimationMethod:Seemingly UnrelatedRegression EstimationMethod:Seemingly Unrelated Regression -. Coeff-ia—entT-Sta.t-istic Coeffiaent T-Statistic C(l) 2.55 5.16 c(1) 0.40 1.30 DMEXERL(EQ.1) 39.09 1.91 RESID1(-1) -9.21 -5.25 DAMEXDEV -21.76 -18.46 D(MEFDISCL(-1)) -0.28 -4.48 C(n) 3.64 7.52 D(MEFDISCL(-2)) -0.19 -3.29 DMEXERL(EQ.2) -23.77 -1.16 D(MEXERL(-l))(EQ.1) -66.28 -5.28 C(21) 7.79 21.13 D(MEXERL(-2))(EQ.1)-12.67 -0.94 DMEXERL(EQ.3) 1.59 0.09 C(n) 0.11 0.36 RESID2(-1) -0.13 -3.35 D(MXEDISCL(-1)) -0.26 -3.66 Equation1:MEFDISCL D(MXEDISCL(-2)) -0.06 -0.66 Observations: 237 S.E.ofregression 7.74 D(MEXERL(-l))(EQ.2) -16.51 -1.39 AdjustedR-squared 0.53 Durbin-Watsonstat 0.51 D(MEXERL(-2))(EQ.2) -9.00 -0.77 C(21) 0.10 0.40 Equation2:MXEDISCL RESID3(-1) -9.22 -5.63 Observations: 234 S.E.ofregression 7.94 D(MXFDISCL(-1)) -0.20 -3.56 AdjustedR-squared 0.51 Durbin-Watsonstat 0.42 D(MXFDISCL(-2)) -0.02 -0.35 D(MEXERL(-l))(EQ.3) -19.57 -1.69 Equation3:MXFDISCL D(MEXERL(-2))(EQ.3) 3.20 0.28 Observations: 314 S.E.ofregression 6.59 AdjustedR-squared 0.37 Durbin-Watsonstat 0.56 Equation1:D(MEFDISCL) Observations: 225 S.E.ofregression 4.87 AdjustedR-squared 0.23 Durbin-Watsonstat 1.69 Equation2:D(MXEDISCL) Obsewations: 216 S.E.ofregression 4.50 AdjustedR-squared 0.11 Durbin-Watsonstat 1.99 Equation3:D(MXFDISCL) Obsewations: 286 S.E.ofregression 4.42717 AdjustedR-squared 0.16 Durbin-Watsonstat 2.027928 36

Table 6: Pairwise Granger Causality Tests Between First Difference of Mexican Fund Prices and NAVS(2 lags) SamplePeriod:1/5/90-3/8/96 NullHypothesis Obs F-St. Prob. DMXEPRICELdoesnotGrangerCauseDMEFNAVL 225 0.12 0.89 DMEFNAVLdoesnotGrangerCauseDMXEPRICEL 11.13 0.00 DMXEPRICELdoesnotGrangerCauseDMEFPRICEL 281 0.20 0.82 DMEFPRICELdoesnotGrangerCauseDMXEPRICEL 17.42 0.00 DMXFPRICELdoesnotGrangerCauseDMEFPRICEL 281 4.88 0.01 DMEFPRICELdoesnotGrangerCauseDMXFPRICEL 0.75 0.47 DMXEPRICELdoesnotGrangerCauseDMXENAVL 216 0.38 0.69 DMXENAVLdoesnotGrangerCauseDMXEPRICEL 11.71 0.00 DMXFNAVLdoesnotGrangerCauseDMXENAVL 186 7.99 0.00 DMXENAVLdoesnotGrangerCauseDMXFNAVL 1.94 0.15 DMXFNAVLdoesnotGrangerCauseDMXEPRICEL 254 14.52 0.00 DMXEPRICELdoesnotGrangerCauseDMXFNAVL 1.53 0.22 DMXFPRICELdoesnotGrangerCauseDMXEPRICEL 288 17.70 0.00 DMXEPRICELdoesnotGrangerCauseDMXFPRICEL 1.85 0.16 37

Table 7: IterativeSeemingly Unrelated Regression Convergence achieved after 6 iterations Sample Period:1/5/90 - 3/8/96 Coefficient T-Statistic C(EQ.1) 0.00 0.03 C(EQ.2) 0.00 0.49 C(EQ.3) 0.00 0.47 DMEFNAVL 0.34 2.83 DMXENAVL 0.19 1.84 DMXFNAVL 0.35 3.83 DMEXERL 0.32 2.35 DTBILLI ML -0.04 -1.22 DPOLSTAB 0.03 2.32 DMEFPRICEL(-1) 0.14 2.05 DMXEPRICEL(-1) -0.17 -2.51 DMXFPRICEL(-1) -0.14 -1.84 DMEFPRICEL(-2) 0.00 0.04 DMXEPRICEL(-2) -0.06 -0.93 DMXFPRICEL(-2) 0.05 0.67 Equation1: DMEFPRICEL Observations: 205 S.E. of regression 0.05 Adjusted R-squared 0.28 Durbin-Watsonstat 2.29 Equation2: DMXEPRICEL Observations: 205 S.E. of regression 0.04 Adjusted R-squared 0.49 Durbin-Watsonstat 1.95 Equation3: DMXFNAVL Observations: 205 S.E. of regression 0.05 Adjusted R-squared 0.37 Durbin-Watsonstat 2.16 38

International Finance Discussion Papers IFDP Number Titles 1996 563 CountryFundDiscountsandtheMexicanCrisisof JeffreyA. Frankei December1994: DidLocalResidentsTurn SergioL. Schmukler PessimisticBeforeInternationalInvestors? 562 EasternEuropeanExportPerformanceduring NathanSheets theTransition SimonaBoata 561 Inflation-AdjustedPotentialOutpu~ JaneT. Haltmaier 560 TheManagementof FinancialRisksatGerman AllenB.Frankel NonfinancialFirms: TheCaseof Metallgeseilschafi DavidE. Palmer 559 BroadMoneyDemandandFinanciaiLiberalization NeilR. Ericsson inGreece SuniiSharma 558 StockholdingBehaviorof U.S.Households:Evidence Carol C. Bertaut fromthe 1983-89Surveyof ConsumerFinances 557 Firm SizeandtheImpactof Profit-MarginUncertainty VivekGhosal onInvestment: DoFinancingConstraintsPlayaRole? PrakashLoungani 556 RegulationandtheCostof CapitalinJapan: A Case JohnAmmer Study MichaelS.Gibson 555 me Sovereign~Option: TheQuebecReferendumand MichaelP. Leahy MarketViewsontheCanadianDollar CharlesP.Thomas 554 ReaiExchangeRatesandInflationin Exchange-Rate StevenB. Kamin Based Stabilizations: An Empirical Examination 553 Macroeconomic State Variables as Determinants John Ammer of Asset Price Covariances 552 The Tequila Effect: Theory and Evidence from Martin Uribe Argentina 551 -TheAccumulation of Human Capital: Alternative Murat F. Iyigun Methods and Why They Matter Ann L. Owen Please address requests for copiesto International Finance Discussion Papers, Division of International Finance, Stop 24, Board of Governors of the Federal Reserve System, Washington, DC 20551. 39

International Finance Discussion Papers IFDP Number Titles Autho~s\ 1996 550 Alternatives in Human Capital Accumulation: MuratF. Iyigun Implications for Economic Growth AnnL.Owen 549 More Evidence on the Link between Bank MichaelS.Gibson Health and Investment in Japan 548 The Syndrome of Exchange-Rate-Based EnriqueG. Mendoza Stabilization and the Uncertain Duration of MartinUribe Currency Pegs 547 German Unification: What Have We Learned JosephE. Gagnon fi-omMulti-Country Models? PaulR.Masson WarwickJ. McKibbin 546 Returns to Scale in U.S. Production: Estimates SusantoBasu and Implications JohnG. Femald 545 Mexico’sBalance-of-Payments Crisis: A Chronicle GuillermoA. Calvo of Death Foretold EnriqueG.Mendoza 544 The Twin Crises: The Causes of Banking and GracielaL.Kaminsky Balance-of-Payments Problems CarmenM.Reinhart 543 High Real Interest Rates in the Afiermath of GracielaL. Kaminsky Disinflation: Is it a Lack of Credibility? LeonardoLeiderman 542 Precautionary Portfolio Behavior from a Life-Cycle CarolC. Bertaut Perspective MichaelHaliassos 541 Using Options Prices to Infer PDF’s for Asset Prices: WilliamR.Melick An Application to Oil Prices During the Gulf Crisis CharlesP. Thomas 540 Monetary Policy in the End-Game to Exchange-Rate StevenB.Kamin Based Stabilizations: The Case of Mexico JohnH.Rogers 539 Comparing the Welfare Costs and the Initial Dynamics MartinUribe of Alternative Temporary Stabilization Policies 538 Long Memory in Inflation Expectations: Evidence JosephE. Gagnon from International Financial Markets 40

InternationalFinanceDiscussionPapers IFDP Number Titles Autho~s) 1996 537 Using Measures of Expectations to Identi& the AlIan D. Brunner Effects of a Monetary Policy Shock 536 Regime Switching in the Dynamic Relationship Chan Huh between the Federal Funds Rate and Innovations in Nonborrowed Reserves 535 The Risks and Implications of External Financial Edwin M. Truman Shocks: Lessons from Mexico 534 Currency Crashes in Emerging Markets: An JeffreyA. Frankel Empirical Treatment Andrew K. Rose 533 Regional Patterns in the Law of One Price: The Charles Engel Roles of Geography Vs. Currencies John H. Rogers